{
 "article": "The Chip Companies Trying to Break the von Neumann Bottleneck",
 "url": "https://knowm.ai/blog/chip-companies-von-neumann-bottleneck/",
 "snapshot": "2026-11-10",
 "notes": "One record per profiled company (74), extracted from Literature/CompanyTaxonomy/companies/<slug>.md in the Knowm workspace: frontmatter, the company's own framing of the von Neumann problem, technical approach, every funding round with its source URL, trajectory, dated news-monitor lines (sweeps 2026-08-23, 2026-09-22, 2026-10-27, 2026-11-09, 2026-11-10) and the source list. total_funding_usd_m is disclosed capital only, ~USD M, including IPO proceeds where noted; 0, null or a string = undisclosed or conflicting (see the profile text); Chinese totals are partial by construction and Houmo's, Weina Hexin's, MemSilicon's and Zhizi Guangyuan's are floors; Ceryx Medical's medical-device capital is excluded from the article's dollar totals. Five companies (Weina Hexin, MemSilicon, Arago, Opticore, Zhizi Guangyuan) were added on 2026-11-10 from a census-gap discovery sweep; the patent table is the 2026-11-10 re-sweep, which covers them. The charts in the post (CompanyCharts.tsx) are baked from the same profiles. patents_table_raw = Google Patents assignee sweep 2026-11-10 (families = distinct inventions, floors; the 2026-09-22 sweep is in _patents_2026-09/). Boundary cases that were catalogued but not profiled live in _boundary.md and are not included here.",
 "companies": [
  {
   "slug": "aistorm",
   "name": "AIStorm",
   "founded": 2016,
   "hq": "Chandler, AZ, USA (originally reported San Jose, CA, with Phoenix-area office)",
   "category": "analog-imc",
   "status": "active",
   "origin": "corporate alumni (CEO David Schie — ex-Maxim/Micrel/Semtech; founder-CEO of Linear Dimensions Semiconductor, which Jon Peddie Research credits with the groundwork for AIStorm's core tech and PitchBook lists as a 2019 investor; co-founders Barker, Sibrai, Matias ex-WSI/Toshiba/Arm)",
   "total_funding_usd_m": 29.2,
   "last_round": "Series B, $16M, 2020-12",
   "key_investors": [
    "Egis Technology",
    "TowerJazz/Tower Semiconductor",
    "AsusTek",
    "Knowles",
    "Meyer Corporation",
    "Senvest Management"
   ],
   "website": "https://aistorm.ai",
   "pitch": "AIStorm attacks the digitization step itself: edge AI wastes power and time converting sensor\nsignals to digital before computing. Its \"AI-in-Sensor\" charge-domain processors accept charge\ndirectly from the sensor — \"electrons from a pixel or MEMs microphone\" — and multiply that\ncharge directly, skipping data converters, promising 5–10x cost savings for real-time edge AI\n([BusinessWire, 2019-02-11](https://www.businesswire.com/news/home/20190211005279/en/AIStorm-Raises-13.2M-Bring-Real-Time-AI-in-Sensor-Technology)).",
   "technical_approach": "Analog **charge-domain** computation (switched-charge multipliers with analog AI \"neurons\"),\ninference-only, integrated into or beside the sensor. Products: Cheetah family of AI-in-imager\nchips built on Tower Semiconductor's charge-domain imaging platform — **Cheetah HS**\n(Aug 2025, up-to-260K fps AI-in-imager for inspection/robotics/sports;\n[GlobeNewswire](https://www.globenewswire.com/news-release/2025/08/12/3131504/0/en/AIStorm-Tower-Semiconductor-Introduce-Cheetah-HS-World-s-First-Up-to-260K-FPS-AI-in-Imager-Chip-for-Inspection-Robotics-Sports.html));\na people-counting solution on the Cheetah AISC11C shipped Dec 2025 and sells via DigiKey.",
   "funding_rounds": [
    {
     "date": "2019-02",
     "round": "Series A",
     "amount": "$13.2M",
     "investors": "Egis Technology, TowerJazz",
     "source": "[BusinessWire](https://www.businesswire.com/news/home/20190211005279/en/AIStorm-Raises-13.2M-Bring-Real-Time-AI-in-Sensor-Technology)"
    },
    {
     "date": "2020-12",
     "round": "Series B (oversubscribed)",
     "amount": "$16M",
     "investors": "AsusTek, Egis, Knowles, Meyer Corporation, Senvest Management",
     "source": "[BusinessWire](https://www.businesswire.com/news/home/20201215005434/en/AIStorm-Raises-%2416M-in-Oversubscribed-Series-B-Funding-to-Commercialize-AI-in-Sensor-Chips)"
    }
   ],
   "funding_notes": "Total disclosed: ~$29.2M — strategics-heavy (foundry, biometrics, MEMS-microphone makers)\nrather than classic VC. No round disclosed since 2020.",
   "trajectory": "Founded 2016 by CEO David Schie and team; long product gestation, then real commercialization\nwith Tower Semiconductor: Cheetah HS launch (Aug 2025) and distribution of shipping parts\nthrough DigiKey (Dec 2025). Status mid-2026: active and shipping product — one of the few\nanalog-IMC edge startups with parts in distribution.\n\n**Founders (sweep 2026-09-22):** David Schie (CEO/chairman; University of Toronto, analog IC design; ex-senior exec Maxim, Micrel, Semtech; founder-CEO of Linear Dimensions Semiconductor, which he still runs), Robert Barker (CFO; ex-Micrel), Andreas Sibrai and Cesar Matias — described as veterans of WSI, Toshiba and Arm. Jon Peddie Research says Schie's Linear Dimensions work \"laid the groundwork for AIStorm's core technology\" and PitchBook records Linear Dimensions as an early-stage investor in AIStorm (11 Feb 2019) (https://www.jonpeddie.com/news/aistorm-puts-ai-inside-the-image-sensor/ ; https://pitchbook.com/profiles/company/233525-17 ; https://councils.forbes.com/profile/David-Schie-CEO-AIStorm-Linear-Dimensions-AISTORM-INC/47a34e4a-11d3-4dd8-82fc-601d00b33f9a ; http://image-sensors-world.blogspot.com/2019/02/aistorm-raises-132m-to-develop-ai-on.html).",
   "relevance": "AIStorm shows the analog edge-inference wedge working commercially with conventional charge-\ndomain CMOS — analog computation without emerging devices and without on-chip learning,\na useful contrast to memristor-based approaches on both device risk and adaptability.",
   "news": [],
   "sources": [
    {
     "title": "AIStorm raises $13.2M — BusinessWire",
     "url": "https://www.businesswire.com/news/home/20190211005279/en/AIStorm-Raises-13.2M-Bring-Real-Time-AI-in-Sensor-Technology"
    },
    {
     "title": "AIStorm raises $16M Series B — BusinessWire",
     "url": "https://www.businesswire.com/news/home/20201215005434/en/AIStorm-Raises-%2416M-in-Oversubscribed-Series-B-Funding-to-Commercialize-AI-in-Sensor-Chips"
    },
    {
     "title": "Cheetah HS launch — GlobeNewswire",
     "url": "https://www.globenewswire.com/news-release/2025/08/12/3131504/0/en/AIStorm-Tower-Semiconductor-Introduce-Cheetah-HS-World-s-First-Up-to-260K-FPS-AI-in-Imager-Chip-for-Inspection-Robotics-Sports.html"
    },
    {
     "title": "AIStorm puts AI inside the image sensor — Jon Peddie Research",
     "url": "https://www.jonpeddie.com/news/aistorm-puts-ai-inside-the-image-sensor/"
    }
   ]
  },
  {
   "slug": "alpsentek",
   "name": "AlpsenTek (Shenzhen Ruishizhixin Technology)",
   "founded": 2019,
   "hq": "Beijing / Shenzhen, China (offices in Nanjing, Switzerland)",
   "category": "event-sensing",
   "status": "active",
   "origin": "corporate alumni (Deng Jian, founder/CEO — ex-CSEM senior designer & project manager, ex-HMT microelectronic; Swiss arm AlpsenTek GmbH, Zurich) — no documented tech transfer from CSEM",
   "total_funding_usd_m": 100,
   "last_round": "Series B+, Mar 2026 (see conflict note)",
   "key_investors": [
    "Lenovo",
    "Allwinner Technology",
    "Hikvision",
    "CRRC Investment",
    "GAC Capital",
    "Changjiang Capital"
   ],
   "website": "https://en.alpsentek.com",
   "pitch": "AlpsenTek argues that neither conventional frame-based CMOS image sensors (CIS) nor pure event-based\nvision sensors (EVS) alone are sufficient: frame cameras produce redundant data and blur under fast\nmotion, while pure event sensors lose absolute intensity/texture. Its \"Hybrid Vision Sensor\" (HVS)\nfuses both at the pixel level so a single sensor delivers full-frame imagery plus microsecond,\nlow-redundancy event data. (https://en.alpsentek.com/)\nNote: event-based *sensing*, not a compute substrate.",
   "technical_approach": "Hybrid biomimetic vision chips (ALPIX series) integrating a CMOS image-sensor pixel path and an\nevent-based pixel path at the architectural, pixel, data, and algorithm levels. Products:\n**ALPIX-Pilates** (first hybrid fusion sensor, Sep 2021); **ALPIX-Eiger (APX003)** — high-resolution\nHVS for mobile, co-designed with OPPO and optimized with Qualcomm for Snapdragon; **ALPIX-Pizol\n(APX014)** hybrid sensor for edge-AI perception (2025); **ALPIX-Maloja (APX002)** pure event sensor\n(2025). Sensing only; downstream AI runs on host SoCs. Targets smartphones, robotics, wearables,\nsmart home, and automotive. (https://en.alpsentek.com/new/alpsentek-unveils-alpix-eiger-apx003-the-high-resolution-hybrid-vision-sensor-for-premium-imaging/)",
   "funding_rounds": [
    {
     "date": "2019-08",
     "round": "Seed",
     "amount": "undisclosed",
     "investors": "—",
     "source": "Tracxn"
    },
    {
     "date": "early 2022",
     "round": "Series A",
     "amount": "~$30M (≈¥200M)",
     "investors": "reported: Lenovo, Allwinner, Hikvision among backers",
     "source": "https://pandaily.com/machine-vision-chip-firm-alpsentek-secures-nearly-30m-in-a-round-funding/"
    },
    {
     "date": "2025",
     "round": "Series B",
     "amount": "undisclosed",
     "investors": "Smart IoT Industry Fund, Puxin Capital, CRRC Investment, Zhichen Investment, Yiling Capital",
     "source": "https://www.yolegroup.com/industry-news/alpsentek-announced-the-completion-of-series-b-financing/"
    },
    {
     "date": "2026-03-17",
     "round": "Series B+",
     "amount": "see note",
     "investors": "BEIDM, Guangdong Finance Fund, GAC Capital, Circumference Capital, Changjiang Capital, Bluetrum, others",
     "source": "http://image-sensors-world.blogspot.com/2026/04/alpsentek-raises-another-round-of.html"
    }
   ],
   "funding_notes": "**Funding conflict:** the company states cumulative fundraising has surpassed **US$100M** after the\nSeries B+; however, commenters note AlpsenTek's official Chinese release describes the B+ round itself\nas \"hundreds of millions of yuan\" (~US$15M for that round). Both figures reported; the >$100M is the\nclaimed cumulative total.",
   "trajectory": "Founded 2019 by Deng Jian (founder/CEO), a Beijing-headquartered fabless vision-sensor startup\n(legal entity Shenzhen Ruishizhixin Technology Co., Ltd) with a Swiss office. Built the CIS+EVS\nhybrid-pixel category and pushed into consumer mobile via the OPPO/Qualcomm collaboration on\nALPIX-Eiger. Named an \"Outstanding Innovative Company\" at the 2024 World Electronics Achievement\nAwards. Active and expanding (robotics, wearables, automotive) as of mid-2026 following the March\n2026 Series B+.\n\n**Founders (sweep 2026-09-22):** Jian Deng / Deng Jian (founder & CEO). LinkedIn lists him as Senior Designer & Project Manager at CSEM (Swiss Center for Electronics and Microtechnology) from Feb 2015, previously Project Manager & Senior Electrical Engineer at HMT microelectronic, and Director of AlpsenTek GmbH (Zurich) Dec 2019–Mar 2024 (https://cn.linkedin.com/in/jian-d-04a9673 ; https://rocketreach.co/jian-deng-email_86726636). Degree not found in English-language sources. No source states that the hybrid-vision-sensor technology was transferred from CSEM.",
   "relevance": "AlpsenTek, like Prophesee and iniVation, is a sensing-side player attacking data redundancy at\ncapture rather than reinventing the compute substrate; its distinctive wedge is fusing conventional\nand event pixels on one die. No compute or on-chip learning, so it contrasts with Knowm's\nmemristor/AHaH learn-in-the-substrate approach as a purely front-end complement.",
   "news": [],
   "sources": [
    {
     "title": "AlpsenTek secures ~$30M A-round (Pandaily)",
     "url": "https://pandaily.com/machine-vision-chip-firm-alpsentek-secures-nearly-30m-in-a-round-funding/"
    },
    {
     "title": "AlpsenTek completes Series B (Yole)",
     "url": "https://www.yolegroup.com/industry-news/alpsentek-announced-the-completion-of-series-b-financing/"
    },
    {
     "title": "AlpsenTek raises another round / Series B+ (Image Sensors World)",
     "url": "http://image-sensors-world.blogspot.com/2026/04/alpsentek-raises-another-round-of.html"
    },
    {
     "title": "AlpsenTek unveils ALPIX-Eiger",
     "url": "https://en.alpsentek.com/new/alpsentek-unveils-alpix-eiger-apx003-the-high-resolution-hybrid-vision-sensor-for-premium-imaging/"
    },
    {
     "title": "OPPO, AlpsenTek, Qualcomm collaboration",
     "url": "https://www.thefastmode.com/technology-solutions/34565-oppo-alpsentek-and-qualcomm-collaborate-to-enhance-hybrid-vision-sensing-for-smartphones"
    }
   ]
  },
  {
   "slug": "ambient-scientific",
   "name": "Ambient Scientific",
   "founded": 2017,
   "hq": "San Jose, CA, USA",
   "category": "analog-imc",
   "status": "active",
   "origin": "corporate alumni (CEO Gajendra \"GP\" Singh — ex-VP Engineering, Wave Computing)",
   "total_funding_usd_m": 17,
   "last_round": "Seed extension, undisclosed, 2023 (Dholakia Ventures); $17M Series A reported earlier",
   "key_investors": [
    "Dholakia Ventures"
   ],
   "website": "https://www.ambientscientific.ai",
   "pitch": "Ambient Scientific pitches \"AI-native\" processors for battery-powered edge devices: digital\nMCUs burn too much power shuttling data for always-on AI, so its analog compute-in-memory\narchitecture claims up to 100x improvements in power efficiency, performance, and area over\ntraditional 32-bit microcontrollers — years of always-on AI on a coin cell\n([PR Newswire, 2025-09-17](https://tools.prnewswire.com/en-us/live/20823/release/20250917EN75692);\n[CNX Software, 2026-07-13](https://www.cnx-software.com/2026/07/13/ambient-scientific-gpx10-pro-mcu-delivers-years-of-always-on-ai-on-a-coin-cell-battery/)).",
   "technical_approach": "Proprietary **DigAn** architecture: analog in-memory compute blocks execute the matrix-multiply\nand activation flows, wrapped in a programmable SoC (SDK-programmable, inference-only).\nProducts: **GPX-10** SoC (announced ~2021) with 10 programmable analog AI cores for always-on\nvoice/keyword spotting and sensing at micro-watt levels; moved toward mass production Jan 2024\n([Edge IR](https://www.edgeir.com/ambient-scientifics-edge-processor-gears-up-for-mass-production-20240110));\n**GPX10 Pro** (Sept 2025) — 10 MX8 cores in two power domains, 2,560 MACs/cycle, 512 GOPS peak,\n2MB SRAM, <100 µW always-on\n([Embedded.com](https://www.embedded.com/ambient-scientific-unveils-gpx10-pro-soc-bringing-ai-native-performance-to-battery-powered-edge-devices/)).",
   "funding_rounds": [
    {
     "date": "~2019 (reported at GPX-10 launch)",
     "round": "Series A",
     "amount": "$17M",
     "investors": "investors not named in coverage",
     "source": "search-result reporting citing launch coverage; see [TechInsights vendor page](https://www.techinsights.com/vendor/ambient-scientific)"
    },
    {
     "date": "2023",
     "round": "Seed (per Tracxn labeling)",
     "amount": "undisclosed",
     "investors": "Dholakia Ventures",
     "source": "[Tracxn](https://tracxn.com/d/companies/ambientscientific/__7ftkA-HxFkFM7x5ajx_hbQn7gRNmysUmb5qXOAsbIkY/funding-and-investors)"
    }
   ],
   "funding_notes": "**Sourcing caveat:** funding is thinly documented — the $17M Series A is repeated in trade\ncoverage without a primary press release surfacing in this sweep, and investor names are\nlargely undisclosed. Treat total (~$17M+) as a floor.",
   "trajectory": "Founded 2017 in Silicon Valley by Gajendra \"GP\" Singh (ex-Wave Computing VP Engineering).\nSlow burn from architecture (2017–2020) to GPX-10 (2021), mass-production readiness (2024),\nGPX10 Pro (2025), with fresh trade coverage into July 2026. Status mid-2026: active, shipping/\nsampling silicon for always-on edge AI.",
   "relevance": "A conventional-CMOS analog compute-in-memory player proving the micro-watt always-on niche —\nthe market wedge closest to where memristor-based edge chips would compete — but inference-only:\nno on-device learning claim, unlike the AHaH/GEMESYS end of the taxonomy.",
   "news": [],
   "sources": [
    {
     "title": "GPX10 Pro launch — Embedded.com",
     "url": "https://www.embedded.com/ambient-scientific-unveils-gpx10-pro-soc-bringing-ai-native-performance-to-battery-powered-edge-devices/"
    },
    {
     "title": "GPX10 Pro on coin cell — CNX Software",
     "url": "https://www.cnx-software.com/2026/07/13/ambient-scientific-gpx10-pro-mcu-delivers-years-of-always-on-ai-on-a-coin-cell-battery/"
    },
    {
     "title": "AI-native processor PR — PR Newswire",
     "url": "https://tools.prnewswire.com/en-us/live/20823/release/20250917EN75692"
    },
    {
     "title": "Mass production — Edge IR",
     "url": "https://www.edgeir.com/ambient-scientifics-edge-processor-gears-up-for-mass-production-20240110"
    },
    {
     "title": "Funding rounds — Tracxn",
     "url": "https://tracxn.com/d/companies/ambientscientific/__7ftkA-HxFkFM7x5ajx_hbQn7gRNmysUmb5qXOAsbIkY/funding-and-investors"
    }
   ]
  },
  {
   "slug": "anaflash",
   "name": "ANAFLASH",
   "founded": 2017,
   "hq": "San Jose, CA, USA (Korean subsidiary SEMIBRAIN)",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic lineage (University of Minnesota — Prof. Chris Kim's single-poly eFlash, the founder's PhD group; exclusive UMN license) + corporate alumni (Samsung Electronics, HGST, SK Hynix)",
   "total_funding_usd_m": 7.3,
   "last_round": "Venture round (undisclosed details); acquired Legato Logic 2025-02",
   "key_investors": [
    "L&S Venture Capital",
    "Stonebridge Ventures",
    "Hatcher+",
    "Plug and Play",
    "Bluepoint",
    "NSF SBIR"
   ],
   "website": "https://www.anaflash.com",
   "pitch": "ANAFLASH's non-volatile edge-AI processors \"run computation at the place where the\nnecessary data is stored,\" while zero-standby-power flash weight memory plus power\ngating eliminates idle-mode dissipation — targeting battery-powered intelligent\nsensors that cannot afford data movement or standby leakage.\nSource: https://www.anaflash.com (company positioning, mirrored in Tracxn profile:\nhttps://tracxn.com/d/companies/anaflash/__DZuFDPVDore5VduMciOObOKyvOp4G1hhlYHyTTrMu98)",
   "technical_approach": "- Device: **logic-compatible embedded flash (eFlash)** — single-poly flash cells built\n  in standard CMOS logic processes (no extra masks), used for analog compute-in-memory\n  and zero-standby-power weight storage. Core eFlash IP exclusively licensed from the\n  **University of Minnesota** (2023)\n  (https://www.storagenewsletter.com/2023/03/14/silicon-valley-start-up-anaflash-acquired-exclusive-license-for-single-poly-based-embedded-flash-memory-technology-from-university-of-minnesota/).\n- Demonstrated silicon: a 28nm AI microcontroller with tightly coupled zero-standby\n  power weight memory using standard-logic-compatible 4Mb 4-bits/cell embedded flash\n  (arXiv preprint by the team: https://arxiv.org/pdf/2503.11660).\n- Feb 2025: **acquired Legato Logic's time-based compute-in-memory** technology to\n  extend its non-volatile CIM roadmap for battery-powered intelligent sensors\n  (https://www.sdxcentral.com/news/anaflash-acquires-legato-logic-to-advance-compute-in-memory-technology/).\n- Samsung Foundry collaboration for Logic-EFLASH development via Korean subsidiary\n  SEMIBRAIN (https://www.anaflash.com/post/anaflash-and-samsung-collaborate).\n- Inference-only edge AI; grant-supported (NSF SBIR, DoD, NASA).",
   "funding_rounds": [
    {
     "date": "various",
     "round": "Seed/venture rounds",
     "amount": "$7.33M total",
     "investors": "L&S Venture Capital, Stonebridge Ventures, Hatcher+, Plug and Play, Bluepoint",
     "source": "https://tracxn.com/d/companies/anaflash/__DZuFDPVDore5VduMciOObOKyvOp4G1hhlYHyTTrMu98"
    },
    {
     "date": "—",
     "round": "NSF SBIR Phase II + supplement",
     "amount": "$0.5M supplement",
     "investors": "U.S. National Science Foundation",
     "source": "https://www.anaflash.com/post/new-funding-from-the-nsf"
    },
    {
     "date": "—",
     "round": "SBIR \"Reliable Logic Compatible Embedded Flash\"",
     "amount": "per award",
     "investors": "NSF SBIR",
     "source": "https://www.sbir.gov/sbirsearch/detail/1623811"
    }
   ],
   "funding_notes": "Total disclosed: **~$7.3M equity** (Tracxn) plus federal grants (NSF/DoD/NASA).\nRound-by-round dates/amounts are not publicly broken out; PitchBook lists further\ndetail behind paywall (https://pitchbook.com/profiles/company/277299-55).",
   "trajectory": "Founded 2017 (per company/StorageNewsletter; Tracxn says 2021 Sunnyvale — conflicting\nrecords) by **Seung-Hwan \"Peter\" Song**, a 15+-year nonvolatile-memory veteran and\nfounding member of Samsung's 4-bits/cell NAND controller team; PhD work at the\nUniversity of Minnesota underlies the licensed eFlash IP\n(https://www.anaflash.com/about-2/leadership ;\nhttps://www.storagenewsletter.com/2023/03/14/silicon-valley-start-up-anaflash-acquired-exclusive-license-for-single-poly-based-embedded-flash-memory-technology-from-university-of-minnesota/).\nSmall, grant-leveraged trajectory: test chips → Samsung collaboration → Legato Logic\nacquisition (2025-02-05). **Status mid-2026: active**, smallest-funded pure eFlash-CIM\nplayer, pre-volume commercial products.\n\n**Founders (sweep 2026-09-22):** Seung-Hwan \"Peter\" Song (CEO, founder; PhD Electrical & Computer Engineering, University of Minnesota — a co-author in Prof. Chris H. Kim's single-poly eFlash work; previously Samsung Electronics 4-bit/cell NAND controller founding team, HGST, SK Hynix Memory Solutions America). The eFlash IP \"was originally developed by Professor Chris Kim at the University of Minnesota\" and exclusively licensed to ANAFLASH; CTO Shahrzad Naraghi (Michigan PhD) joined via the 2025 Legato Logic acquisition, not as a founder (https://www.anaflash.com/about-2/leadership ; https://www.anaflash.com/post/embedded-non-volatile-memory-technology-to-market- ; https://experts.umn.edu/en/publications/a-comparative-study-of-single-poly-embedded-flash-memory-disturba).",
   "relevance": "ANAFLASH shows the minimal-capital path in flash-analog IMC: standard-logic eFlash,\nuniversity IP and SBIR grants rather than nine-figure venture rounds — closer in scale\nto Knowm's own bootstrap trajectory than to Mythic/EnCharge. Inference-only; weight\nstorage is nonvolatile but not adaptive in operation.",
   "news": [],
   "sources": [
    {
     "title": "Tracxn profile",
     "url": "https://tracxn.com/d/companies/anaflash/__DZuFDPVDore5VduMciOObOKyvOp4G1hhlYHyTTrMu98"
    },
    {
     "title": "UMinn eFlash license",
     "url": "https://www.storagenewsletter.com/2023/03/14/silicon-valley-start-up-anaflash-acquired-exclusive-license-for-single-poly-based-embedded-flash-memory-technology-from-university-of-minnesota/"
    },
    {
     "title": "Legato Logic acquisition",
     "url": "https://www.sdxcentral.com/news/anaflash-acquires-legato-logic-to-advance-compute-in-memory-technology/"
    },
    {
     "title": "NSF funding post",
     "url": "https://www.anaflash.com/post/new-funding-from-the-nsf"
    },
    {
     "title": "28nm eFlash AI MCU preprint",
     "url": "https://arxiv.org/pdf/2503.11660"
    }
   ]
  },
  {
   "slug": "anotherbrain",
   "name": "AnotherBrain",
   "founded": 2017,
   "hq": "Paris, France",
   "category": "analog-imc",
   "status": "pivoted",
   "origin": "serial founders (Bruno Maisonnier, ex-Aldebaran Robotics)",
   "total_funding_usd_m": 33.9,
   "last_round": "Series A, €19M (~$21M), 2019-10",
   "key_investors": [
    "Cathay Innovation",
    "360 Capital",
    "Otium Venture",
    "Robert Bosch Venture Capital"
   ],
   "website": "https://anotherbrain.ai",
   "pitch": "AnotherBrain positioned deep learning as too power-hungry and data-hungry to be\nsustainable at the edge, and proposed \"Organic AI\" — a bio-inspired, neuromorphic\nintelligence \"emulating the human brain's cortex\" that is energy-efficient,\ndata-frugal, self-learning, and explainable in real time, meant to be embedded in\nASICs for automotive, industrial and IoT. (Framing per the company site and 2019\nSeries A coverage in Sifted and EE Times Europe.)",
   "technical_approach": "- Originally: a bio-inspired / brain-inspired neuromorphic chip (\"Organic AI\")\n  aimed at low-power, self-supervised learning on-device — pitched as an ASIC\n  target with sub-milliwatt power and <1 ms latency. Classified here as\n  **analog-imc-adjacent / brain-inspired boundary**: the public material describes\n  a cortex-emulating architecture and ASIC ambition, but AnotherBrain never shipped\n  a merchant neuromorphic chip, so the \"in-memory/analog\" substrate claim is\n  unconfirmed at the hardware level — reported honestly as a boundary case.\n- Pivoted to: **PHOSPHOR / Blue Phosphor (PHOSPHOR QUALITY)**, an edge-AI visual\n  industrial-inspection software+systems product for defect detection in luxury\n  goods, pharma and automotive, learning only from \"good\" parts (no defect library).\n  The company now describes itself as an \"integrated Edge AI Software & Systems\n  provider\" rather than a chip startup.",
   "funding_rounds": [
    {
     "date": "2017",
     "round": "Seed",
     "amount": "~€10–11M",
     "investors": "Founder + early backers",
     "source": "Sifted / Domoclick"
    },
    {
     "date": "2019-10",
     "round": "Series A",
     "amount": "€19M (~$21M)",
     "investors": "Cathay Innovation, 360 Capital, Otium, Bosch Venture Capital",
     "source": "Tech.eu / EE Times Europe"
    },
    {
     "date": "—",
     "round": "Total (reported)",
     "amount": "~$33.9M",
     "investors": "—",
     "source": "Crunchbase / NeuromorphicCore"
    }
   ],
   "funding_notes": "",
   "trajectory": "Founded 2017 in Paris by Bruno Maisonnier — the founder/chairman of Aldebaran\nRobotics (Nao, Pepper), acquired by SoftBank in 2015 — making this a serial-founder\nplay. It raised ~€10–11M seed (2017) and a €19M Series A (Oct 2019) for the Organic\nAI chip vision. Over the following years it shifted away from custom neuromorphic\nsilicon toward the PHOSPHOR industrial visual-inspection product line, now in full\ndeployment; the current self-description is an Edge AI software & systems company.\nStatus mid-2026: active but pivoted — surviving as an industrial machine-vision /\ninspection software vendor rather than a neuromorphic chipmaker. (No public\nliquidation or shutdown reports were found; total reported funding ~$33.9M.)\n\n**Founders (sweep 2026-09-22):** Bruno Maisonnier (founder & CEO) is the only founder named by the company, LinkedIn, Dealroom or press; founder of Aldebaran Robotics (2005, sold to SoftBank) — immediately prior employer. No academic lab or corporate parent for the Organic AI technology is documented (https://anotherbrain.ai/en/about-us ; https://www.linkedin.com/company/another-brain/).",
   "relevance": "Another first-wave brain-inspired hardware ambition that ended up as vision\nsoftware. The contrast with Knowm is the substrate honesty question: AnotherBrain\nmarketed \"cortex-emulating Organic AI\" but never demonstrated a shipping analog/\nin-memory learning chip, whereas Knowm's claim is anchored in a physical memristor\ndevice and the AHaH/MSS model. Reported straight: AnotherBrain's neuromorphic\nhardware claim is unverified at the device level, and the company's durable\nbusiness became conventional edge-AI inspection software.",
   "news": [],
   "sources": [
    {
     "title": "France's AnotherBrain raises €19 million Series A for their \"OrganicAI\" — Tech.eu",
     "url": "https://tech.eu/2019/10/01/frances-anotherbrain-raises-e19-million-series-a-for-their-organicai/"
    },
    {
     "title": "AnotherBrain Raises €19m To Speed Up Organic AI Chip — EE Times Europe",
     "url": "https://www.eetimes.eu/anotherbrain-raises-e19m-to-speed-up-organic-ai-chip/"
    },
    {
     "title": "Paris-based AnotherBrain has raised €19m — Sifted",
     "url": "https://sifted.eu/articles/another-brain-startup-ai"
    },
    {
     "title": "AnotherBrain — NeuromorphicCore.ai (status, PHOSPHOR pivot, ~$33.9M)",
     "url": "https://neuromorphiccore.ai/insights/anotherbrain/"
    },
    {
     "title": "Another Brain — Crunchbase Company Profile",
     "url": "https://www.crunchbase.com/organization/another-brain"
    },
    {
     "title": "AnotherBrain — company site",
     "url": "https://anotherbrain.ai/en/about-us"
    }
   ]
  },
  {
   "slug": "applied-brain-research",
   "name": "Applied Brain Research (ABR)",
   "founded": 2014,
   "hq": "Waterloo, Ontario, Canada",
   "category": "digital-snn",
   "status": "active",
   "origin": "academic spinoff (Centre for Theoretical Neuroscience, University of Waterloo — Chris Eliasmith's lab; co-founded with Peter Suma)",
   "total_funding_usd_m": "~4 disclosed pre-2026 (trackers: $3.6–4.5M); Jan 2026 seed round amount undisclosed",
   "last_round": "Seed (oversubscribed), undisclosed amount, 2026-01",
   "key_investors": [
    "Two Small Fish Ventures (lead)",
    "Silicon Catalyst (incubator)"
   ],
   "website": "https://www.appliedbrainresearch.com",
   "pitch": "\"Voice is becoming the default interface for the next wave of edge devices but using cloud\nvoice AI solutions is a terrible experience, plagued by delays, reliability and privacy\nconcerns\" — CEO Kevin Conley; ABR's answer is brain-derived time-series models running on a\npurpose-built low-power chip at the edge\n(https://www.prnewswire.com/news-releases/applied-brain-research-closes-seed-funding-round-to-accelerate-edge-ai-voice-processing-302658022.html).",
   "technical_approach": "- Roots in **Nengo**, the neural-engineering-framework (NEF) simulator from Eliasmith's lab —\n  long the standard toolchain for programming spiking hardware (Loihi, SpiNNaker)\n  (https://www.digitimes.com/news/a20210531PR200.html&chid=9,\n  https://semiwiki.com/semiconductor-services/silicon-catalyst/322705-ceo-interview-dr-chris-eliasmith-and-peter-suma-of-applied-brain-research-inc/).\n- Invented the Legendre Memory Unit lineage that evolved into **state-space models**; their\n  silicon bet is an SSM accelerator rather than a literal spiking chip.\n- **TSP1** (\"Time Series Processor\"): first-generation state-space accelerator, digital,\n  inference-only; full-vocabulary on-device speech-to-text and TTS at **under 30 mW** (~10–100x\n  less power than competing solutions, per company)\n  (https://www.prnewswire.com/news-releases/applied-brain-research-closes-seed-funding-round-to-accelerate-edge-ai-voice-processing-302658022.html).\n  TSP concept was a Best of Sensors Awards finalist in 2022\n  (https://www.newswire.com/news/applied-brain-research-inc-abr-named-best-of-sensors-awards-finalist-21749703).\n- Taxonomy note: ABR sits at the boundary of \"digital SNN\" — spiking-research origin, but the\n  shipping product accelerates SSMs (the post-SNN formulation of the same temporal-compression\n  math).",
   "funding_rounds": [
    {
     "date": "pre-2026",
     "round": "grants/angel/early capital (incl. Silicon Catalyst incubator)",
     "amount": "trackers: $3.6M (Caplight) / ~$4.5M (NeuromorphicCore) — no primary source",
     "investors": "—",
     "source": "https://www.caplight.com/company/appliedbrainresearch ; https://neuromorphiccore.ai/insights/applied-brain-research-abr/"
    },
    {
     "date": "2026-01",
     "round": "Seed (oversubscribed), post-TSP1-release",
     "amount": "undisclosed",
     "investors": "led by Two Small Fish Ventures (Eva Lau joins board)",
     "source": "https://www.prnewswire.com/news-releases/applied-brain-research-closes-seed-funding-round-to-accelerate-edge-ai-voice-processing-302658022.html ; https://www.finsmes.com/2026/01/applied-brain-research-closes-seed-funding.html ; https://www.canadianmanufacturing.com/manufacturing/applied-brain-research-closes-seed-funding-round-for-edge-ai-voice-processing-315272/"
    }
   ],
   "funding_notes": "Total raised: undisclosed overall; among the smallest war-chests in this cohort.",
   "trajectory": "Launched 2014 by Prof. Chris Eliasmith and Peter Suma out of Waterloo's Centre for Theoretical\nNeuroscience (https://uwaterloo.ca/news/global-impact/building-brain; one profile says 2013 —\nhttps://ciobulletin.com/magazine/profile/applied-brain-research-integrated-ai-systems-for-real-world-applications).\nA decade as a neuromorphic software/services company (Nengo) before pivoting to its own\nsilicon (TSP1); leadership added semiconductor veteran Kevin Conley as CEO. As of mid-2026:\n**active**, early commercial stage — TSP1 released, seed round closed Jan 2026 to scale edge\nvoice AI.",
   "relevance": "ABR is the strongest \"algorithms-first\" data point: a top academic spiking group whose\ncommercial chip abandoned literal spikes for state-space math in digital CMOS, inference-only.\nIt illustrates that on-chip learning — Knowm's AHaH territory — remains unclaimed even by the\nmost neuroscience-native player in the SNN cohort.",
   "news": [
    {
     "date": "2026-09-22",
     "text": "**ABR SDK general availability (2026-09-21):** production release of the Niagara streaming-ASR and Nith streaming-TTS model families behind one API, running **entirely on edge CPUs** (Linux x86-64/ARM64, Android ARM64; NPU/DSP acceleration on some partner platforms; Cortex-M + RTOS targets promised by year-end). Claimed first-text latency 115 ms and first-audio 147 ms on embedded application-class CPUs; English, Spanish, Mandarin, Japanese, Korean; free evaluation, commercial license for deployment. Note: this is a software product on commodity silicon — no mention of the TSP1 chip in the release — reinforcing the read that ABR's near-term business is models/SDK rather than its own neuromorphic hardware. (https://www.prnewswire.com/news-releases/applied-brain-research-releases-the-abr-sdk-bringing-real-time-on-device-voice-interfaces-to-edge-applications-302883453.html)"
    },
    {
     "date": "2026-11-09",
     "text": "**Speech AI SDK generally available for Zephyr RTOS on Alif Ensemble MCUs (PR 2026-11-02)** — the software-on-commodity-silicon line continues; no TSP1 chip date in the release title. (https://www.prnewswire.com/news-releases/applied-brain-research-announces-general-availability-of-its-speech-ai-sdk-for-zephyr-rtos-on-the-alif-ensemble-family-of-processors-302918000.html)"
    }
   ],
   "sources": [
    {
     "title": "ABR seed round PR (PRNewswire, Jan 12, 2026)",
     "url": "https://www.prnewswire.com/news-releases/applied-brain-research-closes-seed-funding-round-to-accelerate-edge-ai-voice-processing-302658022.html"
    },
    {
     "title": "FinSMEs: ABR closes seed (Jan 2026)",
     "url": "https://www.finsmes.com/2026/01/applied-brain-research-closes-seed-funding.html"
    },
    {
     "title": "Waterloo News: Building a brain (2019)",
     "url": "https://uwaterloo.ca/news/global-impact/building-brain"
    },
    {
     "title": "SemiWiki CEO interview (2023)",
     "url": "https://semiwiki.com/semiconductor-services/silicon-catalyst/322705-ceo-interview-dr-chris-eliasmith-and-peter-suma-of-applied-brain-research-inc/"
    },
    {
     "title": "Caplight funding profile",
     "url": "https://www.caplight.com/company/appliedbrainresearch"
    },
    {
     "title": "NeuromorphicCore.ai ABR insight",
     "url": "https://neuromorphiccore.ai/insights/applied-brain-research-abr/"
    }
   ]
  },
  {
   "slug": "arago",
   "name": "Arago",
   "founded": 2024,
   "hq": "Paris, France (and Silicon Valley; hiring in Paris and Tel Aviv)",
   "category": "photonic",
   "status": "active",
   "origin": "serial founders via Entrepreneurs First (Nicolas Muller-Vorburger, Eliott Sarrey, Ambroise Müller)",
   "total_funding_usd_m": 26,
   "last_round": "Seed, $26M, 2025-07",
   "key_investors": [
    "Earlybird",
    "Protagonist",
    "Visionaries Tomorrow",
    "Generative IQ",
    "C4 Ventures",
    "Bertrand Serlet",
    "Olivier Pomel",
    "Thomas Wolf"
   ],
   "website": "https://www.arago.inc",
   "pitch": "\"Instead of transistors, it uses lasers to process data with photons\": a photonic AI\naccelerator, codenamed **JEF**, \"designed to deliver 10× lower energy consumption than today's\nleading GPUs at equivalent performance and cost\"\n([Finopotamus 2025-07-08](https://www.finopotamus.com/post/arago-raises-26m-in-seed-funding-to-slash-ai-energy-consumption-with-breakthrough-photonic-chip);\n[PIC Magazine 2025-07-08](https://picmagazine.net/article/122081/Arago_raises_26_million_for_photonic_AI_chips)).\nThe website raises the target to \"10×–30× reductions in energy consumption\" and\n\"multi-PetaOp/s performance\" ([arago.inc](https://www.arago.inc/)).",
   "technical_approach": "- **Hybrid digital/analog/photonic.** A co-founder quote describes a \"hybrid architecture\"\n  spanning \"digital, analog, and photonic approaches\". The site splits the chip into an\n  \"AI unit\" for n³-complexity work (matrix multiplication, per its diagram) and a \"Control\n  unit\" for n²-complexity pointwise operations, on a \"multi-physics computing core\".\n  Integrated vs free-space optics is not stated; an April 2026 LinkedIn post by a backer\n  calls it \"the world's first digital-optical chip in standard silicon\"\n  ([LinkedIn, 2026-04](https://www.linkedin.com/posts/thong-le-hoang_bravo-nicolas-eliott-ambroise-and-team-activity-7447181962458415104-h37M)).\n- **Software:** \"CARLOTA\" stack, PyTorch interface. Target workload: AI inference (\"more\n  tokens, fewer watts\").\n- **Hardware status:** \"Early results demonstrate that 'JEF' can run AI models from\n  industry-standard software frameworks\" (seed PR). No chip specs, benchmarks, or test\n  setup are public; the 10× figure is a design target. Inference focus; no learning claim.",
   "funding_rounds": [
    {
     "date": "2025-07",
     "round": "Seed (oversubscribed)",
     "amount": "$26M",
     "investors": "co-led by Earlybird, Protagonist, Visionaries Tomorrow; Generative IQ, C4 Ventures; angels Bertrand Serlet (ex-Apple), Christophe Frey (Arm), Olivier Pomel (Datadog), Thomas Wolf (Hugging Face), Jack Abraham",
     "source": "https://www.finopotamus.com/post/arago-raises-26m-in-seed-funding-to-slash-ai-energy-consumption-with-breakthrough-photonic-chip ; https://picmagazine.net/article/122081/Arago_raises_26_million_for_photonic_AI_chips"
    }
   ],
   "funding_notes": "**Total:** $26M disclosed. No later round found as of 2026-11-10; no valuation reported.",
   "trajectory": "Founded 2024 at Entrepreneurs First (EF co-founder Alice Bentinck: \"just a year after\nfounding the company at EF\" at the July 2025 seed;\n[LinkedIn](https://www.linkedin.com/posts/alicebentinck_congrats-to-the-team-at-arago-on-their-26m-activity-7348384973520879616-MZy9);\n[L'Embarqué 2025-07-09](https://lembarque.com/article/la-start-up-francaise-arago-leve-26-millions-dollars-pour-son-processeur-photonique-destine-a-lia))\nby Nicolas Muller-Vorburger (CEO), Eliott Sarrey, and Ambroise Müller. ~20 people at the\nseed, described as a 40-person company by late 2025 (yespress.io, blocked). Status: active,\npre-product.",
   "relevance": "A 2024-founded photonic inference accelerator with a $26M seed: a late entrant in the\nphotonic cohort, well below Lightmatter/OLIX scale but part of the 2024–2026 wave of\n\"LLM inference without HBM\" photonic pitches. Inference-only; nothing adaptive.",
   "news": [],
   "sources": [
    {
     "title": "Finopotamus: Arago Raises $26M in Seed Funding (2025-07-08)",
     "url": "https://www.finopotamus.com/post/arago-raises-26m-in-seed-funding-to-slash-ai-energy-consumption-with-breakthrough-photonic-chip"
    },
    {
     "title": "PIC Magazine: Arago raises $26 million for photonic AI chips (2025-07-08)",
     "url": "https://picmagazine.net/article/122081/Arago_raises_26_million_for_photonic_AI_chips"
    },
    {
     "title": "arago.inc (company site)",
     "url": "https://www.arago.inc/"
    },
    {
     "title": "L'Embarqué: La start-up française Arago lève 26 millions (2025-07-09)",
     "url": "https://lembarque.com/article/la-start-up-francaise-arago-leve-26-millions-dollars-pour-son-processeur-photonique-destine-a-lia"
    },
    {
     "title": "Photonics Spectra: Optical Computing Company Arago Raises $26M",
     "url": "https://www.photonics.com/Articles/Optical-Computing-Company-Arago-Raises-26M-in/a71200"
    }
   ]
  },
  {
   "slug": "aspinity",
   "name": "Aspinity",
   "founded": 2015,
   "hq": "Pittsburgh, PA, USA (origin: Morgantown, WV)",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (David Graham's lab, West Virginia University)",
   "total_funding_usd_m": 22,
   "last_round": "Series B, $5M, 2023-09",
   "key_investors": [
    "Anzu Partners",
    "Amazon Alexa Fund",
    "Birchmere Ventures",
    "Mountain State Capital",
    "Riverfront Ventures",
    "Unitrontech"
   ],
   "website": "https://www.aspinity.com",
   "pitch": "Aspinity attacks the \"digitize-first\" architecture of always-on sensing: conventional\nsystems burn power digitizing and DSP-processing mostly-irrelevant sensor data, so\nAspinity's analogML core \"determines the relevance of data while it is still analog,\"\nkeeping the MCU and digital processors asleep until meaningful events occur and cutting\nalways-on system power by up to 95%.\nSource: https://www.aspinity.com/press-releases/aspinity-launches-new-aml100-near-zero-power-monitoring-solutions-for-automotive-security/",
   "technical_approach": "- **analogML / RAMP**: a reconfigurable, trainable analog machine-learning processor\n  built from programmable analog blocks (standard CMOS with nonvolatile parameter\n  storage); performs feature extraction and classification directly on raw analog\n  sensor signals before any ADC. Inference-only on-chip (models trained offline).\n  (https://www.aspinity.com ; https://www.techinsights.com/blog/aspinity-tackles-analog-variability)\n- Named product: **AML100** — first commercial analog ML chip (launched at CES, Jan\n  2022; https://www.pghtech.org/news-and-publications/Aspinity_Chip). Runs always-on\n  detection at tens of microwatts (\"near-zero power\").\n- 2024: AML100-based near-zero-power monitoring solutions for automotive security\n  (glass-break/intrusion detection while parked), plus Renesas Quick-Connect IoT\n  integration (https://www.aspinity.com/press-releases/aspinity-launches-new-aml100-near-zero-power-monitoring-solutions-for-automotive-security/ ;\n  https://www.pghtech.org/news-and-publications/Aspinity_Renesas).",
   "funding_rounds": [
    {
     "date": "2020-09",
     "round": "Series A",
     "amount": "$5.3M",
     "investors": "Anzu Partners (first invested 2020-06); Amazon Alexa Fund, Birchmere Ventures, Mountain State Capital, Riverfront Ventures",
     "source": "https://www.einpresswire.com/article/525366177/aspinity-raises-5-3-million-series-a-funding-to-stop-battery-drain-in-always-on-smart-sensing-products"
    },
    {
     "date": "2023-09",
     "round": "Series B",
     "amount": "$5M",
     "investors": "Anzu Partners, Birchmere, Mountain State, Riverfront + strategic investor Unitrontech (Korean automotive-semi distributor); total then >$19M",
     "source": "https://www.aspinity.com/press-releases/aspinity-raises-5-million-series-b-funding-to-bring-breakthrough-ai-ml-technology-to-a-global-market/"
    }
   ],
   "funding_notes": "Total raised: **$22.1M over 8 rounds** per Tracxn\n(https://tracxn.com/d/companies/aspinity/__XZgrNkTSy3eu4g_riJOoPnZV0PN8tA5Q9-cNDeKGCqA/funding-and-investors);\ncompany PR said \">$19M\" after the 2023 Series B (link above). Earlier seed/grant\nrounds were small and mostly undisclosed.",
   "trajectory": "Spun out to commercialize WVU analog-computing research led by co-founder and CSO\n**David Graham**, with **Brandon Rumberg** (CTO, WVU PhD) and CEO **Tom Doyle**;\nfounding year reported as 2015 (technical.ly, company) though Tracxn lists 2013 —\nlikely incorporation vs. spinoff dates\n(https://technical.ly/growth/aspinity-semiconductor-product-launch-pittsburgh/ ;\nhttps://tracxn.com/d/companies/aspinity/__XZgrNkTSy3eu4g_riJOoPnZV0PN8tA5Q9-cNDeKGCqA).\nTrajectory: consumer/smart-home always-on audio → industrial monitoring → automotive\n(Unitrontech partnership, 2023-2024). No 2025-2026 funding or layoff news found;\ncompany continued marketing activity into 2025 (Audi Minds & Makers, CES demos).\n**Status mid-2026: active**, small, niche always-on-sensing wedge.",
   "relevance": "Aspinity is the purest \"keep it analog as long as possible\" sensing play — analog\nfeature extraction and classification, standard devices, microwatt budgets. It shares\nKnowm's low-power edge framing but sidesteps adaptive devices: learning is offline,\nand the analog fabric is programmed, not plastic.",
   "news": [],
   "sources": [
    {
     "title": "Series A PR",
     "url": "https://www.einpresswire.com/article/525366177/aspinity-raises-5-3-million-series-a-funding-to-stop-battery-drain-in-always-on-smart-sensing-products"
    },
    {
     "title": "Series B PR",
     "url": "https://www.aspinity.com/press-releases/aspinity-raises-5-million-series-b-funding-to-bring-breakthrough-ai-ml-technology-to-a-global-market/"
    },
    {
     "title": "technical.ly product launch/origin",
     "url": "https://technical.ly/growth/aspinity-semiconductor-product-launch-pittsburgh/"
    },
    {
     "title": "AML100 automotive PR",
     "url": "https://www.aspinity.com/press-releases/aspinity-launches-new-aml100-near-zero-power-monitoring-solutions-for-automotive-security/"
    },
    {
     "title": "eeNews: Korean automotive investor",
     "url": "https://www.eenewseurope.com/en/analog-ai-startup-aspinity-gains-korean-automotive-investor/"
    }
   ]
  },
  {
   "slug": "aspirare-semi",
   "name": "Aspirare Semi",
   "founded": "unknown (~2022-2023, undisclosed)",
   "hq": "Ottawa, Canada (company copy) / Toronto, Canada (directory listings) — conflicting",
   "category": "analog-imc",
   "status": "active",
   "origin": "unknown (founders not publicly named in sources found)",
   "total_funding_usd_m": 0,
   "last_round": "none disclosed",
   "key_investors": [],
   "website": "https://www.aspirare.io",
   "pitch": "Aspirare frames compute and energy as the binding constraints on AI and pitches\n\"sustainable AI computing\": analog compute cores for edge devices where \"on-device,\nlow-latency processing is critical,\" claiming \"up to 2x higher performance at 10x\nreduction in energy consumption\" versus traditional (digital) solutions.\nSource: https://www.aspirare.io/",
   "technical_approach": "- **Substrate:** analog compute cores for AI acceleration at the edge; Compute Compass\n  categorizes the company as \"analog neuromorphic accelerators\"\n  (https://computecompass.com/companies/aspirare-semi/). No public detail on the analog\n  storage/compute element (SRAM-, flash-, or capacitor-based IMC vs other) was found.\n- **Products:** \"Gen 1\" AI chip aimed at edge applications (wearables, IoT, robotics);\n  directory listing additionally mentions Gen 2 and \"Edge\" product lines — no dates,\n  process nodes, or benchmarks published.\n- **Training vs inference:** positioned as an edge inference accelerator; no on-chip\n  learning claim found.\n- **Ecosystem:** member of ventureLAB (Markham, Ontario hardware-tech incubator\n  community), per company site.",
   "funding_rounds": [
    {
     "date": "—",
     "round": "none disclosed",
     "amount": "—",
     "investors": "—",
     "source": "https://www.aspirare.io/ ; https://computecompass.com/companies/aspirare-semi/"
    }
   ],
   "funding_notes": "Total raised: **undisclosed**. No press releases, funding announcements, or investor\npages were found in searches (2026-07-23).",
   "trajectory": "Canadian startup; company marketing copy (mirrored on deepenrich) says \"Headquartered in\nOttawa, Canada\" (https://deepenrich.com/company/0175/aspirare-semi-00112j9vg) while\nCompute Compass lists Toronto — both reported here per corpus conflict rule. Founders and\nfounding year are not publicly stated in any source found; contact on the site is\njosh@aspirare.io. Site copy describes a \"rapidly growing\" team building a Gen 1 analog\nedge-AI chip; ventureLAB affiliation suggests pre-seed/incubator stage. Active website as\nof mid-2026; no 2025-2026 news coverage found.\n\n**Founders (sweep 2026-09-22):** still not publicly named. The company site names no people (only a ventureLAB membership and an investment contact), and the 2026-09 searches surfaced nothing beyond directory boilerplate (https://www.aspirare.io).",
   "relevance": "A representative of the long tail of analog edge-AI startups: same energy-efficiency\nmotivation as the memristor/in-memory crowd, but with no disclosed device technology,\nfunding, or silicon — useful in the taxonomy as evidence of how low the barrier to\n*claiming* analog-AI advantage is versus demonstrating it.",
   "news": [],
   "sources": [
    {
     "title": "Aspirare Semi — company site",
     "url": "https://www.aspirare.io/"
    },
    {
     "title": "Compute Compass listing",
     "url": "https://computecompass.com/companies/aspirare-semi/"
    },
    {
     "title": "Deepenrich listing (Ottawa HQ per company copy)",
     "url": "https://deepenrich.com/company/0175/aspirare-semi-00112j9vg"
    }
   ]
  },
  {
   "slug": "bio-inspired-technologies",
   "name": "Bio Inspired Technologies",
   "founded": 2010,
   "hq": "Boise, ID, USA",
   "category": "emerging-device-imc",
   "status": "pivoted",
   "origin": "academic lineage (licensed Kris Campbell's memristor technology, Boise State University)",
   "total_funding_usd_m": 0,
   "last_round": "none disclosed (SBIR/STTR grants only, incl. Air Force)",
   "key_investors": [
    "SBIR/STTR (US Air Force)"
   ],
   "website": "https://www.bioinspired.net",
   "pitch": "BIT \"began in 2010, with the unified purpose of developing memristor based circuits for neural\nnetworking, adaptive intelligence, and neuromorphic applications\" — memristors as integrated\ndevices \"designed to emulate the biological neuron\"\n([bioinspired.net/our-story](https://www.bioinspired.net/our-story)).",
   "technical_approach": "- **Device:** discrete **chalcogenide ion-conducting memristors** sold under the trade name\n  **Neuro-Bit** (16-pin DIP, 44-pin PLCC, raw test die), licensed from **Dr. Kris Campbell's\n  lab at Boise State University** — the same lab whose devices later became Knowm's SDC\n  memristors ([Knowm.org](https://knowm.org/bio-inspired-technologies-neuro-bit-memristors-acquired-by-knowm/);\n  [Hackaday, 2015-07](https://hackaday.com/2015/07/02/new-part-day-memristors/)).\n- **SBIR work:** Air Force SBIR Phase II designed and fabricated a \"soft processor\" chip — a\n  hardware neural network based on the chalcogenide memristor for fuzzy-logic classification\n  ([SBIR.gov](https://www.sbir.gov/awards/142535)).\n- 2014: expanded into adaptive sense-and-avoid control systems for autonomous UAVs.\n- No on-chip-learning product was commercialized; the memristors were discrete research\n  components.",
   "funding_rounds": [
    {
     "date": "~2012–2015",
     "round": "SBIR/STTR grants (Air Force)",
     "amount": "award amounts per SBIR records",
     "investors": "US Air Force",
     "source": "[SBIR.gov](https://www.sbir.gov/awards/142535)"
    }
   ],
   "funding_notes": "No venture funding disclosed anywhere found.",
   "trajectory": "The BSU/Campbell lineage matters and should be stated precisely: BIT held an earlier license to\nCampbell-group memristor technology and sold the Neuro-Bit. **As of May 31, 2016, Boise State\nUniversity exclusively licensed all memristor technology from Dr. Campbell's research group\n(all territories) to Knowm Inc.**; BIT lost manufacturing rights, and the Neuro-Bit line was\nrebranded as Knowm's **SDC Carbon** memristor\n([Knowm.org](https://knowm.org/bio-inspired-technologies-neuro-bit-memristors-acquired-by-knowm/)).\nBIT then pivoted away from hardware — its site cites GPU progress and mature AI software as the\nreason — to AI/ML/computer-vision software services (OSINT, autonomous infrastructure\ninspection). Website is live as of 2026-07-23; SBIR databases list the company as inactive.\nStatus: **pivoted** (out of memristor hardware entirely), not defunct.",
   "relevance": "BIT is Knowm's closest historical sibling: both commercialized discrete memristors from the same\nBSU/Campbell lab, and BIT's Neuro-Bit is literally the predecessor of Knowm's SDC Carbon part.\nIts 2016 exit from hardware makes it the cautionary first chapter of the discrete-memristor\nmarket that Knowm continued.",
   "news": [],
   "sources": [
    {
     "title": "Neuro-Bit license acquired by Knowm — Knowm.org",
     "url": "https://knowm.org/bio-inspired-technologies-neuro-bit-memristors-acquired-by-knowm/"
    },
    {
     "title": "Our Story — bioinspired.net",
     "url": "https://www.bioinspired.net/our-story"
    },
    {
     "title": "New Part Day: Memristors — Hackaday",
     "url": "https://hackaday.com/2015/07/02/new-part-day-memristors/"
    },
    {
     "title": "Air Force SBIR award — SBIR.gov",
     "url": "https://www.sbir.gov/awards/142535"
    },
    {
     "title": "Bio Inspired Technologies — Crunchbase",
     "url": "https://www.crunchbase.com/organization/bio-inspired-technologies"
    },
    {
     "title": "Kris Campbell — Boise State ECE",
     "url": "https://www.boisestate.edu/coen-ece/directory/kris-campbell/"
    }
   ]
  },
  {
   "slug": "biological-computing-co",
   "name": "The Biological Computing Co. (TBC)",
   "founded": 2022,
   "hq": "San Francisco, California, USA",
   "category": "biological",
   "status": "active",
   "origin": "clinician-founders (neurosurgeons Alex Ksendzovsky — concurrently UMD School of Medicine neurosurgery faculty — and Jon Pomeraniec); Dealroom tags a UMD School of Medicine spinout, not confirmed by a primary source",
   "total_funding_usd_m": 25,
   "last_round": "Seed, $25M, 2026-02-10",
   "key_investors": [
    "Primary Venture Partners",
    "Builders VC",
    "Refactor Capital",
    "Wonder Ventures",
    "Tusk Ventures"
   ],
   "website": "https://www.tbc.bio",
   "pitch": "TBC's argument is energetic and architectural: the human brain performs\nexaflop-scale computation on ~20 watts, versus roughly 120,000 watts for comparable\nsilicon, and today's LLMs advance only through \"brute-force scaling and repeated\noptimization cycles\" that grow exponentially more expensive and are ultimately\nunsustainable. Rather than emulate neurons in silicon (neuromorphic), TBC proposes\nto compute with actual living neurons and pair them with AI models. (Framing per\nCEO Alex Ksendzovsky in SiliconANGLE and the Primary Ventures announcement,\nFeb 2026.)",
   "technical_approach": "- Substrate: **living, lab-grown biological neurons** cultured on/against\n  microelectrode interfaces — a wetware compute substrate, not neuromorphic silicon\n  and not in-memory devices. This is the true \"biological\" category of the taxonomy.\n- Product (\"Bionode\" / the biological compute platform): encodes real-world data —\n  text, images, video — directly into living neuron cultures, then decodes the\n  resulting neural activity into representations that map onto frontier AI models via\n  modular \"adapters\" (e.g. VAE efficiency adapters). A parallel \"Inspired Compute\"\n  track uses biologically-derived principles to inform new AI architectures.\n- Learning locus: the biological network provides high-dimensional, continually-\n  adapting feature extraction claimed to learn from far fewer examples; the AI-model\n  side consumes those features. TBC reports its first product delivered a **23×\n  retained improvement in video-model efficiency**, and says it is the first company\n  ever paid by a customer for biological compute. Roadmap: hybrid neuro-silicon\n  cloud clusters targeted for 2027.",
   "funding_rounds": [
    {
     "date": "2026-02-10",
     "round": "Seed",
     "amount": "$25M",
     "investors": "Primary Venture Partners (lead); Builders VC, Refactor Capital, Wonder Ventures, E1 Ventures, Proximity, Tusk Ventures",
     "source": "PRNewswire / SiliconANGLE / DCD / Primary.vc"
    },
    {
     "date": "—",
     "round": "Total (reported)",
     "amount": "$25M",
     "investors": "—",
     "source": "—"
    }
   ],
   "funding_notes": "Reported as the largest-funded dedicated biological-compute play to date. No earlier\npriced round surfaced; the company operated relatively quietly under its former name\nbefore this seed.",
   "trajectory": "Founded 2022 in San Francisco as **Biological Black Box (BBB)**, later renamed **The\nBiological Computing Co. (TBC)**, by neurosurgeon-neuroscientists **Alex Ksendzovsky\n(CEO)** and **Jon Pomeraniec (COO)** — longtime collaborators trained at the\nUniversity of Maryland and University of Pennsylvania. It raised a **$25M seed led by\nPrimary Ventures, announced 10 Feb 2026**, and opened a flagship lab in San\nFrancisco's Mission Bay to support customer deployment. It claims first commercial\nrevenue for biological compute and a 23× video-model efficiency gain. Status\nmid-2026: active, early commercial stage, hiring; hybrid neuro-silicon cloud clusters\npromised for 2027.\n\n**Founders (sweep 2026-09-22):** Alex Ksendzovsky, MD PhD (CEO & co-founder; concurrently Adjunct Clinical Associate Professor of Neurosurgery, University of Maryland School of Medicine, listed there as CEO of TBC) and Jon Pomeraniec, MD MBA (COO & co-founder). Dealroom tags the company a \"spinout\" and lists UMD School of Medicine, and the biology team sits in Baltimore, but no university or company source describes a lab spinout or licence (https://www.tbc.co/about ; http://neurosurgery.umaryland.edu/faculty/neurosurgery-clinical-faculty/ ; https://www.primary.vc/articles/the-biological-computing-company-raises-25m-seed).",
   "relevance": "TBC is the far end of the non-von-Neumann spectrum from Knowm: where Knowm's\nmemristor/AHaH bet is that engineered device physics can do adaptive on-chip\nlearning, TBC bets that *actual biology* is the most energy-efficient learning\nsubstrate and that the job is to interface neurons with AI, not to build a better\ndevice. Both share the core premise — that continual, low-power on-substrate learning\nbeats brute-force digital scaling — but TBC's substrate is wetware with the attendant\nviability, reproducibility and scaling questions, reported here without endorsement of\nthe efficiency claims (which are company-stated).",
   "news": [],
   "sources": [
    {
     "title": "TBC Secures $25M Seed, First to Deploy Neuron-Based Alternative to Silicon AI — PRNewswire",
     "url": "https://www.prnewswire.com/news-releases/the-biological-computing-co-tbc-secures-25-million-seed-round-first-to-deploy-neuron-based-alternative-to-silicon-ai-302685429.html"
    },
    {
     "title": "AI startup The Biological Computing Co. raises $25M to swap silicon for lab-grown neurons — SiliconANGLE, 2026-02-12",
     "url": "https://siliconangle.com/2026/02/12/ai-startup-biological-computing-co-raises-25m-swap-silicon-living-lab-grown-neurons/"
    },
    {
     "title": "The Biological Computing Company raises $25M Seed — Primary.vc",
     "url": "https://www.primary.vc/articles/the-biological-computing-company-raises-25m-seed"
    },
    {
     "title": "The Biological Computing Co. raises $25m for world's first biological computing platform — DCD",
     "url": "https://www.datacenterdynamics.com/en/news/the-biological-computing-co-raises-25m-to-support-launch-of-worlds-first-biological-computing-platform/"
    },
    {
     "title": "TBC secures $25M seed and becomes first company paid for biological compute — Dealroom",
     "url": "https://app.dealroom.co/news/feed/tbc-secures-25m-seed-and-becomes-first-company-paid-for-biological-compute"
    }
   ]
  },
  {
   "slug": "blumind",
   "name": "Blumind",
   "founded": 2020,
   "hq": "Toronto/Ottawa, Ontario, Canada (Toronto per company PR; Ottawa per BetaKit)",
   "category": "analog-imc",
   "status": "active",
   "origin": "serial founders / semiconductor-industry veterans (Niraj Mathur, John Gosson)",
   "total_funding_usd_m": 24,
   "last_round": "Series A, C$20M (~US$14M), 2025-04",
   "key_investors": [
    "Cycle Capital",
    "BDC Capital",
    "Fusion Fund",
    "Two Small Fish Ventures",
    "Real Ventures"
   ],
   "website": "https://www.blumind.ai",
   "pitch": "Blumind pitches AMPL as \"the AI chip we'd have if computers didn't exist\" — arguing\nthat digitizing sensor data before analyzing it wastes nearly all the energy, and that\nits all-analog neural processing \"consumes up to 1000x lower power than available\nalternatives\" for always-on edge AI.\nSources: https://betakit.com/blumind-secures-20-million-series-a-to-build-the-ai-chip-wed-have-if-computers-didnt-exist/ ;\nhttps://www.businesswire.com/news/home/20250408503288/en/Blumind-Secures-Series-A-Funding-to-Accelerate-Analog-AI-Revolution",
   "technical_approach": "- **AMPL™**: an all-analog neural-network fabric implemented in **standard CMOS\n  transistors** (no exotic memory device), processing sensor signals in the analog\n  domain; inference-only, targeting always-on audio/time-series first, with a roadmap\n  from RNN-style audio toward CNN vision and small transformer-class models\n  (https://www.businesswire.com/news/home/20250408503288/en/Blumind-Secures-Series-A-Funding-to-Accelerate-Analog-AI-Revolution ;\n  https://www.eetimes.com/blumind-harnesses-analog-for-ultra-low-power-intelligence/).\n- Test silicon for a keyword-spotting chip shown at electronica 2024: ~10 nJ per\n  inference, a few microwatts in always-on operation, connecting directly to an analog\n  microphone and waking an MCU on detection\n  (https://www.eetimes.com/blumind-hints-at-big-chip-on-roadmap/).\n- Named product: **BM110** always-on analog AI audio-inference SoC — CES 2026\n  Innovation Award honoree; company cites <5% of the power of digital solutions\n  (https://www.ces.tech/ces-innovation-awards/2026/blumind-bm110-always-on-tiny-analog-ai-audio-inferencing-chip/).\n  Demonstrated at ~60 µW for always-on audio/wearables/vision (ARMdevices, 2026-03-15:\n  https://armdevices.net/2026/03/15/blumind-ampl-analog-ai-at-60-microwatts-for-always-on-audio-edge-wearables-and-vision/).",
   "funding_rounds": [
    {
     "date": "2023-01",
     "round": "Early round",
     "amount": "undisclosed",
     "investors": "(existing investors incl. Fusion Fund, Two Small Fish, Real Ventures)",
     "source": "https://www.digitimes.com/news/a20230529VL204/interview-ai-adas-semiconductor-software-big-data.html"
    },
    {
     "date": "2025-04",
     "round": "Series A",
     "amount": "C$20M (~US$14M)",
     "investors": "Cycle Capital + BDC Capital (co-leads); Fusion Fund, Two Small Fish Ventures, Real Ventures",
     "source": "https://www.businesswire.com/news/home/20250408503288/en/Blumind-Secures-Series-A-Funding-to-Accelerate-Analog-AI-Revolution"
    }
   ],
   "funding_notes": "Total raised: **C$34M (~US$24M)** to date per BetaKit\n(https://betakit.com/blumind-secures-20-million-series-a-to-build-the-ai-chip-wed-have-if-computers-didnt-exist/).\nPre-Series-A round amounts were not disclosed.",
   "trajectory": "Founded 2020 by **Niraj Mathur** (CEO) and **John Gosson** (CTO), with Gosson's\npatented analog NN architecture as the basis; Roger Levinson has also been reported as\nCEO in earlier coverage\n(https://www.ept.ca/features/blumind-reimagines-ai-processing-with-breakthrough-analog-chip/ ;\nhttps://betakit.com/blumind-secures-20-million-series-a-to-build-the-ai-chip-wed-have-if-computers-didnt-exist/).\nSeries A (Apr 2025) funds commercialization of the first-generation processor with a\ntier-1 lead customer (wearable), with first product production planned for 2026.\n**Status mid-2026: active**, pre-volume-production, CES 2026 award visibility.\n\n**Founders (sweep 2026-09-22):** Niraj Mathur (co-founder, CEO; BE McGill, MBA Cornell; ex-VP marketing Rambus, ex-VP Product GigaIO) and John Gosson (co-founder, CTO; Ottawa; holder of the patented analog architecture the chips are built on — prior employer not confirmed); Roger Levinson joined the founders as CEO, later COO (https://ca.linkedin.com/in/niraj-mathur ; https://magazine.annexbusinessmedia.com/article/BLUMIND+WANTS+TO+BLOW+YOUR+MIND/4941258/842362/article.html ; https://www.eenewseurope.com/en/blumind-startup-pursues-analog-ai-at-the-edge/ ; https://www.crunchbase.com/organization/blumind-ai).",
   "relevance": "Blumind takes the \"physics does the compute\" position using nothing but standard CMOS —\na deliberate contrast to both memristor startups and flash-analog players, avoiding new\ndevices entirely. Inference-only and tiny-model-focused; no on-chip learning claim.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "CEO **Niraj Mathur appointed to the Canada Semiconductor Council board** (announced ~2026-07-14), alongside coverage of Canada's homegrown AI-hardware wave; Blumind also won the Siemens/AWS \"Industry Disruptor\" pitch challenge at CES 2026. (https://blumind.ai/blog/blumind-wins-siemens-and-aws-industry-disruptor-pitch-challenge-at-ces-2026/)"
    },
    {
     "date": "2026-11-09",
     "text": "**BM110 evaluation kit launched (2026-11-09)** — first customer-bench hardware for its analog always-on inference chip (EIN Presswire release; specs/price behind a blocked page). (https://www.einnews.com/pr_news/954953237/blumind-launches-bm110-evaluation-kit-putting-ultra-low-power-analog-ai-compute-on-the-customer-s-bench)"
    }
   ],
   "sources": [
    {
     "title": "Series A PR",
     "url": "https://www.businesswire.com/news/home/20250408503288/en/Blumind-Secures-Series-A-Funding-to-Accelerate-Analog-AI-Revolution"
    },
    {
     "title": "BetaKit Series A story",
     "url": "https://betakit.com/blumind-secures-20-million-series-a-to-build-the-ai-chip-wed-have-if-computers-didnt-exist/"
    },
    {
     "title": "EE Times: Blumind Harnesses Analog",
     "url": "https://www.eetimes.com/blumind-harnesses-analog-for-ultra-low-power-intelligence/"
    },
    {
     "title": "CES 2026 Innovation Award: BM110",
     "url": "https://www.ces.tech/ces-innovation-awards/2026/blumind-bm110-always-on-tiny-analog-ai-audio-inferencing-chip/"
    }
   ]
  },
  {
   "slug": "brainchip",
   "name": "BrainChip",
   "founded": 2004,
   "hq": "Laguna Hills, CA, USA (Sydney, Australia holding company; founded in Perth)",
   "category": "digital-snn",
   "status": "public",
   "origin": "serial founder (Peter van der Made); pre-wave incumbent — predates the 2015+ startup wave",
   "total_funding_usd_m": "n/a (ASX-listed since 2015; funded via successive placements — see funding history)",
   "last_round": "US$25M placement, 2025-12",
   "key_investors": [
    "ASX public market",
    "LDA Capital (put-option facility)",
    "retail/institutional placements"
   ],
   "website": "https://brainchip.com",
   "pitch": "\"Pioneering the future of edge AI with neuromorphic computing ... Real-time intelligence that\nruns locally. Ultra-low power. No data leaves the device.\" Akida is positioned as\n\"Neuromorphic AI, Built for the Edge\" — always-on, private, sensor-to-action processing\ninstead of shipping data to cloud GPUs. Source: https://brainchip.com/ (accessed 2026-07-23).",
   "technical_approach": "- Fully **digital CMOS** event-based spiking/neuromorphic processor (\"Akida\"); marketed for\n  edge inference with **on-chip incremental (one-shot/few-shot) learning** as a differentiator.\n- **AKD1000**: first-gen Akida silicon; dev kits announced Oct 2021, PCIe boards Jan 2022;\n  1.2M neurons / 10B synapses (https://en.wikipedia.org/wiki/BrainChip).\n- **AKD1500**: design completed Jan 2023 on GlobalFoundries 22nm; \"now shipping\" per company\n  site; initial order of 1,200 AKD1500 chips received 12 Dec 2025\n  (https://www.otcmarkets.com/file/company/financial-report/522904/content).\n- **Akida 2.0** (announced Mar 2023): adds 8-bit weights/activations, Vision Transformer\n  support, and Temporal Event-based Neural Networks (TENNs)\n  (https://en.wikipedia.org/wiki/BrainChip).\n- **Akida Pico** (2024): ultra-low-power variant; also IP-core licensing model, Akida Cloud\n  (https://brainchip.com/).\n- Business model is chip + IP licensing (Arm/Intel Foundry/Edge Impulse partnerships).\n- **Akida GenAI**: 2026 technology roadmap presented by Chief Development Officer Jonathon\n  Tapson centers on \"Akida GenAI,\" an on-device LLM platform, alongside AKD1500 and Akida\n  Pico (https://brainchip.com/brainchips-2026-technology-roadmap/); launch slated H2 2026\n  per the 2026-07 who's-who research report (_research/2026-07_whos_who_deep_research.md)\n  and German market coverage (https://www.aktiencheck.de/news/Artikel-BrainChip_Akida_GenAI_H2_2026-19738051).\n- **ASICLAND** (Korean ASIC-services firm): non-exclusive worldwide Akida IP licensing\n  agreement announced May 2026 — Akida integration into customer SoC designs, with\n  evaluation/production license fees plus royalties\n  (https://brainchip.com/brainchip-strikes-ip-licensing-deal-with-asicland/).",
   "funding_rounds": [
    {
     "date": "2020–",
     "round": "LDA Capital put-option agreement (POA), repeatedly amended; capital call notices through 2025",
     "amount": "tranche-based",
     "investors": "LDA Capital",
     "source": "https://investor.brainchip.com/wp-content/uploads/2025/09/Half-Yearly-Report-June-2025.pdf"
    },
    {
     "date": "2025-11",
     "round": "Institutional placement + Share Purchase Plan",
     "amount": "A$35M + ~A$3M SPP (SPP overs of A$0.976M above A$2M target)",
     "investors": "institutional/retail",
     "source": "https://www.listcorp.com/asx/brn/brainchip-holdings/news/35m-placement-and-2m-share-purchase-plan-3273643.html ; https://www.intelligentinvestor.com.au/shares/asx-brn/brainchip-holdings-ltd/announcements?page=2"
    },
    {
     "date": "2025-12",
     "round": "Capital raise announced ahead of CES 2026",
     "amount": "US$25M",
     "investors": "not disclosed in PR",
     "source": "https://brainchip.com/brainchip-announces-25-million-usd-funding-ahead-of-ces-to-power-next-gen-edge-ai/ ; https://www.nasdaq.com/press-release/brainchip-announces-25-million-usd-funding-ahead-ces-power-next-gen-edge-ai-2025-12"
    }
   ],
   "funding_notes": "Public company (ASX: BRN; OTCQX: BRCHF) — funded by market raises rather than VC rounds.\nListed Sep 2015 via reverse merger with Australian mining company Aziana\n(https://en.wikipedia.org/wiki/BrainChip).\nCumulative lifetime raises (many smaller placements since 2015) are not tabulated here; no\nsingle authoritative total exists in the sources reviewed. One third-party tracker cites\n~$60M raised historically (https://neuromorphiccore.ai/insights/applied-brain-research-abr/),\nwhich predates the late-2025 raises — treat as approximate.",
   "trajectory": "Founded 2004 by Peter van der Made (Perth, Australia); acquired by/reverse-merged into ASX\nshell Aziana in March–September 2015 (https://en.wikipedia.org/wiki/BrainChip). Van der Made\nhas since retired as CTO and sits on the board / Scientific Advisory Board; Anil Mankar is\ncredited as co-founder and former Chief Development Officer (https://brainchip.com/company/,\nhttps://markets.financialcontent.com/concordmonitor/article/bizwire-2024-8-7-brainchip-appoints-new-cmo-enhances-scientific-advisory-board).\nManagement expects first meaningful revenue in 2026–2027 per the 2026-07 who's-who research\nreport (_research/2026-07_whos_who_deep_research.md), unverified. Two decades of\nR&D preceded first commercial silicon (2021). As of mid-2026: **active, public**; shipping\nAKD1500, promoting Akida 2.0/TENNs and Akida Pico; still pre-profit and raising capital\nregularly (Nov 2025 A$35M placement, Dec 2025 US$25M raise). Long-running retail-investor\nattention on ASX; revenue remains small relative to market cap (see half-yearly report above).\n\n**Ecosystem note (added 2026-07-23, found by Tim):** independent practitioner traction exists\noutside BrainChip's own marketing — e.g. Kevin D. Johnson (IBM Field CTO, HPC cloud), whose\npersonal site documents 65+ demos (Dec 2025–Jul 2026) pairing Akida AKD1000/AKD1500 boards\nwith LLMs and distributed orchestration (IBM Spectrum Symphony/LSF, vLLM), claiming\napplications from neuromorphic music generation to missile-defense prototypes\n(https://kevindjohnson.org/). Personal-site claims, unverified and not BrainChip-affiliated —\nuseful as evidence of third-party developer uptake and of the neuromorphic+LLM integration\npattern (article §4), not as product validation.\n\n**Founders (sweep 2026-09-22):** Peter van der Made (founder; earlier founded vCIS Technology — CTO, then Chief Scientist after its acquisition by Internet Security Systems and then IBM — and PolyGraphic Systems) and Anil Mankar (co-founder, ex-Chief Development Officer; previously VP Engineering at Conexant Systems). No academic lab lineage (https://brainchip.com/company/).",
   "relevance": "Closest commercial analogue to Knowm's on-chip-learning claim: Akida advertises on-chip\nincremental learning, but implements it in digital CMOS SRAM logic rather than in an analog\nmemory element like a memristor. Its 20-year, still-loss-making trajectory is the cautionary\nbaseline for how long neuromorphic commercialization takes.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "BrainChip **open-sourced its Akida software stack for AKD1500** and launched the \"Symphony Community Akida Bundle\" integrating with IBM Spectrum Symphony workload management (2026-08-20). (https://www.businesswire.com/news/home/20260820171305/en/BrainChip-Launches-Symphony-Community-Akida-Bundle-For-IBMs-Workload-Management-Solution ; https://www.opensourceforu.com/2026/08/brainchip-software-stack-akd1500/)"
    },
    {
     "date": "2026-08-23",
     "text": "BrainChip + Neuromorphyx launched the **BrainBoard1500** AKD1500 embedded developer board (2026-08-18/20; already detailed in neuromorphyx.md). (https://www.hpcwire.com/aiwire/2026/08/20/brainchip-and-neuromorphyx-launch-akd1500-developer-board/)"
    },
    {
     "date": "2026-08-23",
     "text": "BrainChip and **Orama.AOI** target AI-based industrial defect detection (2026-08-19). (https://www.embedded.com/brainchip-orama-aoi-target-ai-based-industrial-defect-detection/)"
    },
    {
     "date": "2026-09-22",
     "text": "BrainChip now sells **AKD1500 evaluation hardware in every common slot**: the AKD1500 PCIe development card (US$149, launched ~2026-09-17) joins the AKD1500 M.2 B+M card (US$129, July 2026) and the US$99 BrainBoard 1500 SPI module (Aug 2026, with Neuromorphyx). Card spec: 800 GOPS at <300 mW, 22 nm FD-SOI, on-chip learning. (https://www.cnx-software.com/2026/09/21/brainchip-akd1500-m-2-and-pcie-edge-ai-cards-brainboard-1500-spi-module/ ; https://www.hpcwire.com/aiwire/2026/09/21/brainchip-launches-akd1500-pcie-card-for-edge-ai-evaluation-everywhere/)"
    },
    {
     "date": "2026-09-22",
     "text": "**H1/Q2 2026 earnings call (~2026-09-10):** cash ~US$20M at 30 June 2026; ~25% of the initial AKD1500 production run sold into customer projects; **Parsons placed an initial order for \"several thousand\" Akida chips** (framed as a multi-year defense partnership); Akida 2500 in tape-out; first licensee Edge AI to tape out its own chip later in 2026; management admitted it missed the prior-year bookings goal, gave no financial guidance, and said it will \"assess capital options\" — the war-chest question from the August digest is now explicit. (https://www.theglobeandmail.com/investing/markets/markets-news/Tipranks/4576608/brainchip-earnings-call-strong-pipeline-execution-tests/ ; https://www.investing.com/news/transcripts/earnings-call-transcript-brainchip-q2-2026-loss-narrows-as-chip-demand-builds-93CH-4897154)"
    },
    {
     "date": "2026-10-27",
     "text": "**Sequora Core IP launched (2026-10-14):** a state-space-model processor core for on-device generative AI/LLMs 'from a few million parameters up to 3B', fixed memory regardless of sequence length, paired with Akida for sensor streams; available for licensing, **no signed licence at launch** (company says adoption pace 'cannot be reliably forecast'; shares fell on the news). Quotes from ForwardEdge ASIC and edgeAI. (https://finance.yahoo.com/technology/ai/articles/brainchip-launches-sequora-edge-ai-130000333.html ; https://www.ad-hoc-news.de/boerse/news/unternehmensnachrichten/brainchip-slides-as-sequora-ip-hits-the-market-without-a-single-signed-license/70374854)"
    },
    {
     "date": "2026-10-27",
     "text": "**AKD1500 SPI (2026-10-22):** a co-processor variant for microcontroller-based wearables/smart-home devices over SPI, with on-chip learning; Neuromorphyx (BrainBoard1500, NeuroReflex) quoted as the reference user. The AKD1500 family now spans SPI, M.2 and PCIe. (https://finance.yahoo.com/technology/ai/articles/brainchip-unveils-akd1500-spi-chip-130000577.html)"
    },
    {
     "date": "2026-10-27",
     "text": "**Q3 (Sept-quarter) Appendix 4C: not retrieved.** Kalkine's 2026-10-26/27 pieces ('priorities shifting', 'nearer commercial scale despite its loss') suggest the quarterly landed ~2026-10-26, but the figures (cash vs ~US$20M at 30 June, receipts, any placement) were behind bot-blocked pages. **Verify on ASX before quoting cash.**"
    },
    {
     "date": "2026-11-09",
     "text": "**September-quarter Appendix 4C was released 2026-10-30** (HotCopper and Renew BRN timelines confirm the filing) but every host we can reach (investor.brainchip.com, HotCopper, Market Index, Kalkine) blocks automated readers, so **cash at 30 Sep 2026, receipts and any capital raise remain unrecorded here**. Last verified cash: ~US$20M at 30 June 2026. Do not quote a newer figure without the ASX PDF."
    },
    {
     "date": "2026-11-10",
     "text": "**New Akida SPI processor** announced, broadening access to Akida for low-cost microcontroller hosts over SPI (Kalkine, 2026-11-10; page blocked to bots, headline-level only). (https://kalkine.com.au/news/penny-stocks/brainchip-holdings-asxbrn-broadens-akida-chip-access-with-new-spi-processor)"
    }
   ],
   "sources": [
    {
     "title": "BrainChip homepage",
     "url": "https://brainchip.com/"
    },
    {
     "title": "BrainChip — Wikipedia",
     "url": "https://en.wikipedia.org/wiki/BrainChip"
    },
    {
     "title": "A$35M Placement and A$2M SPP (Listcorp, 10 Nov 2025)",
     "url": "https://www.listcorp.com/asx/brn/brainchip-holdings/news/35m-placement-and-2m-share-purchase-plan-3273643.html"
    },
    {
     "title": "BrainChip $25M USD funding PR (Dec 2025)",
     "url": "https://brainchip.com/brainchip-announces-25-million-usd-funding-ahead-of-ces-to-power-next-gen-edge-ai/"
    },
    {
     "title": "Half-Yearly Report June 2025 (LDA Capital POA)",
     "url": "https://investor.brainchip.com/wp-content/uploads/2025/09/Half-Yearly-Report-June-2025.pdf"
    },
    {
     "title": "Quarterly/annual filing mentioning AKD1500 order + capital raise (OTC Markets, Jan 2026)",
     "url": "https://www.otcmarkets.com/file/company/financial-report/522904/content"
    },
    {
     "title": "Kevin D. Johnson — independent Akida+LLM orchestration demos",
     "url": "https://kevindjohnson.org/"
    },
    {
     "title": "BrainChip 2026 Technology Roadmap (AGM report, Tapson)",
     "url": "https://brainchip.com/brainchips-2026-technology-roadmap/"
    },
    {
     "title": "BrainChip–ASICLAND IP licensing deal (May 2026)",
     "url": "https://brainchip.com/brainchip-strikes-ip-licensing-deal-with-asicland/"
    },
    {
     "title": "BrainChip company page (Mankar co-founder; van der Made board/SAB)",
     "url": "https://brainchip.com/company/"
    }
   ]
  },
  {
   "slug": "ceryx-medical",
   "name": "Ceryx Medical",
   "founded": 2016,
   "hq": "Cardiff / Bristol, UK",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (University of Bath physics — Alain Nogaret; Bristol/Auckland cardiac physiology — Julian Paton)",
   "total_funding_usd_m": 15,
   "last_round": "$15M announced 2025-06 (company headline; likely cumulative — see funding notes)",
   "key_investors": [
    "Development Bank of Wales",
    "angel syndicates (SETsquared)"
   ],
   "website": "https://www.ceryxmedical.com",
   "pitch": "Not a von-Neumann pitch: conventional pacemakers pace metronomically, while real hearts vary\nbeat-to-beat with breathing (respiratory sinus arrhythmia, RSA). Ceryx builds \"bioelectronic\ntechnology that could change the way diseases such as heart failure are treated\" — analog\ncircuits that behave like the neurons that naturally pace the heart, \"designed to work in\nharmony with the body's natural rhythms\" (https://www.ceryxmedical.com;\nhttps://www.ceryxmedical.com/post/ceryx-medical-raises-15m-to-revolutionise-heart-failure-treatment).",
   "technical_approach": "Solid-state analog neurons / artificial central pattern generators — hardware neuron circuits\n(Nogaret's Bath group is known for analog silicon neurons that reproduce biological neuron\ndynamics) that generate adaptive cardiac pacing synchronized to respiration. Lead product\n**Cysoni**, an RSA pacing device for heart failure: paces the heart in dialogue with breathing\nrather than at fixed rate. Prototyped as artificial CPG systems tested on large animals\n(https://www.setsquared.co.uk/case-studies/ceryx-medical-angel-investment-helps-medical-device-company-achieve-key-milestones/).\nFirst-in-human study of Cysoni initiated (announced via company news, ~mid-2025,\nhttps://www.ceryxmedical.com/post/ceryx-medical-initiates-first-in-human-study-of-the-cysoni-rsa-pacemaker),\nrun with Prof. Julian Paton's translational group (University of Auckland,\nhttps://www.translationalcardiorespiratory.auckland.ac.nz/translation/). This is inference-free,\nlearning-free neuromorphic hardware: the neuron dynamics themselves are the product.",
   "funding_rounds": [
    {
     "date": "~2017-2019",
     "round": "Angel / early",
     "amount": "undisclosed",
     "investors": "SETsquared-network angels; Development Bank of Wales",
     "source": "https://www.setsquared.co.uk/case-studies/ceryx-medical-angel-investment-helps-medical-device-company-achieve-key-milestones/ ; https://developmentbank.wales/case-studies/ceryx-medical"
    },
    {
     "date": "~2025",
     "round": "Round to fund in-human trials",
     "amount": "£3.8M",
     "investors": "undisclosed",
     "source": "https://www.ceryxmedical.com/post/ceryx-raises-3-8m-to-bring-cardiac-device-to-in-human-trials"
    },
    {
     "date": "2025-06-20 (announced)",
     "round": "\"Raises $15M\"",
     "amount": "$15M",
     "investors": "undisclosed",
     "source": "https://www.ceryxmedical.com/post/ceryx-medical-raises-15m-to-revolutionise-heart-failure-treatment"
    }
   ],
   "funding_notes": "**Conflict note (facts straight):** the company's own post headline says \"$15 million in new\nfunding\"; Cardiac Rhythm News describes it as \"a new funding round, bringing the company's\ntotal raised to US$15 million\" (https://cardiacrhythmnews.com/tag/ceryx-medical/), and CEO\nStuart Plant's LinkedIn says \"raised over $15m in investment across multiple rounds\"\n(https://uk.linkedin.com/in/stuart-plant-36551aa). $15M is treated here as the cumulative\ncitable total; the June-2025 round's own size is not cleanly separable from the sources.\nAggregator startupintros lists \"$12.0M total across 2 rounds\" — inconsistent, unverified.",
   "trajectory": "Formed 2016 as a joint venture exploiting neuronal technology patented at the University of\nBath (Prof. Alain Nogaret, condensed-matter physics) and cardiac know-how of Prof. Julian Paton\n(then Bristol, now Auckland) (https://www.bath.ac.uk/announcements/enhanced-pacemaker-developer-ceryx-medical-wins-prestigious-institute-of-physics-award/;\nDealroom adds a third co-founder, https://dealroom.co/companies/ceryx-medical/). CEO Stuart\nPlant since December 2019. Institute of Physics award 2022. Milestones: large-animal CPG\nprototypes → £3.8M to reach in-human trials → Cysoni first-in-human study initiated → $15M\nposition announced June 2025 for next-generation biomimetic pacemaker development. Status\nmid-2026: active, clinical-stage; still circulating in med-device trade roundups (STAT Health\nTech item recirculated 2026-06).\n\n**Founders (sweep 2026-09-22):** Prof. Alain Nogaret (Condensed Matter Physics, University of Bath — designed the nonlinear/neuronal electronics), Prof. Julian Paton (cardiac physiology; then Bristol, now Auckland) and Dr. Stuart Plant (CEO; ex-GE Healthcare and IP Group). The University of Bath describes Ceryx as \"a spin-out from the Universities of Bath and Bristol\" (https://www.bath.ac.uk/announcements/enhanced-pacemaker-developer-ceryx-medical-wins-prestigious-institute-of-physics-award/ ; https://dealroom.co/companies/ceryx-medical/ ; https://www.auckland.ac.nz/en/news/2025/12/02/pacemaker-could-help-heart-heal.html).",
   "relevance": "The clearest commercial proof that analog hardware neurons have value entirely outside AI\ncompute: Ceryx monetizes neuron dynamics as therapy, sidestepping the accelerator market that\nmost of the corpus (and Knowm's memristor/AHaH work) competes in.",
   "news": [
    {
     "date": "2026-10-27",
     "text": "**Watch, unverified:** 'First implantable medical device manufactured at King's College London's MAISI facility' (Today's Medical Developments, 2026-10-27) surfaced on the Ceryx query; the page returned empty and Ceryx's own news page lists nothing after 2026-01. Could be Cysoni or an unrelated device — do not attribute. (https://www.todaysmedicaldevelopments.com/news/first-implantable-medical-device-manufactured-kings-college-london-maisi-facility/)"
    }
   ],
   "sources": [
    {
     "title": "Ceryx Medical Raises $15M (company post, 2025-06-20)",
     "url": "https://www.ceryxmedical.com/post/ceryx-medical-raises-15m-to-revolutionise-heart-failure-treatment"
    },
    {
     "title": "Cardiac Rhythm News: Ceryx Medical tag (total-raised reading)",
     "url": "https://cardiacrhythmnews.com/tag/ceryx-medical/"
    },
    {
     "title": "Ceryx raises £3.8m to bring cardiac device to in-human trials (company post)",
     "url": "https://www.ceryxmedical.com/post/ceryx-raises-3-8m-to-bring-cardiac-device-to-in-human-trials"
    },
    {
     "title": "First-in-human study of Cysoni RSA pacemaker (company post)",
     "url": "https://www.ceryxmedical.com/post/ceryx-medical-initiates-first-in-human-study-of-the-cysoni-rsa-pacemaker"
    },
    {
     "title": "University of Bath: Ceryx wins Institute of Physics award (2022-11-03; 2016 formation)",
     "url": "https://www.bath.ac.uk/announcements/enhanced-pacemaker-developer-ceryx-medical-wins-prestigious-institute-of-physics-award/"
    },
    {
     "title": "SETsquared case study (angel funding, large-animal CPG prototypes)",
     "url": "https://www.setsquared.co.uk/case-studies/ceryx-medical-angel-investment-helps-medical-device-company-achieve-key-milestones/"
    },
    {
     "title": "Development Bank of Wales case study",
     "url": "https://developmentbank.wales/case-studies/ceryx-medical"
    },
    {
     "title": "Dealroom: Ceryx Medical profile (2016 spin-out, founders)",
     "url": "https://dealroom.co/companies/ceryx-medical/"
    },
    {
     "title": "Stuart Plant LinkedIn (CEO since 2019-12; >$15m across multiple rounds)",
     "url": "https://uk.linkedin.com/in/stuart-plant-36551aa"
    }
   ]
  },
  {
   "slug": "cortical-labs",
   "name": "Cortical Labs",
   "founded": 2019,
   "hq": "Melbourne, Australia",
   "category": "biological",
   "status": "active",
   "origin": "serial founders (Hon Weng Chong, clinician-entrepreneur, and Andy Kitchen — both came out of CliniCloud, Chong's earlier startup)",
   "total_funding_usd_m": 11,
   "last_round": "Series A, $10M, 2023-04",
   "key_investors": [
    "Horizons Ventures",
    "Blackbird Ventures",
    "In-Q-Tel",
    "LifeX",
    "Radar Ventures"
   ],
   "website": "https://corticallabs.com",
   "pitch": "Cortical Labs argues that today's deep-learning hardware is enormously energy-hungry and still fails\nto match the adaptability of biological brains, which learn continuously on a few watts. Its answer is\n\"Synthetic Biological Intelligence\" (SBI) — using real, lab-grown human neurons as the learning\nsubstrate, claiming they self-organize and adapt in ways silicon networks cannot, at a fraction of the\nenergy. (https://corticallabs.com; DCD coverage)",
   "technical_approach": "Living neurons cultured on a high-density multi-electrode array (MEA); the silicon reads neuronal\nactivity and writes stimulation in sub-millisecond closed-loop feedback, training the culture via\nstructured electrical reward/penalty signals. Origin demo: **DishBrain** (2022) — ~800,000 cortical\nneurons taught to play Pong, published in *Neuron*. Product: **CL1**, launched in Barcelona on\n**2 March 2025**, marketed as the \"world's first commercial biological computer\" — ~800,000 lab-grown\nhuman neurons on a chip in a self-contained life-support unit, ~$35,000 per unit, ~850–1,000 W,\nrunning without an external computer. Cortical also planned a cloud \"biological neural network\"\nserver stack (30 units per stack) for remote access (\"Wetware-as-a-Service\"). This is genuine\nbiological computation with on-substrate learning. (https://www.datacenterdynamics.com/en/news/australian-startup-cortical-labs-unveils-worlds-first-commercial-biological-computer/)",
   "funding_rounds": [
    {
     "date": "2019/2020",
     "round": "Seed",
     "amount": "~$1.6M",
     "investors": "Blackbird Ventures (lead)",
     "source": "Crunchbase"
    },
    {
     "date": "2023-04",
     "round": "Series A",
     "amount": "$10M",
     "investors": "Horizons Ventures (lead); In-Q-Tel, LifeX, Radar Ventures",
     "source": "https://techcrunch.com/2023/04/19/cortical-labs-raises-10m-for-its-pong-playing-stem-cells-which-eventually-could-power-ai/"
    }
   ],
   "funding_notes": "Total raised reported ~$11M–$11.6M across ~3 rounds. In-Q-Tel (CIA venture arm) participation noted.",
   "trajectory": "Founded 2019 in Melbourne by Dr. **Hon Weng Chong** (with Andy Kitchen among the early team), a\nclinician-turned-entrepreneur. The 2022 DishBrain Pong result drew global attention and grounded the\nscience. Commercial pivot delivered the **CL1** in March 2025 and a developer/cloud program targeting\nresearch labs, drug discovery, and disease modeling as first markets. Active as of mid-2026, shipping\nCL1 units and standing up cloud access; still early-stage and pre-revenue at scale.\n(https://en.wikipedia.org/wiki/Cortical_Labs)\n\n**Founders (sweep 2026-09-22):** Hon Weng Chong (CEO; medical doctor trained at the University of Melbourne and Johns Hopkins; co-founder & CTO of CliniCloud before this) and Andy Kitchen (co-founder & CTO; Head of AI Research at CliniCloud Oct 2017–Jun 2019 per his LinkedIn, i.e. both founders came out of Chong's earlier company). Founding was \"inspired by\" a DeepMind paper, not a university lab (https://snfnostos.org/en/participants/hon-weng-chong/ ; https://de.linkedin.com/in/andy-kitchen ; https://www.forbes.com/sites/zinnialee/2023/06/21/cortical-labs-brain-computer/ ; https://blackbird.vc/blog/spotlight-on-cortical-labs).",
   "relevance": "Cortical Labs is the strongest \"on-substrate learning\" claim among the biological startups — the\nneurons themselves adapt in a closed loop, directly analogous in spirit to Knowm's AHaH\nlearn-in-the-device thesis, but implemented in living tissue rather than memristors. Both reject the\ntrain-off-chip/infer-on-chip split; Knowm's substrate is manufacturable solid-state hardware, whereas\nCortical's requires cultured cells and life support — a sharp contrast in scalability and stability.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "DayOne, Cortical Labs and NUS Medicine launched Singapore's first **biological data center prototype** — a server rack of CL1 units (~16M living human neurons) deployed inside a commercial data-center operator's facility; announced 2026-08-17/18. First deployment of Cortical's tech in a commercial DC setting; neurons reported to last ~6 months per culture. (https://www.prnewswire.com/apac/news-releases/dayone-launches-singapores-first-biological-data-center-prototype-with-cortical-labs-and-nus-medicine-302852460.html ; https://www.datacenterdynamics.com/en/news/dayone-partners-with-cortical-labs-nus-medicine-for-deployment-of-singapores-first-biological-data-center-prototype/)"
    },
    {
     "date": "2026-10-27",
     "text": "**HIVE (presented at the first International Conference on Organoid Intelligence, Washington DC, 2026-10-26/27): 15 CL1 units linked** so cultures divide tasks. Company-reported results: symbol recognition ~80% → 99% with seven units; grid-navigation task — DQN baseline ~32% of max score, best single CL1 ~40%, voting ensemble ~70%, specialist groups ~80%. Preliminary, company release, no preprint cited. Also: AFR named Cortical its 2026 'Gamechanger'; the NUS/DayOne Singapore rack (20 CL1 units, ~16M neurons quoted, not measured) got a second wave of coverage 2026-09-30. (https://newatlas.com/ai-humanoids/living-computers-forge-a-hive-mind-and-work-together-to-outsmart-ai/ ; https://www.sify.com/data-centers/singapore-just-built-a-data-center-with-16-million-living-human-neurons/)"
    },
    {
     "date": "2026-11-09",
     "text": "**Robot dog steered by a CL1 culture** (New Scientist 2026-10-26, wide pickup 2026-10-31 → 11-05): a quadruped controlled by lab-grown human neurons finding a target; company demo, no paper. (https://www.sciencealert.com/scientists-have-built-a-robot-dog-powered-by-human-brain-cells)"
    }
   ],
   "sources": [
    {
     "title": "Cortical Labs unveils world's first commercial biological computer (DCD)",
     "url": "https://www.datacenterdynamics.com/en/news/australian-startup-cortical-labs-unveils-worlds-first-commercial-biological-computer/"
    },
    {
     "title": "Cortical Labs raises $10M (TechCrunch)",
     "url": "https://techcrunch.com/2023/04/19/cortical-labs-raises-10m-for-its-pong-playing-stem-cells-which-eventually-could-power-ai/"
    },
    {
     "title": "Cortical Labs launches $35K biological computer (BiopharmaTrend)",
     "url": "https://www.biopharmatrend.com/news/cortical-labs-introduces-biological-computer-built-on-human-brain-cells-1156/"
    },
    {
     "title": "Cortical Labs (Wikipedia)",
     "url": "https://en.wikipedia.org/wiki/Cortical_Labs"
    }
   ]
  },
  {
   "slug": "crossbar",
   "name": "Crossbar, Inc.",
   "founded": 2010,
   "hq": "Santa Clara, California, USA",
   "category": "emerging-device-imc (ReRAM memory/IP incumbent; pre-wave)",
   "status": "active (private, quiet)",
   "origin": "academic spinoff (University of Michigan — Wei Lu's lab; UMich-licensed RRAM patents)",
   "total_funding_usd_m": 120,
   "last_round": "$5.17M, 2023-11 (per CB Insights)",
   "key_investors": [
    "Artiman Ventures",
    "Kleiner Perkins Caufield & Byers",
    "Northern Light Venture Capital",
    "University of Michigan MINTS",
    "China/HK investors (2015 round)"
   ],
   "website": "https://www.crossbar-inc.com",
   "pitch": "Crossbar is a pre-wave ReRAM company pitched primarily against NAND flash rather than the\nvon Neumann bottleneck: its ReRAM promised faster writes, lower power, and higher endurance\nthan NAND at half the die size (https://en.wikipedia.org/wiki/Crossbar,_Inc.). In later\nyears its marketing added ReRAM for \"in-memory computing\" / AI and secure keys, but the\ncore business is memory IP licensing.",
   "technical_approach": "- Device: filamentary ReRAM — the 2013 prototype used \"a crosspoint array to stack\n  multiple silver-ion-based memory layers\" (Ag/a-Si metal-filament cells from Wei Lu's\n  UMich work), three layers targeting 1TB on a 200mm² chip; CMOS-compatible; 1T1R\n  configurations for embedded use (https://en.wikipedia.org/wiki/Crossbar,_Inc.).\n- Business model: **licensing ReRAM IP** to foundries/OEM SoC designers (consumer,\n  enterprise, mobile, industrial, IoT); notable licensing/foundry engagements incl. SMIC\n  (40nm) and Microsemi in the mid-2010s (widely reported trade press; see RRAM-info\n  company page https://www.rram-info.com/companies/crossbar). More recent positioning\n  emphasizes ReRAM-based physical-unclonable-function secure keys.\n- Compute: **digital memory only** — no analog IMC product, no on-chip learning; adjacent\n  to neuromorphics only through the resistive device itself.",
   "funding_rounds": [
    {
     "date": "2012",
     "round": "early rounds (thru Series C era)",
     "amount": "$25M",
     "investors": "Artiman Ventures, Kleiner Perkins, Northern Light VC, UMich MINTS (UMich's first startup investment)",
     "source": "https://en.wikipedia.org/wiki/Crossbar,_Inc. ; http://www.eecs.umich.edu/eecs/about/articles/2012/crossbar_wei_lu_MINTS_funding.html"
    },
    {
     "date": "2015-09",
     "round": "growth round",
     "amount": "~$35M",
     "investors": "investors in China and Hong Kong",
     "source": "https://en.wikipedia.org/wiki/Crossbar,_Inc."
    },
    {
     "date": "2023-11",
     "round": "late round",
     "amount": "$5.17M",
     "investors": "undisclosed",
     "source": "https://www.cbinsights.com/company/crossbar-vs-supermem (CB Insights profile data via search)"
    }
   ],
   "funding_notes": "**Conflicting totals — report all:** Wikipedia ≈ $60M; CB Insights $120.38M total;\nTracxn $136M over 7 rounds from 15 investors\n(https://en.wikipedia.org/wiki/Crossbar,_Inc.;\nhttps://tracxn.com/d/companies/crossbar/__yUXElZhOmC6ngUZtPhTlbAe5pt390Wq00GyNXlRjVIQ).\nThe $120–136M figures include rounds Wikipedia does not itemize.",
   "trajectory": "Founded 2010 by George Minassian, Hagop Nazarian, and UMich professor **Wei Lu** (who later\nalso co-founded MemryX), with RRAM patents licensed from the University of Michigan\n(https://en.wikipedia.org/wiki/Crossbar,_Inc.). High-profile 2013 announcement of\nterabyte-scale 3D ReRAM; multiple licensing deals mid-decade; never reached mass-market\nstandalone memory; shifted emphasis to embedded/secure-key niches. Status mid-2026:\nprivate, ~20+ employees, no major public announcements since the early 2020s — a\n\"pre-wave incumbent\" that predates and outlived the 2015+ neuromorphic startup wave\nwithout joining the AI-compute race directly.",
   "relevance": "Baseline for how hard ReRAM commercialization is even with tier-1 VCs and a decade of\nrunway: strong device IP, licensing model, yet no in-memory-compute product. Useful\ncontrast to both Knowm (discrete devices + learning theory) and TetraMem (analog IMC on\nthe same device class).",
   "news": [],
   "sources": [
    {
     "title": "Wikipedia — Crossbar, Inc.",
     "url": "https://en.wikipedia.org/wiki/Crossbar,_Inc."
    },
    {
     "title": "UMich EECS — MINTS investment in Crossbar (2012)",
     "url": "http://www.eecs.umich.edu/eecs/about/articles/2012/crossbar_wei_lu_MINTS_funding.html"
    },
    {
     "title": "Tracxn — Crossbar profile ($136M/7 rounds)",
     "url": "https://tracxn.com/d/companies/crossbar/__yUXElZhOmC6ngUZtPhTlbAe5pt390Wq00GyNXlRjVIQ"
    },
    {
     "title": "CB Insights — Crossbar financials ($120.38M; $5.17M 2023-11)",
     "url": "https://www.cbinsights.com/compare/crossbar-vs-supermem"
    },
    {
     "title": "RRAM-info — Crossbar company page",
     "url": "https://www.rram-info.com/companies/crossbar"
    }
   ]
  },
  {
   "slug": "cyberswarm",
   "name": "CyberSwarm, Inc.",
   "founded": 2017,
   "hq": "Ploiești, Romania (US-incorporated entity \"CyberSwarm, Inc.\")",
   "category": "emerging-device-imc",
   "status": "active",
   "origin": "serial founder (Mihai Raneti — cybersecurity background; collaborates with Romanian research institutions)",
   "total_funding_usd_m": 50,
   "last_round": "Series A, $50M, 2026-05 (Falanga Invest Perton lead)",
   "key_investors": [
    "Falanga Invest Perton (Series A lead)",
    "Draper Associates (since 2018)",
    "Capital Factory (per PitchBook)",
    "AVANT investiční společnost (per PitchBook)"
   ],
   "website": "https://www.cyber-swarm.net",
   "pitch": "CyberSwarm's technology page opens with Moore's Law running out and the von Neumann split:\n\"the design of these computers is based on what's called the Von Neumann architecture… data\nstorage is on one side of the room and the processing unit… on the other. There's a constant\nback and forth, which takes time and energy.\" Its answer is a memristor-based \"brain-on-a-chip\"\nthat \"integrates these elements, just like our brains integrate memory and processing\", using\n\"analog methods instead of binary\" and \"self-learning\" that \"assimilate[s] algorithms directly at\nthe hardware level\" (https://www.cyber-swarm.net/technology/ ; https://www.cyber-swarm.net/).\nThe Series A release frames it as \"a new computational layer between traditional computing and\nquantum — designed for problems that neither solves efficiently\" (Raneti, 2026-05-15,\nhttps://start-up.ro/cyberswarm-raises-50-million-series-a-to-scale-neuromorphic-computing-hardware/).",
   "technical_approach": "- **Substrate:** memristor crossbars used as analog artificial synapses — \"hardware designed to\n  process and store information simultaneously, similar to biological neurons\" (start-up.ro).\n  The company shows its own \"Memristor matrix created at CyberSwarm, Inc.\" and a \"Neuromorphic\n  Crossbar\" photo (https://www.cyber-swarm.net/blog/the-future-of-artificial-intelligence-neuromorphic/ ;\n  https://www.cyber-swarm.net/technology/). Draper describes it as \"developing an artificial\n  nervous system using memristor technology\" (https://www.draper.vc/portfolio/cyberswarm).\n  Device chemistry, process, array size and any measured device data are **not public**.\n- **Learning locus:** on-chip / in-hardware learning is the stated goal (\"self-learning devices\",\n  \"learns and evolves\"); no published demonstration of on-chip learning at scale.\n- **First product — Myrmex (2021 product sheet):** a security-first edge SoC: quad-core Arm\n  Cortex-A53 + Arm CryptoCell-312 + Arm Ethos-U65 NPU, with an added \"analog neuromorphic\"\n  block of **48 neurons × 100 analog states (48,000 parameters)** fed by 6×12-bit DACs and\n  6×16-bit ADCs, used for \"neuromorphic encryption and key distribution\" and a \"keyless\n  encryption scheme\" (product sheet hosted at https://www.cbinsights.com/company/cyberfog).\n  I.e., the shipped/demoed neuromorphic content is small and bolted onto a conventional\n  digital SoC; the 2026 raise is to \"scale\" the memristor architecture into a\n  \"neuromorphic compute layer for autonomous systems\" (start-up.ro).\n- **Wedge:** cybersecurity first (the 2017–2021 positioning: \"un-hackable by digital means\",\n  dual-channel analog communication; https://thesiliconreview.com/magazine/profile/cyber-swarm-is-securing-both-the-device-and-communication),\n  now broadened to automotive, aerospace, defense, healthcare and robotics deployments\n  (Series A release).",
   "funding_rounds": [
    {
     "date": "2018",
     "round": "Early (seed)",
     "amount": "undisclosed",
     "investors": "Draper Associates (\"backed by Draper Associates since 2018\")",
     "source": "https://start-up.ro/cyberswarm-raises-50-million-series-a-to-scale-neuromorphic-computing-hardware/ ; https://www.draper.vc/portfolio/cyberswarm"
    },
    {
     "date": "2026-05 (announced 05-15)",
     "round": "Series A",
     "amount": "**$50M**",
     "investors": "Falanga Invest Perton (lead; CEO Ian Imrisek)",
     "source": "https://start-up.ro/cyberswarm-raises-50-million-series-a-to-scale-neuromorphic-computing-hardware/ ; https://www.romaniajournal.ro/business/romanian-ai-chip-startup-cyberswarm-raises-50-million/"
    }
   ],
   "funding_notes": "Total disclosed: **$50M** (Series A). Trackers put cumulative at $51.58M (CB Insights,\nhttps://www.cbinsights.com/company/cyberfog) to $54.6M (PitchBook, listing AVANT investiční\nspolečnost, Capital Factory and Draper Associates; https://pitchbook.com/profiles/company/168252-40),\nimplying ~$1.6–4.6M of undisclosed pre-Series-A capital. Vestbee notes the $50M is \"recorded as\na Series A\" although it is seed-stage-sized for CEE (https://vestbee.com/insights/articles/cee-venture-funding-trends).\nNo valuation reported.",
   "trajectory": "Founded 2017 by Mihai Raneti in Ploiești, Romania, initially as a memristor-based\ncybersecurity company (The Silicon Review profile; Romania Journal); Draper Associates invested\nin 2018. Product sheet for the Myrmex secure SoC dates from 2021. Re-positioned by 2025 as an\n\"AI-enabling neuromorphic engineering company\"; on 2026-05-15 announced a $50M Series A led by\nCzech investment firm Falanga Invest Perton to fund R&D, engineering hiring, industry\npartnerships and \"transition toward commercial deployment\". Status: active, pre-commercial for\nthe neuromorphic compute layer. The census missed it in July 2026 (surfaced in the 2026-08-23\nwatchlist sweep; profiled 2026-09-22).",
   "relevance": "CyberSwarm is, in its own words, building memristor crossbars as analog synapses with learning\nin the hardware — the same substrate-and-learning-locus claim Knowm made for AHaH/kT-RAM. It is\nalso the only European memristor-neuromorphic startup with a $50M round. The public technical\nrecord is thin (a 48-neuron analog block on an Arm SoC, no device data), so treat the $50M as a\nbet on the claim, not evidence the claim has been demonstrated.",
   "news": [],
   "sources": [
    {
     "title": "CyberSwarm — Technology",
     "url": "https://www.cyber-swarm.net/technology/"
    },
    {
     "title": "CyberSwarm — home",
     "url": "https://www.cyber-swarm.net/"
    },
    {
     "title": "CyberSwarm blog: The future of AI is neuromorphic computing (memristor matrix photo)",
     "url": "https://www.cyber-swarm.net/blog/the-future-of-artificial-intelligence-neuromorphic/"
    },
    {
     "title": "start-up.ro: CyberSwarm raises $50M Series A (2026-05-15)",
     "url": "https://start-up.ro/cyberswarm-raises-50-million-series-a-to-scale-neuromorphic-computing-hardware/"
    },
    {
     "title": "Romania Journal: Romanian AI chip startup CyberSwarm raises $50 million",
     "url": "https://www.romaniajournal.ro/business/romanian-ai-chip-startup-cyberswarm-raises-50-million/"
    },
    {
     "title": "Draper Associates portfolio page",
     "url": "https://www.draper.vc/portfolio/cyberswarm"
    },
    {
     "title": "Myrmex product sheet (2021), via CB Insights",
     "url": "https://www.cbinsights.com/company/cyberfog"
    },
    {
     "title": "PitchBook profile ($54.6M total)",
     "url": "https://pitchbook.com/profiles/company/168252-40"
    },
    {
     "title": "The Silicon Review: Cyber Swarm profile / Raneti interview",
     "url": "https://thesiliconreview.com/magazine/profile/cyber-swarm-is-securing-both-the-device-and-communication"
    },
    {
     "title": "Vestbee: CEE venture funding 2026",
     "url": "https://vestbee.com/insights/articles/cee-venture-funding-trends"
    }
   ]
  },
  {
   "slug": "encharge-ai",
   "name": "EnCharge AI",
   "founded": 2022,
   "hq": "Santa Clara, CA, USA (R&D roots in Princeton, NJ)",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (Naveen Verma's lab, Princeton University) + corporate alumni (IBM)",
   "total_funding_usd_m": 144,
   "last_round": "Series B, $100M+, 2025-02",
   "key_investors": [
    "Tiger Global",
    "Anzu Partners",
    "Samsung Ventures",
    "RTX Ventures",
    "VentureTech Alliance",
    "HH-CTBC",
    "Constellation Energy",
    "Foxconn/Hon Hai",
    "AlleyCorp",
    "Scout Ventures"
   ],
   "website": "https://www.enchargeai.com",
   "pitch": "EnCharge frames AI's cost/energy crisis as a data-movement problem that digital\narchitectures cannot escape: its \"analog in-memory computing\" does the\nmultiply-accumulate where the weights live, claiming \"~20x better performance per\nwatt\" than leading digital alternatives and \"orders-of-magnitude higher compute\nefficiency and density compared to today's best-in-class solutions.\"\nSources: https://www.businesswire.com/news/home/20250529108055/en/EnCharge-AI-Announces-EN100-First-of-its-Kind-AI-Accelerator-for-On-Device-Computing ;\nhttps://www.businesswire.com/news/home/20250213156276/en/EnCharge-AI-Closes-Oversubscribed-$100M-Series-B-Funding-Round-Led-by-Tiger-Global-to-Produce-Transformative-AI-Accelerator-Solutions-with-Unprecedented-Efficiency-for-Client-Computing",
   "technical_approach": "- Device: **switched metal-layer capacitors** (charge-domain analog compute) rather\n  than current-domain resistive devices — precise, geometry-defined capacitors make\n  the analog MAC robust to noise/temperature, the key differentiator EnCharge claims\n  over flash/memristor IMC. Built on standard CMOS (TSMC); based on Verma-lab research\n  2015–2022 funded by NSF and DARPA\n  (https://techcrunch.com/2022/12/14/encharge-ai-emerges-from-stealth-with-21-7m-to-develop-ai-accelerator-hardware/ ;\n  https://www.fastcompany.com/91278505/encharge-ai-banked-100-million-for-its-energy-slashing-analog-chips).\n- Inference-only accelerator (client/edge focus), full software stack.\n- Named product: **EN100** (announced 2025-05-29) — M.2 card for laptops, 200+ TOPS at\n  8.25W; PCIe workstation card with 4 NPUs at ~1 PetaOPS\n  (https://www.businesswire.com/news/home/20250529108055/en/EnCharge-AI-Announces-EN100-First-of-its-Kind-AI-Accelerator-for-On-Device-Computing).",
   "funding_rounds": [
    {
     "date": "2022-12",
     "round": "Series A",
     "amount": "$21.7M",
     "investors": "Anzu Partners (lead); AlleyCorp, Scout Ventures, Silicon Catalyst Angels, Schams Ventures, E14 Fund, Alumni Ventures",
     "source": "https://www.businesswire.com/news/home/20221214005242/en/EnCharge-AI-Launches-With-$21.7M-Series-A-to-Enable-Edge-AI-at-Scale"
    },
    {
     "date": "2023-12",
     "round": "Series A extension",
     "amount": "$22.6M",
     "investors": "VentureTech Alliance, RTX Ventures, ACVC Partners; total then $45M",
     "source": "https://www.prnewswire.com/news-releases/encharge-ai-closes-additional-22-6m-from-venturetech-alliance-rtx-ventures-and-acvc-partners-to-unlock-ai-compute-302005460.html"
    },
    {
     "date": "2024-03",
     "round": "DARPA OPTIMA grant (with Princeton)",
     "amount": "$18.6M",
     "investors": "DARPA (part of $78M OPTIMA program)",
     "source": "https://www.hpcwire.com/off-the-wire/darpa-awards-encharge-ai-and-princeton-18-6m-to-pioneer-next-gen-in-memory-ai-processors/"
    },
    {
     "date": "2025-02",
     "round": "Series B (oversubscribed)",
     "amount": "$100M+",
     "investors": "Tiger Global (lead); Samsung Ventures, HH-CTBC; RTX, Constellation Energy, Foxconn also reported",
     "source": "https://www.businesswire.com/news/home/20250213156276/en/EnCharge-AI-Closes-Oversubscribed-$100M-Series-B-Funding-Round-Led-by-Tiger-Global-to-Produce-Transformative-AI-Accelerator-Solutions-with-Unprecedented-Efficiency-for-Client-Computing"
    }
   ],
   "funding_notes": "Total equity raised: **>$144M** after Series B (BusinessWire, above), plus the $18.6M\nDARPA award.",
   "trajectory": "Co-founded March 2022 by **Naveen Verma** (CEO; Princeton EE professor since 2009, whose\nlab developed the charge-domain IMC), **Kailash Gopalakrishnan** (chief product officer,\nex-IBM AI-hardware leader) and **Echere Iroaga** (COO)\n(https://techcrunch.com/2022/12/14/encharge-ai-emerges-from-stealth-with-21-7m-to-develop-ai-accelerator-hardware/).\nScaled quickly: DARPA OPTIMA award 2024, $100M+ Series B Feb 2025, EN100 product launch\nMay 2025 with an early-access program for laptop/workstation OEMs.\n**Status mid-2026: active**, arguably the current momentum leader in analog IMC for\nclient computing.",
   "relevance": "EnCharge's bet is that the analog-IMC idea fails on device physics unless the analog\nelement is a precise capacitor rather than a variable resistive device — a direct\ncounterpoint to memristor-based IMC. Like the rest of the category it is\ninference-only; the adaptivity lives in software, not the substrate.",
   "news": [],
   "sources": [
    {
     "title": "Series A PR",
     "url": "https://www.businesswire.com/news/home/20221214005242/en/EnCharge-AI-Launches-With-$21.7M-Series-A-to-Enable-Edge-AI-at-Scale"
    },
    {
     "title": "$22.6M PR",
     "url": "https://www.prnewswire.com/news-releases/encharge-ai-closes-additional-22-6m-from-venturetech-alliance-rtx-ventures-and-acvc-partners-to-unlock-ai-compute-302005460.html"
    },
    {
     "title": "DARPA OPTIMA award",
     "url": "https://www.hpcwire.com/off-the-wire/darpa-awards-encharge-ai-and-princeton-18-6m-to-pioneer-next-gen-in-memory-ai-processors/"
    },
    {
     "title": "Series B PR",
     "url": "https://www.businesswire.com/news/home/20250213156276/en/EnCharge-AI-Closes-Oversubscribed-$100M-Series-B-Funding-Round-Led-by-Tiger-Global-to-Produce-Transformative-AI-Accelerator-Solutions-with-Unprecedented-Efficiency-for-Client-Computing"
    },
    {
     "title": "EN100 launch PR",
     "url": "https://www.businesswire.com/news/home/20250529108055/en/EnCharge-AI-Announces-EN100-First-of-its-Kind-AI-Accelerator-for-On-Device-Computing"
    },
    {
     "title": "TechCrunch stealth exit",
     "url": "https://techcrunch.com/2022/12/14/encharge-ai-emerges-from-stealth-with-21-7m-to-develop-ai-accelerator-hardware/"
    }
   ]
  },
  {
   "slug": "eta-compute",
   "name": "Eta Compute (now ModelCat)",
   "founded": 2015,
   "hq": "Westlake Village, California, USA",
   "category": "digital-snn",
   "status": "pivoted",
   "origin": "serial founders / corporate alumni (Inphi — Gopal Raghavan and Tim Semones were Inphi co-founders, Paul Washkewicz an early Inphi executive)",
   "total_funding_usd_m": 31.5,
   "last_round": "Series C, $12.5M, 2020-12",
   "key_investors": [
    "Synaptics",
    "Fathom Group",
    "Lockheed Martin Ventures",
    "Steve Sanghi"
   ],
   "website": "https://www.modelcat.ai",
   "pitch": "Eta Compute framed always-on edge AI as fundamentally power-bound: conventional\nsynchronous processors burn too much energy for sensor nodes that must run on a\ncoin cell or harvested power. Its answer was event-driven, near-/sub-threshold\nasynchronous logic plus spiking neural networks, cutting the multiply-accumulate\ncost of inference by keeping voltage low and computing only when events arrive.\n(Original framing archived via IEEE Spectrum, \"Eta Compute Debuts Spiking Neural\nNetwork Chip for Edge AI,\" and eenewseurope coverage of its asynchronous Cortex-M3.)",
   "technical_approach": "- Delay-Insensitive Asynchronous Logic (DIAL): clockless Arm Cortex-M3 + NXP\n  CoolFlux DSP cores run at voltages as low as ~0.2 V (vs. the usual ~0.9 V).\n- ECM3532: the production \"neural sensor processor\" SoC, an ultra-low-power AI\n  chip for edge devices; paired with the TENSAI Flow / TENSAI software stack.\n- Positioned around spiking neural networks where weights reduce to 1s/0s so the\n  dominant MAC operation becomes mostly addition — a digital-SNN, inference-first\n  design (on-device inference, not on-chip learning).",
   "funding_rounds": [
    {
     "date": "2018-ish",
     "round": "Early / Series A-B (multiple)",
     "amount": "~$19M cumulative reported by 2019",
     "investors": "Undisclosed; incl. angel Steve Sanghi (Microchip)",
     "source": "EE Times / Electronic Design"
    },
    {
     "date": "2020-12",
     "round": "Series C",
     "amount": "$12.5M",
     "investors": "Synaptics (lead), existing investors",
     "source": "BusinessWire, 2020-12-15"
    },
    {
     "date": "—",
     "round": "Total reported",
     "amount": "~$31.5M",
     "investors": "—",
     "source": "BusinessWire (Series C release)"
    }
   ],
   "funding_notes": "The originally-briefed \"~$19M raised\" figure matches the pre-2020 cumulative;\nafter the Dec-2020 Series C the total reported was ~$31.5M. Both are recorded here\nto resolve the conflict.",
   "trajectory": "Founded 2015 by Gopal Raghavan, Tim Semones, and Paul Washkewicz (semiconductor /\nmixed-signal veterans). In December 2020, concurrent with the Synaptics-led Series\nC, Eta Compute exited the silicon roadmap: it announced no further generations of\nEta Compute chips beyond the ECM3532 and exclusively licensed its Tensai Flow\nsoftware to Synaptics (for Synaptics' Katana SoC), repositioning as an edge-AI\nsoftware/systems company rather than a chip designer. It continued down the\nsoftware path and in August 2025 rebranded as **ModelCat**, an \"AI-first\" company\nthat automates ML model creation (\"AI-in-the-loop\" / agentic model building) — e.g.\nNXP's eIQ Model Creator (Sept 2025) and a Silicon Labs partnership. The\nneuromorphic/SNN silicon ambition is effectively retired; the surviving business is\nmodel-tooling software. Status mid-2026: active as ModelCat, software-only.\n\n**Founders (sweep 2026-09-22):** Gopal Raghavan, PhD, and Tim Semones — both co-founders of Inphi Corporation — and Paul Washkewicz, an early Inphi executive (company launch release). The SNN work came later with CTO Narayan Srinivasa (2017–2019, ex-Intel Labs chief scientist), not from the founders' prior employer (https://www.globenewswire.com/news-release/2017/04/11/958783/0/en/eta-compute-debuts-with-world-s-lowest-power-microcontroller-ip-targeting-energy-harvesting-segment.html ; https://loop.frontiersin.org/people/63753/bio).",
   "relevance": "A clean first-wave cautionary tale for the \"digital SNN at the edge\" thesis: Eta\nCompute built real near-threshold asynchronous silicon (ECM3532) but concluded the\ndefensible value was in software, exiting custom chips within ~5 years. Unlike\nKnowm's memristor/AHaH bet on analog on-chip learning in the device physics, Eta's\napproach was digital and inference-first, and even that proved hard to sustain as a\nmerchant-silicon business.",
   "news": [],
   "sources": [
    {
     "title": "Eta Compute Debuts Spiking Neural Network Chip for Edge AI — IEEE Spectrum",
     "url": "https://spectrum.ieee.org/eta-compute-debuts-spiking-neural-network-chip-for-edge-ai"
    },
    {
     "title": "Neuromorphic computing runs on asynchronous Cortex-M3 — eeNews Europe",
     "url": "https://www.eenewseurope.com/en/neuromorphic-computing-runs-on-asynchronous-cortex-m3/"
    },
    {
     "title": "Eta Compute Closes $12.5M Series C Led By Synaptics — BusinessWire, 2020-12-15",
     "url": "https://www.businesswire.com/news/home/20201215005404/en/Eta-Compute-Closes-%2412.5-Million-Series-C-Funding-Led-By-Synaptics"
    },
    {
     "title": "Eta Compute Pivots Away From Silicon, Signs Deal with Synaptics — EE Times",
     "url": "https://www.eetimes.com/eta-compute-pivots-away-from-silicon-signs-deal-with-synaptics/"
    },
    {
     "title": "Eta Compute is Changing its Name to ModelCat — PRNewswire, Aug 2025",
     "url": "https://www.prnewswire.com/news-releases/eta-compute-is-changing-its-name-to-modelcat--launching-a-bold-new-era-in-ai-development-302541000.html"
    },
    {
     "title": "ModelCat (Formerly Eta Compute) — NeuromorphicCore.ai",
     "url": "https://neuromorphiccore.ai/insights/eta-compute/"
    }
   ]
  },
  {
   "slug": "extropic",
   "name": "Extropic",
   "founded": 2022,
   "hq": "Austin, TX, USA",
   "category": "probabilistic-thermodynamic",
   "status": "active",
   "origin": "corporate alumni (Alphabet X / Google Quantum AI)",
   "total_funding_usd_m": 14.1,
   "last_round": "Seed, $14.1M, 2023-12",
   "key_investors": [
    "Kindred Ventures",
    "Valor Equity Partners",
    "OSS Capital",
    "HOF Capital",
    "Buckley Ventures"
   ],
   "website": "https://extropic.ai",
   "pitch": "Extropic argues AI's scaling is hitting an energy wall because digital deterministic\nhardware fights thermal noise instead of using it: the company builds \"thermodynamic\nsampling units (TSUs) — hardware that accelerates probabilistic inference by leveraging\nthermodynamic fluctuations in electronics,\" positioning noise itself as the compute\nresource for generative (sampling-based) AI. Sources: https://extropic.ai/,\nhttps://siliconangle.com/2023/12/04/extropic-raises-14-1m-build-physics-based-computing-hardware-generative-ai/",
   "technical_approach": "- **Substrate:** standard CMOS transistor circuits operated stochastically — \"probabilistic\n  circuits\" that sample directly from probability distributions. Circuit primitives include\n  the pbit (Bernoulli sampler) plus pdit/pmode/pmog samplers (https://extropic.ai/hardware).\n  Early exploration included superconducting Josephson-junction devices, but the productized\n  path is room-temperature CMOS.\n- **What the physics does:** thermal fluctuations drive sampling from energy-based models;\n  a TSU is an array of coupled pbits forming a physical probabilistic graphical model.\n  Inference = physical relaxation/sampling, not matrix multiplication.\n- **Training vs inference:** chips perform sampling (generative inference); model\n  design/training is done in software — the open-source JAX library **THRML** (released\n  Oct 2025) simulates TSUs and energy-based models (https://github.com/extropic-ai/thrml).\n  No on-chip learning claim.\n- **Algorithm claim:** the \"Denoising Thermodynamic Model\" (DTM), a diffusion-like\n  generative algorithm; Extropic claims *simulated* DTMs on TSUs could be ~10,000x more\n  energy-efficient than comparable algorithms on GPUs — a projection, not a measured\n  system-level benchmark (https://www.vktr.com/ai-news/extropic-claims-10000x-energy-savings-with-new-probabilistic-ai-chip/).\n- **Products:** X0 proof-of-concept chip (Q1 2025; \"dozens\" of probabilistic circuits);\n  XTR-0 development platform (Q3 2025 to select partners; FPGA host + two X0 daughter\n  chips); Z1 production-scale chip planned for early access 2026 with ~250,000 pbits per\n  chip (https://extropic.ai/hardware, https://winbuzzer.com/2025/10/30/extropic-unveils-thermodynamic-ai-chips-to-combat-the-industrys-energy-crisis-xcxwbn/).",
   "funding_rounds": [
    {
     "date": "2023-12",
     "round": "Seed",
     "amount": "$14.1M",
     "investors": "Kindred Ventures (Steve Jang) lead; Buckley Ventures, HOF Capital, Julian Capital, Marque VC, OSS Capital, Valor Equity Partners, Weekend Fund; angels incl. Aidan Gomez, Amjad Masad, Arash Ferdowsi, Balaji Srinivasan, Garry Tan, Naval Ravikant, Scott Belsky, Tobias Lütke",
     "source": "https://thequantuminsider.com/2023/12/05/extropic-announces-14-1-million-seed-round-building-entropy-computer-for-generative-ai/"
    }
   ],
   "funding_notes": "Total disclosed: **$14.1M** (Tracxn also lists $14.1M over 1 round as of 2026:\nhttps://tracxn.com/d/companies/extropic/__vwWnk7wGdSIGd5J9vh2-o5dOzD-nYCUKxeOVSzx0msc/funding-and-investors).\nNo Series A had been announced as of 2026-07.",
   "trajectory": "Founded 2022 by Guillaume Verdon (ex-Alphabet X / Google Quantum AI tech lead; publicly\nknown as \"Beff Jezos\" of the effective-accelerationism movement) with CTO Trevor McCourt.\nEmerged from stealth Dec 2023 with the seed round. Public hardware unveiling\n2025-10-29/30: X0 silicon, XTR-0 dev kits shipping to early partners (reported to include\nfrontier AI labs and government users), THRML open-sourced, DTM paper. As of mid-2026 the\nZ1 production chip is in the \"early access 2026\" stage; the 10,000x figure remains a\nsimulation-based claim awaiting silicon validation. HQ Austin, TX\n(https://www.zoominfo.com/c/extropic-corp/5000009652).",
   "relevance": "Extropic is the highest-profile revival of the idea Knowm branded \"Thermodynamic RAM\"\n(~2014): treating thermal noise/dissipation as a computational resource. Differences are\nfactual and significant — Extropic uses stochastic CMOS sampling circuits for generative\ninference with no in-device adaptation, whereas kT-RAM/AHaH centered on memristive analog\nadaptation (on-chip learning). Neither approach has demonstrated production-scale silicon.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Extropic signed a **$75M letter of intent with the U.S. Department of Commerce** (2026-07-30) to scale and onshore thermodynamic computing. LOI, not a closed contract — report as such. (https://finance.yahoo.com/technology/ai/articles/extropic-signs-75-million-letter-163200858.html ; https://interestingengineering.com/ai-robotics/thermodynamic-smart-chips-extropic-deal)"
    },
    {
     "date": "2026-09-22",
     "text": "**\"From One to One Billion\" summer update (2026-08-04):** the **Z1 chip is taped out** (not shipped — the August watchlist sweep's \"shipped Z1\" was premature); planned form factors are a Z1 \"thermo compute stick\" (2 chips, >500k pbits, M.2) and Z1 PCIe accelerator cards (16 chips, >4M pbits), with cards/sticks/clusters slated for **early access in 2027** and a \"billion-pbit\" cluster targeted for 2027. Software: Torx (stochastic differentiable programming framework), a Thermalizers compiler preview, and an early-access GPU-simulator API for Torx/THRML. Correction to the watchlist sweep: the $75M letter of intent is with the Department of **Commerce**, not DOE. (https://extropic.ai/writing/from-one-to-one-billion)"
    },
    {
     "date": "2026-09-22",
     "text": "**Z1T (2026-09-04):** research post proposing sparse Transformer-like models designed to run on Z1's probabilistic hardware — Extropic explicitly pivoting its algorithm work toward the dominant Transformer workload rather than only energy-based/diffusion-style models. Simulation-stage. (https://extropic.ai/writing/z1t)"
    },
    {
     "date": "2026-10-27",
     "text": "**Prime Intellect case study (2026-10-03):** Extropic RL-post-trained an open Qwen3.6-35B-A3B model as a 'thermodynamic ML research agent' (≈50 coding tasks reproducing classic thermodynamic-ML experiments; 100 GRPO steps, ~25 h; held-out reward 0.127→0.361). Notable admission: TSUs 'currently run a family of mid-2000s models, so few modern algorithms exist for them'. No Z1 hardware update. (https://www.primeintellect.ai/case-study/extropic)"
    }
   ],
   "sources": [
    {
     "title": "Extropic — Hardware",
     "url": "https://extropic.ai/hardware"
    },
    {
     "title": "Extropic unveils thermodynamic AI chips (WinBuzzer)",
     "url": "https://winbuzzer.com/2025/10/30/extropic-unveils-thermodynamic-ai-chips-to-combat-the-industrys-energy-crisis-xcxwbn/"
    },
    {
     "title": "Extropic announces $14.1M seed (Quantum Insider)",
     "url": "https://thequantuminsider.com/2023/12/05/extropic-announces-14-1-million-seed-round-building-entropy-computer-for-generative-ai/"
    },
    {
     "title": "SiliconANGLE seed coverage",
     "url": "https://siliconangle.com/2023/12/04/extropic-raises-14-1m-build-physics-based-computing-hardware-generative-ai/"
    },
    {
     "title": "10,000x claim coverage (VKTR)",
     "url": "https://www.vktr.com/ai-news/extropic-claims-10000x-energy-savings-with-new-probabilistic-ai-chip/"
    },
    {
     "title": "THRML on GitHub",
     "url": "https://github.com/extropic-ai/thrml"
    },
    {
     "title": "Tracxn funding profile",
     "url": "https://tracxn.com/d/companies/extropic/__vwWnk7wGdSIGd5J9vh2-o5dOzD-nYCUKxeOVSzx0msc/funding-and-investors"
    }
   ]
  },
  {
   "slug": "femtosense",
   "name": "femtoAI (fka Femtosense)",
   "founded": 2018,
   "hq": "San Bruno / Silicon Valley, CA, USA",
   "category": "digital-snn",
   "status": "active",
   "origin": "academic spinoff (Stanford Brains in Silicon Lab — Kwabena Boahen's group; founders Sam Fok, Alex Neckar)",
   "total_funding_usd_m": "8 disclosed (Series A); plus a reported €20M 2025 round and an undisclosed late-2025 round — see conflicts",
   "last_round": "undisclosed round, closed 2025 (announced 2025-12-30)",
   "key_investors": [
    "Fine Structure Ventures",
    "Quest Venture Partners",
    "Ubiquity Ventures"
   ],
   "website": "https://femto.ai",
   "pitch": "femtoAI targets \"the critical need for high performance, low latency, and energy efficiency\nwithin an extremely small footprint,\" enabling \"advanced AI capabilities on the tiniest\ndevices\" — exploiting brain-like **sparsity** (only compute on non-zero, only store what\nmatters) rather than brute-force dense matrix math\n(https://femto.ai/press_releases/2025-12-30/, https://www.eejournal.com/article/is-the-future-of-ai-sparse/).",
   "technical_approach": "- **Digital sparse-processing accelerator** (\"Sparse Processing Unit\", SPU): dual sparsity\n  (weights + activations) descended from Stanford Brains in Silicon SNN research; positioned\n  by neuromorphic trackers as a true event-driven SNN-capable processor at sub-milliwatt power\n  (https://neuromorphiccore.ai/insights/femtosense/). Inference-only; models trained off-chip.\n- **SPU-001**: first production chip; shipping in volume — 100,000+ units shipped in 2025;\n  claims \"100x lower power consumption, 10x smaller silicon footprint\" vs alternatives\n  (https://femto.ai/press_releases/2025-12-30/).\n- Built with Arm Flexible Access for Startups (https://newsroom.arm.com/blog/flexible-access-femtosense).\n- Flagship market: AI hearing — NewSound OTC hearing aids run \"Clara AI\" all-day on SPU-001 at\n  sub-mW (https://femto.ai/press_releases/2025-03-26/). SPU-001 won a CES 2026 award\n  (https://femto.ai/news/, https://sg.finance.yahoo.com/news/femtoai-unlocks-5x-growth-full-155500048.html).",
   "funding_rounds": [
    {
     "date": "2022-09",
     "round": "Series A",
     "amount": "$8M",
     "investors": "led by Fine Structure Ventures (Fidelity-affiliated); TechCrunch also lists prior backers",
     "source": "https://www.businesswire.com/news/home/20220928005770/en/Edge-AI-Innovator-Femtosense-Raises-%248-Million-Series-A-Funding ; https://techcrunch.com/2022/09/28/femtosense-raises-capital-to-power-the-ai-inside-of-consumer-electronics/"
    },
    {
     "date": "2025-06",
     "round": "Round reported alongside femtoAI rebrand",
     "amount": "\"€20 million injection\" (secondary report; no primary PR found)",
     "investors": "undisclosed",
     "source": "https://www.linkedin.com/posts/femtoai_from-femtosense-to-femtoai-activity-7344058486546137090-7lzA"
    },
    {
     "date": "2025 (announced 2025-12-30)",
     "round": "New funding round",
     "amount": "undisclosed",
     "investors": "undisclosed",
     "source": "https://femto.ai/press_releases/2025-12-30/"
    }
   ],
   "funding_notes": "**Total:** only the $8M Series A is firmly disclosed with amount+lead; the €20M figure appears\nonly in a reposted article snippet — treat as unconfirmed. Company reports \"5x growth\" into\n2026 (https://sg.finance.yahoo.com/news/femtoai-unlocks-5x-growth-full-155500048.html).",
   "trajectory": "Founded 2018 by researchers from Stanford's Brains in Silicon Lab (Sam Fok, CEO; Alex Neckar)\n— the lab behind Neurogrid/Braindrop (https://femto.ai/about-us/,\nhttps://newsroom.arm.com/blog/flexible-access-femtosense). Trajectory: SNN research → sparse\ndigital accelerator → volume shipments in hearing/voice devices. Rebranded **femtoAI** in\nJune 2025 (https://www.linkedin.com/posts/femtoai_from-femtosense-to-femtoai-activity-7344058486546137090-7lzA).\nAs of mid-2026: **active**, one of the few neuromorphic-lineage startups with six-figure unit\nshipments in consumer products; NVIDIA Inception member; exec team expanded (Chief AI\nOfficer, CBO) Dec 2025 (https://femto.ai/press_releases/2025-12-30/).",
   "relevance": "Femtosense shows the pragmatic endpoint of SNN research once commercialized: the \"neuromorphic\"\npart that survived is sparsity exploitation in digital CMOS, with no adaptive synapses or\non-chip learning — the opposite bet from Knowm's physical-learning memristor approach, and\ncurrently the one with real consumer-product volume.",
   "news": [
    {
     "date": "2026-10-27",
     "text": "**Developer community opened (2026-10-06)** after a beta: access to the SPU eval kit, compressed models, compiler and 'Sparsity Studio'; '10x memory & power savings' claim. No funding figure. (https://finance.yahoo.com/technology/ai/articles/femtoai-opens-developer-community-build-153200962.html)"
    }
   ],
   "sources": [
    {
     "title": "femtoAI about us",
     "url": "https://femto.ai/about-us/"
    },
    {
     "title": "femtoAI growth PR (Dec 30, 2025)",
     "url": "https://femto.ai/press_releases/2025-12-30/"
    },
    {
     "title": "BusinessWire: $8M Series A (Sep 2022)",
     "url": "https://www.businesswire.com/news/home/20220928005770/en/Edge-AI-Innovator-Femtosense-Raises-%248-Million-Series-A-Funding"
    },
    {
     "title": "TechCrunch: Femtosense raises capital (Sep 2022)",
     "url": "https://techcrunch.com/2022/09/28/femtosense-raises-capital-to-power-the-ai-inside-of-consumer-electronics/"
    },
    {
     "title": "NewSound AI hearing aids PR (Mar 2025)",
     "url": "https://femto.ai/press_releases/2025-03-26/"
    },
    {
     "title": "femtoAI rebrand post (Jun 2025)",
     "url": "https://www.linkedin.com/posts/femtoai_from-femtosense-to-femtoai-activity-7344058486546137090-7lzA"
    },
    {
     "title": "Yahoo Finance: femtoAI 5x growth / CES 2026 award (Jul 2026)",
     "url": "https://sg.finance.yahoo.com/news/femtoai-unlocks-5x-growth-full-155500048.html"
    }
   ]
  },
  {
   "slug": "finalspark",
   "name": "FinalSpark",
   "founded": 2014,
   "hq": "Vevey, Switzerland",
   "category": "biological",
   "status": "active",
   "origin": "serial founders (Fred Jordan & Martin Kutter; ex-AlpVision co-founders)",
   "total_funding_usd_m": 1,
   "last_round": "Bootstrapped/self-funded ~CHF 1M to date; seeking CHF 50M growth round (not yet closed)",
   "key_investors": [
    "self-funded / founders"
   ],
   "website": "https://finalspark.com",
   "pitch": "FinalSpark's premise is the energy wall of digital AI: training large models emits as much CO2 as\nseveral cars over their lifetimes and datacenter silicon is straining under demand. It claims living\nneurons are on the order of **one million times more energy efficient** than digital hardware, so a\n\"bioprocessor\" built from human neurons could compute AI workloads at a tiny fraction of the power.\n(https://www.techjournal.uk/p/finalspark-explores-living-neurons)",
   "technical_approach": "\"Organic computing\" using **human brain organoids** — 3D cell masses grown from stem-cell-derived\nneurons — interfaced to electrodes for stimulation and recording. The **Neuroplatform** (launched\nMay 2024) is a remote, cloud-accessible research platform: **16 organoids (~160,000 living human\nneurons)** housed across four Multi-Electrode Arrays (four organoids per MEA, eight electrodes each),\nkept alive and accessible around the clock so researchers worldwide can run experiments in vitro.\nPositioned as the first remote biocomputing research platform; still a research substrate rather than\na deployed computer. Dopamine-based reward signaling is used to explore training/plasticity.\n(https://www.businesswire.com/news/home/20240515701469/en/FinalSpark-Launches-the-First-Remote-Research-Platform-Using-Human-Neurons-for-Biocomputing;\nhttps://www.tomshardware.com/pc-components/cpus/worlds-first-bioprocessor-uses-16-human-brain-organoids)",
   "funding_rounds": [
    {
     "date": "2014–present",
     "round": "Self-funded",
     "amount": "~CHF 1M invested since founding",
     "investors": "founders (Jordan & Kutter)",
     "source": "https://www.techjournal.uk/p/finalspark-explores-living-neurons"
    },
    {
     "date": "(sought, not closed)",
     "round": "Growth",
     "amount": "CHF 50M target",
     "investors": "— (to hire ~30 scientists; ~7-yr cloud-compute goal)",
     "source": "techjournal / interviews"
    }
   ],
   "funding_notes": "No priced venture rounds or named institutional investors publicly reported. The founders — who\npreviously built and self-funded image-authentication company AlpVision — have bootstrapped FinalSpark;\ntotal external funding is effectively **undisclosed/negligible** (~CHF 1M self-invested).",
   "trajectory": "Founded 2014 in Vevey, Switzerland by Dr. **Fred Jordan** and Dr. **Martin Kutter** (also co-founders\nof AlpVision) around the idea of a \"thinking machine\" built from biology rather than silicon. Team of\n~6 core members plus advisers; three patents; reports ~10 million neurons tested and ~20 TB of\nexperimental data collected. The Neuroplatform serves ~10 selected universities free of charge plus\npaying commercial clients who develop proprietary applications. Active as of mid-2026, operating as a\nresearch-platform business while seeking a larger raise to scale toward a practical living processor.\n(https://infinitefrontiers.io/conversations/brain-organoids-for-ai-biocomputing-with-finalsparks-fred-jordan/)\n\n**Founders (sweep 2026-09-22):** Dr. Fred Jordan (co-CEO; PhD signal processing, EPFL, 1999) and Dr. Martin Kutter (co-CEO; PhD EPFL, 2000). Both co-founded AlpVision (anti-counterfeiting, ~2001) and still run it (Jordan CEO, Kutter president) alongside FinalSpark — the founding team came out of AlpVision, the technology did not (https://finalspark.com/team/ ; https://petermbach.com/index.php/podcast/056-fred-jordan/ ; https://www.techjournal.uk/p/finalspark-explores-living-neurons ; https://dld-conference.com/speaker/fred-jordan).",
   "relevance": "FinalSpark shares the biological-substrate, radical-energy-efficiency thesis with Cortical Labs and\nKoniku — living neurons instead of Knowm's memristors. Its wedge is a remote research platform\n(access-as-a-service) rather than a shipped computer, and unlike Knowm's manufacturable solid-state\ndevices it depends on keeping organoids alive. A biological counterpoint to the memristor/AHaH\nlow-power learning claim, but far earlier-stage and not a hardware product.",
   "news": [],
   "sources": [
    {
     "title": "FinalSpark explores living neurons (Tech Journal UK)",
     "url": "https://www.techjournal.uk/p/finalspark-explores-living-neurons"
    },
    {
     "title": "FinalSpark launches first remote human-neuron platform (BusinessWire)",
     "url": "https://www.businesswire.com/news/home/20240515701469/en/FinalSpark-Launches-the-First-Remote-Research-Platform-Using-Human-Neurons-for-Biocomputing"
    },
    {
     "title": "World's first bioprocessor uses 16 brain organoids (Tom's Hardware)",
     "url": "https://www.tomshardware.com/pc-components/cpus/worlds-first-bioprocessor-uses-16-human-brain-organoids-for-a-million-times-less-power-consumption-than-a-digital-chip"
    },
    {
     "title": "Interview: Fred Jordan (Infinite Frontiers)",
     "url": "https://infinitefrontiers.io/conversations/brain-organoids-for-ai-biocomputing-with-finalsparks-fred-jordan/"
    }
   ]
  },
  {
   "slug": "fractile",
   "name": "Fractile",
   "founded": 2022,
   "hq": "London, UK",
   "category": "boundary-digital",
   "status": "active",
   "origin": "academic spinoff (Oxford Robotics Institute — founder's PhD, not a formal university spinout)",
   "total_funding_usd_m": 958.5,
   "last_round": "Series C, ~£540M (~US$721M; allotment sums up to ~£577M), 2026-09-09→2026-10-23 per Companies House; valuation undisclosed",
   "key_investors": [
    "Accel",
    "Founders Fund",
    "Factorial Funds",
    "NATO Innovation Fund",
    "Kindred Capital",
    "Oxford Science Enterprises",
    "Pat Gelsinger (angel)"
   ],
   "website": "https://www.fractile.ai",
   "pitch": "Fractile frames memory bandwidth — not compute — as the rate limiter on AI inference:\n\"Inference is both the revenue engine of the AI industry and the rate-limiting factor on\nexpanding it.\" Goodwin notes frontier LLMs can emit ~100M tokens on complex problems but run at\n~40 tokens/s on existing chips, i.e. roughly a month per output. Moving model weights between\nDRAM/HBM and compute is the bottleneck; running calculations directly in memory removes it.\n([Tech.eu, 2026-05-13](https://tech.eu/2026/05/13/uk-ai-chip-startup-fractile-raises-220m-to-tackle-the-growing-inference-bottleneck/);\n[DCD, 2024-07](https://www.datacenterdynamics.com/en/news/uk-ai-chip-startup-fractile-emerges-from-stealth-with-15m-funding-round/))",
   "technical_approach": "- **Substrate:** in-memory compute built on Fractile's own custom CMOS SRAM cell design;\n  RISC-V control; no emerging memory device. Digital vs analog officially undisclosed\n  (Goodwin declined to answer), widely read as digital IMC.\n  ([eeNews Europe](https://www.eenewseurope.com/en/ceo-interview-walter-goodwin-fractile-ai/))\n- **Memory architecture:** \"DRAM-less\" — model weights held on-chip in SRAM rather than external\n  HBM/DRAM, an angle that gained salience during the 2025–26 DRAM pricing/shortage crunch.\n  ([Tom's Hardware](https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropic-in-early-talks-to-buy-inference-chips-from-uk-startup-fractile))\n- **Workload:** datacenter LLM **inference only** (no on-chip learning claim — off-chip training).\n- **Products:** no commercial chip yet as of mid-2026; taping out test chips from London and\n  Bristol teams ([eeNews Europe](https://www.eenewseurope.com/en/ceo-interview-walter-goodwin-fractile-ai/)).\n  Feb 2026: announced £100M ($135M) UK investment over 3 years, incl. a new Bristol hardware\n  engineering facility ([DCD](https://www.datacenterdynamics.com/en/news/fractile-raises-220m-to-accelerate-development-of-ai-inference-chips/)).",
   "funding_rounds": [
    {
     "date": "~2022–23",
     "round": "Pre-seed",
     "amount": "~$2.5M (implied: seed brought total to $17.5M)",
     "investors": "undisclosed",
     "source": "[Tech.eu](https://tech.eu/2024/07/26/fractile-secures-15m-funding-for-better-hardware-for-ai-inference/)"
    },
    {
     "date": "2024-07",
     "round": "Seed",
     "amount": "$15M",
     "investors": "co-led Kindred Capital, NATO Innovation Fund, Oxford Science Enterprises; Cocoa, Inovia; angels Hermann Hauser, Stan Boland, Amar Shah",
     "source": "[DCD](https://www.datacenterdynamics.com/en/news/uk-ai-chip-startup-fractile-emerges-from-stealth-with-15m-funding-round/), [Fortune](https://fortune.com/2024/07/26/fractile-ai-chip-startup-nvidia-15-million-funding-seed-round)"
    },
    {
     "date": "2025",
     "round": "Angel (add-on)",
     "amount": "undisclosed",
     "investors": "Pat Gelsinger (ex-Intel CEO, Playground Global)",
     "source": "[DCD](https://www.datacenterdynamics.com/en/analysis/fractile-ai-chip-startup-inference-pat-gelsinger/), [EE Times](https://www.eetimes.com/gelsinger-invests-in-british-ai-chip-startup-fractile/)"
    },
    {
     "date": "2026-05",
     "round": "Series B",
     "amount": "$220M",
     "investors": "led Accel, Founders Fund, Factorial Funds; Gigascale, Conviction, Buckley Ventures, O1A, 8VC, Felicis",
     "source": "[Tech.eu](https://tech.eu/2026/05/13/uk-ai-chip-startup-fractile-raises-220m-to-tackle-the-growing-inference-bottleneck/), [Fractile PR](https://www.fractile.ai/news/fractile-raises-220m-to-build-the-next-generation-of-inference-hardware)"
    }
   ],
   "funding_notes": "Total: ~$237.5M. Valuation: ~$1B reported around the Series B (Mar 2026 reports of raising\n\"$200M at a $1B valuation\"; final round was $220M —\n[BeBeez](https://bebeez.eu/2026/03/31/uk-chip-startup-fractile-looking-to-raise-200m-at-1bn-valuation-report/),\n[Electronics Weekly](https://www.electronicsweekly.com/news/business/uk-ai-chip-startup-in-talks-to-raise-200m-2026-05/)).",
   "trajectory": "Founded 2022 by Dr. Walter Goodwin, then a PhD student at the Oxford Robotics Institute, after\nconcluding LLM inference speed was the untapped hardware market. Emerged from stealth Jul 2024.\nOperations in London, Bristol, San Francisco and Taipei (~mid-2026). **Anthropic is reported to\nbe in early talks to buy Fractile inference chips** — reported (Tom's Hardware / The Next Web,\nMay 2026), not a confirmed deal; no formal agreement announced. Status mid-2026: active,\npre-product, one of the best-funded UK chip startups.\n\n**Founders (sweep 2026-09-22):** Walter Goodwin (sole founder & CEO; MA and PhD, University of Oxford — Oxford Robotics Institute; founded Fractile in 2021/22 while still on his doctorate). Oxford Science Enterprises is an investor, but no source describes a formal spinout or a lab technology transfer (https://ori.ox.ac.uk/people/walter-goodwin ; https://www.forbes.com/profile/walter-goodwin/ ; https://www.oxfordscienceenterprises.com/portfolio/fractile).",
   "relevance": "Fractile shows how far \"in-memory compute\" branding now stretches: a ~$1B-valued company whose\nIMC is (apparently) conventional-CMOS SRAM, inference-only, with no emerging device and no\non-chip learning — the opposite end of the taxonomy from Knowm's memristor/AHaH on-chip-learning\napproach, and a useful boundary marker between digital memory-wall plays and physics-based ones.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Bloomberg (2026-08-19): Fractile reached an **initial ~$250M deal to sell inference chips to Anthropic** (chips not expected ready until 2027) and is in advanced talks to raise **~$600M at a $6.5B valuation** — up from the ~$1B target reported in March. (https://www.bloomberg.com/news/articles/2026-08-19/ai-chip-startup-fractile-in-talks-for-6-5-billion-value-after-anthropic-deal ; https://finance.yahoo.com/technology/ai/articles/chip-firm-fractile-seeks-6-183158545.html)"
    },
    {
     "date": "2026-11-09",
     "text": "**Series C of ~£540M (~US$721M) assembled through share allotments 2026-09-09 → 2026-10-23, per Companies House filings (Sifted 2026-10-29); Tech Funding News' sum of the nine SH01 returns is ~£577M (~US$774M), which Nordic9 reconciles as £540M in September plus October top-ups.** Three price tiers (C-1 £801.54, C-2 £707.25, C-3 £1,452.35/share); ~£450.5M allotted on 2026-09-11 in cash. **Not announced by the company.** Valuation: no figure in the filings; the Telegraph reports ~US$6.5B, Nordic9 implies ~£3.84B (~US$5.1B) post-money on issued shares; Series B (2026-05) was $220M at ~$1B. Investors: Lightspeed and Redpoint reported as leads and the **UK Sovereign AI Fund** reported among new investors (Telegraph, unnamed sources; Fractile and the government declined comment). Reported, unconfirmed: an initial **~US$250M chip order from Anthropic** (Bloomberg via secondary coverage). Product: no production chip shipped; data-centre deployment expected 2027; claimed target 1,200 tok/s on long workloads. **Working total ~US$960M † (237.5 + ~721)** — the largest raise in the census, as a *boundary* (digital IMC) case ahead of Lightmatter. (https://sifted.eu/articles/exclusive-fractile-series-c ; https://ultrathink.ai/news/fractile-577m-series-c-companies-house-anthropic-chip-order ; https://nordic9.com/news/fractile-raises-540m-in-series-c/ ; https://sanderbosmanai.substack.com/p/uk-chip-start-up-fractile-raises)"
    }
   ],
   "sources": [
    {
     "title": "Fractile raises $220m — DCD",
     "url": "https://www.datacenterdynamics.com/en/news/fractile-raises-220m-to-accelerate-development-of-ai-inference-chips/"
    },
    {
     "title": "Fractile raises $220M — Tech.eu",
     "url": "https://tech.eu/2026/05/13/uk-ai-chip-startup-fractile-raises-220m-to-tackle-the-growing-inference-bottleneck/"
    },
    {
     "title": "Fractile $15M seed — DCD",
     "url": "https://www.datacenterdynamics.com/en/news/uk-ai-chip-startup-fractile-emerges-from-stealth-with-15m-funding-round/"
    },
    {
     "title": "Fractile seed, total $17.5M — Tech.eu",
     "url": "https://tech.eu/2024/07/26/fractile-secures-15m-funding-for-better-hardware-for-ai-inference/"
    },
    {
     "title": "CEO interview: Walter Goodwin — eeNews Europe",
     "url": "https://www.eenewseurope.com/en/ceo-interview-walter-goodwin-fractile-ai/"
    },
    {
     "title": "Anthropic in early talks — Tom's Hardware",
     "url": "https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropic-in-early-talks-to-buy-inference-chips-from-uk-startup-fractile"
    },
    {
     "title": "Gelsinger invests — EE Times",
     "url": "https://www.eetimes.com/gelsinger-invests-in-british-ai-chip-startup-fractile/"
    },
    {
     "title": "$1bn valuation report — BeBeez",
     "url": "https://bebeez.eu/2026/03/31/uk-chip-startup-fractile-looking-to-raise-200m-at-1bn-valuation-report/"
    },
    {
     "title": "Fractile PR: $220M round",
     "url": "https://www.fractile.ai/news/fractile-raises-220m-to-build-the-next-generation-of-inference-hardware"
    }
   ]
  },
  {
   "slug": "gemesys",
   "name": "GEMESYS",
   "founded": 2021,
   "hq": "Bochum, Germany",
   "category": "emerging-device-imc",
   "status": "active",
   "origin": "academic spinoff (Ruhr-University Bochum)",
   "total_funding_usd_m": 9.1,
   "last_round": "Pre-seed, €8.6M (~$9.1M), 2024-11",
   "key_investors": [
    "Amadeus APEX Technology Fund",
    "Atlantic Labs",
    "NRW.BANK",
    "Sony Innovation Fund",
    "Plug and Play"
   ],
   "website": "https://gemesys.tech",
   "pitch": "GEMESYS argues AI's potential is capped by hardware: \"current hardware infrastructure will not\nbe able to unlock this potential because of its energy and computing power limits\"\n([gemesys.tech](https://gemesys.tech/)). Their answer is \"The New ANALOG PARADIGM\" enabled by\nthe memristor — adapting \"the information processing mechanism of the human brain\" so that\nneural networks can be **trained**, not just executed, on edge devices, with claimed training\nefficiency gains up to a factor of 10,000 and what they call \"the world's first hardware for\nedge AI training.\"",
   "technical_approach": "- **Device/substrate:** memristors — \"novel electronic components that enable highly efficient\n  neural network processing\" — in a **fully analog** chip architecture\n  ([gemesys.tech](https://gemesys.tech/); [Tech.eu](https://tech.eu/2024/11/19/gemesys-secures-8-6m-pre-seed-for-ai-chip-tech/)).\n  Public materials do not specify the memristor chemistry/stack.\n- **Learning locus:** the distinguishing claim — **on-device training AND inference** at the\n  edge, pitched on data efficiency, energy savings, privacy, and small chip size\n  ([Sifted](https://sifted.eu/articles/ai-chip-startup-gemesys-raise-news);\n  [Amadeus Capital PR](https://www.amadeuscapital.com/gemesys-secures-e8-6m-pre-seed-funding-to-transform-ai-on-edge-devices/)).\n  This makes GEMESYS one of very few funded startups claiming memristor-based on-chip *learning*\n  rather than inference-only in-memory compute.\n- **Products/status:** no named commercial chip yet. A working hardware demonstrator exists;\n  silicon testing was planned for 2025 following MVP development\n  ([Future of Computing profile](https://www.future-of-computing.com/gemesys-shaping-the-future-of-ai-chips-inspired-by-the-human-brain/)).\n  No public tape-out or dev-kit announcement found as of 2026-07-23.\n- **Market wedge:** ultra-low-power edge AI (on-device training for personalization/adaptation).",
   "funding_rounds": [
    {
     "date": "(pre-2024)",
     "round": "Grant",
     "amount": "undisclosed",
     "investors": "German government EXIST \"transfer of research\" programme",
     "source": "[Tech.eu](https://tech.eu/2024/11/19/gemesys-secures-8-6m-pre-seed-for-ai-chip-tech/)"
    },
    {
     "date": "2024-11-19",
     "round": "Pre-seed",
     "amount": "€8.6M (~$9.1M); figure includes the EXIST grant component",
     "investors": "co-led Amadeus APEX Technology Fund + Atlantic Labs; NRW.BANK, Sony Innovation Fund, Plug and Play Tech Center",
     "source": "[Amadeus PR](https://www.amadeuscapital.com/gemesys-secures-e8-6m-pre-seed-funding-to-transform-ai-on-edge-devices/), [Sifted](https://sifted.eu/articles/ai-chip-startup-gemesys-raise-news), [GEMESYS PR (PDF)](https://gemesys.tech/wp-content/uploads/2024/11/241118_Press-Release-Pre-Seed-Round-Website_DM.pdf)"
    }
   ],
   "funding_notes": "Total disclosed: €8.6M (~$9.1M). One of the larger European pre-seeds of 2024 for chip hardware.\nNo later round found as of 2026-07-23.",
   "trajectory": "Founded 2021 as a spin-off from **Ruhr-University Bochum** by **Dr. Dennis Michaelis (CEO),\nDr. Enver Solan, Moritz Schmidt, and Daniel Krüger** ([Sifted](https://sifted.eu/articles/ai-chip-startup-gemesys-raise-news);\n[NeuromorphicCore.ai](https://neuromorphiccore.ai/insights/gemesys/)). Team of ~19 at the time\nof the pre-seed, with plans to double and to accelerate R&D on the chip architecture. Backing\nmixes deep-tech VC (Amadeus APEX, Atlantic Labs), a state bank (NRW.BANK), and a strategic\ncorporate (Sony Innovation Fund). Status mid-2026: active, pre-product, demonstrator-stage.",
   "relevance": "GEMESYS is the closest wave-2 European analogue to Knowm's own thesis: memristor-based analog\nhardware where the differentiator is **on-device learning**, not inference acceleration — the\nsame territory as Knowm's AHaH/on-chip-adaptation angle, arriving roughly a decade later with\ninstitutional VC and a university-lab device lineage (Bochum rather than Boise State). Direct\ncomparison of device maturity is not yet possible: GEMESYS has not published device chemistry,\ndatasheets, or commercial parts, whereas Knowm has shipped discrete SDC memristors since 2015.",
   "news": [],
   "sources": [
    {
     "title": "GEMESYS website",
     "url": "https://gemesys.tech/"
    },
    {
     "title": "GEMESYS press release, pre-seed (PDF)",
     "url": "https://gemesys.tech/wp-content/uploads/2024/11/241118_Press-Release-Pre-Seed-Round-Website_DM.pdf"
    },
    {
     "title": "GEMESYS secures €8.6M — Amadeus Capital",
     "url": "https://www.amadeuscapital.com/gemesys-secures-e8-6m-pre-seed-funding-to-transform-ai-on-edge-devices/"
    },
    {
     "title": "GEMESYS secures €8.6M pre-seed — Tech.eu",
     "url": "https://tech.eu/2024/11/19/gemesys-secures-8-6m-pre-seed-for-ai-chip-tech/"
    },
    {
     "title": "Gemesys raises €8.6m pre-seed — Sifted",
     "url": "https://sifted.eu/articles/ai-chip-startup-gemesys-raise-news"
    },
    {
     "title": "Gemesys raises €8.6M for human brain-like AI chips — TFN",
     "url": "https://techfundingnews.com/gemesys-raises-e8-6m-for-human-brain-like-ai-chips/"
    },
    {
     "title": "GEMESYS — Future of Computing profile",
     "url": "https://www.future-of-computing.com/gemesys-shaping-the-future-of-ai-chips-inspired-by-the-human-brain/"
    },
    {
     "title": "GEMESYS — NeuromorphicCore.ai",
     "url": "https://neuromorphiccore.ai/insights/gemesys/"
    }
   ]
  },
  {
   "slug": "general-vision",
   "name": "General Vision Inc.",
   "founded": 1999,
   "hq": "Petaluma, California, USA (team page now lists 201 Spear St, San Francisco; roots in Sonoma County)",
   "category": "digital-snn (brain-inspired digital — but associative-memory RBF/kNN classifier, NOT spiking)",
   "status": "active (quiet)",
   "origin": "serial founders (Guy Paillet — ZISC co-inventor with IBM France; Anne Menendez — machine-vision founder, GTFS 1987)",
   "total_funding_usd_m": 0,
   "last_round": "none disclosed — 'unfunded' per Tracxn; self-funded via licensing and chip sales",
   "key_investors": [],
   "website": "https://general-vision.com",
   "pitch": "General Vision does not lead with \"von Neumann bottleneck\" language; its framing is \"Real\nIntelligence at the edge\" — trainable neurons in memory that learn \"on the spot\" with \"low power\nconsumption, blazing speed performance, knowledge traceability\" (https://general-vision.com/).\nThe implicit anti-von-Neumann claim: NeuroMem neurons \"learn and recognize patterns ... with\nless energy and complexity than modern microprocessors,\" all neurons evaluating an input in\nparallel at constant latency regardless of network size\n(https://general-vision.com/ip-and-chips/nm500/).",
   "technical_approach": "- **What NeuroMem actually is (classify honestly):** a fully digital, massively parallel\n  **associative-memory pattern classifier** — each identical \"neuron\" stores a prototype vector\n  (up to 256 bytes) plus category, computes L1/Lsup distance to the input in parallel, and fires\n  by **RBF (Radial Basis Function / RCE-style, adjustable influence fields)** or **kNN** decision\n  logic. On-chip learning = committing new neurons and shrinking influence fields. It is **NOT a\n  spiking neural network** and does no deep-learning matrix math — GV itself contrasts RBF with\n  kNN and Deep Learning\n  (https://general-vision.com/neuromem-is-yet-another-brain-inspired-chip-but-it-is-shipping/, 2019).\n- **Chip lineage** (per GV's own \"NeuroMem Odyssey,\" rev 01-18,\n  https://general-vision.com/publications/PR_NeuroMemOdyssey.pdf):\n  - **ZISC36** (1993) — Zero Instruction Set Computer, 36 neurons, 1 µm; Paillet's concept built\n    with IBM's Corbeil-Essonnes lab; patents co-assigned IBM/Paillet.\n  - **ZISC78** (2001) — 78 neurons, 0.25 µm; IBM closed the Corbeil foundry Dec 2001, ending ZISC.\n  - **CM1K** (2007–08) — 1,024 neurons, 130 nm (OKI), via Hong Kong entity CogniMem Ltd.\n  - **Intel QuarkSE/Curie** (2015) — 128-neuron NeuroMem pattern-recognition engine licensed to\n    Intel (2011 via IP-holding Norlitech); shipped in the Curie module until Intel killed its IoT\n    line (Jul 2017).\n  - **NM500** (2017–18) — 576 neurons, 110 nm WLCSP; manufacturing license to Korea's **nepes\n    Corp** (collaboration Aug 2016, license Jan 2017, production samples + NeuroShield/NeuroBrick\n    boards Jan 2018). nepes manufactures and sells under GV license\n    (https://general-vision.com/nm500release/).\n- **Business model:** IP licensing (Intel, nepes, ANURAG/India DoD) + chips/boards/software\n  (Knowledge Builder), not a fabbed-product startup. Learning locus: genuine **on-chip learning**\n  (its main claim since 1993).",
   "funding_rounds": [
    {
     "date": "—",
     "round": "none disclosed",
     "amount": "—",
     "investors": "—",
     "source": "Tracxn lists General Vision as \"unfunded\" (https://tracxn.com/d/companies/general-vision/__teOUSeFpN8LhnfHwBCxw4mcmNxPwZbAb3eZFv-jTIP0)"
    }
   ],
   "funding_notes": "Effectively **bootstrapped** on licensing/sales revenue. Per GV's own Odyssey PDF: CogniMem Ltd\n(2006) was funded by the founders plus two angel investors; a 2014 spin-out \"NeuroMem Inc.\" (CA)\nwas **abandoned after an 18-month investor campaign \"due to lack of funding.\"** A separate\nGeneva-based \"NeuroMem\" on Crunchbase shows an undisclosed equity-crowdfunding round\n(https://www.crunchbase.com/organization/neuromem) — relationship to GV unverified. No VC round,\namount, or valuation for General Vision Inc. itself is disclosed anywhere reviewed.",
   "trajectory": "Guy Paillet co-invented the IBM ZISC (1993); after IBM discontinued it (2001) he and Menendez\ncarried the architecture through a chain of small entities (Silicon Recognition — bankrupt 2004,\nassets bought by OmniVision for $1.4M; Norlitech; CogniMem; NeuroMem Inc.) to today's General\nVision (Odyssey PDF). Peak visibility: Intel Curie at CES 2015–16, nepes NM500 launch 2017–18.\nMenendez left day-to-day work in 2019 (advisor/shareholder; https://general-vision.com/team/).\n**Status mid-2026: active but quiet** — website up but copyright ©2022, latest dated blog/product\nnews ~2019–2023, the general-vision.tech mirror has an expired TLS certificate, and no 2024–2026\nannouncements, awards, or funding news surfaced in searches. NM500 remains purchasable via\nnepes. Unverified: current employee count and revenue.",
   "relevance": "The 30-year baseline for the whole wave: on-chip learning in silicon has been commercially\nshippable since 1993, yet stayed a niche — GV survived by licensing, not scaling. Its RBF/kNN\nassociative memory is an honest contrast to both spiking chips and Knowm's analog memristor/AHaH\napproach: digital, deterministic, explainable, but no synaptic physics or deep-network capability.",
   "news": [],
   "sources": [
    {
     "title": "General Vision homepage",
     "url": "https://general-vision.com/"
    },
    {
     "title": "The NeuroMem Odyssey (GV company history PDF, rev 01-18)",
     "url": "https://general-vision.com/publications/PR_NeuroMemOdyssey.pdf"
    },
    {
     "title": "About General Vision / Team",
     "url": "https://general-vision.com/team/"
    },
    {
     "title": "NM500 product page",
     "url": "https://general-vision.com/ip-and-chips/nm500/"
    },
    {
     "title": "nepes launches the NeuroMem NM500",
     "url": "https://general-vision.com/nm500release/"
    },
    {
     "title": "NeuroMem is yet another brain-inspired chip, but it is shipping (GV blog, May 2019)",
     "url": "https://general-vision.com/neuromem-is-yet-another-brain-inspired-chip-but-it-is-shipping/"
    },
    {
     "title": "EE Times — Startup's Tech is Intel's Quark Neural Network (2015)",
     "url": "https://www.eetimes.com/startups-tech-is-intels-quark-neural-network/"
    },
    {
     "title": "Tracxn — General Vision profile (\"unfunded\", founded 1999)",
     "url": "https://tracxn.com/d/companies/general-vision/__teOUSeFpN8LhnfHwBCxw4mcmNxPwZbAb3eZFv-jTIP0"
    },
    {
     "title": "Crunchbase — NeuroMem (Geneva; equity crowdfunding, relationship unverified)",
     "url": "https://www.crunchbase.com/organization/neuromem"
    }
   ]
  },
  {
   "slug": "grai-matter-labs",
   "name": "GrAI Matter Labs (GML)",
   "founded": 2016,
   "hq": "Paris, France + Eindhoven, Netherlands (+ Silicon Valley office)",
   "category": "digital-snn",
   "status": "acquired",
   "origin": "startup-studio spinoff (iBionext, Paris) with academic founders (Ryad Benosman — Institut de la Vision/UPMC; Giacomo Indiveri — INI Zurich)",
   "total_funding_usd_m": 29,
   "last_round": "acquired by Snap Inc., 2023-10 (terms undisclosed)",
   "key_investors": [
    "iBionext",
    "360 Capital",
    "3T Finance",
    "Bpifrance"
   ],
   "website": "https://grokipedia.com/page/GrAI_Matter_Labs (original site offline post-acquisition)",
   "pitch": "\"GrAI Matter Labs, a pioneer of ultra-low power neuromorphic computing\" promising \"the fastest\nAI per watt to every device on the edge\" — its NeuronFlow event-driven dataflow only computes\non change, cutting the latency and energy wasted re-processing redundant sensor frames\n(https://www.dutchnews.nl/businesswire/grai-matter-labs-raises-14m-to-bring-fastest-ai-per-watt-to-every-device-on-the-edge/,\nhttps://www.businesswireindia.com/neuromorphic-computing-pioneer-grai-matter-labs-raises-usd15m-to-bring-ai-to-every-device-on-the-edge-57843.html).",
   "technical_approach": "- **Digital event-driven near-memory dataflow** (\"NeuronFlow\"): neuron-level sparsity /\n  compute-on-change derived from spiking principles, executing standard NN workloads;\n  inference-only (no on-chip learning claimed).\n- **GrAI One** (Oct 2019): first silicon, \"world's first AI processor optimized for ultra-low\n  latency and low-power processing at the edge\"\n  (https://www.actuia.com/en/news/deep-tech-grai-matter-labs-raises-e13-million-to-bring-to-market-its-embedded-ai-chip/).\n- **GrAI VIP** (~2022): full SoC (\"Vision Inference Processor\") targeting near-sensor vision;\n  profiled by Forbes as \"brain-inspired AI for the edge\"\n  (https://www.forbes.com/sites/karlfreund/2022/05/27/grai-matter-labs-brain-inspired-ai-for-the-edge/,\n  https://www.eetimes.com/has-grai-matter-labs-been-snapped-up-by-snap-inc/).",
   "funding_rounds": [
    {
     "date": "2018-04",
     "round": "Series A",
     "amount": "$15M",
     "investors": "led by iBionext",
     "source": "https://www.wsj.com/articles/ai-startup-grai-matter-labs-completes-15-million-series-a-1523565354 ; https://ibionext.com/en/telesurveillance-medicale-chronolife-et-be-ys-concluent-un-partenariat-pour-optimiser-le-suivi-des-patients-atteints-de-pathologies-chroniques-3-3/"
    },
    {
     "date": "2020-11",
     "round": "Series A2 / second round",
     "amount": "$14M (€13M)",
     "investors": "360 Capital, 3T Finance, Bpifrance (with iBionext)",
     "source": "https://www.dutchnews.nl/businesswire/grai-matter-labs-raises-14m-to-bring-fastest-ai-per-watt-to-every-device-on-the-edge/ ; https://www.actuia.com/actualite/snap-fait-lacquisition-de-la-start-up-francaise-grai-matter-labs/"
    }
   ],
   "funding_notes": "**Total raised:** $29M across two rounds (ActuIA, IO+); EE Times: \"close to $30 million\"\n(https://www.eetimes.com/has-grai-matter-labs-been-snapped-up-by-snap-inc/,\nhttps://ioplus.nl/archive/en/french-media-snap-inc-snaps-up-french-dutch-ai-trailblazer-grai-matter-labs/).",
   "trajectory": "Founded 2016 as \"Brainiac\" inside the iBionext startup studio (Paris) by Ryad Benosman,\nBernard Gilly, Giacomo Indiveri and colleagues\n(https://www.businesswireindia.com/neuromorphic-computing-pioneer-grai-matter-labs-raises-usd15m-to-bring-ai-to-every-device-on-the-edge-57843.html);\nIngolf Held CEO from Jan 2018 (https://www.kisacoresearch.com/content/ingolf-held). R&D split\nParis/Eindhoven. **Acquisition confirmed:** Snap Inc. acquired GrAI Matter Labs SAS on\n31 October 2023, completed quietly; terms undisclosed\n(https://www.marketscreener.com/quote/stock/SNAP-INC-34091150/news/Snap-Inc-acquired-GrAI-Matter-Labs-SAS-46151707/,\nhttps://bits-chips.com/article/grai-matter-labs-quietly-snapped-up-by-snap/). French press\n(L'Informé, via Bits&Chips) reported GML \"was driven into American arms because it couldn't\nfind funding in Europe.\" As of mid-2026: absorbed into Snap (AR/edge-AI work); no independent\noperations.",
   "relevance": "GML shows the exit path for the digital-SNN cohort: a decade of neuromorphic IP ending as an\nacqui-hire into a consumer-AR company after ~$29M — event-driven efficiency was valued, but\nnot as a standalone chip business. No physical-synapse learning; contrast with Knowm's\ndevice-level adaptation thesis.",
   "news": [],
   "sources": [
    {
     "title": "WSJ: $15M Series A (Apr 2018)",
     "url": "https://www.wsj.com/articles/ai-startup-grai-matter-labs-completes-15-million-series-a-1523565354"
    },
    {
     "title": "BusinessWire India: Series A + founding story",
     "url": "https://www.businesswireindia.com/neuromorphic-computing-pioneer-grai-matter-labs-raises-usd15m-to-bring-ai-to-every-device-on-the-edge-57843.html"
    },
    {
     "title": "DutchNews/BusinessWire: $14M round (Nov 2020)",
     "url": "https://www.dutchnews.nl/businesswire/grai-matter-labs-raises-14m-to-bring-fastest-ai-per-watt-to-every-device-on-the-edge/"
    },
    {
     "title": "MarketScreener: Snap acquired GML SAS 2023-10-31",
     "url": "https://www.marketscreener.com/quote/stock/SNAP-INC-34091150/news/Snap-Inc-acquired-GrAI-Matter-Labs-SAS-46151707/"
    },
    {
     "title": "EE Times: Has GrAI Matter Labs Been Snapped Up By Snap? (Mar 2024)",
     "url": "https://www.eetimes.com/has-grai-matter-labs-been-snapped-up-by-snap-inc/"
    },
    {
     "title": "Bits&Chips: quietly snapped up by Snap",
     "url": "https://bits-chips.com/article/grai-matter-labs-quietly-snapped-up-by-snap/"
    },
    {
     "title": "ActuIA (FR): acquisition + investor list",
     "url": "https://www.actuia.com/actualite/snap-fait-lacquisition-de-la-start-up-francaise-grai-matter-labs/"
    }
   ]
  },
  {
   "slug": "grayscale-ai",
   "name": "Grayscale AI",
   "founded": 2020,
   "hq": "London, UK",
   "category": "digital-snn",
   "status": "active",
   "origin": "academic-adjacent founder (Dragos Stanciu — Edinburgh PhD not completed; ETH Zurich guest researcher on neuromorphic robot navigation; Entrepreneur First alumnus)",
   "total_funding_usd_m": 0,
   "last_round": "undisclosed (Techstars accelerator backing)",
   "key_investors": [
    "Techstars"
   ],
   "website": "https://grayscale.ai",
   "pitch": "Grayscale's framing is edge autonomy rather than the memory wall per se: neuromorphic AI\nthat \"mimics a human's neural network\" to recognize complex edge cases \"with human-like\nprecision,\" analyze them in under 100 ms without cloud dependency, and do so with a\nclaimed \"up to 500x less energy consumption than traditional computing.\"\nSource: https://grayscale.ai/",
   "technical_approach": "- **Approach:** neuromorphic-powered autonomous robots and drones — \"neural processing\n  units that replicate human brain activity to analyze sensory data and make decisions\n  instantaneously\" (press profile: https://www.automotiveworld.com/articles/grayscale-ai-paving-the-way-for-the-future-of-avs/).\n  Founder Dragos Stanciu applies mathematical models from neuroscience to robot\n  perception/decision-making.\n- **Substrate:** not disclosed; no custom chip announced. The 500x energy claim is a\n  company claim with no published benchmark.\n- **Applications:** autonomous vehicles (edge-case recognition), logistics automation,\n  security monitoring, disaster response; dual-use/defense direction via NATO DIANA and\n  U.S. Army xTech.\n- **Programs:** Techstars alumnus; selected for the NATO DIANA acceleration programme\n  (https://grayscale.ai/blog/grayscaleai-nato-diana); xTechDetect winner; Zenzic CAM\n  Scale-Up programme (UK connected & automated mobility)\n  (https://zenzic.io/cam-scale-up/grayscale-ai/).",
   "funding_rounds": [
    {
     "date": "—",
     "round": "Accelerator (Techstars)",
     "amount": "undisclosed",
     "investors": "Techstars",
     "source": "https://www.linkedin.com/company/grayscaleai ; https://www.crunchbase.com/organization/grayscale-ai"
    }
   ],
   "funding_notes": "Total raised: **undisclosed**. No priced rounds found; NATO DIANA and xTech participation\nimply grant/prize funding of undisclosed size.",
   "trajectory": "Founded 2020 (UK Companies House no. 12551623, consistent with a 2020 incorporation;\nEU-Startups directory also lists 2020: https://www.eu-startups.com/directory/grayscale-ai/)\nby Dragos Stanciu (founder/CEO, neuroscience background), London. Trajectory has run\nthrough accelerator and government-innovation programmes (Techstars → Zenzic CAM →\nxTechDetect → NATO DIANA, announced on their blog) rather than VC rounds. As of mid-2026\nthe company is actively \"seeking forward-thinking partners\" for commercialization — i.e.,\npre-scale, programme-funded. Adequately sourced to include, but the technical record is\nthin: no chip, no published benchmarks, no disclosed revenue or funding.\n\n**Founders (sweep 2026-09-22):** Dragos Stanciu (founder & CEO). Per his personal site (unverified): BSc Essex; EPSRC MRes/PhD scholarship at Edinburgh 2013 (RocketReach lists the PhD 2014–2016 as \"dropout\"); academic guest at ETH Zurich 2017–18 on neuromorphic chips for robot navigation; Entrepreneur First LD12 cohort 2019; co-founded MeshFlow; Grayscale AI from Jan 2020 (https://dragosstanciu.com/ ; https://rocketreach.co/dragos-stanciu-email_80205544).",
   "relevance": "Illustrates \"neuromorphic\" as a capability/branding layer for robotics software on\nconventional hardware — the opposite end of the spectrum from device-level physics\napproaches like memristors. Useful in the taxonomy to delimit what counts as a\nneuromorphic *substrate* claim versus a neuromorphic *algorithm* claim.",
   "news": [],
   "sources": [
    {
     "title": "Grayscale AI — home",
     "url": "https://grayscale.ai/"
    },
    {
     "title": "Grayscale AI blog: NATO DIANA selection",
     "url": "https://grayscale.ai/blog/grayscaleai-nato-diana"
    },
    {
     "title": "Automotive World: Grayscale AI and AVs",
     "url": "https://www.automotiveworld.com/articles/grayscale-ai-paving-the-way-for-the-future-of-avs/"
    },
    {
     "title": "Zenzic CAM Scale-Up profile",
     "url": "https://zenzic.io/cam-scale-up/grayscale-ai/"
    },
    {
     "title": "EU-Startups directory",
     "url": "https://www.eu-startups.com/directory/grayscale-ai/"
    },
    {
     "title": "LinkedIn company page",
     "url": "https://www.linkedin.com/company/grayscaleai"
    }
   ]
  },
  {
   "slug": "great-sky",
   "name": "Great Sky",
   "founded": 2024,
   "hq": "Boulder, CO, USA",
   "category": "photonic",
   "status": "active",
   "origin": "government-lab spinoff (NIST, Boulder)",
   "total_funding_usd_m": 14,
   "last_round": "Seed, $14M, 2026-03",
   "key_investors": [
    "Bison Ventures (lead)",
    "Matchstick Ventures",
    "Range Ventures"
   ],
   "website": "https://www.greatsky.ai",
   "pitch": "\"AI from first principles\": today's chips were designed for arithmetic, not intelligence, and\nwhen adapted for AI they consume \"energy orders of magnitude beyond that required by biological\nintelligence\" (https://www.greatsky.ai). At launch, CEO Jeff Shainline framed the goal as\n\"hardware that enables more sophisticated architectures while performing operations near the\nphysical limits of speed and energy efficiency... a completely different roadmap for scaling\nthat doesn't require hundreds of billions in capex and gigawatt data centers\"\n(https://pulse2.com/great-sky-14-million-seed-funding/).",
   "technical_approach": "Superconducting Optoelectronic Networks (SOENs), commercializing ~12 years of Shainline's NIST\nBoulder research program (23 peer-reviewed publications, 8 patents from the team's NIST tenure).\nThree pillars: superconducting circuits (compute, extreme energy efficiency), optical\ninterconnect sensitive down to single photons (communication/fan-out), and semiconductor\nbridging circuitry — all operating at cryogenic temperatures. Computation is classical analog,\nin parallel distributed neural-network-style architectures with memory and processing\nco-located, claimed to support continuous learning and on-device adaptation\n(https://www.greatsky.ai/news/great-sky-ai-emerges-from-nist).\n\nHonest classification: pre-product frontier hardware. As of the March 2026 launch the company\nreports first-generation chips taped out and \"first chips are now running inference\"\n(company site); headline claim is >60M video frames/sec analyzed at far lower energy than a\nGPU (~30 fps), from company materials — no independent benchmark, no product, and the\ncryogenic cooling requirement confines near-term deployment to datacenter/facility settings.\nTarget applications named: defense, energy (fusion prediction), large-scale AI infrastructure,\nsignal processing/video intelligence, molecular simulation.",
   "funding_rounds": [
    {
     "date": "2026-03-13",
     "round": "Seed",
     "amount": "$14M",
     "investors": "Bison Ventures (lead); Matchstick Ventures, Range Ventures; angels Mark Leslie, Adam Pritzker, Ivan Vendrov",
     "source": "https://pulse2.com/great-sky-14-million-seed-funding/"
    }
   ],
   "funding_notes": "Total verified: $14M. Aggregator StartupHub.ai lists \"total funding $19.6M\" with no round\nbreakdown or source — unverified; possibly includes an unannounced pre-seed or grants. Treat\n$14M as the citable figure.\nNon-dilutive: selected 2026-07-22 for Phase I of the U.S. DOE's Genesis Mission (a $293M\nprogram per neuromorphiccore.ai), one of only 19 private companies funded as principal\ninvestigators; award amount to Great Sky not disclosed\n(https://www.greatsky.ai/news/genesis-mission).",
   "trajectory": "Founded 2024 in Boulder, CO by four ex-NIST researchers: Jeff Shainline (CEO, physics PhD;\nleft NIST September 2024), Jeff Chiles (CTO, optics PhD), Saeed Khan (VP Fabrication,\nelectronics PhD), Bryce Primavera (VP Architecture, physics PhD). Operated quietly until the\npublic debut on 2026-03-13, announced together with the $14M seed (BusinessWire press release\n+ launch coverage). Status mid-2026: active, pre-product; first-gen chips taped out and\nreportedly running inference internally; seed capital earmarked for platform development and\ncommercialization in defense, energy, and AI infrastructure.\n\nUpdate 2026-08-23: on 2026-07-22 the DOE announced Genesis Mission Phase I awards; Great Sky\nwas selected as one of only 19 private companies funded as principal investigators, partnering\nwith Brookhaven National Laboratory to scale its superconducting optoelectronic neuromorphic\nsystems (SOENs) (https://www.greatsky.ai/news/genesis-mission; Brookhaven's project list\nhttps://www.bnl.gov/newsroom/news.php?a=223070). Shainline also appeared on EE Times' Brains\nand Machines podcast the same week\n(https://www.eetimes.com/podcasts/can-great-skys-light-connected-ai-enable-bigger-brains/).\nFirst federal-program validation of the SOEN approach post-NIST.",
   "relevance": "Extends the census's photonic category into superconducting territory — cryogenic single-photon\noptoelectronics rather than room-temperature silicon photonics — and stands as the newest\nphysics-frontier entrant alongside the thermodynamic cohort (Extropic, Normal Computing);\nnotably it claims on-device continuous learning, the same on-chip-learning locus Knowm's AHaH\napproach occupies, but via an entirely different (and far colder) substrate.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "SEC Form D reported ~2026-07-23: Great Sky **raised $25.9M of a $30.7M offering**; Dealroom (2026-08-06) headlines it as a \"$30.7M seed\". This is on top of / superseding the $14M seed announced 2026-03 and may explain the $44.7M total-funding claims flagged above. Round details (investors, whether it extends the March seed) undisclosed. (https://dealroom.co/news/143599-great-sky-raises-30-7m-seed-to-build-ai-chips-from-superconductors-and-l/ ; https://app.fundz.net/fundings/great-sky-funding-round-c20298)"
    },
    {
     "date": "2026-10-27",
     "text": "**First application announcement (2026-10-19): superconducting AI chip processing 'real tokamak data at one million samples per second'** with Brookhaven, Princeton Plasma Physics Laboratory and the DIII-D National Fusion Facility (ties to its Genesis Mission Phase I role). No product name, latency or commercial terms; Business Wire page bot-blocked, summary from greatsky.ai/news. (https://www.greatsky.ai/news ; https://www.businesswire.com/news/home/20261019055591/en/Great-Skys-High-Speed-AI-Chips-Analyze-Fusion-Plasmas)"
    }
   ],
   "sources": [
    {
     "title": "Great Sky — company site",
     "url": "https://www.greatsky.ai"
    },
    {
     "title": "Great Sky AI Emerges from NIST (company launch post)",
     "url": "https://www.greatsky.ai/news/great-sky-ai-emerges-from-nist"
    },
    {
     "title": "Pulse2: Great Sky: $14 Million Seed Raised",
     "url": "https://pulse2.com/great-sky-14-million-seed-funding/"
    },
    {
     "title": "BusinessWire: Great Sky Unveils Brain-like Computing Architecture (2026-03-13)",
     "url": "https://www.businesswire.com/news/home/20260313838393/en/Great-Sky-Unveils-Brain-like-Computing-Architecture-Built-for-the-Next-Era-of-AI"
    },
    {
     "title": "StartupHub.ai: Great Sky profile",
     "url": "https://www.startuphub.ai/startups/great-sky"
    },
    {
     "title": "FinSMEs: Great Sky Raises $14M in Seed Funding",
     "url": "https://www.finsmes.com/2026/03/great-sky-raises-14m-in-seed-funding.html"
    },
    {
     "title": "Great Sky Selected for DOE's Genesis Mission (2026-07-22)",
     "url": "https://www.greatsky.ai/news/genesis-mission"
    },
    {
     "title": "Brookhaven Lab: Complete List of Genesis Mission Phase I projects (2026-07-27)",
     "url": "https://www.bnl.gov/newsroom/news.php?a=223070"
    },
    {
     "title": "EE Times podcast: Can Great Sky's Light-Connected AI Enable Bigger Brains? (2026-07-22)",
     "url": "https://www.eetimes.com/podcasts/can-great-skys-light-connected-ai-enable-bigger-brains/"
    }
   ]
  },
  {
   "slug": "houmo-ai",
   "name": "Houmo.ai (后摩智能)",
   "founded": 2020,
   "hq": "Nanjing / Shanghai, China",
   "category": "analog-imc",
   "status": "active",
   "origin": "corporate alumni (Horizon Robotics CTO; Princeton PhD)",
   "total_funding_usd_m": 430,
   "last_round": "Series B, 数十亿元 (several billion CNY, exact amount and investors undisclosed), reported 2026-11-04",
   "key_investors": [
    "Sequoia China (红杉中国)",
    "Qiming Venture Partners (启明创投)",
    "Matrix Partners China (经纬)",
    "Lenovo Capital (联想创投)",
    "Jinpu Yueda Auto Fund",
    "China Mobile industry funds"
   ],
   "website": "https://www.houmoai.com",
   "pitch": "Houmo argues AI compute demand has outrun von Neumann chips, where shuttling data between\nmemory and compute dominates energy (\"memory wall / power wall\"); its 存算一体\n(compute-in-memory) architecture computes where the data sits, promising tens-of-times\nbetter TOPS/W for large-model inference at the edge — CEO Wu Qiang predicts \"90% of data\nprocessing may happen at the device/edge\" (https://www.houmoai.com/AboutUs.html;\nhttps://www.guancha.cn/economy/2025_07_30_784867.shtml).",
   "technical_approach": "- **Device/substrate:** **SRAM-based, fully digital compute-in-memory** on standard CMOS\n  (TSMC 12nm for H30) — conventional memory, digital CIM macros, no emerging devices.\n- **鸿途 H30** (released 2023-05-10): billed as China's first large-compute CIM chip for\n  autonomous driving; up to 256 TOPS at 35W typical\n  (https://www.eet-china.com/news/202305111167.html). A successor H50 was announced as in\n  development at launch (https://www.pingwest.com/w/281952).\n- **漫界 M30** (2024): edge/on-device LLM chip; strategic cooperation with Lenovo for AI\n  PC / NAS use (https://capital.lenovo.com/news/detail/id/1012/s/1.html).\n- **漫界 M50** (released 2025-07-25 at WAIC 2025): next-gen edge LLM inference chip, with\n  力擎 M.2 cards, 力谋 accelerator cards and compute boxes\n  (https://zhuanlan.zhihu.com/p/1932446472898584601).\n- **Inference-only** accelerators; training off-chip.",
   "funding_rounds": [
    {
     "date": "2020-09 (announced 2021-03)",
     "round": "Angel",
     "amount": "tens of $M (\"数千万美元\")",
     "investors": "Sequoia China lead; 经纬中国, 联想创投, 弘毅创投",
     "source": "https://startup.aliyun.com/info/104342.html; https://www.houmoai.com/AboutUs.html"
    },
    {
     "date": "2022-01",
     "round": "Pre-A",
     "amount": "CNY 300M (~$47M)",
     "investors": "Qiming Venture Partners lead; Matrix China add-on; MSA Capital (和玉) et al.",
     "source": "https://m.cyzone.cn/article/647501"
    },
    {
     "date": "2022-04",
     "round": "Pre-A+",
     "amount": "several CNY 100Ms",
     "investors": "经纬创投 + 金浦悦达汽车基金 co-lead; Qiming, Lenovo Capital, MSA follow",
     "source": "https://d1ev.com/news/qiye/173165; https://www.iyiou.com/company/houmozhineng"
    },
    {
     "date": "2024-07",
     "round": "Strategic",
     "amount": "several CNY 100Ms",
     "investors": "China Mobile funds (北京中移数字新经济基金, 上海中移数字转型基金)",
     "source": "https://www.eet-china.com/info/72178.html"
    }
   ],
   "funding_notes": "Cumulative funding reported as \"十余亿元\" (**over CNY 1B, ~$140–150M** at ~7 CNY/USD) by\n2026 (https://zhuanlan.zhihu.com/p/2017900893715073015). Treat as **partial** — several\nround amounts are ranges (\"数亿元\") and trackers under-report; earlier third-party\nestimates put the total at ~$109M+.",
   "trajectory": "Founded Nov 2020 in Nanjing by **Wu Qiang (吴强)**: Southeast Univ. BS, Princeton CS PhD\n(2006; thesis on energy-efficient compute chips/compilers, MICRO best paper), then Intel,\nAMD, Facebook, and CTO of Horizon Robotics before leaving to start Houmo\n(https://m.36kr.com/p/1124929144336898; https://baike.baidu.com/item/吴强/60960628).\nTeam drawn from Nvidia, TI, AMD, Intel, HiSilicon, Horizon. Trajectory: first CIM\nlarge-compute chip lit up 2022; H30 for smart driving 2023; then a visible **pivot from\nautonomous driving to edge/on-device LLM compute** (M30, M50) — press noted a quiet period\nin driving before the WAIC 2025 re-emergence\n(https://www.leiphone.com/category/chips/r2pEdaGO0tftSPTZ.html). As of mid-2026: active;\nnamed a 2025–2026 \"quasi-unicorn TOP50\"; disclosed intended customers include Lenovo's\nnext-gen AI PC and iFlytek (https://www.guancha.cn/economy/2025_07_30_784867.shtml).",
   "relevance": "Houmo shows the pragmatic end of the CIM spectrum: digital SRAM in-memory computing on a\nfoundry CMOS process — no analog devices, no on-chip learning — trading the density/energy\nceiling of emerging-device approaches (memristor/ReRAM) for manufacturability and fast\ntime-to-market at 256-TOPS scale.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Houmo's **M50 CIM chip shipped inside Lenovo's AI host P7** (\"global first launch\", 2026-05-21) claiming 120B-parameter-class local LLM support, and multiple M50-based AI terminals showed at WAIC 2026 (2026-07-19/20) — the M50 moving from launch to commercial deployments. (https://www.leiphone.com/category/chips/aUvdVj39lWtZWgJL.html ; https://www.guancha.cn/economy/2026_07_20_824409.shtml)"
    },
    {
     "date": "2026-09-22",
     "text": "Two more **M50 design wins**: (1) the **FN008 \"fully domestic\" AI laptop** from 国科亿道 (Guoke Yidao) pairs a Phytium D3000M CPU with Houmo's M50 dNPU (160 TOPS INT8 / 100 TFLOPS bFP16, ~10 W, 24 GB, runs up to ~35B-parameter models locally on KylinOS), announced 2026-09-03 (https://tech.ifeng.com/c/8w8p4wQDfUP); (2) **UGREEN HomeAgent HA100 Pro**, a home AI hub/NAS built around the M50, launched at UGREEN's global event 2026-09-12 (https://www.leiphone.com/category/chips/OWi6OEGUYbDYwZof.html). Consumer and PC OEM adoption of the CIM chip continues to broaden beyond the Lenovo P7."
    },
    {
     "date": "2026-10-27",
     "text": "**First ecosystem-partner conference (2026-10-14): dual roadmap on an SRAM-CIM root** — '2D CIM' (SRAM-CIM + 2D packaging, capacity-driven loads: M30/M50 in production, **M6X** successor in development at 3–6x M50) and '3D CIM' (SRAM-CIM as the logic die under stacked memory, bandwidth-driven dense models: **first 3D-CIM chip A20 'now in production'**, up to 16-die full interconnect, +50% utilization). M50 claims: **150+ design wins, 100+ customers** in under a year; AI-core area −3/5 and power −4/5 vs conventional. Co-creation program launched with Tencent Hunyuan, SenseTime et al. Lenovo AI Host P7 (300 g, up to 122B-parameter local model, 190 TOPS, 128K context, ~50 tok/s offline) re-previewed with early access (Pandaily 2026-10-25). Earlier (2026-09-22) Sina item confirms the next-gen edge-LLM chip uses the 3D-CIM architecture. (https://m.zhidx.com/p/601617.html ; https://finance.sina.com.cn/tech/digi/2026-09-22/doc-inistrnv5537036.shtml ; https://pandaily.com/lenovo-ai-host-p7-preview-houmo-m50-cix-p1-local-100b-models-early-access)"
    },
    {
     "date": "2026-11-09",
     "text": "**Series B of '数十亿元' (several billion CNY), 'recently completed' per reports of 2026-11-04** (科创板日报 / 投资界 / DoNews; sourced to 知情人士, company confirmed the use of funds). Exact amount, lead and investor names not disclosed: 'upstream/downstream industrial capital, state-owned capital and international investors, with several existing shareholders adding'. Use of funds: scale M50 delivery, develop the M6X and next-gen 3D-CIM (A20 line), software toolchain. Restated: M50 (160 TOPS INT8, 150 GB/s, ~10 W) in mass production <1 year with 150+ design wins and 100+ customers. **Working total raised to ~US$430M † as a floor** (prior ~150 + ≥CNY 2B ≈ US$280M for 数十亿); partial by construction. Also 2026-11-04: Lenovo Kaitian AI host pairs a Zhaoxin KX-7000 CPU with Houmo's LQ50 NPU for local 35B models (Pandaily, headline only). (https://news.pedaily.cn/202611/571313.shtml ; https://www.donews.com/news/detail/1/6764338.html ; https://pandaily.com/lenovo-kaitian-ai-host-zhaoxin-kx-7000-houmo-lq50-npu-160-tops-local-35b-qwen)"
    }
   ],
   "sources": [
    {
     "title": "Houmo About Us",
     "url": "https://www.houmoai.com/AboutUs.html"
    },
    {
     "title": "EET-China: H30 launch",
     "url": "https://www.eet-china.com/news/202305111167.html"
    },
    {
     "title": "Zhihu/company PR: M50 launch WAIC 2025",
     "url": "https://zhuanlan.zhihu.com/p/1932446472898584601"
    },
    {
     "title": "Aliyun/36kr: angel round",
     "url": "https://startup.aliyun.com/info/104342.html"
    },
    {
     "title": "Cyzone: Pre-A CNY 300M",
     "url": "https://m.cyzone.cn/article/647501"
    },
    {
     "title": "d1ev: Pre-A+ round",
     "url": "https://d1ev.com/news/qiye/173165"
    },
    {
     "title": "EET-China: China Mobile strategic round",
     "url": "https://www.eet-china.com/info/72178.html"
    },
    {
     "title": "Guancha interview with Wu Qiang",
     "url": "https://www.guancha.cn/economy/2025_07_30_784867.shtml"
    },
    {
     "title": "Zhihu: company profile, cumulative funding",
     "url": "https://zhuanlan.zhihu.com/p/2017900893715073015"
    },
    {
     "title": "Leiphone WAIC 2025 analysis",
     "url": "https://www.leiphone.com/category/chips/r2pEdaGO0tftSPTZ.html"
    }
   ]
  },
  {
   "slug": "ihw",
   "name": "iHW (Intelligent HW)",
   "founded": 2024,
   "hq": "South Korea",
   "category": "analog-imc",
   "status": "active",
   "origin": "corporate alumni (CEO Kim Tae-hoon — ex-SK Hynix VP, ex-Samsung Electronics, ex-Micron; Arizona State PhD; team of Samsung/SK Hynix/Intel alumni)",
   "total_funding_usd_m": 46,
   "last_round": "Series A, ₩52B (~$35.5–37.5M), 2026-08-19",
   "key_investors": [
    "Korea Development Bank",
    "Kolon Investment",
    "LIG Defense & Aerospace + IBK Capital",
    "Woori Venture Partners",
    "Mirae Asset Venture Investment",
    "DSC Investment",
    "HB Investment"
   ],
   "website": "",
   "pitch": "\"Computing directly on data\": large amounts of weight data are stored directly inside the chip\nand processed where they are stored, minimizing data movement — enabling AI inference without\nexternal DRAM, heat sinks, or cooling (The Elec / edaily coverage of the Series A,\nhttps://www.thelec.net/news/articleView.html?idxno=13188;\nhttps://en.edaily.co.kr/news/eda202608225071/).",
   "technical_approach": "Ultra-low-power **Analog Compute-in-Memory (ACiM)** chip, **Infertron**, described by The Elec\nas applying \"neuromorphic computing technology to its computing structure.\" Weights stored in\nnonvolatile embedded flash (no external DRAM/NAND); mature 28–40nm process nodes; fanless\noperation. Fabless model; performance goal stated as comparable to Nvidia and Hailo edge\nproducts. Target: embedded/on-device AI first, with an LLM-chip prototype on the roadmap\n(https://www.thelec.net/news/articleView.html?idxno=13188). Dealroom summary: \"running\ncalculations where weights are stored rather than moving data to a separate circuit\"\n(https://dealroom.co/news/145707-ihw-raises-37m-series-a-for-ultra-low-power-ai-chips/).\nInference accelerator — no on-chip learning claim found.",
   "funding_rounds": [
    {
     "date": "late 2024",
     "round": "Pre-Series A",
     "amount": "₩12B",
     "investors": "undisclosed",
     "source": "https://www.thelec.net/news/articleView.html?idxno=13188"
    },
    {
     "date": "~2025",
     "round": "Government funding",
     "amount": "~$10M",
     "investors": "Korean government programs",
     "source": "https://www.thelec.net/news/articleView.html?idxno=13188"
    },
    {
     "date": "2026-08-19",
     "round": "Series A",
     "amount": "₩52B (~$35.5M–$37.5M)",
     "investors": "13 investors: Korea Development Bank (new), LIG Defense & Aerospace + IBK Capital consortium; follow-ons Kolon Investment, Woori Venture Partners, Mirae Asset Venture Investment, DSC Investment, HB Investment",
     "source": "https://en.sedaily.com/finance/2026/08/19/low-power-ai-chip-startup-ihw-raises-52-billion-won-in ; https://www.thelec.net/news/articleView.html?idxno=13188"
    }
   ],
   "funding_notes": "Cumulative equity investment ₩64B (~$46M) per The Elec/Seoul Economic Daily; USD conversions\nvary by outlet (Wowtale $35.5M \"led by Kolon Investment\", sedaily $37.5M for the Series A —\nreported as-is). Series A funds: mass production of edge AI chips, LLM-chip prototype,\nmarketing, hiring.",
   "trajectory": "Founded June 2024 (The Elec), CEO Kim Tae-hoon; South Korean fabless AI-semiconductor company\n\"pioneering Analog Compute-in-Memory (ACiM) technology\" (Dealroom company page,\nhttps://uzbekistan.dealroom.co/companies/intelligent_hw). Unusually fast capital accumulation\nfor an analog-CIM startup (₩64B within ~26 months of founding), with defense-adjacent investors\n(LIG D&A) suggesting a defense/edge wedge. Status: active, pre-mass-production.\n\n**Founders (sweep 2026-09-22):** Kim Tae-hoon (founder & CEO; born 1970; engineering PhD, Arizona State University; worked at Samsung Electronics, Micron Technology and SK Hynix — ten years as an SK Hynix vice-president leading AI memory, per Hankook Ilbo). The 2024 founding team is described as Samsung, SK Hynix and Intel alumni (https://www.hankookilbo.com/news/article/A2026081915440002141 ; https://www.mt.co.kr/future/2026/08/19/2026081909164378071 ; https://dealroom.co/news/145707-ihw-raises-37m-series-a-for-ultra-low-power-ai-chips/).",
   "relevance": "A 2024-vintage confirmation that the analog in-memory inference thesis (Mythic, Sagence,\nEnCharge territory) is still raising large rounds — here on embedded flash rather than\nmemristors, and squarely inference-only; contrast with Knowm's on-chip-learning ambition.",
   "news": [
    {
     "date": "2026-10-27",
     "text": "**Defense MOU (2026-10-06) with Qunion and Quniworks** to apply the sub-1 W INFERTRON chip to drones/unmanned systems; iHW supplies silicon + tools, Qunion integrates/tests, Quniworks the embedded platform. First post-Series-A news. (https://www.mk.co.kr/en/business/12170491)"
    }
   ],
   "sources": [
    {
     "title": "Seoul Economic Daily: Low-Power AI Chip Startup iHW Raises 52 Billion Won (2026-08-19)",
     "url": "https://en.sedaily.com/finance/2026/08/19/low-power-ai-chip-startup-ihw-raises-52-billion-won-in"
    },
    {
     "title": "The Elec: iHW Raises 52 Billion Won in Series A Funding",
     "url": "https://www.thelec.net/news/articleView.html?idxno=13188"
    },
    {
     "title": "edaily: \"Computing Directly on Data\" — VC's Pick (2026-08-22)",
     "url": "https://en.edaily.co.kr/news/eda202608225071/"
    },
    {
     "title": "Dealroom news: iHW raises $37M Series A",
     "url": "https://dealroom.co/news/145707-ihw-raises-37m-series-a-for-ultra-low-power-ai-chips/"
    },
    {
     "title": "Dealroom company page: Intelligent HW",
     "url": "https://uzbekistan.dealroom.co/companies/intelligent_hw"
    },
    {
     "title": "Wowtale: iHW Secures $35M Series A led by Kolon Investment",
     "url": "https://en.wowtale.net/2026/08/19/234780/"
    }
   ]
  },
  {
   "slug": "inivation",
   "name": "iniVation",
   "founded": 2015,
   "hq": "Zurich, Switzerland",
   "category": "event-sensing",
   "status": "acquired",
   "origin": "academic spinoff (Institute of Neuroinformatics, University of Zurich / ETH Zurich)",
   "total_funding_usd_m": 0,
   "last_round": "Acquired by SynSense (100% shares), Feb 2024 — deal terms undisclosed",
   "key_investors": [
    "Samsung Catalyst Fund"
   ],
   "website": "https://inivation.com",
   "pitch": "iniVation frames conventional frame-based cameras as fundamentally inefficient: they capture and\ntransmit redundant full frames at fixed rates, blowing power and bandwidth and suffering from motion\nblur and limited dynamic range. Its Dynamic Vision Sensors (DVS) instead have pixels that respond\nasynchronously and independently only to changes in the scene, \"the way biological retinas do,\"\ndelivering microsecond latency, high dynamic range, and sparse data. (https://inivation.com/technology/)\nNote: event-based *sensing*, not a compute substrate.",
   "technical_approach": "Neuromorphic event/dynamic-vision sensors descended directly from the original academic DVS work of\nTobi Delbruck's group at the Institute of Neuroinformatics. Products: **DVS** camera families\n(DAVIS346, DVXplorer), the DV software stack, and the **Aeveon** \"Adaptive Event Vision\" sensor\nlaunched March 2025 — adding multi-bit change events, full-frame readout, and region-based HDR beyond\nclassic binary DVS. Pixels asynchronously emit ON/OFF brightness-change events; sensing only, paired\nwith conventional or (post-merger) SynSense neuromorphic processors. (https://inivation.com/dvs/;\nhttps://inivation.com/company/news/)",
   "funding_rounds": [
    {
     "date": "~2015–2018",
     "round": "Seed",
     "amount": "undisclosed",
     "investors": "Samsung Catalyst Fund",
     "source": "Crunchbase / grokipedia"
    },
    {
     "date": "2018-01",
     "round": "— (asset consolidation)",
     "amount": "n/a",
     "investors": "absorbed DVS business & team of iniLabs GmbH",
     "source": "https://inivation.com/inivation-incorporates-dynamic-vision-sensor-business-from-inilabs/"
    },
    {
     "date": "2024-02",
     "round": "Acquisition",
     "amount": "undisclosed",
     "investors": "acquired 100% by SynSense (forms SynSense Group)",
     "source": "https://www.synsense.ai/synsense-and-inivation-join-forces-to-form-leading-neuromorphic-technology-provider/"
    }
   ],
   "funding_notes": "Total disclosed venture funding is essentially undisclosed; no priced rounds publicly reported beyond\nseed backing. Acquisition terms not disclosed.",
   "trajectory": "Founded 2015 in Zurich as a spinoff from the Institute of Neuroinformatics (UZH/ETH Zurich) by Tobi\nDelbruck, Rodney Douglas, Kynan Eng and Sven-Erik Jacobsen — commercializing decades of academic DVS\nresearch. In January 2018 iniVation incorporated the DVS business and engineering team of sister\ncompany **iniLabs GmbH**, with Kynan Eng becoming CEO. In **February 2024, SynSense acquired 100% of\niniVation**, forming the **SynSense Group** and marketing itself as the first fully neuromorphic\nend-to-end sensing-plus-processing company; the Zurich entity continues to operate under the group.\n(This confirms the merger/acquisition question in the task: SynSense acquired iniVation, not a\nmerger of equals, though both parties framed it as \"joining forces.\") Post-merger the Aeveon sensor\nshipped in 2025. Active as a subsidiary of SynSense Group as of mid-2026.\n(https://www.startupticker.ch/en/news/synsense-acquires-inivation-to-boost-its-market-position)",
   "relevance": "Like Prophesee, iniVation is the sensing side of the neuromorphic stack — no compute substrate or\non-chip learning of its own — but its post-2024 tie-up with SynSense (a digital SNN inference\nprocessor) makes the combined group a full end-to-end neuromorphic vision pipeline, a different\nintegration path than Knowm's memristor/AHaH compute-and-learn substrate.",
   "news": [],
   "sources": [
    {
     "title": "iniVation technology",
     "url": "https://inivation.com/technology/"
    },
    {
     "title": "iniVation incorporates DVS business from iniLabs",
     "url": "https://inivation.com/inivation-incorporates-dynamic-vision-sensor-business-from-inilabs/"
    },
    {
     "title": "SynSense and iniVation join forces",
     "url": "https://www.synsense.ai/synsense-and-inivation-join-forces-to-form-leading-neuromorphic-technology-provider/"
    },
    {
     "title": "SynSense acquires iniVation (startupticker)",
     "url": "https://www.startupticker.ch/en/news/synsense-acquires-inivation-to-boost-its-market-position"
    },
    {
     "title": "iniVation news (Aeveon)",
     "url": "https://inivation.com/company/news/"
    }
   ]
  },
  {
   "slug": "innatera",
   "name": "Innatera Nanosystems",
   "founded": 2018,
   "hq": "Delft, Netherlands (offices also in Bengaluru, India)",
   "category": "digital-snn",
   "status": "active",
   "origin": "academic spinoff (TU Delft — Sumeet Kumar et al.)",
   "total_funding_usd_m": "conflicting: ~27 disclosed (seed ~$6M + $21M Series A); Tracxn $43.3M; PitchBook ~$32.2M (per 2026-07 report); CEO-cited ~$55M post-Series-B (unconfirmed)",
   "last_round": "Series A (oversubscribed extension), $21M cumulative, 2024-06",
   "key_investors": [
    "Invest-NL Deep Tech Fund",
    "EIC Fund",
    "MIG Capital",
    "Matterwave Ventures",
    "btov Partners"
   ],
   "website": "https://www.innatera.com",
   "pitch": "\"Innatera's Spiking Neural Processor architecture delivers up to 500x lower energy than\nconventional edge AI, enabling truly always-on sensing in battery devices\"\n(https://www.innatera.com/technology/). Seed investor btov framed it as a \"paradigm shift at\nthe very edge\": 10,000x higher performance/watt vs conventional digital alternatives\n(https://medium.com/brains-to-ventures/paradigm-shift-at-the-very-edge-4483c26b592b).",
   "technical_approach": "- Mixed-signal/digital **spiking neural processor** for the \"sensor edge\": analog-mixed-signal\n  SNN compute fabric plus conventional cores; inference-focused (training off-chip via their\n  Talamo SDK). Sub-milliwatt always-on pattern recognition\n  (https://www.innatera.com/technology/, https://www.reddit.com/r/RISCV/comments/1kvlkz6/innatera_t1_neural_processor/).\n- **Spiking Neural Processor T1** — unveiled CES 2024; single-chip SNN + RISC-V MCU for\n  always-on sensing (https://www.innatera.com/about-us/).\n- **Pulsar** — launched 21 May 2025, marketed as \"the world's first mass-market neuromorphic\n  microcontroller\"; Best-in-Show at COMPUTEX 2025 (Microcontrollers/Microprocessors & IP category)\n  (https://finance.yahoo.com/news/innatera-unveils-pulsar-worlds-first-125900736.html,\n  https://www.electronicspecifier.com/news/awards/innatera-s-pulsar-wins-best-in-show-at-computex-2025/).\n- CES 2026: showed commercial deployments across smart home, industrial IoT, wearables\n  (https://www.innatera.com/newsroom/redefining-the-cutting-edge-innatera-debuts-real-world-neuromorphic-edge-ai-at-ces-2026/).",
   "funding_rounds": [
    {
     "date": "2020-11",
     "round": "Seed",
     "amount": "~$6M (~€5M)",
     "investors": "MIG Capital, btov Partners",
     "source": "https://xpu.pub/2025/07/22/innatera-pulsar/ ; https://www.mig.ag/en/cases/innatera-en/ ; https://medium.com/brains-to-ventures/paradigm-shift-at-the-very-edge-4483c26b592b"
    },
    {
     "date": "2024-03",
     "round": "Series A",
     "amount": "€15M (~$16M)",
     "investors": "Invest-NL Deep Tech Fund, EIC Fund, MIG Capital, Matterwave Ventures",
     "source": "https://www.prnewswire.com/news-releases/dutch-innatera-raises-15-million-for-mass-production-of-processor-that-mimics-human-brain-302092079.html"
    },
    {
     "date": "2024-06",
     "round": "Series A extension (oversubscribed)",
     "amount": "+$5M → $21M cumulative",
     "investors": "new investors (unnamed in PR)",
     "source": "https://www.datacenterdynamics.com/en/news/neuromorphic-processor-startup-innatera-raises-21m/ ; https://ec.europa.eu/newsroom/eismea/items/837657/ ; https://www.electronicsweekly.com/news/business/innatera-raises-21m-for-neuromorphic-processing-2024-07/"
    }
   ],
   "funding_notes": "**Total raised:** ~$27M disclosed (seed + Series A). **Conflict:** Tracxn reports \"$43.3M over\n4 rounds\" (https://tracxn.com/d/companies/innatera/__NYtziF1PlRrQFkdKkc6aXEaWjRg3qMPh2iGldZ-a21U);\nno press release for the difference was found — treat $27M as the sourced floor.\nNow a multi-way conflict: per the 2026-07 who's-who research report\n(_research/2026-07_whos_who_deep_research.md), unverified, PitchBook lists ~$32.2M, and a\n~$55M post-Series-B total cited by the CEO could not be independently confirmed. Record all\nfour figures side by side; $27M remains the sourced floor.\nEU support: EIC Fund equity participation (Series A) per PRNewswire above.",
   "trajectory": "Spun out of TU Delft in 2018; CEO Sumeet Kumar (named Fellow of the Netherlands Academy of\nEngineering, https://www.innatera.com/newsroom/innatera-ceo-sumeet-kumar-named-fellow-of-the-netherlands-academy-of-engineering/).\nProgression: research chips → T1 (CES 2024/2025) → Pulsar mass-market MCU (May 2025) →\ncustomer deployments shown at CES 2026. As of mid-2026: **active**, in early commercial ramp;\npartnerships include VLSI EXPERT training deployments of Pulsar\n(https://siliconsemiconductor.net/article/123026/VLSI_EXPERT_selects_Innatera_Spiking_Neural_Processors_to_build_industry-led_neuromorphic_talent_pool).\nMission line: \"a billion sensors intelligent by 2030\" (https://www.innatera.com/about-us/).\nPer the 2026-07 who's-who research report (_research/2026-07_whos_who_deep_research.md),\nunverified: ~110 employees and pursuing a Series B.",
   "relevance": "Innatera monetizes spiking dynamics in mixed-signal CMOS for always-on sensing — the same\nultra-low-power wedge Knowm's AHaH/memristor work targets — but with conventional silicon\nmemory and off-chip training, not adaptive physical synapses or on-chip learning.",
   "news": [],
   "sources": [
    {
     "title": "Innatera technology page",
     "url": "https://www.innatera.com/technology/"
    },
    {
     "title": "Innatera about us",
     "url": "https://www.innatera.com/about-us/"
    },
    {
     "title": "PRNewswire €15M Series A (Mar 2024)",
     "url": "https://www.prnewswire.com/news-releases/dutch-innatera-raises-15-million-for-mass-production-of-processor-that-mimics-human-brain-302092079.html"
    },
    {
     "title": "DCD: Innatera raises $21m (Jul 2024)",
     "url": "https://www.datacenterdynamics.com/en/news/neuromorphic-processor-startup-innatera-raises-21m/"
    },
    {
     "title": "Pulsar launch (May 2025)",
     "url": "https://finance.yahoo.com/news/innatera-unveils-pulsar-worlds-first-125900736.html"
    },
    {
     "title": "CES 2026 newsroom post (Dec 2025)",
     "url": "https://www.innatera.com/newsroom/redefining-the-cutting-edge-innatera-debuts-real-world-neuromorphic-edge-ai-at-ces-2026/"
    },
    {
     "title": "Tracxn Innatera profile",
     "url": "https://tracxn.com/d/companies/innatera/__NYtziF1PlRrQFkdKkc6aXEaWjRg3qMPh2iGldZ-a21U"
    }
   ]
  },
  {
   "slug": "intrinsic-semiconductor",
   "name": "Intrinsic Semiconductor Technologies Ltd (branding also \"Intrinsic AI Solutions\")",
   "founded": 2017,
   "hq": "London, UK",
   "category": "emerging-device-imc (ReRAM/embedded NVM IP for edge AI)",
   "status": "active",
   "origin": "academic spinoff (UCL — Prof. Tony Kenyon & Dr. Adnan Mehonic, Electronic & Electrical Engineering)",
   "total_funding_usd_m": 11.6,
   "last_round": "£7M equity + £1M Innovate UK grants, 2023-03, led by Octopus Ventures",
   "key_investors": [
    "Octopus Ventures",
    "IP Group",
    "UCL Technology Fund",
    "Innovate UK (grants)"
   ],
   "website": "https://www.intrinsicsemi.com",
   "pitch": "Intrinsic frames a \"memory bottleneck for data-hungry applications\": putting \"efficient\nnon-volatile memory directly on advanced processor chips\" removes the power and latency\ncost of external flash for embedded and edge-AI systems\n(https://www.intrinsicsemi.com/;\nhttps://www.semiconductor-digest.com/intrinsic-semiconductor-technologies-secures-7m-investment-to-solve-the-memory-bottleneck-for-data-hungry-applications/).\nCEO Mark Dickinson: \"RRAM has the potential to become the backbone for the next generation\nof edge and IoT computers\" (https://www.theregister.com/2023/03/13/ucl_spinout_reram_funding/).",
   "technical_approach": "- Device: **SiOx (silicon-oxide) ReRAM** — the Kenyon/Mehonic UCL device line; standard\n  CMOS-friendly materials, BEOL-integrable on logic\n  (https://www.uclb.com/portfolio/our-spinouts/intrinsic-semiconductor-technologies-ltd/).\n- Claims: reads 10–100x faster and writes 1,000x faster than existing embedded NVM\n  solutions (https://www.theregister.com/2023/03/13/ucl_spinout_reram_funding/).\n- Business model: embedded NVM IP for next-gen microcontrollers and **edge-AI processors**\n  (weights-on-die); collaboration with sureCore on RRAM-based NVM macros\n  (https://www.intrinsicsemi.com/). Digital memory IP — no analog IMC product or on-chip\n  learning claims; current branding (\"Intrinsic AI Solutions\") leans toward the edge-AI\n  memory use case.\n- Academic note (pointer only): Mehonic & Kenyon are also authors of the widely cited\n  \"Memristors — from in-memory computing... \" perspective literature; the company itself\n  sells memory IP.",
   "funding_rounds": [
    {
     "date": "2021-03",
     "round": "Seed",
     "amount": "£1.35M",
     "investors": "UCL Technology Fund, IP Group",
     "source": "https://www.uktech.news/ai/ucl-spin-out-intrinsic-closes-1-35m-seed-funding-to-develop-next-gen-memory-devices-20210322 ; https://www.ucltf.co.uk/insight/intrinsic-closes-135m-seed-funding-round-prototype-next-generation-memory-devices"
    },
    {
     "date": "2023-03",
     "round": "Venture round (~$8.5M)",
     "amount": "£7M equity + £1M Innovate UK grants",
     "investors": "Octopus Ventures (lead); IP Group, UCL Technology Fund",
     "source": "https://www.theregister.com/2023/03/13/ucl_spinout_reram_funding/ ; https://www.uktech.news/deep-tech/intrinsic-receives-7m-20230313"
    }
   ],
   "funding_notes": "Total ≈ **$11.6M** across rounds (CB Insights,\nhttps://www.cbinsights.com/company/intrinsic-semiconductor-technologies). No 2024–2026\nround found in public sources as of 2026-07-23; an Albion VC page referencing the £7M\nraise suggests possible additional investor involvement, but no separate later round could\nbe confirmed (https://albion.vc/intrinsic-semiconductor-technologies-secures-7m/ — page\nnow 404).",
   "trajectory": "Spun out of UCL (incorporated December 2017, Companies House 11102458:\nhttps://find-and-update.company-information.service.gov.uk/company/11102458) to\ncommercialize Kenyon & Mehonic's SiOx RRAM; professional CEO Mark Dickinson brought in.\nTrajectory: seed (2021) → nanometer-scale device demonstrations → £7M round (2023) to\ngrow the engineering team toward market → sureCore macro collaboration; website rebrand\ntoward \"Intrinsic AI Solutions.\" Status mid-2026: active, pre-revenue IP developer;\nnotably smaller war chest than peers (Weebit has raised ~20x more for the same SiOx\ndevice class). Gartner analysts have publicly questioned the embedded-RRAM-vs-stacked-die\ntradeoff (https://www.theregister.com/2023/03/13/ucl_spinout_reram_funding/).",
   "relevance": "Second independent SiOx ReRAM commercialization (after Weebit), from the UK's leading\nacademic memristor group — evidence that university memristor labs converge on embedded\nmemory IP, not learning hardware, as the fundable first market.",
   "news": [],
   "sources": [
    {
     "title": "The Register — UCL spinout raises £7M (2023-03-13)",
     "url": "https://www.theregister.com/2023/03/13/ucl_spinout_reram_funding/"
    },
    {
     "title": "UKTN — £1.35M seed (2021-03)",
     "url": "https://www.uktech.news/ai/ucl-spin-out-intrinsic-closes-1-35m-seed-funding-to-develop-next-gen-memory-devices-20210322"
    },
    {
     "title": "UCLB spinout profile — founders",
     "url": "https://www.uclb.com/portfolio/our-spinouts/intrinsic-semiconductor-technologies-ltd/"
    },
    {
     "title": "Intrinsic website",
     "url": "https://www.intrinsicsemi.com/"
    },
    {
     "title": "CB Insights — total raised $11.6M",
     "url": "https://www.cbinsights.com/company/intrinsic-semiconductor-technologies"
    },
    {
     "title": "Companies House 11102458",
     "url": "https://find-and-update.company-information.service.gov.uk/company/11102458"
    }
   ]
  },
  {
   "slug": "koniku",
   "name": "Koniku",
   "founded": 2015,
   "hq": "San Rafael, California, USA",
   "category": "biological",
   "status": "active",
   "origin": "student startup (Osh Agabi — founded during his Imperial College London bioengineering PhD; IndieBio accelerator alumnus)",
   "total_funding_usd_m": 1.65,
   "last_round": "Undisclosed round (Presight Capital, SoftBank, Platform Capital, others) — amount not disclosed",
   "key_investors": [
    "IndieBio",
    "Plug and Play",
    "Platform Capital",
    "Presight Capital",
    "SoftBank",
    "IDO Investments"
   ],
   "website": "https://koniku.com",
   "pitch": "Koniku's premise is that biology already solves problems — especially olfaction/chemical sensing —\nthat silicon cannot do at comparable size, power, or sensitivity. Rather than simulate neurons, it\nuses *actual* living neurons as the processing element, arguing that \"biology is technology\" and that\nwetware can sense and compute in ways conventional electronics can't. (https://koniku.com;\nhttps://en.wikipedia.org/wiki/Koniku)",
   "technical_approach": "\"Wetware\" neuron-silicon hybrid. The **Konikore** device fuses engineered living biological neurons\n(reported as mouse stem-cell-derived) expressing **olfactory receptors** with a silicon substrate that\nreads and writes neuronal activity via electrodes. The neurons detect and classify volatile organic\ncompounds (VOCs) — explosives, narcotics such as fentanyl and amphetamines, and disease markers — in\neffect a programmable electronic \"nose.\" The device is reported at <600 g, comprising a\nneuron-silicon core, chemical sensors, and read/write electrodes. This is genuine biological\ncomputation/sensing, not a neuromorphic-silicon emulation. (https://www.whipsaw.com/work/konikore-bioelectric-sensor;\nCNN, TEDGlobal coverage)",
   "funding_rounds": [
    {
     "date": "~2015–2017",
     "round": "Seed / accelerator",
     "amount": "~$1.65M (cumulative reported)",
     "investors": "IndieBio, Plug and Play, Platform Capital",
     "source": "Tracxn / Crunchbase"
    },
    {
     "date": "2017",
     "round": "Commercial contract",
     "amount": "$8M contract (not equity)",
     "investors": "Airbus (threat-detection development)",
     "source": "CNN / Airbus"
    },
    {
     "date": "recent (undisclosed)",
     "round": "Growth round",
     "amount": "undisclosed",
     "investors": "Presight Capital, SoftBank, IDO Investments, Platform Capital, Andon Okapi (Oman), Flying Doctors Healthcare",
     "source": "CB Insights / company"
    }
   ],
   "funding_notes": "Only ~$1.65M of equity funding is publicly quantified; later rounds and the Airbus relationship are\ndisclosed by name but not by amount. Total raised is effectively undisclosed.",
   "trajectory": "Founded 2015 in the San Francisco Bay Area (San Rafael, CA) by Nigerian-born physicist/computational\nneuroscientist **Osh (Oshiorenoya) Agabi** — with co-founders reported as Christopher McAndrew and\nChristopher Hedvall — out of the IndieBio accelerator. Agabi's 2017 TEDGlobal talk in Tanzania put the\n\"smell cyborg\" on the map. Koniku partnered with **Airbus** from 2017 on aviation security, expanding\nthe partnership in 2020/2022, with field trials of bomb/threat-sniffing devices reported at Changi\n(Singapore) and San Francisco International airports. In 2025 Koniku became a founding member of the\nOracle Defense Ecosystem. The company remains small and R&D/defense-oriented; active as of mid-2026,\nthough revenue and headcount are not publicly disclosed.\n(https://tsi-mag.com/koniku-and-airbus-expand-partnership-to-re-imagine-aviation-security/)\n\n**Founders (sweep 2026-09-22):** Oshiorenoya \"Osh\" Agabi (founder, CEO & CTO; PhD candidate in Bioengineering, Imperial College London, Apr 2011–Oct 2014, supervised by Dr Simon Schultz, visitor in Vincenzo De Paola's MRC group). Imperial's 2015 news piece calls Koniku \"a startup company he founded during his time at Imperial\" — then a drug-trial modelling platform — with team members Chris McAndrew (BS ME Tulane) and Christina Salys; Tracxn also lists Christopher Hedvall. Student startup, not a lab spinout (https://www.imperial.ac.uk/news/165372/imperial-student-start-up-tackles-debilitating-diseases/ ; https://www.linkedin.com/in/agabi ; https://tracxn.com/d/companies/koniku-inc/__z1QycQCREVt72sQrUQpGs1ToUnNmlXgfgtUJyI0bRFY).",
   "relevance": "Koniku sits at the far \"wetware\" end of the non-von-Neumann spectrum — literal living neurons rather\nthan Knowm's solid-state memristor synapses. Both reject the digital-simulation paradigm, but Koniku's\nwedge is chemical *sensing* (an electronic nose) where biology's sensitivity is unmatched, versus\nKnowm's general-purpose analog learning substrate. Useful as the biological extreme of the taxonomy.",
   "news": [],
   "sources": [
    {
     "title": "Koniku (Wikipedia)",
     "url": "https://en.wikipedia.org/wiki/Koniku"
    },
    {
     "title": "Konikore bioelectric sensor (Whipsaw)",
     "url": "https://www.whipsaw.com/work/konikore-bioelectric-sensor"
    },
    {
     "title": "Nigerian creates explosive/cancer-detecting device (CNN, TEDGlobal)",
     "url": "https://www.cnn.com/2017/08/28/africa/koniku-kore-tedglobal/index.html"
    },
    {
     "title": "Koniku and Airbus expand partnership (TSI Magazine)",
     "url": "https://tsi-mag.com/koniku-and-airbus-expand-partnership-to-re-imagine-aviation-security/"
    },
    {
     "title": "Koniku company profile (CB Insights)",
     "url": "https://www.cbinsights.com/company/koniku"
    }
   ]
  },
  {
   "slug": "lightelligence",
   "name": "Lightelligence",
   "founded": 2017,
   "hq": "Shanghai, China (founded in Boston, MA, USA)",
   "category": "photonic",
   "status": "public",
   "origin": "academic spinoff (MIT — Yichen Shen, lead author of 2017 Nature Photonics \"Deep learning with coherent nanophotonic circuits\", Soljačić/Englund labs)",
   "total_funding_usd_m": 247,
   "last_round": "IPO (HKEX 1879), $323M, 2026-04",
   "key_investors": [
    "Baidu Ventures",
    "ZhenFund",
    "Matrix Partners China",
    "Tencent",
    "CICC Capital",
    "Sequoia China",
    "China Mobile"
   ],
   "website": "https://www.lightelligence.ai",
   "pitch": "Lightelligence pitches optical computing as the way past the slowing of Moore's law for AI:\nperforming the linear algebra of neural networks with photons promises orders-of-magnitude\ngains in speed and energy over electronics — its PACE demo was advertised as \"25–100x faster\nthan NVIDIA's industry-leading GPU\" on certain compute-intensive (Ising/optimization)\nworkloads. (https://www.lightelligence.ai/, https://www.csis.org/analysis/controlling-light-silicon-photonics-emerging-front-us-china-tech-competition)",
   "technical_approach": "- Substrate: silicon photonics co-packaged with CMOS. Roots in the 2017 MIT Nature Photonics\n  paper (coherent nanophotonic MZI meshes for deep learning) — the same paper behind\n  Lightmatter; Shen was first author, Lightmatter's Harris a co-author.\n- **PACE** (Photonic Arithmetic Computing Engine, Dec 2021): ~12,000-device photonic-electronic\n  package doing analog optical matrix multiplication, demoed on Ising/NP-hard problems.\n  (https://www.globenewswire.com/news-release/2023/08/22/2729261/0/en/)\n- **Hummingbird** (Hot Chips 2023): pivot in emphasis — an *optical network-on-chip* (oNOC)\n  processor: a 64-core electronic AI inference ASIC (TSMC 28nm) whose cores are linked by a\n  programmable optical broadcast layer; light moves data, electronics compute.\n  (https://www.lightelligence.ai/index.php/product/hummingbird.html)\n- **Photowave** (2023): CXL-over-optics interconnect card line for datacenter composability.\n- **LightSphere X** (2025, with partners): billed as the first distributed optical\n  interconnect + optical switching GPU supernode solution. Caixin now describes the company\n  simply as an \"optical interconnect maker\" — i.e., like Lightmatter, the shipping business\n  is photonic interconnect, not photonic matmul.\n  (https://www.caixinglobal.com/2026-03-31/optical-interconnect-maker-lightelligence-files-for-hong-kong-ipo-102429646.html)\n- Inference only; no on-chip learning claims.",
   "funding_rounds": [
    {
     "date": "2018-02",
     "round": "Seed",
     "amount": "~$10M",
     "investors": "Baidu Ventures (lead), ZhenFund",
     "source": "https://www.dealstreetasia.com/stories/baidu-ventures-leads-10m-funding-round-us-based-ai-startup-lightelligence-91850"
    },
    {
     "date": "2018",
     "round": "Series A",
     "amount": "CNY 64.5M",
     "investors": "ZhenFund, Matrix Partners China, Fengrui, Baidu Ventures",
     "source": "https://eu.36kr.com/en/p/3825622558757768"
    },
    {
     "date": "2021",
     "round": "Series B / B+",
     "amount": "CNY 400M+",
     "investors": "Matrix Partners China, Tencent, CICC Capital",
     "source": "https://eu.36kr.com/en/p/3825622558757768"
    },
    {
     "date": "2024-03",
     "round": "Series C",
     "amount": "CNY 1.265B",
     "investors": "Sequoia China, Tencent, Baidu, Zhongke Chuangxing, China Mobile",
     "source": "https://eu.36kr.com/en/p/3825622558757768"
    },
    {
     "date": "2026-04-28",
     "round": "IPO (HKEX 1879)",
     "amount": "HK$2.5B (~$323M)",
     "investors": "cornerstones incl. Alibaba, Temasek, BlackRock (reported)",
     "source": "https://www.scmp.com/tech/tech-trends/article/3351551/"
    }
   ],
   "funding_notes": "Conflicts: Tracxn puts pre-IPO total at **$247M over 4 rounds**; Caproasia dates a \"$210M\n(CNY 1.5B) Series C\" to Sept 2025 while 36kr dates Series C (CNY 1.265B) to March 2024 —\nlikely the same round tranche-counted differently. Wikipedia/CSIS say \">$220M\" pre-IPO.\nReport range: **~$220–250M private + $323M IPO**.",
   "trajectory": "Founded 2017 in Boston by MIT physics PhD Yichen Shen with MIT collaborators; over time the\ncompany re-centered on China (now Shanghai-based) and Chinese state/strategic capital.\nListed on HKEX 2026-04-28 at HK$183.20 (top of range, ~600x oversubscribed retail book);\nopened at HK$880 — a 380–400% first-day pop, the best record-setting debut for a $100M+ HK\nIPO — market cap above HK$80B (~$10.4B). 70% of proceeds earmarked for silicon-photonics R&D\nand optoelectronic hybrid computing cards. World's first listed \"AI silicon photonics chip\"\nstock. (https://www.caixinglobal.com/2026-04-28/lightelligence-sets-record-ipo-gain-with-383-surge-on-hong-kong-debut-102438977.html)",
   "relevance": "The other MIT twin from the same 2017 paper as Lightmatter — and the same arc: founded on\noptical matrix multiplication, monetizing optical interconnect. Its record IPO shows public\nmarkets pricing the photonic *plumbing* of AI, not analog optical compute, and there is no\nin-substrate learning — useful contrast to memristor/AHaH claims.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Lightelligence (HKEX 1879) announced a **strategic partnership with Centec Communications on co-packaged optics** at WAIC (2026-07-20); Pandaily analysis (2026-08-20) frames the business as optical-interconnect-first with photonic computing still pre-revenue. (https://www.moomoo.com/news/post/73253544/hk-stock-market-movement-lightelligence-holdings-limited-01879-hk-rises ; https://pandaily.com/lightelligence-photonic-computing-pivot-optical-interconnect-hkex-listing-aug2026)"
    },
    {
     "date": "2026-11-09",
     "text": "**A-share listing tutoring registered 2026-11-03** with the Shanghai securities bureau (sponsor Guotai Haitong) — a second, mainland listing six months after the HKEX IPO. HK IPO net proceeds restated at **HK$2.741B (~US$349.5M)** incl. the May over-allotment (HK$2.69B still unused at 2026-06-30). H1 2026: revenue CNY 79.8M (+284%; optical interconnect CNY 60.7M vs optical computing CNY 19.1M), R&D CNY 350M, adjusted net loss CNY 147M. Note: the interconnect line is now 3x the computing line. (https://finance.biggo.com/news/676ad181-aac0-4f40-9d0d-9b9c6e808911)"
    }
   ],
   "sources": [
    {
     "title": "SCMP: 400% HK debut",
     "url": "https://www.scmp.com/tech/tech-trends/article/3351551/lightelligence-jumps-hong-kong-debut-amid-ai-driven-demand-photonics-chips"
    },
    {
     "title": "Caixin: record IPO gain",
     "url": "https://www.caixinglobal.com/2026-04-28/lightelligence-sets-record-ipo-gain-with-383-surge-on-hong-kong-debut-102438977.html"
    },
    {
     "title": "Caixin: files for HK IPO",
     "url": "https://www.caixinglobal.com/2026-03-31/optical-interconnect-maker-lightelligence-files-for-hong-kong-ipo-102429646.html"
    },
    {
     "title": "36kr: Six Light Chasers (funding history)",
     "url": "https://eu.36kr.com/en/p/3825622558757768"
    },
    {
     "title": "Caproasia: Series C $210M",
     "url": "https://www.caproasia.com/2025/09/07/"
    },
    {
     "title": "Hummingbird at Hot Chips 2023",
     "url": "https://www.globenewswire.com/news-release/2023/08/22/2729261/0/en/Lightelligence-to-Demonstrate-Hummingbird-Optical-Network-on-Chip-Processor-at-Hot-Chips-2023.html"
    },
    {
     "title": "CSIS silicon photonics analysis",
     "url": "https://www.csis.org/analysis/controlling-light-silicon-photonics-emerging-front-us-china-tech-competition"
    },
    {
     "title": "Tracxn profile",
     "url": "https://tracxn.com/d/companies/lightelligence/__73h11l8AckdiWJwRQ1ycBSwvFg0f-Dc2MJ_cHG9bqRM"
    }
   ]
  },
  {
   "slug": "lightmatter",
   "name": "Lightmatter",
   "founded": 2017,
   "hq": "Mountain View, CA, USA (founded in Boston)",
   "category": "photonic",
   "status": "active",
   "origin": "academic spinoff (MIT — Dirk Englund's Quantum Photonics group; Nick Harris PhD work)",
   "total_funding_usd_m": 850,
   "last_round": "Series D, $400M, 2024-10",
   "key_investors": [
    "T. Rowe Price",
    "Fidelity",
    "GV",
    "Viking Global",
    "SIP Global",
    "Matrix Partners",
    "Spark Capital",
    "HPE Pathfinder",
    "Lockheed Martin"
   ],
   "website": "https://lightmatter.co",
   "pitch": "Lightmatter's framing: transistor scaling (Moore's law / Dennard scaling) has ended while AI\ncompute demand grows exponentially, and electrical signaling hits heat and resistance walls —\n\"by using light instead of electrical signals for computing, Lightmatter's chips don't\nexperience the same heat or resistance as traditional chip architectures.\"\n(https://www.businesswire.com/news/home/20241016498931/en/, https://lightmatter.co/vision/)",
   "technical_approach": "- Substrate: silicon photonics. Original product line was **photonic compute** — analog\n  matrix-vector multiplication in Mach-Zehnder interferometer meshes: *Mars* test chip\n  (Hot Chips 2020), *Envise* photonic AI inference accelerator (announced 2021), *Idiom*\n  software stack. Inference-focused; weights loaded electronically, MACs done optically.\n- **Pivot (verified): commercial focus moved from photonic compute to photonic\n  interconnect.** Flagship is now *Passage*, a 3D photonic interposer/co-packaged-optics\n  line linking conventional XPUs with an on-chip programmable optical network:\n  Passage M1000 \"superchip\" (114 Tbps optical bandwidth, up to 34 chiplets on a photonic\n  interposer, announced OFC March 2025, available summer 2025) and Passage L200/L200X\n  3D CPO (32/64 Tbps, 2026, customer chip integration), fabbed on GlobalFoundries GF\n  Fotonix with Amkor/ASE packaging. Commercial volume shipments reported targeted ~2027.\n  (https://lightmatter.co/press-release/lightmatter-unveils-passage-m1000-photonic-superchip-worlds-fastest-ai-interconnect/,\n  https://www.datacenterdynamics.com/en/news/lightmatter-unveils-m1000-and-l200-passage-photonic-interconnects/,\n  https://www.digitimes.com/news/a20251121PD230/silicon-photonics-globalfoundries-shipments-commercial.html)\n- The compute thread survives as R&D: April 2025 Nature paper \"Universal photonic\n  artificial intelligence acceleration\" — first photonic processor running ResNet/BERT-class\n  networks at near-FP32 accuracy. (https://www.nature.com/articles/s41586-025-08854-x)\n- No on-chip learning claim; off-chip-trained inference (compute line) and pure data\n  movement (interconnect line).",
   "funding_rounds": [
    {
     "date": "2018-02",
     "round": "Series A",
     "amount": "$11M",
     "investors": "Matrix Partners + Spark Capital (co-led)",
     "source": "https://www.globenewswire.com/de/news-release/2018/02/05/1333152/0/en/"
    },
    {
     "date": "2019-02",
     "round": "Series A ext.",
     "amount": "$22M (A total $33M)",
     "investors": "GV joined",
     "source": "https://sacra.com/c/lightmatter/"
    },
    {
     "date": "2021-05",
     "round": "Series B",
     "amount": "$80M",
     "investors": "Viking Global; GV, HPE, Lockheed Martin, Matrix, Spark, SIP Global",
     "source": "https://www.businesswire.com/news/home/20210506005271/en"
    },
    {
     "date": "2023-05",
     "round": "Series C",
     "amount": "$154M",
     "investors": "SIP Global; Fidelity, Viking, GV, HPE Pathfinder",
     "source": "https://www.businesswire.com/news/home/20230531005414/en/"
    },
    {
     "date": "2023-12",
     "round": "Series C-2",
     "amount": "$155M (C total ~$309M)",
     "investors": "at $1.2B valuation",
     "source": "https://siliconangle.com/2023/12/19/lightmatter-raises-155m-photonic-computing-1-2b-valuation/"
    },
    {
     "date": "2024-10",
     "round": "Series D",
     "amount": "$400M",
     "investors": "T. Rowe Price-advised funds; Fidelity, GV",
     "source": "https://www.datacenterdynamics.com/en/news/photonic-computing-company-lightmatter-achieves-44bn-valuation-from-400m-series-d-funding-round/"
    }
   ],
   "funding_notes": "Total raised: **$850M**; valuation **$4.4B** (Oct 2024, quadrupled from $1.2B Dec 2023).",
   "trajectory": "Founded 2017 by Nick Harris (CEO), Darius Bunandar, Thomas Graham out of MIT (Harris's\nphotonic-processor PhD work; MIT $100K competition winners). Best-funded photonic computing\nstartup in the West. Honest read of the trajectory: the Envise photonic-compute accelerator\nnever became the shipping business; from ~2023 onward the company repositioned around the\nPassage interconnect for AI datacenters, where light carries data rather than performs the\nmatmul. Active and private as of mid-2026; M1000 sampling, L200 in customer integration for\n2026, revenue-scale shipments projected 2027.",
   "relevance": "Same von Neumann/energy diagnosis as Knowm, opposite prescription: photons for analog matmul\n(and now mostly for bandwidth) vs memristive in-memory compute. Lightmatter is also the\ncautionary data point that even $850M has so far monetized photonic *interconnect*, not\nphotonic *compute* — and there is no learning in the substrate (contrast AHaH on-chip\nadaptation).",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Lightmatter is leading a **19-company CPO (co-packaged optics) System Architecture initiative formally launched within the Open Compute Project** (2026-08-13), plus a white paper on the fragmented photonic-interconnect supply chain — cementing the interconnect (not compute) direction. (https://www.datacenterdynamics.com/en/news/lightmatter-launches-19-company-strong-initiative-to-standardize-silicon-photonics-ready-infrastructure/ ; https://www.hpcwire.com/off-the-wire/lightmatter-industry-partners-launch-cpo-architecture-initiative-within-ocp/)"
    },
    {
     "date": "2026-09-22",
     "text": "**Joined the Open CPX multi-source agreement and introduced the Passage L20 CPX module (2026-09-17)**, described as the first bidirectional (transmit+receive on one fiber) Open CPX optical engine: 12.8 Tbps aggregate (6.4 Tbps each way), 32 lanes per direction; company analysis claims a 512-GPU scale-up pod drops from ~131k to ~65k fibers and ~15% lower scale-up network cost. Product line is near-package optics for AI clusters — the interconnect pivot is now a shipping-product roadmap, with CPO framed as the follow-on. (https://www.unite.ai/lightmatter-joins-open-cpx-msa-with-bidirectional-passage-l20-module/ ; https://lightmatter.co/press-release/lightmatter-joins-open-cpx-msa-introduces-the-industrys-first-bidirectional-cpx-optical-engine)"
    },
    {
     "date": "2026-10-27",
     "text": "**OCP Global Summit (2026-10-13): 204.8 Tbps CPO switch tray (1RU, with Flex) and a 102.4 Tbps four-XPU tray** — first ORv3/MHS systems carrying Passage CPO, Guide DR lasers and vClick detachable optics; the OCP photonics initiative now >20 companies; roadmap: NPO form factors 2027–28, **Passage M1000 interposer late 2028**. Also Time 'Best Inventions 2026' (Guide) and a Taiwan Tech Day. Interconnect, not compute, throughout. (https://futurumgroup.com/insights/lightmatter-and-flex-bring-2048-tbps-co-packaged-optics-into-open-racks/ ; https://time.com/collection/best-inventions/2026/lightmatter-guide/)"
    }
   ],
   "sources": [
    {
     "title": "Businesswire Series D",
     "url": "https://www.businesswire.com/news/home/20241016498931/en/"
    },
    {
     "title": "DCD $4.4B valuation",
     "url": "https://www.datacenterdynamics.com/en/news/photonic-computing-company-lightmatter-achieves-44bn-valuation-from-400m-series-d-funding-round/"
    },
    {
     "title": "Sacra funding history",
     "url": "https://sacra.com/c/lightmatter/"
    },
    {
     "title": "TechCrunch Series C",
     "url": "https://techcrunch.com/2023/05/31/lightmatters-photonic-ai-hardware-is-ready-to-shine-with-154m-in-new-funding/"
    },
    {
     "title": "M1000 press release",
     "url": "https://lightmatter.co/press-release/lightmatter-unveils-passage-m1000-photonic-superchip-worlds-fastest-ai-interconnect/"
    },
    {
     "title": "Nature: Universal photonic AI acceleration",
     "url": "https://www.nature.com/articles/s41586-025-08854-x"
    },
    {
     "title": "MIT News profile",
     "url": "https://news.mit.edu/2024/startup-lightmatter-accelerates-progress-toward-light-speed-computing-0301"
    },
    {
     "title": "Digitimes: 2027 shipments via GlobalFoundries",
     "url": "https://www.digitimes.com/news/a20251121PD230/silicon-photonics-globalfoundries-shipments-commercial.html"
    }
   ]
  },
  {
   "slug": "lighton",
   "name": "LightOn",
   "founded": 2016,
   "hq": "Paris, France",
   "category": "photonic",
   "status": "public",
   "origin": "academic spinoff (ENS / Sorbonne — Laboratoire Kastler Brossel, Sylvain Gigan; Florent Krzakala's ENS statistical-physics group; Laurent Daudet, Institut Langevin — the OPU grew out of the founders' 2015 optical random-projection paper)",
   "total_funding_usd_m": 20,
   "last_round": "IPO on Euronext Growth Paris, €11.9M raised, 2024-11-26 (+ €12.5M post-IPO Dec 2024)",
   "key_investors": [
    "Quantonation",
    "Anorak Ventures",
    "Axon Partners Group"
   ],
   "website": "https://www.lighton.ai",
   "pitch": "LightOn's original thesis was that certain massive linear-algebra operations in ML\n— especially the random projections used in kernel methods, reservoir computing and\ndimensionality reduction — can be computed optically in a single pass through\nscattering media, at large scale and low energy, sidestepping the memory-bandwidth\nand power cost of doing the same multiplies on electronic hardware. (Framing per\nLightOn's OPU material and its integration into the Jean Zay supercomputer, 2021.)",
   "technical_approach": "- Originally: the **Optical Processing Unit (OPU)** — an analog photonic co-processor\n  that performs very large fixed random projections by shining modulated light\n  through a scattering (diffusive) medium and reading out the transformed signal.\n  Used for random-feature ML, reservoir computing, and high-dimensional transforms;\n  a physical OPU was installed in France's Jean Zay supercomputer (IDRIS) in 2021.\n  Analog, inference/transform-only (the projection is random/fixed, not trained).\n- Pivoted to: **generative AI software**. Sensing the LLM wave (~2020 onward),\n  LightOn moved off hardware to build Paradigm, an enterprise GenAI platform\n  (on-prem LLMs, RAG, agents) sold via software licenses to aerospace/defense,\n  banking, insurance and public-sector customers (Safran, Groupama, the French\n  space agency). The OPU photonic hardware is effectively legacy.",
   "funding_rounds": [
    {
     "date": "2018-12",
     "round": "Seed",
     "amount": "~$3.3M",
     "investors": "Anorak Ventures, Quantonation",
     "source": "Tracxn / Crunchbase"
    },
    {
     "date": "2024-11-26",
     "round": "IPO (Euronext Growth Paris, ticker ALTAI)",
     "amount": "€11.9M raised; ~€62M market cap at listing",
     "investors": "Public float (issue €10.35/share)",
     "source": "Euronext / Sifted / Reuters"
    },
    {
     "date": "2024-12-11",
     "round": "Post-IPO equity",
     "amount": "€12.5M",
     "investors": "Axon Partners Group",
     "source": "Tracxn"
    },
    {
     "date": "—",
     "round": "Total (reported)",
     "amount": "~$20M+ incl. IPO proceeds",
     "investors": "—",
     "source": "Crunchbase / Tracxn"
    }
   ],
   "funding_notes": "",
   "trajectory": "Founded 2016 in Paris by Igor Carron, Laurent Daudet, Florent Krzakala and Sylvain\nGigan — physicists/ML researchers (ENS / Sorbonne lineage). Built the OPU and\ndemonstrated it at supercomputer scale (Jean Zay, 2021), but pivoted around 2020 to\ngenerative-AI software as LLMs became the market. On **26 November 2024** LightOn\nlisted on Euronext Growth Paris (ticker ALTAI), becoming Europe's first pure-play\nlisted generative-AI company, raising ~€11.9M at a ~€62M market cap (above its ~€50M\ntarget), followed by a €12.5M post-IPO placement from Axon in December 2024. Status\nmid-2026: public company, software-only GenAI vendor (Paradigm); the photonic OPU is\npart of its origin story, not its current product.\n\n**Founders (sweep 2026-09-22):** Igor Carron (CEO; PhD Texas A&M, engineering degree INPG; earlier space/nuclear engineering management), Laurent Daudet (CTO; ENS graduate, PhD applied mathematics Marseille; professor Sorbonne/Institut Langevin), Sylvain Gigan (professor, Laboratoire Kastler Brossel, ENS/Sorbonne) and Florent Krzakala (professor ENS, now EPFL). Gigan's official bios call LightOn \"a spin-off\" he co-founded, and the four founders co-authored the 2015 optical random-projection paper the OPU is built on (https://lighton.ai/team ; https://spie.org/photonics-europe/event/hot-topics-iii/2626610 ; https://arxiv.org/abs/1510.06664 ; https://www.quantonation.com/2018/12/20/lighton-seed/).",
   "relevance": "The starkest first-wave photonic-compute pivot: LightOn built a real analog optical\nprocessor that ran on a national supercomputer, then abandoned hardware entirely for\nGenAI software once LLMs dominated. It illustrates how quickly a physics-based\ncompute substrate can be outrun by the software market — a contrast to Knowm's\ncontinued hardware/device commitment, and a data point on the commercial fragility\nof analog-accelerator businesses whose primitive (random projection) had no on-chip\nlearning story.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "LightOn (Euronext Growth: ALTAI) **confirmed the total amount of its bond financing** via EQS regulatory release (2026-06-23); amount not stated in the headline — fetch the release for figures before quoting. (https://www.eqs-news.com/news/fr-regulatory/lighton-lighton-confirms-the-total-amount-of-its-bond-financing/f50564ba-54a5-42ac-819c-85e6f80018d7_en)"
    }
   ],
   "sources": [
    {
     "title": "LightOn lists on Euronext Growth — Euronext press release",
     "url": "https://www.euronext.com/en/about/media/euronext-press-releases/lighton-lists-euronext-growth"
    },
    {
     "title": "France's LightOn says valued at $65 million in Euronext IPO — Reuters/Yahoo",
     "url": "https://finance.yahoo.com/news/frances-lighton-says-valued-65-175306902.html"
    },
    {
     "title": "Paris-based GenAI startup LightOn to IPO at a €50m valuation — Sifted",
     "url": "https://sifted.eu/articles/lighton-ipo-50m-news"
    },
    {
     "title": "LightOn — Wikipedia (OPU history, Jean Zay, pivot)",
     "url": "https://en.wikipedia.org/wiki/LightOn"
    },
    {
     "title": "LightOn funding rounds & investors — Tracxn",
     "url": "https://tracxn.com/d/companies/lighton/__3L9DaKoHAWKPObAeSfMiHgmbBkfnmn_L9qwrNlCmRe8/funding-and-investors"
    }
   ]
  },
  {
   "slug": "lumai",
   "name": "Lumai",
   "founded": 2022,
   "hq": "Oxford, UK",
   "category": "photonic",
   "status": "active",
   "origin": "academic spinoff (Prof. Alex Lvovsky's experimental optics group, University of Oxford)",
   "total_funding_usd_m": 14,
   "last_round": "Venture round, >$10M (reported £8.9M), announced 2025-04",
   "key_investors": [
    "Constructor Capital",
    "IP Group",
    "Runa Capital",
    "PhotonVentures",
    "State Farm Ventures"
   ],
   "website": "https://lumai.ai",
   "pitch": "Lumai's framing: silicon accelerators are hitting energy and scaling walls for AI's core\noperation, matrix-vector multiplication, so the multiplication should be done by light itself —\ncomputation \"performed as light beams traverse a three-dimensional geometry\" rather than on a\nplanar chip — targeting 50× the performance of silicon-based accelerators at ~10% of the energy\ncost ([optics.org, 2025-04](https://optics.org/news/photonic-computing-startup-lumai-lands-10m);\n[Optica OPN](https://www.optica-opn.org/home/industry/2025/april/lumai_lands_funds_to_push_3d_optical_computing/)).",
   "technical_approach": "- **Substrate:** **3D free-space optical matrix-vector multiplication (MVM)** — analog optical\n  compute in free space, distinct from integrated-photonic (on-chip waveguide) approaches;\n  claimed potential of 10^17 operations/s, ~1000× current electronic technology\n  ([optics.org](https://optics.org/news/photonic-computing-startup-lumai-lands-10m)).\n- **Learning locus:** productizing datacenter **inference** acceleration; the founding research\n  also demonstrated *optical training* — an optical implementation of backpropagation (Oxford/\n  Lumai work, Spall & Guo; [SemiEngineering](https://semiengineering.com/the-optical-implementation-of-backpropagation-oxford-lumai/)) —\n  so on-hardware learning is in the research lineage, not yet the product.\n- **Products:** no commercial product shipping as of mid-2026; funding is for product\n  development, doubling headcount, and US expansion\n  ([Tech Startups](https://techstartups.com/2025/04/02/lumai-an-oxford-spinout-and-ai-accelerator-startup-raises-10m-to-cut-ai-compute-costs-with-optical-acceleration/)).",
   "funding_rounds": [
    {
     "date": "~2022",
     "round": "Seed (at spinout)",
     "amount": "£1.75M",
     "investors": "IP Group, Runa Capital",
     "source": "[LinkedIn/Wolba profile](https://www.linkedin.com/pulse/lumai-shaping-future-optical-computing-artificial-benjamin-wolba-bxnve)"
    },
    {
     "date": "~2023",
     "round": "Grant",
     "amount": "£1.1M",
     "investors": "UK government (all-optical neural network commercialization)",
     "source": "[eeNews Europe](https://www.eenewseurope.com/en/uk-funds-startup-lumai-for-optical-ai/)"
    },
    {
     "date": "2024-12 / announced 2025-04 (OFC)",
     "round": "Venture",
     "amount": "reported £8.9M (Preqin) / \">$10M\" (company)",
     "investors": "led Constructor Capital; IP Group, PhotonVentures, Journey Ventures, LIFTT, Qubits Ventures, State Farm Ventures, TIS Inc.",
     "source": "[Lumai PR](https://lumai.ai/resources/lumai-secures-more-than-10m-investment-2), [optics.org](https://optics.org/news/photonic-computing-startup-lumai-lands-10m), [Preqin](https://www.preqin.com/data/profile/asset/lumai-limited/531430)"
    }
   ],
   "funding_notes": "Total: roughly $14M equity + grant (£-figures converted approximately; the Dec-2024 £8.9M and\nthe Apr-2025 \">$10M\" announcements appear to be the same round — both recorded).",
   "trajectory": "Spun out of the University of Oxford in 2022 from Prof. **Alex Lvovsky's** experimental optics\ngroup. Founders: **Tim Weil (CEO), Dr. Xianxin Guo** (primary inventor of the optical training\ntechnology; later CEO per a 2026 interview), **Prof. Alex Lvovsky, Dr. Thomas Barrett, and\nDr. James Spall** (co-developer of the core optical computing technology)\n([LinkedIn/Wolba](https://www.linkedin.com/pulse/lumai-shaping-future-optical-computing-artificial-benjamin-wolba-bxnve);\n[The Sunday Signal interview](https://www.thesundaysignal.ai/p/the-full-sunday-signal-interview)).\nStatus mid-2026: active, pre-product, expanding into the US.",
   "relevance": "Lumai does the multiply-accumulate in physics (light propagation) rather than in devices —\nanalog computation without an emerging memory element, and with credible published work on\ntraining *in* the physical substrate (optical backprop), which parallels Knowm's argument that\nlearning, not just inference, should live in the hardware.",
   "news": [
    {
     "date": "2026-11-09",
     "text": "**ARIA's Scaling Inference Lab (run by CommonAI CIC, £50M ARIA programme) deploys Lumai Iris inference servers at rack scale (PR 2026-11-03)** — the lab's first optical inference platform; ARIA's David Lewis calls it 'the most mature optical compute platform we evaluated'. Vendor claims: up to 90% lower datacentre power, ≥16x energy per inference with optical prefill + silicon decode, ~1/10 TCO, '50x' performance roadmap; Iris Nova (April 2026) runs Llama 8B/70B. Contract value and node count not stated. (https://www.hpcwire.com/off-the-wire/arias-scaling-inference-lab-deploys-lumais-optical-compute-tech/ ; https://www.unite.ai/aria-lumai-optical-servers-ai-inference-testing/)"
    }
   ],
   "sources": [
    {
     "title": "Photonic computing startup Lumai lands $10M — optics.org",
     "url": "https://optics.org/news/photonic-computing-startup-lumai-lands-10m"
    },
    {
     "title": "Lumai Lands Funds — Optica OPN",
     "url": "https://www.optica-opn.org/home/industry/2025/april/lumai_lands_funds_to_push_3d_optical_computing/"
    },
    {
     "title": "Lumai secures more than $10M — Lumai PR",
     "url": "https://lumai.ai/resources/lumai-secures-more-than-10m-investment-2"
    },
    {
     "title": "Lumai raises $10M+ — Tech Startups",
     "url": "https://techstartups.com/2025/04/02/lumai-an-oxford-spinout-and-ai-accelerator-startup-raises-10m-to-cut-ai-compute-costs-with-optical-acceleration/"
    },
    {
     "title": "UK funds startup Lumai — eeNews Europe",
     "url": "https://www.eenewseurope.com/en/uk-funds-startup-lumai-for-optical-ai/"
    },
    {
     "title": "Optical implementation of backpropagation — SemiEngineering",
     "url": "https://semiengineering.com/the-optical-implementation-of-backpropagation-oxford-lumai/"
    },
    {
     "title": "Lumai profile — Benjamin Wolba, LinkedIn",
     "url": "https://www.linkedin.com/pulse/lumai-shaping-future-optical-computing-artificial-benjamin-wolba-bxnve"
    }
   ]
  },
  {
   "slug": "luminous-computing",
   "name": "Luminous Computing",
   "founded": 2018,
   "hq": "Santa Clara / Mountain View, CA, USA",
   "category": "photonic",
   "status": "defunct",
   "origin": "academic spinoff (Princeton — Mitchell Nahmias's neuromorphic-photonics PhD) + serial founders (Marcus Gomez, Michael Gao)",
   "total_funding_usd_m": 123,
   "last_round": "Series A, $105M, 2022-03",
   "key_investors": [
    "Bill Gates",
    "Gigafund",
    "Neo",
    "8090 Partners",
    "Third Kind VC",
    "10100 (Travis Kalanick)",
    "Dara Khosrowshahi"
   ],
   "website": "https://luminous-co.squarespace.com/ (stale; copyright 2021)",
   "pitch": "\"Our models are growing at an exponential rate, and digital electronics isn't keeping up\" —\nsemiconductor incumbents are stuck in \"a cycle of incrementalism,\" and real physical\nbottlenecks need \"new physical solutions.\" The stated moonshot: an AI supercomputer with the\npower of 3,000 TPU/GPU boards \"in a single chip,\" using photonics to blow past memory and\nI/O walls. (https://luminous-co.squarespace.com/, https://techcrunch.com/2019/06/04/bill-gates-neo-gigafund-backing-luminous-in-photonics-supercomputer-moonshot/)",
   "technical_approach": "- Origin tech: Nahmias's Princeton work on laser neurons / neuromorphic photonics; early\n  positioning was a light-based AI accelerator chip (optical compute).\n- **Pivot #1 (~2021-2022, pre-shutdown, publicly framed):** away from optical matmul toward\n  a full AI supercomputer architecture in which photonics solves *data movement* (memory and\n  chip-to-chip bandwidth) while compute stays digital electronic — \"the bottleneck is\n  communication, not FLOPs.\" Silicon-photonics pioneer Michael Hochberg joined as President\n  (March 2022) to build that interconnect. (https://www.nextplatform.com/connect/2022/03/17/luminous-shines-a-light-on-optical-architecture-for-future-ai-supercomputer/1636867, https://www.forbes.com/sites/brucerogers/2022/08/19/luminous-computing-aims-to-accelerate-ai-with-photonic-computing/)\n- No product ever shipped; no named commercial chip.",
   "funding_rounds": [
    {
     "date": "~2018",
     "round": "Pre-seed",
     "amount": "~$9M incl. in early total",
     "investors": "Class 5 Global, Ozmen Ventures, Third Kind, Schox",
     "source": "https://techcrunch.com/2019/06/04/bill-gates-neo-gigafund-backing-luminous-in-photonics-supercomputer-moonshot/"
    },
    {
     "date": "2019-06",
     "round": "Seed",
     "amount": "$9M",
     "investors": "Bill Gates, Neo (Ali Partovi), Gigafund; 10100, BoxGroup, Dara Khosrowshahi",
     "source": "https://techcrunch.com/2019/06/04/bill-gates-neo-gigafund-backing-luminous-in-photonics-supercomputer-moonshot/"
    },
    {
     "date": "2022-03",
     "round": "Series A",
     "amount": "$105M",
     "investors": "Gigafund; Bill Gates, 8090 Partners, Neo, Third Kind, Alumni Ventures, Horsley Bridge",
     "source": "https://www.businesswire.com/news/home/20220303005033/en/"
    }
   ],
   "funding_notes": "Total raised: **~$115–123M** (Tracxn: $123M over 3 rounds). No valuation disclosed.",
   "trajectory": "Founded 2018 by Marcus Gomez (CEO), Mitchell Nahmias (CTO), Michael Gao (left early).\nWhat happened, as best the record shows (no formal shutdown announcement found):\n- **May 2023:** laid off half the company — including the entire photonics team — per\n  ex-employee accounts. (https://medium.com/@mpchang17/making-the-leap-from-hardware-to-machine-learning-d85853b59a05)\n- **Late 2023:** Enosemi, founded by ex-Luminous photonics leads (Streshinsky, Novack,\n  Ardalan), licensed Luminous's silicon-photonics design IP. **AMD acquired Enosemi June\n  2025** — the clearest surviving legacy of Luminous's technology.\n  (https://optics.org/news/amd-gets-into-co-packaged-optics-with-enosemi-deal)\n- 2023-2024: surplus-asset auctions (Silicon Valley Disposition), \"Luminous Computing is\n  dead?\" threads on Blind, Glassdoor reviews describing the hardware team's shutdown; Forge\n  Global marks the company as having ceased (shares no longer tradeable); website frozen at\n  2021. (https://www.svdisposition.com/auction-detail?id=594, https://forgeglobal.com/luminous_ipo/)\n- Gomez and Hochberg resurfaced as commentators (NIPP essay Dec 2024; Photonics Spectra\n  \"Photonic Supercomputing\" podcast Sept 2025), not with a shipping Luminous product.\nBest characterization as of mid-2026: **effectively defunct after a 2023 wind-down; IP lives\non inside AMD via Enosemi.**",
   "relevance": "The wave's flame-out case: ~$120M, Gates-backed, \"3,000 GPUs in one chip\" — then the physics\nand unit economics of optical compute pushed it (like the survivors) toward interconnect,\nand it didn't survive the transition. A caution against equating funding with substrate\nviability; its neuromorphic-photonics pedigree (laser neurons) never reached silicon.",
   "news": [],
   "sources": [
    {
     "title": "TechCrunch seed round",
     "url": "https://techcrunch.com/2019/06/04/bill-gates-neo-gigafund-backing-luminous-in-photonics-supercomputer-moonshot/"
    },
    {
     "title": "Businesswire Series A",
     "url": "https://www.businesswire.com/news/home/20220303005033/en/Luminous-Computing-Raises-%24105M-in-Series-A-Round-to-Build-Worlds-Most-Powerful-AI-Supercomputer"
    },
    {
     "title": "Forbes 2022 profile",
     "url": "https://www.forbes.com/sites/brucerogers/2022/08/19/luminous-computing-aims-to-accelerate-ai-with-photonic-computing/"
    },
    {
     "title": "Ex-employee on May 2023 layoffs",
     "url": "https://medium.com/@mpchang17/making-the-leap-from-hardware-to-machine-learning-d85853b59a05"
    },
    {
     "title": "optics.org: AMD acquires Enosemi (ex-Luminous team/IP)",
     "url": "https://optics.org/news/amd-gets-into-co-packaged-optics-with-enosemi-deal"
    },
    {
     "title": "Cignal AI startup tracker (Enosemi origin)",
     "url": "https://cignal.ai/2025/11/optical-component-startup-tracker/"
    },
    {
     "title": "Luminous site (stale, 2021)",
     "url": "https://luminous-co.squarespace.com/"
    }
   ]
  },
  {
   "slug": "lynxi",
   "name": "Lynxi Technology (灵汐科技)",
   "founded": 2018,
   "hq": "Beijing, China",
   "category": "digital-snn",
   "status": "active",
   "origin": "academic spinoff (Tsinghua Center for Brain-Inspired Computing Research — Tianjic lineage)",
   "total_funding_usd_m": 0,
   "last_round": "Series B+, undisclosed, 2024-04 (国鼎资本)",
   "key_investors": [
    "Tsinghua Holdings (清华控股)",
    "Huakong Jishi Fund (华控基石)",
    "Hikvision (海康威视)",
    "National Military-Civil Fusion Industry Investment Fund",
    "CETC Investment (电科投资)",
    "Guoding Capital (国鼎资本)"
   ],
   "website": "https://www.lynxi.com",
   "pitch": "Lynxi's lineage press framed the Tsinghua \"Tianjic\" hybrid chip as \"overturning von\nNeumann\" (颠覆冯诺伊曼): one chip that natively runs both computer-science deep learning\nand neuroscience-derived spiking models, on a compute-in-memory, many-core,\n\"heterogeneous-fusion\" architecture — pitched as an opening move toward general AI rather\nthan another NPU (https://zhidx.com/p/156860.html; https://www.lynxi.com/).",
   "technical_approach": "- **Architecture:** digital brain-inspired many-core chip with 存算一体 (in/near-memory\n  compute), 众核并行 (many-core parallelism), 异构融合 (hybrid ANN+SNN in one fabric) —\n  the commercialized descendant of Tsinghua's Tianjic (Nature cover, Aug 2019,\n  self-driving-bicycle demo).\n- **领启 KA200** (mass production announced Sept 2021; billed as the world's first\n  heterogeneous-fusion brain-inspired chip in volume production): 250k neurons / 25M\n  synapses per chip in dense mode, scalable to 2M neurons / 2B synapses in sparse mode;\n  mixed-precision ANN math plus SNN support (https://zhuanlan.zhihu.com/p/408994045;\n  http://www.pinevc.com.cn/article/1297.html).\n- **Products around the chip:** KA200(-S) chips, PCIe boards, modules, edge boxes, the\n  SL800 brain-simulation server, LynOS software stack, and (by 2024–2026) LLM all-in-one\n  machines (大模型一体机) (https://www.saiia.org.cn/index.php/2023/07/20/7-19/;\n  https://www.lynxi.com/).\n- **Markets:** brain simulation / neuroscience research platforms, cloud and edge AI\n  inference. Primarily **inference/simulation**; on-chip learning is not the commercial\n  claim.",
   "funding_rounds": [
    {
     "date": "2018-08",
     "round": "Angel",
     "amount": "undisclosed",
     "investors": "华控基石基金, 优选资本, Tsinghua Holdings (清华控股)",
     "source": "https://zhidx.com/p/156860.html"
    },
    {
     "date": "2021-08",
     "round": "A+",
     "amount": "undisclosed",
     "investors": "Hikvision (海康威视), 卓源资本, 新鼎资本, 无锡武吉士, 宁波灵知",
     "source": "https://www.eet-china.com/news/202108201435.html"
    },
    {
     "date": "2023-11",
     "round": "Series B",
     "amount": "undisclosed",
     "investors": "National Military-Civil Fusion Industry Investment Fund (国家军民融合产业投资基金), CETC Investment (电科投资) et al.",
     "source": "https://baike.baidu.com/item/北京灵汐科技有限公司/24542560"
    },
    {
     "date": "2024-04",
     "round": "Series B+",
     "amount": "undisclosed",
     "investors": "Guoding Capital (国鼎资本)",
     "source": "https://news.jinglingshuju.com/article/78552779600"
    }
   ],
   "funding_notes": "**No round amounts have been disclosed publicly — total raised is unknown** (frontmatter\n0 = unknown, not zero). The investor roster (Tsinghua, Hikvision, military-civil-fusion\nand CETC state funds) signals strategic/state backing more than reported dollar volume.",
   "trajectory": "Registered Jan 2018 as the commercialization vehicle of Tsinghua's Center for\nBrain-Inspired Computing Research (CBICR, Prof. Shi Luping's Tianjic team); chairman & CEO\n**Zhu Yaolong (祝夭龙)**, with Shi Luping listed among co-founders\n(https://zhidx.com/p/156860.html; https://www.zhihu.com/question/333778254). Milestones:\nTianjic Nature cover 2019 → KA200 volume production 2021 (\"13th Five-Year\" national\ntech-achievement exhibition) → B/B+ rounds 2023–24 → product line extended to brain-scale\nsimulation servers and LLM appliances; a Hangzhou subsidiary was profiled in 2025\n(https://tidenews.com.cn/news.html?id=3086446). Status mid-2026: active; the most\nresearch-platform-oriented company in this batch, straddling brain simulation and\ncommercial edge/cloud AI.",
   "relevance": "Lynxi is the \"hybrid\" answer to the neuromorphic question: rather than analog synapses or\non-chip learning (Knowm/AHaH territory), it keeps everything digital but makes SNN and ANN\nfirst-class citizens on one many-core die — closer to Loihi/TrueNorth lineage than to\nmemristor computing, and funded largely by state-strategic rather than venture capital.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "At WAIC 2026 Lynxi unveiled **LynAInfra (灵琍), a brain-inspired \"supernode\" rack-scale system for large-model inference** (2026-07-19) — moving the KA200/brain-inspired line up to datacenter-scale LLM serving. (https://m.zhidx.com/p/577241.html)"
    },
    {
     "date": "2026-09-22",
     "text": "**China Mobile Cloud launched a \"domestic GPU + brain-inspired chip\" heterogeneous LLM inference system built with Lynxi** (with CETC Nanhu Institute, Iluvatar CoreX 天数智芯, Jiangsu Mobile, Tsinghua and Peking University) at the 2026 China Computing Power Conference, Langfang, 2026-09-11/13. The Transformer is split: prefill/attention on the domestic GPU, latency-sensitive FFN/MoE-expert layers on Lynxi's chips (on-chip SRAM, compute-in-memory). On DeepSeek V4 Flash with 3 Iluvatar GPU servers + 3 brain-inspired racks, the team claims >1x higher throughput and energy efficiency and >40% lower operating cost than an equal-investment pure-GPU cluster; 15 granted patents; \"small-batch trial production\" stage; explicitly benchmarked against Nvidia's Vera Rubin + Groq heterogeneous plan. (https://finance.sina.com.cn/tech/roll/2026-09-14/doc-iniruhxe4091558.shtml ; https://cn.chinadaily.com.cn/a/202609/16/WS6aaa3297e4b09a165c78a679.html ; English: https://pandaily.com/china-mobile-cloud-gpu-neuromorphic-hetero-llm-inference)"
    }
   ],
   "sources": [
    {
     "title": "zhidx: Tianjic/Lynxi origin story",
     "url": "https://zhidx.com/p/156860.html"
    },
    {
     "title": "Zhihu answer: KA200 mass production",
     "url": "https://zhuanlan.zhihu.com/p/408994045"
    },
    {
     "title": "Pine Capital: KA200 specs",
     "url": "http://www.pinevc.com.cn/article/1297.html"
    },
    {
     "title": "EET-China: 2021 round",
     "url": "https://www.eet-china.com/news/202108201435.html"
    },
    {
     "title": "Baidu Baike: company entry, B/B+ rounds",
     "url": "https://baike.baidu.com/item/北京灵汐科技有限公司/24542560"
    },
    {
     "title": "Jingling Data: B+ round",
     "url": "https://news.jinglingshuju.com/article/78552779600"
    },
    {
     "title": "SAIIA: product line incl. SL800",
     "url": "https://www.saiia.org.cn/index.php/2023/07/20/7-19/"
    },
    {
     "title": "Lynxi site",
     "url": "https://www.lynxi.com/"
    },
    {
     "title": "Tidenews: Hangzhou Lynxi profile 2025",
     "url": "https://tidenews.com.cn/news.html?id=3086446"
    }
   ]
  },
  {
   "slug": "memcomputing",
   "name": "MemComputing, Inc.",
   "founded": 2016,
   "hq": "San Diego, California, USA",
   "category": "adjacent paradigm — physics-based/memcomputing (batch: emerging-device-imc; see notes)",
   "status": "active",
   "origin": "academic spinoff (UC San Diego — Prof. Massimiliano Di Ventra & Dr. Fabio Traversa) + entrepreneur John Beane",
   "total_funding_usd_m": 4.45,
   "last_round": "Series A, closed 2025-11 (Form D target ~$5M; amount closed not disclosed)",
   "key_investors": [
    "angel/seed investors (largely undisclosed)",
    "US DoD SBIR programs (AFWERX/AFRL/Space Force — non-dilutive)"
   ],
   "website": "https://memcpu.com",
   "pitch": "The founders describe memcomputing as \"a radically new computing paradigm inspired by\nphysics and the human brain,\" with circuit methods \"completely antithetical to the von\nNeumann architecture that traditional semiconductor companies can't seem to break away\nfrom\" (https://memcpu.com/our-story/). Since a late-2024 rebrand (\"The New Chip Company\")\nthe emphasis is energy: a coming chip that will \"drastically reduce energy consumption\nwhile maintaining exceptional performance\"\n(https://quantumzeitgeist.com/memcomputing-unveils-new-chip/).",
   "technical_approach": "- Paradigm: **memcomputing** (Di Ventra/Traversa) — computation with self-organizing logic\n  gates / dynamical systems where memory participates in computation; conceptually related\n  to memristive systems, though the commercial implementation is **digital circuit\n  emulation, not memristor hardware**.\n- Product to date: **MEMCPU Platform** (2019) — cloud SaaS (\"Virtual MemComputing\n  Machine\") that emulates the self-organizing circuits in software to solve\n  industrial-scale optimization/MILP, big-data analytics, and ML problems\n  (https://memcpu.com/our-story/;\n  https://www.airforce-technology.com/news/memcomputing-satellite-computing-processing-contracts/).\n  San Diego Supercomputer Center benchmarking reported >10^6x speedups on target problem\n  classes vs conventional solvers (company-reported: https://memcpu.com/our-story/).\n- Hardware pivot: announced November 2024 — chip portfolio of MEMAI (AI), MEMGPU\n  (image/signal), MEM5G, and MEMCPU (optimization/encryption) lines; first AI-focused chip\n  promised for H1 2025 (https://quantumzeitgeist.com/memcomputing-unveils-new-chip/). No\n  public confirmation of shipped silicon found as of 2026-07-23.\n- Not an inference accelerator and not on-chip learning: it targets combinatorial\n  optimization — included in this corpus as an adjacent non-von-Neumann paradigm.",
   "funding_rounds": [
    {
     "date": "2016–2020",
     "round": "seed/angel rounds",
     "amount": "~$3M over 3 rounds (Tracxn)",
     "investors": "6 investors, undisclosed",
     "source": "https://tracxn.com/d/companies/memcomputing/__ITnoSI4vCQ3H__KgRf4XX2zurK3LiJwVwNyMd_5rtDs"
    },
    {
     "date": "2021-04",
     "round": "AFWERX Phase II SBIR (non-dilutive)",
     "amount": "undisclosed",
     "investors": "AFRL / US Space Force — satellite moving-target tracking",
     "source": "https://www.prnewswire.com/news-releases/memcomputing-awarded-afwerx-phase-ii-sbir-related-to-the-automated-tracking-of-a-moving-object-from-space-301255589.html"
    },
    {
     "date": "2021-11",
     "round": "Two Air Force Phase I SBIRs (non-dilutive)",
     "amount": "undisclosed",
     "investors": "AFRL/Space Force (RPO motion planning; on-orbit distributed compute)",
     "source": "https://www.prnewswire.com/news-releases/memcomputing-awarded-two-air-force-phase-i-sbir-contracts-301419817.html"
    },
    {
     "date": "2025-11-20",
     "round": "Series A (closed)",
     "amount": "Form D targeted ~$5M; closed amount undisclosed",
     "investors": "undisclosed",
     "source": "https://www.crunchbase.com/organization/memcomputing-inc ; https://memcpu.com/memcomputing-secures-new-funding/"
    }
   ],
   "funding_notes": "**Conflicting totals:** Tracxn reports $3M total; CB Insights reports $4.45M total\n(https://www.cbinsights.com/company/memcomputing). Either way, MemComputing is by far the\nleast-capitalized company in this batch relative to its claims.",
   "trajectory": "Founded 2016 by UCSD theoretical physicists **Massimiliano Di Ventra** and **Fabio\nTraversa** (CTO) with serial entrepreneur **John Beane**, spinning out exclusive rights to\nthe memcomputing patents (https://memcpu.com/our-story/). Trajectory: theory (universal\nmemcomputing machines, ~2015) → SaaS emulation (MEMCPU Platform, 2019) → DoD SBIR traction\n(2021) → chip pivot + rebrand (Nov 2024) → Series A closed (Nov 2025). Status mid-2026:\nactive, small, chip roadmap announced but unverified in silicon.",
   "relevance": "Di Ventra's memcomputing is the optimization-solver cousin of memristive computing: memory\nthat computes, but commercialized as digital emulation rather than physical memristors.\nUseful in the article as the \"paradigm without the device\" counterpoint — while device\ncompanies (Knowm, TetraMem) sell physics and MemComputing sells the math.",
   "news": [],
   "sources": [
    {
     "title": "MemComputing — Our Story",
     "url": "https://memcpu.com/our-story/"
    },
    {
     "title": "Quantum Zeitgeist — new chip and rebrand (2024-11-27)",
     "url": "https://quantumzeitgeist.com/memcomputing-unveils-new-chip/"
    },
    {
     "title": "MemComputing — Secures New Funding",
     "url": "https://memcpu.com/memcomputing-secures-new-funding/"
    },
    {
     "title": "Crunchbase — MemComputing (Series A 2025-11-20)",
     "url": "https://www.crunchbase.com/organization/memcomputing-inc"
    },
    {
     "title": "Tracxn — MemComputing ($3M/3 rounds)",
     "url": "https://tracxn.com/d/companies/memcomputing/__ITnoSI4vCQ3H__KgRf4XX2zurK3LiJwVwNyMd_5rtDs"
    },
    {
     "title": "PRNewswire — AFWERX Phase II SBIR (2021-04)",
     "url": "https://www.prnewswire.com/news-releases/memcomputing-awarded-afwerx-phase-ii-sbir-related-to-the-automated-tracking-of-a-moving-object-from-space-301255589.html"
    },
    {
     "title": "PRNewswire — two Air Force Phase I SBIRs (2021-11)",
     "url": "https://www.prnewswire.com/news-releases/memcomputing-awarded-two-air-force-phase-i-sbir-contracts-301419817.html"
    }
   ]
  },
  {
   "slug": "memryx",
   "name": "MemryX Inc.",
   "founded": 2019,
   "hq": "Ann Arbor, Michigan, USA",
   "category": "boundary — digital at-memory dataflow (NOT analog/emerging-device IMC; see Technical approach)",
   "status": "active",
   "origin": "academic spinoff (University of Michigan — Wei Lu and Zhengya Zhang)",
   "total_funding_usd_m": 63,
   "last_round": "Series B, $44M, 2025-03",
   "key_investors": [
    "HarbourVest",
    "NEOM Investment Fund",
    "Arm IoT Fund / ARM IoT Capital",
    "eLab Ventures",
    "M Ventures",
    "Motus Ventures",
    "University of Michigan"
   ],
   "website": "https://memryx.com",
   "pitch": "MemryX pitches \"at-memory AI processing\" that removes data movement: its MX4 announcement\nsays the architecture leverages 3D hybrid-bonded memory \"to eliminate the industry's most\npressing bottleneck: the 'memory wall'\"\n(https://www.prnewswire.com/news-releases/memryx-unveils-mx4-roadmap-enabling-distributed-asynchronous-dataflow-for-highly-efficient-data-center-ai-302649698.html).\nCompany framing: \"production-proven 'at-memory' dataflow architecture that delivers\nsuperior efficiency\" for edge and data-center AI (same source; https://memryx.com/company/).",
   "technical_approach": "**Honest classification: digital, not analog.** Despite co-founder Wei Lu's memristor\npedigree (he also co-founded Crossbar Inc.), the shipping MemryX architecture is a\n**digital at-memory dataflow accelerator**, not an analog memristor/ReRAM IMC design:\n- MX3 computes with bfloat16 activations and 4/8/16-bit integer weights in digital MAC\n  cores (M-Cores) and ALU cores (A-Cores), with weights held in distributed on-chip weight\n  memories so models run with no external DRAM\n  (https://developer.memryx.com/architecture/architecture.html;\n  https://www.eetimes.com/adding-low-power-ai-ml-interference-to-edge-devices/).\n- Inference-only; models are trained off-chip and compiled to the dataflow graph.\n- Products: MX1/MX2 internal prototypes 2020–2022; **MX3** edge AI accelerator in\n  production with volume shipments from 2024, incl. a $149 M.2 module rated ~24 TOPS at\n  ~1–2 W per chip (https://memryx.com/company/;\n  https://www.tomshardware.com/tech-industry/artificial-intelligence/memryx-launches-usd149-mx3-m-2-ai-accelerator-module-capable-of-24-tops-compute-power).\n- 2026: Cascade platform expansion to edge servers (June 2026,\n  https://www.prnewswire.com/news-releases/memryx-expands-cascade-platform-bringing-power-efficient-edge-ai-to-servers-302806429.html)\n  and **MX4** roadmap scaling at-memory dataflow to data centers via 3D hybrid-bonded\n  memory; claims >20x perf/W vs mainstream GPUs on targeted inference\n  (https://www.prnewswire.com/news-releases/memryx-unveils-mx4-roadmap-enabling-distributed-asynchronous-dataflow-for-highly-efficient-data-center-ai-302649698.html).",
   "funding_rounds": [
    {
     "date": "2019",
     "round": "Seed",
     "amount": "$5M",
     "investors": "—",
     "source": "https://www.crainsdetroit.com/technology/memryx-raises-44m-series-b-funding"
    },
    {
     "date": "2021",
     "round": "Series A",
     "amount": "$14M",
     "investors": "ARM IoT Capital, HarbourVest, M Ventures, eLab Ventures, Motus Ventures, University of Michigan",
     "source": "https://www.crainsdetroit.com/technology/memryx-raises-44m-series-b-funding ; https://news.engin.umich.edu/2021/07/michigan-startup-memryx-inc-promises-faster-cheaper-ai-processing"
    },
    {
     "date": "2025-03",
     "round": "Series B",
     "amount": "$44M",
     "investors": "HarbourVest, NEOM Investment Fund (NIF), Arm IoT Fund, eLab Ventures, M Ventures, Motus Ventures",
     "source": "https://www.prnewswire.com/news-releases/memryx-raises-44-million-in-series-b-funding-for-advanced-edge-ai-computing-302412842.html"
    }
   ],
   "funding_notes": "Total raised ≈ **$63M** (https://www.crainsdetroit.com/technology/memryx-raises-44m-series-b-funding).",
   "trajectory": "Co-founded February 2019 by University of Michigan EECS professors **Wei Lu** (CTO;\nneuromorphic/in-memory computing, memristor researcher, Crossbar co-founder) and\n**Zhengya Zhang** (VLSI) (https://ece.engin.umich.edu/stories/michigan-startup-memryx-inc-promises-faster-cheaper-ai-processing).\nVeteran Qualcomm executive **Keith Kressin** was hired as President & CEO\n(https://www.m-ventures.com/news/ai-chip-innovator-memryx-hires-veteran-semiconductor-leader-keith-kressin-as-ceo).\nTrajectory: prototypes (2020–2022) → MX3 production and consumer-priced M.2 modules\n(2024) → $44M Series B (Mar 2025) → server (Cascade) and data-center (MX4) push in 2026.\nStatus mid-2026: active, shipping, one of the few in the cohort with real volume product.",
   "relevance": "A useful boundary case: founded by one of the best-known memristor academics, yet its\ncommercial product is fully digital at-memory dataflow — evidence of how startups with\nemerging-device roots have de-risked toward digital while the analog-memristor value\nproposition matures. Should be counted in the \"digital at/near-memory\" contrast group,\nnot as an analog memristor company.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "MemryX formed a strategic partnership with **Ebttikar** to deploy computer-vision edge AI across Saudi Arabia (2026-07-09), and reports expanded partnerships/deployments plus a new Americas sales director (2026-08-22). (https://www.prnewswire.com/news-releases/ebttikar-and-memryx-form-strategic-partnership-to-accelerate-computer-vision-edge-ai-across-saudi-arabia-302822297.html ; https://www.tipranks.com/news/private-companies/memryx-expands-edge-ai-partnerships-and-deployments-adds-americas-sales-director)"
    },
    {
     "date": "2026-09-22",
     "text": "**Lenovo MOU (2026-09-15)** to expand sovereign edge-AI deployments across Saudi Arabia: Lenovo ThinkEdge SE455 V3 servers + MemryX **Cascade 100P** accelerators, already live in a construction-safety analytics program and a port compliance/monitoring deployment; framework covers infrastructure, industrial, smart-city and video-management opportunities. (https://www.prnewswire.com/news-releases/memryx-and-lenovo-advance-sovereign-edge-ai-in-saudi-arabia-302879724.html) Also: Express LUCK selected MemryX for AI-enabled smart-manufacturing operations (2026-08-31; https://www.prnewswire.com/news-releases/express-luck-selects-memryx-for-ai-enabled-smart-manufacturing-operations-302864648.html)."
    }
   ],
   "sources": [
    {
     "title": "MemryX company page",
     "url": "https://memryx.com/company/"
    },
    {
     "title": "PRNewswire — $44M Series B (2025-03)",
     "url": "https://www.prnewswire.com/news-releases/memryx-raises-44-million-in-series-b-funding-for-advanced-edge-ai-computing-302412842.html"
    },
    {
     "title": "Crain's Detroit — Series B, totals, seed/Series A",
     "url": "https://www.crainsdetroit.com/technology/memryx-raises-44m-series-b-funding"
    },
    {
     "title": "UMich ECE — MemryX founding story",
     "url": "https://ece.engin.umich.edu/stories/michigan-startup-memryx-inc-promises-faster-cheaper-ai-processing"
    },
    {
     "title": "MemryX Developer Hub — architecture (digital, bfloat16/INT weights)",
     "url": "https://developer.memryx.com/architecture/architecture.html"
    },
    {
     "title": "PRNewswire — MX4 roadmap / memory wall",
     "url": "https://www.prnewswire.com/news-releases/memryx-unveils-mx4-roadmap-enabling-distributed-asynchronous-dataflow-for-highly-efficient-data-center-ai-302649698.html"
    },
    {
     "title": "Tom's Hardware — $149 MX3 M.2 module",
     "url": "https://www.tomshardware.com/tech-industry/artificial-intelligence/memryx-launches-usd149-mx3-m-2-ai-accelerator-module-capable-of-24-tops-compute-power"
    }
   ]
  },
  {
   "slug": "memsilicon",
   "name": "MemSilicon (铭芯启睿; Beijing MemSilicon Technology Co.)",
   "founded": 2024,
   "hq": "Beijing (Haidian District), China",
   "category": "emerging-device-imc",
   "status": "active",
   "origin": "academic spinoff (Chinese Academy of Sciences RRAM team of academician 刘明 Liu Ming; core team from CAS and Fudan University)",
   "total_funding_usd_m": 56,
   "last_round": "Series A, >CNY 400M (~US$56M+), 2026-10 (博原资本 lead)",
   "key_investors": [
    "博原资本 (Boyuan Capital)",
    "博世创投 (Bosch Ventures)",
    "米哈游 (miHoYo)",
    "兆易创新 (GigaDevice)",
    "小米战投 (Xiaomi)",
    "锦秋基金",
    "中科创星",
    "国开科创",
    "中关村科学城公司"
   ],
   "website": "n/d",
   "pitch": "\"以RRAM为存储地基、LPU为计算引擎\" — RRAM as the storage foundation and an LPU (language\nprocessing unit) as the compute engine, pursued as \"存算协同\" (storage-compute synergy) for\nlarge-model inference; three lines: embedded RRAM IP, a standalone 1 Gb RRAM chip\n(泰山一号), and \"RRAM与DRAM结合的混合存算\" (RRAM+DRAM hybrid compute-in-memory) for edge\ninference ([网易/华尔街见闻 2026-10-19](https://c.m.163.com/news/a/L9KA37BP05198NMR.html);\n[网易/科创板日报 2026-10-19](https://c.m.163.com/news/a/L9K99TB40550B1DU.html)).",
   "technical_approach": "- **Device:** ReRAM. 泰山一号 is a 1 Gb chip, 1T4R cell on 28 nm, 30 Mb/mm², <20 ns access,\n  taped out October 2026 and \"仍在验证\" (still under validation)\n  ([AIFlux 2026-10-19](https://aifluxnews.com/%E9%93%AD%E8%8A%AF%E5%90%AF%E7%9D%BF%E5%AE%8C%E6%88%90%E8%B6%854%E4%BA%BF%E5%85%83a%E8%BD%AE%E8%9E%8D%E8%B5%84%EF%BC%9A%E4%BB%8Erram%E5%AD%98%E5%82%A8%E8%B5%B0%E5%90%91%E5%A4%A7%E6%A8%A1%E5%9E%8B/)).\n- **Compute-in-memory claim:** the third product line is RRAM+DRAM hybrid 存算 with \"hybrid\n  3D integration\", for embodied AI and AI PCs — \"未披露具体芯片型号、流片进度或商业化节点\"\n  (no chip model, tape-out progress, or commercial timing disclosed). Whether multiply-\n  accumulate happens inside the RRAM array is not stated. **Treat as exploratory.**\n- **LPU:** the first-generation LPU \"不采用RRAM\" (does not use RRAM); it is a memory-\n  architecture-centric digital inference chip with targets of trillion-parameter models and\n  1,000 tokens/s per user, \"研发目标，尚无实测性能支撑\" (R&D targets, no measured data); no\n  tape-out schedule. The LPU is near-memory digital, i.e. boundary-class, not CIM.\n- No on-chip learning claim.",
   "funding_rounds": [
    {
     "date": "2024–25",
     "round": "Angel",
     "amount": "\"近亿元\" (~CNY 100M) per a company-site headline cited by AIFlux — unverified, not counted",
     "investors": "锦秋基金 (lead, confirmed by its partner's quote)",
     "source": "https://aifluxnews.com/… (secondary)"
    },
    {
     "date": "2026-01",
     "round": "Pre-A",
     "amount": "\">亿元\" (>CNY 100M) per a 知乎 column headline — unverified (page 403), not counted",
     "investors": "联想创投 (Lenovo Capital) et al.",
     "source": "https://zhuanlan.zhihu.com/p/1991466518786097306"
    },
    {
     "date": "2026-10",
     "round": "Series A",
     "amount": ">CNY 400M (~$56M+)",
     "investors": "博原资本 (lead); new: 博世创投, 创领资本, 米哈游, 中关村科学城公司, 兆易创新; existing: 复容投资, 国开科创, 恒裕资本, 锦秋基金, 顺禧基金, 小米战投, 中科创星",
     "source": "https://c.m.163.com/news/a/L9K99TB40550B1DU.html ; https://new.qq.com/rain/a/20261019A063PB00 ; https://www.163.com/dy/article/L9K5VM2405532TJY.html"
    }
   ],
   "funding_notes": "**Total:** verified floor **>CNY 400M ≈ US$56M** (†). If the angel and Pre-A headlines are\nright, cumulative is >CNY 600M (~$84M+). Valuation undisclosed.",
   "trajectory": "Founded May 2024 in Beijing to commercialize RRAM work from the Chinese Academy of Sciences\nteam of academician 刘明 (Liu Ming); chairman 刘琦 (Liu Qi), 20+ years in memory. On\n2026-10-19 卢涛 (Lu Tao) — former Graphcore president and Greater China GM, former Cavium\nChina GM, former VP at Lightelligence (曦智科技) — joined as CEO and co-founder, and the\ncompany announced a \"strategic upgrade\" from RRAM toward large-model inference. Status:\nactive; RRAM chip in validation, LPU in early R&D.",
   "relevance": "A ReRAM company whose compute story is, so far, a near-memory digital LPU plus an\nundisclosed hybrid-CIM research line — the same memory-first, compute-later shape as Weebit\nand CrossBar, with a Chinese-capital twist (Xiaomi, miHoYo, GigaDevice, Bosch). It adds a\ntenth-plus member and ~$56M to the memristor/ReRAM category without adding any analog-\ncompute or learning hardware. Facts straight: the device is real, the CIM is a plan.",
   "news": [],
   "sources": [
    {
     "title": "网易 (科创板日报): 铭芯启睿完成超4亿元A轮融资 博原资本领投 (2026-10-19)",
     "url": "https://c.m.163.com/news/a/L9K99TB40550B1DU.html"
    },
    {
     "title": "网易 (华尔街见闻): 铭芯启睿完成超4亿元A轮融资，启动LPU推理芯片研发 (2026-10-19)",
     "url": "https://c.m.163.com/news/a/L9KA37BP05198NMR.html"
    },
    {
     "title": "网易: MemSilicon融资超4亿元，芯片老将卢涛出任CEO (2026-10-19)",
     "url": "https://www.163.com/dy/article/L9K5VM2405532TJY.html"
    },
    {
     "title": "腾讯新闻: MemSilicon完成超4亿元A轮融资，宣布战略升级 (2026-10-19)",
     "url": "https://new.qq.com/rain/a/20261019A063PB00"
    },
    {
     "title": "AIFlux: 从RRAM存储走向大模型推理 (2026-10-19)",
     "url": "https://aifluxnews.com/%E9%93%AD%E8%8A%AF%E5%90%AF%E7%9D%BF%E5%AE%8C%E6%88%90%E8%B6%854%E4%BA%BF%E5%85%83a%E8%BD%AE%E8%9E%8D%E8%B5%84%EF%BC%9A%E4%BB%8Erram%E5%AD%98%E5%82%A8%E8%B5%B0%E5%90%91%E5%A4%A7%E6%A8%A1%E5%9E%8B/"
    },
    {
     "title": "知乎专栏: 「铭芯启睿」获联想创投等超亿元Pre-A轮融资 (2026-01-05)",
     "url": "https://zhuanlan.zhihu.com/p/1991466518786097306"
    }
   ]
  },
  {
   "slug": "mentium",
   "name": "Mentium Technologies",
   "founded": 2017,
   "hq": "Santa Barbara / Goleta, CA, USA",
   "category": "emerging-device-imc",
   "status": "active",
   "origin": "academic spinoff (Dmitri Strukov's lab, UC Santa Barbara)",
   "total_funding_usd_m": 0,
   "last_round": "undisclosed; 2 rounds per Crunchbase, most recent incl. Alumni Ventures",
   "key_investors": [
    "Silicon Catalyst",
    "National Science Foundation (SBIR)",
    "Alumni Ventures",
    "Unpopular Ventures"
   ],
   "website": "https://mentium.tech",
   "pitch": "Mentium pitches \"cloud-quality inference at the edge\" for mission-critical uses: a hybrid\nanalog in-memory + digital co-processor claiming <0.5W power, 10x higher speed/lower latency,\nand support for 10x larger neural networks than competing edge AI silicon\n([Edge AI & Vision Alliance profile](https://www.edge-ai-vision.com/companies/mentium-technologies/)).",
   "technical_approach": "Hybrid architecture (\"AIM\"): a digital processor combined with an **analog in-memory-compute**\nprocessor. The founding lineage is memristor crossbar research (Strukov's UCSB group, with\nKonstantin Likharev among the founders), but Crunchbase/trade descriptions of the product\nemphasize the analog IMC co-processor without confirming which memory device shipped —\nfoundry-available floating-gate/Flash-style analog compute is the approach associated with this\nlab lineage; not confirmed from primary Mentium sources in this sweep. Inference-only, edge\nco-processor. Target markets: mission-critical space, robotics, and security applications;\nfirst-generation chip was in engineering as of Aug 2019\n([CB Insights](https://www.cbinsights.com/company/mentium-technologies); [Synopsys cloud case\nstudy](https://www.synopsys.com/cloud/mentium-success.html) — Mentium designs on Synopsys cloud EDA).",
   "funding_rounds": [
    {
     "date": "2017–",
     "round": "Grants/incubation",
     "amount": "undisclosed",
     "investors": "NSF (SBIR), Silicon Catalyst incubator",
     "source": "[Crunchbase](https://www.crunchbase.com/organization/mentium)"
    },
    {
     "date": "n.d. (2 rounds)",
     "round": "Venture, undisclosed",
     "amount": "undisclosed",
     "investors": "Alumni Ventures (most recent), Unpopular Ventures",
     "source": "[Crunchbase](https://www.crunchbase.com/organization/mentium), [Tracxn](https://tracxn.com/d/companies/mentium/__8Az6yx1mf0kknvorGvmIjp1El0inF5CrGx9MjsY0grY)"
    }
   ],
   "funding_notes": "Total raised: **undisclosed** (no dollar amounts published). Frontmatter total left at 0 for lack\nof a disclosed figure.",
   "trajectory": "Founded February 2017 (some directories say 2016) in Santa Barbara/Goleta by **Mirko Prezioso\n(CEO), Farnood Merrikh-Bayat (CTO), Dmitri Strukov, and Konstantin Likharev** — Prezioso and\nMerrikh-Bayat were lead researchers in Strukov's UCSB memristor/analog-computing group.\nIncubated via UCSB CNSI / Silicon Catalyst. **Still active:** headcount grew from ~5 (Feb 2023)\nto ~8 (Oct 2024) to ~12 (June 2026) per Crustdata tracking, with a space/mission-critical edge\nAI focus — small and quiet rather than defunct\n([Crustdata](https://crustdata.com/profiles/company/mentium-technologies-inc)).",
   "relevance": "Mentium is the direct commercial descendant of the UCSB memristor-crossbar school (Strukov/\nLikharev), contemporaneous with Knowm's product launch era — and illustrates the common\ntrajectory of memristor-lab spinoffs toward foundry-compatible analog IMC and inference-only\nproducts, versus Knowm's discrete-device + on-chip-learning path.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Mentium's **Luna-R1 AI chip was selected for the ET-01 constellation mission** — its first multi-satellite deployment (announced ~2026-06-04); next-gen **Luna-2** (4.2 TOPS) in development. (https://www.edge-ai-vision.com/2026/06/mentium-technologies-luna-r1-ai-chip-selected-for-et-01-constellation-mission-first-multi-satellite-deployment-of-mentium-hardware/ ; https://innovation.ucsb.edu/blog/news/mentium-technologies-luna-r1-ai-chip-selected-et-01-constellation-mission-first-multi)"
    }
   ],
   "sources": [
    {
     "title": "Mentium — Crunchbase",
     "url": "https://www.crunchbase.com/organization/mentium"
    },
    {
     "title": "Mentium — Tracxn",
     "url": "https://tracxn.com/d/companies/mentium/__8Az6yx1mf0kknvorGvmIjp1El0inF5CrGx9MjsY0grY"
    },
    {
     "title": "Mentium — Edge AI & Vision Alliance",
     "url": "https://www.edge-ai-vision.com/companies/mentium-technologies/"
    },
    {
     "title": "Mentium — CB Insights",
     "url": "https://www.cbinsights.com/company/mentium-technologies"
    },
    {
     "title": "Mentium success story — Synopsys",
     "url": "https://www.synopsys.com/cloud/mentium-success.html"
    },
    {
     "title": "Mentium headcount — Crustdata",
     "url": "https://crustdata.com/profiles/company/mentium-technologies-inc"
    },
    {
     "title": "UCSB Startup Village 2022 — Mentium",
     "url": "https://tia.ucsb.edu/startup-village-2022/mentium-technologies/"
    }
   ]
  },
  {
   "slug": "mythic",
   "name": "Mythic",
   "founded": 2012,
   "hq": "Austin, TX / Redwood City, CA, USA",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (Michigan Integrated Circuits Lab, University of Michigan)",
   "total_funding_usd_m": 303,
   "last_round": "Growth round, $125M, 2025-12",
   "key_investors": [
    "DCVC",
    "Lux Capital",
    "Atreides Management",
    "BlackRock",
    "Hewlett Packard Enterprise",
    "SoftBank Ventures",
    "Threshold/DFJ",
    "Honda",
    "Lockheed Martin"
   ],
   "website": "https://mythic.ai",
   "pitch": "Mythic says its Analog Processing Units (APUs) are \"a fundamentally different approach\nto AI inference that collapses compute and memory into a single plane,\" and that \"by\novercoming the energy bottlenecks of traditional digital architectures\" it delivers\nperformance-per-watt digital processors cannot — claiming an architecture \"100x more\nenergy-efficient than today's top-of-the-line GPUs.\"\nSource: https://mythic.ai/whats-new/mythic-to-challenge-ais-gpu-pantheon-with-100x-energy-advantage-and-oversubscribed-125m-raise/",
   "technical_approach": "- Analog compute-in-memory using embedded NOR **flash** cells: weights are stored as\n  flash-cell conductances and matrix-vector multiplies happen in the analog domain\n  inside the memory array, with on-chip ADC/DAC conversion around a tiled architecture\n  (\"Analog Matrix Processor\" / AMP). Inference-only; models are trained off-chip and\n  quantized (8-bit) onto the array. (https://venturebeat.com/business/mythic-raises-70m-to-disrupt-ai-chips-with-analog-and-flash-components)\n- Named chips: **M1108 AMP** (first product generation, announced 2020) and the smaller\n  **M1076** (2021) for edge vision/video analytics, sold as chips and PCIe cards.\n  (https://www.prnewswire.com/news-releases/mythic-raises-70-million-in-series-c-funding-led-by-blackrock-and-hewlett-packard-enterprise-301288109.html)\n- Post-recapitalization roadmap (Dec 2025): next-generation APUs aimed at LLM inference\n  from edge to data center; claims air-cooled servers with \"two orders of magnitude\n  better TCO per token\" than Nvidia Blackwell-generation hardware for 70B models.\n  (https://mythic.ai/whats-new/mythic-to-challenge-ais-gpu-pantheon-with-100x-energy-advantage-and-oversubscribed-125m-raise/)",
   "funding_rounds": [
    {
     "date": "2017-03",
     "round": "Series A",
     "amount": "$9M",
     "investors": "DFJ (lead)",
     "source": "https://medium.com/@DFJvc/mythic-enabling-intelligence-at-the-edge-3b08479d1798"
    },
    {
     "date": "2018-03",
     "round": "Series B",
     "amount": "$40M",
     "investors": "SoftBank Ventures (lead); DFJ, Lux, DCVC",
     "source": "https://www.prnewswire.com/news-releases/mythic-closes-40m-series-b-round-to-deliver-a-new-compute-paradigm-bringing-the-power-of-a-datacenter-on-a-tiny-chip-to-any-device-300616318.html"
    },
    {
     "date": "2019",
     "round": "Series B-1",
     "amount": "$30M",
     "investors": "SoftBank Ventures Asia, Threshold, Lux, DCVC, AME Cloud, Lockheed Martin Ventures",
     "source": "https://mythic.ai/whats-new/mythic-secures-30m-in-new-financing-and-strengthens-leadership-team/"
    },
    {
     "date": "2021-05",
     "round": "Series C",
     "amount": "$70M",
     "investors": "BlackRock + HPE (co-leads); total then $165.2M",
     "source": "https://www.prnewswire.com/news-releases/mythic-raises-70-million-in-series-c-funding-led-by-blackrock-and-hewlett-packard-enterprise-301288109.html"
    },
    {
     "date": "2023-03",
     "round": "Recapitalization",
     "amount": "$13M",
     "investors": "Atreides, DCVC, Lux + new Catapult Ventures, Hermann Hauser Investment; total then $178M",
     "source": "https://www.businesswire.com/news/home/20230309005131/en/Mythic-Raises-%2413-Million-to-Bring-Its-Next-generation-Analog-Computing-Solution-to-Market"
    },
    {
     "date": "2025-12",
     "round": "Growth (oversubscribed)",
     "amount": "$125M",
     "investors": "DCVC (lead); Honda, Lockheed Martin strategic; NEA, Atreides et al.",
     "source": "https://siliconangle.com/2025/12/17/compute-memory-chip-startup-mythic-raises-125m-round/"
    }
   ],
   "funding_notes": "Total raised ≈ $303M ($178M through Mar 2023 per BusinessWire, + $125M Dec 2025 per\nSiliconANGLE/Bloomberg: https://www.bloomberg.com/news/articles/2025-12-17/ai-chip-startup-mythic-raises-125-million-in-bid-to-take-on-nvidia).",
   "trajectory": "Founded 2012, incubated at the Michigan Integrated Circuits Lab (University of Michigan)\nby Mike Henry and Dave Fick (https://medium.com/@DFJvc/mythic-enabling-intelligence-at-the-edge-3b08479d1798).\nNear-death at end of 2022: after failing to raise new capital it went through a radical\nrestructure and mass layoffs; existing investors put in $13M in March 2023 \"to keep the\nlights on\" and appointed co-founder and former CTO Dave Fick as CEO\n(https://www.eetimes.com/mythic-rises-from-the-ashes-with-125-million-funding-round/).\nDecember 2025: oversubscribed $125M round to challenge GPUs on datacenter LLM inference.\n**Status mid-2026: active**, the best-funded flash-analog IMC player after its\nrecapitalization.",
   "relevance": "Mythic is the longest-running proof that analog in-memory compute with a conventional\ndevice (NOR flash) can reach silicon products — and its 2022 collapse/2023 recap shows\nthe commercialization risk of the category. Like nearly all analog-IMC startups, it is\ninference-only: weights are programmed, not learned on-chip, in contrast to the\non-chip-learning ambitions of AHaH/memristor approaches.",
   "news": [
    {
     "date": "2026-09-22",
     "text": "**Defense Advisory Board formed (2026-09-16)** — chaired by Stephen Rodriguez (DCVC), with former Under Secretary of Defense Ellen Lord and former Lockheed Skunk Works CTO John Clark among founding members, to move the APU into national-security programs of record. CEO Taner Ozcelik repeats the \"up to 1000x lower SWaP-C\" claim (a marketing figure, not a benchmark). First company news since the Dec-2025 $125M round; the wedge is now explicitly defense/tactical edge alongside the datacenter-LLM messaging. (https://www.businesswire.com/news/home/20260916833376/en/Mythic-Announces-Formation-of-Defense-Advisory-Board-to-Accelerate-AI-Hardware-Adoption-Across-the-U.S.-National-Security-Enterprise)"
    },
    {
     "date": "2026-10-27",
     "text": "**Bengaluru Centre of Excellence opened (~2026-09-25/28); India team to quadruple by Q1 2027** (base headcount undisclosed). Scope: front-end design/verification of the analog compute technology, hybrid analog-digital physical design, DFT for automotive, systems/applications engineering; initial contribution to the **Vanguard** inference platform. VP Manoj Kumar leads. CEO Ozcelik's India messaging: datacentre, automotive and robotics; '100x efficiency' claim repeated (marketing figure). (https://www.ciol.com/news/mythic-ai-bengaluru-centre-quadruple-india-team-2027-12584333 ; https://m.economictimes.com/tech/startups/chip-startup-mythic-ai-enters-india-eyes-data-centre-robotics-opportunities/articleshow/134466112.cms)"
    }
   ],
   "sources": [
    {
     "title": "Mythic $125M PR",
     "url": "https://mythic.ai/whats-new/mythic-to-challenge-ais-gpu-pantheon-with-100x-energy-advantage-and-oversubscribed-125m-raise/"
    },
    {
     "title": "EE Times: Mythic Rises from the Ashes",
     "url": "https://www.eetimes.com/mythic-rises-from-the-ashes-with-125-million-funding-round/"
    },
    {
     "title": "SiliconANGLE 2025-12-17",
     "url": "https://siliconangle.com/2025/12/17/compute-memory-chip-startup-mythic-raises-125m-round/"
    },
    {
     "title": "Series C PR 2021",
     "url": "https://www.prnewswire.com/news-releases/mythic-raises-70-million-in-series-c-funding-led-by-blackrock-and-hewlett-packard-enterprise-301288109.html"
    },
    {
     "title": "$13M recap PR 2023",
     "url": "https://www.businesswire.com/news/home/20230309005131/en/Mythic-Raises-%2413-Million-to-Bring-Its-Next-generation-Analog-Computing-Solution-to-Market"
    }
   ]
  },
  {
   "slug": "neurobus",
   "name": "Neurobus",
   "founded": 2023,
   "hq": "Toulouse (& Paris), France",
   "category": "digital-snn",
   "status": "active",
   "origin": "serial founder (Florian Corgnou — ex-Tesla, HEC Paris; Airbus space-entrepreneurship program) + academic (Gregor Lenz, co-founder/CTO — Institut de la Vision PhD, ex-SynSense)",
   "total_funding_usd_m": 0.12,
   "last_round": "Pre-seed / grants + revenue; ~$120K disclosed; €5–10M round targeted",
   "key_investors": [
    "Creative Destruction Lab",
    "Agoranov",
    "SpaceFounders",
    "Incubateur HEC Paris",
    "SURE5.0"
   ],
   "website": "https://neurobus.ai",
   "pitch": "Neurobus frames space and defense autonomy as impossible under conventional compute\npower budgets: satellites, drones and defense platforms need real-time perception\nbut have tiny energy and thermal envelopes. Its answer is \"frugal AI\" — pairing\nevent-based (neuromorphic) vision sensors with neuromorphic processors so the system\ncomputes only on change, claiming up to ~100× lower energy and sub-millisecond\nreaction versus frame-based pipelines. (Framing per neurobus.ai and HEC Paris /\nUsine Nouvelle profiles.)",
   "technical_approach": "- Neurobus is a **systems integrator**, not a chip designer: it combines third-party\n  event-based vision sensors (e.g. Prophesee-style dynamic-vision sensors) with\n  neuromorphic/SNN processors (e.g. Intel Loihi–class or other event-driven silicon)\n  and its own software for real-time detection and tracking on the edge.\n- Applications: space situational awareness, on-board satellite vision, and — with\n  rapid expansion — drones and defense autonomy. Partners/customers cited include\n  Airbus Defence and Space, the European Space Agency (ESA) and the French space\n  agency (CNES). Inference-oriented, event-driven; the \"neuromorphic\" claim is at the\n  sensor+processor system level rather than a novel device.",
   "funding_rounds": [
    {
     "date": "2023–2024",
     "round": "Grants / accelerator + contract revenue",
     "amount": "~$120K disclosed (PitchBook)",
     "investors": "Creative Destruction Lab, Agoranov, SpaceFounders, Incubateur HEC Paris, SURE5.0",
     "source": "PitchBook / Crunchbase"
    },
    {
     "date": "2024",
     "round": "R&D revenue (not equity)",
     "amount": "~€600K (from defense/aerospace contracts)",
     "investors": "—",
     "source": "Usine Nouvelle / HEC"
    },
    {
     "date": "Targeted (2025–26)",
     "round": "Seed",
     "amount": "€5–10M (planned, not confirmed closed)",
     "investors": "—",
     "source": "Usine Nouvelle / HEC"
    },
    {
     "date": "—",
     "round": "Total equity (disclosed)",
     "amount": "~$0.12M",
     "investors": "—",
     "source": "PitchBook"
    }
   ],
   "funding_notes": "**Sourcing caveat:** Neurobus's equity funding is not well documented. Public\ndatabases show only ~$120K disclosed plus grant/accelerator support and ~€600K of\n2024 contract revenue; the company was reported to be raising a €5–10M round but no\nclose was confirmed in sources as of mid-2026. The briefed \"seed €1–6M\" could not be\nverified — treat the funding total as provisional/under-reported.",
   "trajectory": "Founded 2023 by Florian Corgnou (ex-Tesla) out of an Airbus-linked space\nentrepreneurship program, based in Toulouse with Paris presence; incubated at HEC\nParis / Agoranov / Creative Destruction Lab / SpaceFounders. It started in space\n(ESA, CNES, Airbus DS partnerships), funded largely by grants and R&D contracts\n(~€600K revenue in 2024), and has been expanding into drones and defense while\npreparing a larger raise. Status mid-2026: active, early-stage integrator; still\nlargely grant/contract-funded with no confirmed institutional seed close.\n\n**Founders (sweep 2026-09-22):** Florian Corgnou (CEO; HEC Paris; ex-Tesla Europe, then co-founded a fintech) and Gregor Lenz (co-founder & CTO; PhD Sorbonne Université at the Institut de la Vision, thesis director Sio-Hoï Ieng, defended 2021-07-06; then at SynSense until mid-2023) (https://app.dealroom.co/news/note/neurobus-frugal-ai-for-defense-and-aerospace-autonomy ; https://europe.startupguide.com/interview/florian-corgnou ; https://theses.hal.science/tel-03474197 ; https://open-neuromorphic.org/contributors/gregor-lenz/ ; https://www.eetimes.com/podcasts/synsense-research-head-talks-combined-sensing-processing/).",
   "relevance": "Neurobus represents the \"integrator\" wing of the neuromorphic wave — it buys\nevent-cameras and SNN chips and sells systems, rather than inventing a compute\nsubstrate. That makes it a useful contrast: the low-power/on-device value it markets\nis the same promise Knowm makes for memristor/AHaH hardware, but Neurobus attacks it\nby assembling existing digital-SNN silicon for a defense/space niche, carrying no\ndevice-physics or on-chip-learning claim of its own.",
   "news": [],
   "sources": [
    {
     "title": "Neurobus — company site",
     "url": "https://neurobus.ai/"
    },
    {
     "title": "Neurobus: Bringing Frugal AI to Defense and Aerospace — HEC Paris",
     "url": "https://www.hec.edu/en/dare/innovation-entrepreneurship/neurobus-bringing-frugal-ai-defense-and-aerospace"
    },
    {
     "title": "Fondée par un ex-Tesla, Neurobus veut embarquer l'informatique neuromorphique dans les drones — Usine Nouvelle",
     "url": "https://www.usinenouvelle.com/mes-magazines/magazine-usine-nouvelle/fondee-par-un-ex-tesla-la-pepite-francaise-neurobus-veut-faire-embarquer-linformatique-neuromorphique-dans-les-drones.A2EJS7NVXRJR7GVNBATRXAESTU.html"
    },
    {
     "title": "Neurobus — Crunchbase",
     "url": "https://www.crunchbase.com/organization/neurobus"
    },
    {
     "title": "Neurobus — PitchBook profile",
     "url": "https://pitchbook.com/profiles/company/538219-00"
    }
   ]
  },
  {
   "slug": "neuromorphyx",
   "name": "Neuromorphyx",
   "founded": 2025,
   "hq": "Pula, Croatia",
   "category": "digital-snn",
   "status": "active",
   "origin": "serial founders (Nex Novus d.o.o.; CEO Ivan Projić, CTO Filip Gembec)",
   "total_funding_usd_m": 0,
   "last_round": "none disclosed (as of 2026-08-23)",
   "key_investors": [],
   "website": "https://neuromorphyx.com",
   "pitch": "\"Reflex Intelligence for Machines\": compact, complete neuromorphic vision products for\ndemanding robotics, space, defense, automotive and industrial applications — event-based\nsensing plus spiking-network inference at ultra-low power, instead of frame-based GPU\npipelines (CEO Ivan Projić in the BrainBoard1500 launch release,\nhttps://www.hpcwire.com/aiwire/2026/08/20/brainchip-and-neuromorphyx-launch-akd1500-developer-board/).",
   "technical_approach": "Systems integration of third-party neuromorphic silicon, sold as finished hardware:\n- **BrainBoard1500** (launched with BrainChip, announced 2026-08-17..20): ultra-low-power\n  embedded developer/evaluation board for BrainChip's AKD1500 neural-network accelerator,\n  aimed at bringing neuromorphic AI to embedded design engineers\n  (https://www.newelectronics.co.uk/content/news/brainchip-and-neuromorphyx-present-brainboard1500-expanding-access-to-neuromorphic-ai-development).\n- **NeuroCore**: Efinix Ti60 FPGA + BrainChip AKD1500 board (https://neuromorphyx.com/neurocore).\n- Four Akida-powered products total as of mid-2026 (NeuroCore, Neuroflex, and two others per\n  BrainChip-community reporting, https://x.com/MIAM_Schwan/status/2076971828942123324).\n- Parent/legal entity Nex Novus d.o.o. offers custom SNN implementations with event-based\n  vision (DVS), radar, IMU and audio sensing (https://www.linkedin.com/company/nex-novus).\nDescribed by BrainChip as a lead partner; positioning is inference-only edge AI (no on-chip\nlearning claim of its own — the learning locus is whatever Akida provides).",
   "funding_rounds": [
    {
     "date": "—",
     "round": "none disclosed",
     "amount": "—",
     "investors": "—",
     "source": "checked 2026-08-23: no round in trade press, Tracxn/PitchBook snippets, or company materials"
    }
   ],
   "funding_notes": "Appears bootstrapped / revenue-funded via Nex Novus engineering services. Promote-to-update\nwhen a round appears.",
   "trajectory": "Founded 2025 (company LinkedIn, https://hr.linkedin.com/company/neuromorphyx) in Pula, Croatia,\nas the product brand of Nex Novus d.o.o. (VAT HR27018825320, Kandlerova 34, Pula); CEO Ivan\nProjić, CTO Filip Gembec. Shipped a product line of Akida-based boards/modules through 2026,\nculminating in the co-branded BrainBoard1500 launch with BrainChip (2026-08), which drew\ncoverage from HPCwire, New Electronics, Design&Reuse, Hackster.io and syndicated press —\nunusual visibility for a company this size. Status: active, product-shipping.\n\n**Founders (sweep 2026-09-22):** Ivan Projić (co-founder & CEO from Apr 2025; founder & CEO of Nex Novus since Jul 2023) and Filip Gembec (co-founder & CTO from Dec 2025, embedded-systems designer). Degrees and prior employers not found in public sources (https://hr.linkedin.com/in/ivan-projic ; https://hr.linkedin.com/in/filip-gembec).",
   "relevance": "A downstream integrator monetizing someone else's digital-SNN silicon — the mirror image of\nKnowm's position as a component/IP originator; it shows the neuromorphic market maturing to the\npoint of third-party board vendors, something the memristor/analog-learning side (AHaH) does\nnot yet have.",
   "news": [],
   "sources": [
    {
     "title": "HPCwire: BrainChip and Neuromorphyx Launch AKD1500 Developer Board",
     "url": "https://www.hpcwire.com/aiwire/2026/08/20/brainchip-and-neuromorphyx-launch-akd1500-developer-board/"
    },
    {
     "title": "New Electronics: BrainBoard1500 expands access to neuromorphic AI",
     "url": "https://www.newelectronics.co.uk/content/news/brainchip-and-neuromorphyx-present-brainboard1500-expanding-access-to-neuromorphic-ai-development"
    },
    {
     "title": "Design&Reuse: AKD1500 embedded developer board announcement",
     "url": "https://www.design-reuse.com/news/202531001-brainchip-and-neuromorphyx-announce-an-akd1500-embedded-developer-board-bringing-neuromorphic-ai-to-the-embedded-engineering-community/"
    },
    {
     "title": "Neuromorphyx — NeuroCore product page",
     "url": "https://neuromorphyx.com/neurocore"
    },
    {
     "title": "Neuromorphyx company LinkedIn (founded 2025)",
     "url": "https://hr.linkedin.com/company/neuromorphyx"
    },
    {
     "title": "Nex Novus d.o.o. LinkedIn",
     "url": "https://www.linkedin.com/company/nex-novus"
    }
   ]
  },
  {
   "slug": "neuronova",
   "name": "Neuronova",
   "founded": 2024,
   "hq": "Milan, Italy",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (Politecnico di Milano — Daniele Ielmini's group, via PoliHub; CTO Michele Mastella from CogniGron / University of Groningen, Elisabetta Chicca's group)",
   "total_funding_usd_m": 1.6,
   "last_round": "Pre-seed, €1.5M (~$1.6M), 2024-11",
   "key_investors": [
    "CDP Venture Capital (Tech4Planet)",
    "360 Capital"
   ],
   "website": "https://www.neuronovatech.com",
   "pitch": "Current edge hardware is \"too power-hungry or too limited in computational ability\"; Neuronova's\nchip \"processes and stores data together on individual neurons,\" avoiding \"potential bottlenecks\nwhen transferring data between separate processing and memory components\" — claiming up to\n1,000x lower energy than current solutions for always-on sensing\n(https://tech.eu/2025/01/20/bringing-the-brain-to-silicon-neuronovas-vision-for-energy-efficient-ai-hardware/).",
   "technical_approach": "Analog neuromorphic processor for always-on edge sensing. First chip **Neuronova H1**: \"an\nanalog neuromorphic processor designed for real-time intelligence at sub-microwatt power\nlevels\" (company job posting, 2026-08-04,\nhttps://it.linkedin.com/jobs/view/senior-ic-lead-at-neuronova-4448148231); company site claims\nAI workloads at under 1 µW (https://www.neuronovatech.com). ~3mm x 3mm die, CMOS-compatible\n(standard foundries), compute and memory co-located in analog \"neuron\" elements; target\nworkloads are local voice/audio recognition and always-on sensing on battery devices —\ninference at the edge, no on-chip-learning claim found. ~99% of design and processing done in\nItaly (tech.eu interview). EE Times \"Brains and Machines\" podcast episode on the sub-µW chip,\n2026-01-08 (https://www.eetimes.com/podcasts/green-always-on-sensing-with-neuronovas-sub-%CE%BCwatt-chip/).",
   "funding_rounds": [
    {
     "date": "2024 (founding)",
     "round": "grant",
     "amount": "€30K",
     "investors": "PoliHub pitch competition",
     "source": "https://tech.eu/2025/01/20/bringing-the-brain-to-silicon-neuronovas-vision-for-energy-efficient-ai-hardware/"
    },
    {
     "date": "2024-11",
     "round": "Pre-seed",
     "amount": "€1.5M",
     "investors": "CDP Venture Capital (Tech4Planet fund), 360 Capital",
     "source": "https://www.cdpventurecapital.it/en/news.page?contentId=COM37170 ; https://polihub.it/en/news/neuronova-raises-1-5-million-in-pre-seed-funding-round/"
    }
   ],
   "funding_notes": "Total verified: €1.5M pre-seed + €30K grant (~$1.6M). (tech.eu dates the round October 2024;\nthe CDP/PoliHub announcements ran late November 2024.)",
   "trajectory": "Founded 2024 (Wired Italia: July 2024, https://www.wired.it/article/neuronova-chip-cervello-energia/)\nby three engineers with PhDs in neuromorphic computing: Alessandro Milozzi (CEO) and Michele\nMastella (CTO) per Economyup (https://www.economyup.it/startup/101-startup-innovative-che-le-aziende-dovrebbero-seguire/),\nwith Marco Rasetto also announcing the pre-seed as a team member\n(https://www.linkedin.com/posts/marco-rasetto_im-thrilled-to-announce-neuronovas-pre-seed-activity-7269672088234328064-whCY).\nSpun out of PoliHub, Politecnico di Milano's incubator. Dealroom lists 2–10 employees, Milan.\nMid-2026: actively hiring senior IC designers for the H1 (2026-08), EU Funding & Tenders Portal\nregistered, press profile growing (EE Times podcast, Wired Italia, Tech Europe Foundation\ninterview). Status: active, pre-product (H1 in development).\n\n**Founders (sweep 2026-09-22):** Alessandro Milozzi (CEO; MS 2020 and PhD 2024 at Politecnico di Milano under Daniele Ielmini, whose CV lists him \"now with NeuroNova\"), Michele Mastella (CTO; PhD at CogniGron, University of Groningen, Aug 2020–Mar 2024, Elisabetta Chicca's bio-inspired circuits group) and Marco Rasetto (co-founder; PoliMi electronic-engineering graduate, PhD in neuromorphic computing). Incubated at PoliHub (https://onlineservices.polimi.it/manifesti/manifesti/controller/ricerche/Cv.do?evn_downloadcv=evento&id_cv=1222974&jaf_currentWFID=main ; https://polihub.it/en/news/neuronova-raises-1-5-million-in-pre-seed-funding-round/ ; https://it.linkedin.com/in/michelemastella ; https://www.rug.nl/research/zernike/bio-inspired-circuits-and-systems/chicca-group/news/20200801-new-research-group-bio-inspired-circuits-systems?lang=en).",
   "relevance": "The purest European analog-neuromorphic-edge play in the corpus — sub-µW analog neurons with\nco-located memory and compute, the same \"physics does the math\" wager as memristor approaches,\nbut implemented in standard CMOS rather than emerging devices, and inference-only rather than\non-chip learning.",
   "news": [
    {
     "date": "2026-10-27",
     "text": "**Headline only — New Electronics (2026-10-26): 'Neuronova receives first production run of neuromorphic AI chips.'** Article body was bot-blocked; company site now says 'silicon-validated', eval-kit pre-orders open, investors 360 Capital and tech4planet. Treat as **first silicon of the H1 received**, details (foundry, node, quantity) unverified. (https://www.newelectronics.co.uk/content/news/neuronova-receives-first-production-run-of-neuromorphic-ai-chips ; https://www.neuronovatech.com/)"
    }
   ],
   "sources": [
    {
     "title": "CDP Venture Capital: Neuronova raises €1.5 million",
     "url": "https://www.cdpventurecapital.it/en/news.page?contentId=COM37170"
    },
    {
     "title": "PoliHub: Neuronova raises €1.5 million in pre-seed",
     "url": "https://polihub.it/en/news/neuronova-raises-1-5-million-in-pre-seed-funding-round/"
    },
    {
     "title": "tech.eu: Bringing the brain to silicon (2025-01-20)",
     "url": "https://tech.eu/2025/01/20/bringing-the-brain-to-silicon-neuronovas-vision-for-energy-efficient-ai-hardware/"
    },
    {
     "title": "EE Times podcast: Green Always-On Sensing With Neuronova's Sub-μWatt Chip (2026-01-08)",
     "url": "https://www.eetimes.com/podcasts/green-always-on-sensing-with-neuronovas-sub-%CE%BCwatt-chip/"
    },
    {
     "title": "Wired Italia: Neuronova sta progettando un chip che replica il cervello (2026-06-22)",
     "url": "https://www.wired.it/article/neuronova-chip-cervello-energia/"
    },
    {
     "title": "Neuronova Senior IC Lead job posting (2026-08-04, names the H1)",
     "url": "https://it.linkedin.com/jobs/view/senior-ic-lead-at-neuronova-4448148231"
    },
    {
     "title": "Neuronova Semiconductors site",
     "url": "https://www.neuronovatech.com/"
    },
    {
     "title": "Economyup: 101 startup innovative (founders/CTO)",
     "url": "https://www.economyup.it/startup/101-startup-innovative-che-le-aziende-dovrebbero-seguire/"
    }
   ]
  },
  {
   "slug": "neurophos",
   "name": "Neurophos",
   "founded": 2020,
   "hq": "Austin, TX, USA (spun out of Duke University, Durham, NC)",
   "category": "photonic",
   "status": "active",
   "origin": "academic spinoff (Duke — David R. Smith's metamaterials lab, via the Metacept incubator; founders Patrick Bowen, Tom Driscoll, Andrew Traverso)",
   "total_funding_usd_m": 118,
   "last_round": "Series A, $110M, 2026-01",
   "key_investors": [
    "Gates Frontier",
    "M12 (Microsoft)",
    "Aramco Ventures",
    "Bosch Ventures",
    "Carbon Direct Capital",
    "MetaVC",
    "Mana Ventures",
    "Silicon Catalyst",
    "Space Capital",
    "Tectonic Ventures"
   ],
   "website": "https://www.neurophos.com",
   "pitch": "AI datacenters are hitting a power wall; Neurophos claims metamaterial optical AI chips that\nare \"super-fast, energy-efficient\" — GPU-class and beyond compute at a fraction of the\nenergy, with claims around 300 TOPS/W and \"exaflop-scale\" photonic AI chips for inference in\ndatacenters. (https://www.prweb.com/releases/neurophos-raises-7-2-million-round-to-create-super-fast-energy-efficient-metamaterial-based-optical-ai-chips-for-data-centers-302015103.html, https://otc.duke.edu/news/neurophos-secures-110-million-series-a-to-launch-exaflop-scale-photonic-ai-chips/)",
   "technical_approach": "- Substrate: **metasurface/metamaterial optical in-memory compute** driven by silicon\n  photonics. The core claim: micron-scale metamaterial optical modulators — a ~10,000x area\n  miniaturization vs conventional photonic elements (MZIs) — arranged as a dense metasurface\n  array that performs matrix multiplication optically, \"in-memory\" (weights held in the\n  modulator array), with high-speed silicon photonics feeding it.\n  (https://optics.org/news/neurophos-raises-110m-for-photonic-chips-based-around-metamaterial-modulators, https://www.eetimes.com/can-metamaterials-revolutionize-optical-computing/)\n- This is the density play: where MZI-mesh photonics (Lightmatter/Lightelligence) hit a\n  scaling wall from device size, metamaterial cells aim to make optical matmul arrays big\n  enough to matter — claimed path to exaflop-scale inference chips.\n- Positioning is datacenter **inference** acceleration; training stays off-chip on GPUs. No\n  on-chip learning claim. No shipped product yet; Series A (Jan 2026) is to \"launch\" the\n  first chips.",
   "funding_rounds": [
    {
     "date": "2023-12 / 2024-01",
     "round": "Seed",
     "amount": "$7.2M",
     "investors": "Gates Frontier (lead); MetaVC, Mana Ventures",
     "source": "https://www.prweb.com/releases/neurophos-raises-7-2-million-round-to-create-super-fast-energy-efficient-metamaterial-based-optical-ai-chips-for-data-centers-302015103.html"
    },
    {
     "date": "2026-01",
     "round": "Series A",
     "amount": "$110M (oversubscribed)",
     "investors": "Gates Frontier (lead); M12, Aramco Ventures, Bosch Ventures, Carbon Direct Capital, Gaingels, Silicon Catalyst, Space Capital, Tectonic, DNX, Alumni Ventures, Morgan Creek",
     "source": "https://otc.duke.edu/news/neurophos-secures-110-million-series-a-to-launch-exaflop-scale-photonic-ai-chips/"
    }
   ],
   "funding_notes": "Total raised: **~$118M** (optics.org). No valuation reported. Founding-year note: Preqin/\nTracxn list 2020 incorporation; press variously describes the Duke/Metacept spinout as\n2020-2022.",
   "trajectory": "Founded by Patrick Bowen (CEO, Duke PhD under David R. Smith — the negative-index\nmetamaterials pioneer), with Tom Driscoll and Andrew Traverso, incubated at Metacept and\nspun out with Duke-licensed IP; relocated to Austin, TX. Quiet until the Gates Frontier seed\n(announced Jan 2024), then a step-change $110M Series A in Jan 2026 with corporate-strategic\nmoney (Microsoft's M12, Aramco, Bosch). As of mid-2026: active, pre-revenue, building its\nfirst metasurface photonic inference processors.",
   "relevance": "The purest surviving \"analog optical in-memory compute\" bet in the batch — weights resident\nin a physical array, compute-in-memory, exactly the architectural argument memristor IMC\nmakes, but with photonic modulators instead of resistive synapses. Also the counter-example\nto the compute-to-interconnect slide: fresh 2026 money is still funding optical *matmul*.\nLike the others, no on-chip learning — adaptation remains off-chip.",
   "news": [
    {
     "date": "2026-09-22",
     "text": "**Scale-up after the $110M Series A (PR 2026-09-10):** headcount up 4x in 2026 with another 2x planned by year-end; new Sunnyvale, CA engineering office alongside Austin; Dr. Nathan Kundtz (Kymeta founder/CEO) appointed CTO, Michael Shebanow promoted to SVP Engineering, Kevin Keller General Counsel, Preston Woo adds CCO; **Nigel Toon (Graphcore co-founder/ex-CEO) joins the board.** Company describes itself as heading \"toward scale production\" of metasurface photonic AI chips — no product ship date given. (https://www.prnewswire.com/news-releases/neurophos-scales-team-and-opens-silicon-valley-office-to-accelerate-commercialization-302874375.html)"
    },
    {
     "date": "2026-10-27",
     "text": "**MIT CSAIL Julia Lab collaboration (2026-10-21)** to fit numerical algorithms to its Optical Processing Unit; PR states the first chip targets ~50x lower compute resources and **commercial systems in 2028** — the first public ship-year. (https://www.prnewswire.com/news-releases/collaboration-pairs-julia-labs-advanced-algorithms-with-neurophos-photonic-ai-chips-to-improve-compute-speed-and-energy-efficiency-302914854.html)"
    }
   ],
   "sources": [
    {
     "title": "Duke OTC: $110M Series A",
     "url": "https://otc.duke.edu/news/neurophos-secures-110-million-series-a-to-launch-exaflop-scale-photonic-ai-chips/"
    },
    {
     "title": "optics.org: $110M, metamaterial modulators, $118M total",
     "url": "https://optics.org/news/neurophos-raises-110m-for-photonic-chips-based-around-metamaterial-modulators"
    },
    {
     "title": "PRWeb seed release",
     "url": "https://www.prweb.com/releases/neurophos-raises-7-2-million-round-to-create-super-fast-energy-efficient-metamaterial-based-optical-ai-chips-for-data-centers-302015103.html"
    },
    {
     "title": "EE Times: Can metamaterials revolutionize optical computing?",
     "url": "https://www.eetimes.com/can-metamaterials-revolutionize-optical-computing/"
    },
    {
     "title": "Preqin profile (founded 2020, founders)",
     "url": "https://www.preqin.com/data/profile/asset/neurophos-inc-/570848"
    },
    {
     "title": "Tech Company News Series A",
     "url": "https://www.techcompanynews.com/neurophos-raises-110-million-in-series-a-funding-round/"
    }
   ]
  },
  {
   "slug": "normal-computing",
   "name": "Normal Computing",
   "founded": 2022,
   "hq": "New York, NY, USA",
   "category": "probabilistic-thermodynamic",
   "status": "active",
   "origin": "corporate alumni (Google Brain / Google X / Alphabet)",
   "total_funding_usd_m": 85,
   "last_round": "Series B, $50M, 2026-03",
   "key_investors": [
    "Samsung Catalyst Fund",
    "Celesta Capital",
    "Drive Capital",
    "First Spark Ventures (Eric Schmidt)",
    "Micron Ventures"
   ],
   "website": "https://www.normalcomputing.com",
   "pitch": "CEO Faris Sbahi: \"Datacenters are expected to hit an energy wall around 2030. Finding more\nenergy is not a durable strategy\" — meeting intelligence-per-dollar-per-watt demands\nrequires novel architecture. Normal's thermodynamic-computing thesis is to improve\nefficiency \"by leveraging physical system dynamics rather than suppressing them\" — using\nfluctuations, dissipation and stochasticity as compute instead of engineering them out.\nSources: https://www.prnewswire.com/news-releases/normal-computing-raises-50m-led-by-samsung-catalyst-to-accelerate-silicon-design-and-solve-ai-hardware-energy-crisis-302724819.html,\nhttps://www.normalcomputing.com/blog/normal-computing-announces-tape-out-of-worlds-first-thermodynamic-computing-chip",
   "technical_approach": "- **Substrate:** physics-based ASIC in conventional silicon; stochastic analog circuit\n  dynamics (the company published a \"stochastic processing unit\" prototype and\n  thermodynamic linear-algebra results in 2023 before the ASIC).\n- **What the physics does:** natural relaxation dynamics of coupled stochastic circuits\n  perform sampling and linear-algebra primitives; targeted at probabilistic/diffusion-style\n  AI workloads (image/video generation, uncertainty-aware reasoning).\n- **CN101:** tape-out completed **August 2025**, announced as \"the world's first\n  thermodynamic computing chip,\" targeting multi-modal diffusion AI inference; roadmap\n  claims up to 1000x energy-efficiency gains vs conventional GPUs (a roadmap projection,\n  not a measured product figure). Sources: https://www.prnewswire.com/news-releases/normal-computing-announces-tape-out-of-worlds-first-thermodynamic-computing-chip-302527154.html,\n  https://www.tomshardware.com/tech-industry/semiconductors/worlds-first-thermodynamic-computing-chip-reaches-tape-out-normal-computings-physics-based-asic-changes-lanes-to-train-more-ai\n- **Second business line:** \"Normal EDA,\" an AI-native semiconductor-design platform\n  (agentic AI for RTL/verification) — the near-term revenue product; thermodynamic chips\n  are the 2027 production goal (https://www.normalcomputing.com/blog/normal-computing-corp).\n- **Training vs inference:** CN101 is an inference/sampling accelerator; no on-chip\n  learning claim. Partly funded by the UK's ARIA Scaling Compute programme\n  (https://www.datacenterdynamics.com/en/news/normal-computing-tapes-out-worlds-first-thermodynamic-chip-for-energy-efficient-ai-workloads/).",
   "funding_rounds": [
    {
     "date": "2023",
     "round": "Seed",
     "amount": "$8.5M",
     "investors": "Celesta Capital, First Spark Ventures, Micron Ventures",
     "source": "https://www.datacenterdynamics.com/en/news/normal-computing-tapes-out-worlds-first-thermodynamic-chip-for-energy-efficient-ai-workloads/"
    },
    {
     "date": "2025-03-24",
     "round": "Seed extension (announced as \"$35M raised to date\")",
     "amount": "to $35M cumulative",
     "investors": "Eric Schmidt (First Spark Ventures), Celesta Capital, Drive Capital, ARIA (UK), Micron Ventures, Samsung Next, Intel Ignite, NSIN",
     "source": "https://www.normalcomputing.com/blog/normal-computing-corp"
    },
    {
     "date": "2026-03-25",
     "round": "Series B (strategic)",
     "amount": "$50M",
     "investors": "Samsung Catalyst Fund lead; Galvanize, Brevan Howard Macro Venture Fund, ArcTern Ventures; existing: Celesta, Drive, First Spark, Micron Ventures",
     "source": "https://www.prnewswire.com/news-releases/normal-computing-raises-50m-led-by-samsung-catalyst-to-accelerate-silicon-design-and-solve-ai-hardware-energy-crisis-302724819.html"
    }
   ],
   "funding_notes": "Total raised: **>$85M** (per the March 2026 release). Note: the March 2025 \"$35M\" was a\ncumulative figure including the 2023 seed, not a $35M single round — some press summarize\nit either way.",
   "trajectory": "Founded 2022 in New York by ex-Google Brain / Google X / Google Quantum engineers —\nFaris Sbahi (CEO), Antonio Martinez, and Yui-Hong \"Matthias\" Tan\n(https://www.promptloop.com/directory/what-does-normal-computing-do). Published\nthermodynamic-computing theory and prototype papers 2023-2024, taped out CN101 Aug 2025,\nclosed the Samsung-led $50M in March 2026. Offices in NYC, SF, London, Copenhagen.\nRoadmap: AI-designed hardware logic (2025), thermodynamic chips in production (2027),\nrecursive AI-driven chip design \"Ctrl+R\" (2028). Active as of mid-2026.",
   "relevance": "Normal is the most institutionally-backed thermodynamic-computing company (Samsung, Micron,\nSchmidt, UK ARIA) and validates the \"use physics, don't fight it\" framing Knowm advanced\nwith Thermodynamic RAM a decade earlier. Like Extropic, its substrate is stochastic CMOS\nsampling rather than memristive adaptation; learning stays off-chip, unlike the AHaH\non-chip-learning thesis.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Normal Computing's Head of ML **Thomas Dybdahl Ahle joined the Si2 (Silicon Integration Initiative) Board of Directors** (2026-08-08) — an EDA-standards seat alongside major semiconductor firms, consistent with the AI-for-chip-design side of the business. (https://si2.org/si2-welcomes-normal-computings-thomas-ahle-to-si2-board/)"
    }
   ],
   "sources": [
    {
     "title": "Normal blog: $35M announcement",
     "url": "https://www.normalcomputing.com/blog/normal-computing-corp"
    },
    {
     "title": "PRNewswire: $50M Series B led by Samsung Catalyst (2026-03-25)",
     "url": "https://www.prnewswire.com/news-releases/normal-computing-raises-50m-led-by-samsung-catalyst-to-accelerate-silicon-design-and-solve-ai-hardware-energy-crisis-302724819.html"
    },
    {
     "title": "PRNewswire: CN101 tape-out",
     "url": "https://www.prnewswire.com/news-releases/normal-computing-announces-tape-out-of-worlds-first-thermodynamic-computing-chip-302527154.html"
    },
    {
     "title": "DCD: CN101 tape-out coverage",
     "url": "https://www.datacenterdynamics.com/en/news/normal-computing-tapes-out-worlds-first-thermodynamic-chip-for-energy-efficient-ai-workloads/"
    },
    {
     "title": "Tom's Hardware: CN101",
     "url": "https://www.tomshardware.com/tech-industry/semiconductors/worlds-first-thermodynamic-computing-chip-reaches-tape-out-normal-computings-physics-based-asic-changes-lanes-to-train-more-ai"
    }
   ]
  },
  {
   "slug": "olix-computing",
   "name": "OLIX Computing (fka Flux Computing)",
   "founded": 2024,
   "hq": "London, UK",
   "category": "photonic",
   "status": "active",
   "origin": "serial founder (James Dacombe, also founder of neurotech startup CoMind)",
   "total_funding_usd_m": 250,
   "last_round": "Series A, $220M, 2026-02",
   "key_investors": [
    "Hummingbird Ventures",
    "Plural",
    "Vertex Ventures",
    "LocalGlobe",
    "Entrepreneurs First"
   ],
   "website": "https://www.olix.com",
   "pitch": "OLIX targets \"the memory wall\": AI chips cannot run at full capacity because HBM bandwidth\nbottlenecks starve the compute. Its answer is to eliminate HBM entirely — keep all data in\non-chip SRAM and use integrated photonics to move data between circuits — claiming higher\nthroughput per megawatt and lower TCO than HBM-based designs for inference.\n([SiliconANGLE, 2026-02-11](https://siliconangle.com/2026/02/11/photonic-ai-chip-startup-olix-nabs-220m-investment/);\n[DCD](https://www.datacenterdynamics.com/en/news/uk-chip-startup-olix-raises-220m-for-development-of-photonic-ai-chips-that-forgo-the-need-for-hbm-report/))",
   "technical_approach": "- **Product:** the **OTPU (Optical Tensor Processing Unit)** — an AI accelerator for datacenter\n  **inference only**.\n- **Substrate:** described as an \"optical digital processor\" — photonics is used chiefly for the\n  memory/interconnect fabric (data movement between circuits), combined with tensor-optimized\n  cores plus scalar and vector units; data stored exclusively in on-chip 6T **SRAM**, no HBM.\n  So this is photonic data movement around largely digital compute, not analog optical\n  matrix multiplication (contrast Lumai). ([SiliconANGLE](https://siliconangle.com/2026/02/11/photonic-ai-chip-startup-olix-nabs-220m-investment/))\n- **Compatibility:** building a compiler to map existing AI models onto the OTPU unchanged.\n- **Timeline:** conflicting reports — customer shipments beginning 2026\n  ([SiliconANGLE](https://siliconangle.com/2026/02/11/photonic-ai-chip-startup-olix-nabs-220m-investment/))\n  vs first commercial products reaching customers in 2027\n  ([New Electronics](https://www.newelectronics.co.uk/content/news/uk-startup-olix-secures-220m-to-advance-photonic-ai-chip-development)).",
   "funding_rounds": [
    {
     "date": "2024–25 (as Flux Computing)",
     "round": "Pre-seed/Seed",
     "amount": "~$30M (implied: total $250M minus $220M; per-round breakdown undisclosed)",
     "investors": "Plural, Vertex Ventures, LocalGlobe, Entrepreneurs First",
     "source": "[SiliconANGLE](https://siliconangle.com/2026/02/11/photonic-ai-chip-startup-olix-nabs-220m-investment/), [Tracxn](https://tracxn.com/d/companies/olix/__4TeHKbFTTtuPK_EwLW38sxw1Pvy4MigXlhFmbz6VUxU)"
    },
    {
     "date": "2026-02",
     "round": "Series A",
     "amount": "$220M",
     "investors": "led Hummingbird Ventures; Plural, Vertex, LocalGlobe, EF",
     "source": "[SiliconANGLE](https://siliconangle.com/2026/02/11/photonic-ai-chip-startup-olix-nabs-220m-investment/), [TFN](https://techfundingnews.com/olix-220m-unicorn-photonic-ai-chips-inference/)"
    }
   ],
   "funding_notes": "Total ~$250M; valuation >$1B post-round. **Conflict noted:** most outlets report a $220M round\nled by Hummingbird; EstVCA describes it as \"Plural leads OLIX's $250M Series A\"\n([EstVCA](https://www.estvca.ee/news/plural-leads-olix-s-usd250m-series-a-funding-round)) —\nthe $250M figure matches total-raised, lead attribution differs. Both recorded.",
   "trajectory": "Founded March 2024 in London as **Flux Computing** by James Dacombe (25 at the time of the\nSeries A; previously founded brain-monitoring startup CoMind as a teenager, which has raised\n~$100M). Rebranded **Olix Computing Ltd.** in January 2026 alongside the $220M round. Operated\nlargely in stealth (\"the secretive $1bn AI company hiring in Bristol\" — Yahoo News UK). Plans to\nexpand engineering in the UK and North America and prepare for manufacturing. Status mid-2026:\nactive, pre-revenue, unicorn valuation ~2 years after founding.\n\n**Founders (sweep 2026-09-22):** James Dacombe (sole founder & CEO; Thiel Fellow; founded CoMind at 18 and still runs it — an EF job ad had one chief of staff working \"across Flux and CoMind\"). No academic lab or corporate parent (https://uk.linkedin.com/in/jamesdacombe ; https://portfolio.joinef.com/companies/flux-computing-2/jobs/61465202-chief-of-staff ; https://www.pluralplatform.com/features/why-we-invested-in-olix).",
   "relevance": "OLIX illustrates the 2025–26 capital surge into non-von-Neumann *packaging* of inference: like\nFractile it is SRAM-based and inference-only, adding photonics for interconnect rather than for\nanalog computation — physics-assisted data movement, not physics-based learning, and no on-chip\nadaptation of the kind Knowm's AHaH approach targets.",
   "news": [],
   "sources": [
    {
     "title": "Photonic AI chip startup OLIX nabs $220M — SiliconANGLE",
     "url": "https://siliconangle.com/2026/02/11/photonic-ai-chip-startup-olix-nabs-220m-investment/"
    },
    {
     "title": "Olix raises $220m, forgoes HBM — DCD",
     "url": "https://www.datacenterdynamics.com/en/news/uk-chip-startup-olix-raises-220m-for-development-of-photonic-ai-chips-that-forgo-the-need-for-hbm-report/"
    },
    {
     "title": "Olix nabs $220M for photonic AI inference chips — TechFundingNews",
     "url": "https://techfundingnews.com/olix-220m-unicorn-photonic-ai-chips-inference/"
    },
    {
     "title": "OLIX secures $220m — New Electronics",
     "url": "https://www.newelectronics.co.uk/content/news/uk-startup-olix-secures-220m-to-advance-photonic-ai-chip-development"
    },
    {
     "title": "Plural leads OLIX's $250M Series A — EstVCA",
     "url": "https://www.estvca.ee/news/plural-leads-olix-s-usd250m-series-a-funding-round"
    },
    {
     "title": "Olix — Tracxn profile",
     "url": "https://tracxn.com/d/companies/olix/__4TeHKbFTTtuPK_EwLW38sxw1Pvy4MigXlhFmbz6VUxU"
    }
   ]
  },
  {
   "slug": "optalysys",
   "name": "Optalysys",
   "founded": 2013,
   "hq": "Leeds, United Kingdom",
   "category": "photonic",
   "status": "pivoted",
   "origin": "corporate spinoff / serial founder (Cambridge Correlators — Nick New's company spun out of his Cambridge PhD in optical pattern recognition; co-founder Robert Todd)",
   "total_funding_usd_m": 63,
   "last_round": "Series A extension, £23M (~$31M), 2026-01",
   "key_investors": [
    "Lingotto",
    "imec.xpand",
    "Northern Gritstone",
    "DARPA (grant)"
   ],
   "website": "https://optalysys.com",
   "pitch": "Optalysys argued that key mathematical primitives — Fourier transforms and\nconvolutions at the heart of HPC, ML and signal processing — are enormously\nenergy- and time-expensive on electronic von Neumann machines, but are performed\n\"for free\" by the physics of light: encode data into a laser beam, let the wavefront\ninterfere through optics, and read out the transform at the speed of light with very\nlow power. (Original framing: NextBigFuture 2016 coverage and optics.org on its\noptical CNN; company \"Optical Fourier Engine\" material.)",
   "technical_approach": "- Originally: free-space optical Fourier computing — a photonic co-processor\n  performing Fourier transforms/correlations optically, targeted at HPC, defense/\n  signal processing, genomics sequence alignment (the \"GENESYS\"-style genomics\n  pitch), and convolutional neural networks. Passive, analog optics; inference/\n  transform acceleration, not on-chip learning.\n- Pivoted to: a **photonic accelerator for Fully Homomorphic Encryption (FHE)**.\n  Optalysys now builds an integrated photonic semiconductor (\"Enable\") that\n  accelerates the polynomial arithmetic (large NTT/FFT-heavy operations) that make\n  FHE slow, aiming to process always-encrypted data at near-plaintext speed. The\n  substrate is still photonic; the application moved from general optical compute to\n  privacy/encryption acceleration where the FFT-heavy workload is a natural fit.",
   "funding_rounds": [
    {
     "date": "2015",
     "round": "Seed",
     "amount": "—",
     "investors": "EU (Horizon), Grant Thornton UK",
     "source": "Tracxn / Dealroom"
    },
    {
     "date": "2016-04",
     "round": "Grant",
     "amount": "—",
     "investors": "DARPA",
     "source": "Tracxn"
    },
    {
     "date": "2023-07",
     "round": "Series A",
     "amount": "£21M (~$27M)",
     "investors": "Lingotto, imec.xpand, Northern Gritstone",
     "source": "Tech.eu / UKTN"
    },
    {
     "date": "2026-01",
     "round": "Series A extension",
     "amount": "£23M (~$31M; €26.4M)",
     "investors": "(expansion round, US expansion)",
     "source": "UKTN / EU-Startups / optics.org"
    },
    {
     "date": "—",
     "round": "Total (reported)",
     "amount": "~$63M (7 rounds, Tracxn)",
     "investors": "—",
     "source": "Tracxn"
    }
   ],
   "funding_notes": "",
   "trajectory": "Founded 2013 in West Yorkshire by Dr. Nick New (whose optical-pattern-recognition\nPhD at Cambridge and prior company Cambridge Correlators seeded the technology),\nwith Robert Todd. Early years pursued optical Fourier co-processing for HPC and\ngenomics, backed by EU and DARPA grants. Around 2019–2023 Optalysys refocused its\nphotonic hardware onto FHE acceleration, closing a £21M Series A in July 2023 and a\n£23M (~€26.4M / ~$31M) extension in January 2026 to commercialize the chips and\nexpand into the US. Status mid-2026: active, well-funded, pivoted to photonic FHE\nacceleration; total reported raise ~$63M.\n\n**Founders (sweep 2026-09-22):** Dr. Nick New (co-founder, CEO then CPO; PhD optical pattern recognition, Cambridge; founded Cambridge Correlators in 2001 from that PhD work and led it until Optalysys) and Robert Todd (co-founder & CTO; optics/engineering, ex-Seagate, BAE, CERN; founded an optical manufacturing company). Northern Gritstone: Optalysys \"developed from co-founders Dr. Nick New and Robert Todd's work for Cambridge Correlators, a company spun out of Nick's PhD research on Optical Computing at the University of Cambridge\" (https://optalysys.com/people/ ; https://www.northern-gritstone.com/news/northern-gritstone-invests-in-optalysys-the-chip-developer-aiming-to-revolutionise-secure-computing ; https://www.verve.vc/blog/the-trust-in-our-computing-system-is-being-questioned/).",
   "relevance": "A physics-based (photonic) analog-compute company that survived by re-aiming the\nsame substrate at a workload with a hard, quantifiable bottleneck (FHE), rather than\ncompeting head-on in general AI acceleration. Like Knowm it bets on device physics\ndoing the math, but it is inference/transform acceleration with no on-chip learning\n— the opposite end of the spectrum from AHaH's adaptive on-device learning claim.",
   "news": [],
   "sources": [
    {
     "title": "Optalysys raises £21 million to unlock Fully Homomorphic Encryption — Tech.eu, 2023-07-17",
     "url": "https://tech.eu/2023/07/17/leeds-based-photonic-chip-company-optalysys-raises-ps21-million-to-unlock-its-fully-homomorphic-encryption-process/"
    },
    {
     "title": "Optalysys raises £23m for US expansion — UKTN, 2026-01-22",
     "url": "https://www.uktech.news/deep-tech/optalysys-readies-for-us-expansion-with-23m-funding-injection-20260122"
    },
    {
     "title": "Optalysys raises US$31M to push photonics toward cloud — Optica OPN, Feb 2026",
     "url": "https://www.optica-opn.org/home/industry/2026/february/optalysys_raises_us_$31_million_to_push_photonics_computing_toward_cloud_infrastructure/"
    },
    {
     "title": "Crossover of Optical Computing and Genomics: Interview with Dr. Nick New — GeneOnline",
     "url": "https://www.geneonline.com/crossover-of-optical-computing-and-genomics-exclusive-interview-with-dr-nick-new-founder-and-ceo-of-optalysys/"
    },
    {
     "title": "Optalysys optical processing powers convolutional neural network — optics.org",
     "url": "https://optics.org/news/optalysys-optical-processing-powers-convolutional-neural-network"
    },
    {
     "title": "Optalysys funding & investors — Dealroom",
     "url": "https://app.dealroom.co/companies/optalysys_ltd"
    }
   ]
  },
  {
   "slug": "opticore",
   "name": "Opticore (Opticore AI / Opticore Inc.)",
   "founded": 2024,
   "hq": "Fremont, California, USA",
   "category": "photonic",
   "status": "active",
   "origin": "academic spinoff (founder-CEO Zaijun Chen, photonics academic — affiliation per his own LinkedIn: UC Berkeley assistant professor; unverified)",
   "total_funding_usd_m": 14,
   "last_round": "Seed, $14M (†; $7.5M per one source), 2025-Q3",
   "key_investors": [
    "Origin Ventures",
    "Jetha Global",
    "Sagax Capital",
    "Neotribe Ventures"
   ],
   "website": "https://www.opticore.ai",
   "pitch": "\"Opticore is developing optical processing units (OPUs) designed to handle memory-intensive\nAI tasks\" — converting memory data to optical beams in waveguides and computing with optical\ndevices, \"fabricated using standard foundry processes\", with claimed \"100 TOPS/W compute\nefficiency\" and the ability to encode \"a trillion parameters on a single chip\"\n([Semiconductor Engineering, Startup Funding Q3 2025, 2025-10-06](https://semiengineering.com/startup-funding-q3-2025/)).\nIts X bio: \"photonic AI inference chips that deliver faster throughput at 100x less energy\nthan conventional silicon\" ([x.com/Opticore_AI](https://x.com/Opticore_AI)).",
   "technical_approach": "- **Optical processing unit** using **time-multiplexed** optical computing; data in memory is\n  encoded onto optical beams via waveguides and the arithmetic is done optically. Standard\n  foundry process (silicon photonics implied, not stated).\n- **Training and inference:** the Semi Engineering entry says the OPU targets \"real-time\n  training and inference\" through \"dynamic weights\" — unusual for photonic startups, which\n  are mostly inference-only. Treat as a claim; no demonstration published.\n- **Status:** \"expects to have its first scaled system demonstration in 2026\" (Q3 2025);\n  the company presented at OFC 2026 and the AI Infra Summit 2026 (LinkedIn posts). No\n  silicon results public as of 2026-11-10.",
   "funding_rounds": [
    {
     "date": "2025-Q3 (≈2025-09)",
     "round": "Seed",
     "amount": "$14M † ($7.5M per Semiconductor Engineering; \"$14 million\" per AccessNewswire slug, Dealroom, and a Neotribe-counsel post that calls Neotribe's piece a \"$7.5MM\" tranche)",
     "investors": "co-led by Origin Ventures and Jetha Global; Sagax Capital, Neotribe Ventures",
     "source": "https://semiengineering.com/startup-funding-q3-2025/ ; https://app.dealroom.co/news/feed/origin-ventures-co-leads-opticore-s-14m-seed-round-for-photonic-processors-to-reshape-ai-compute (403 to bots) ; https://www.linkedin.com/posts/silicon-legal-strategy_opticore-raises-14-million-to-advance-photonic-activity-7373390380672806912-38w8"
    }
   ],
   "funding_notes": "**Total:** ~$14M disclosed (†). A VentureRadar/AI Insider mention of a \"$5M seed\" in Dec 2024\ncould not be sourced and is not counted.",
   "trajectory": "Founded 2024 in Fremont, CA by Zaijun Chen (founder and CEO; his LinkedIn headline in 2023\nread \"Founder and CEO of Opticore AI | Assistant Professor of UC Berkeley\"\n([LinkedIn](https://www.linkedin.com/posts/zaijun-chen-138647109_ai-photonics-photonicswest2023-activity-7025932222880784385-n0xT));\nthe company's LinkedIn calls him CEO and places the company in the Bay Area). Status:\nactive, pre-product; system demo promised for 2026.",
   "relevance": "Small photonic entrant; notable only because it claims optical *training* via dynamic\nweights, which almost no photonic company does. Unverified. Adds little money to the cohort.",
   "news": [],
   "sources": [
    {
     "title": "Semiconductor Engineering: Startup Funding Q3 2025 (Opticore entry, 2025-10-06)",
     "url": "https://semiengineering.com/startup-funding-q3-2025/"
    },
    {
     "title": "Dealroom: Origin Ventures co-leads Opticore's $14M seed",
     "url": "https://app.dealroom.co/news/feed/origin-ventures-co-leads-opticore-s-14m-seed-round-for-photonic-processors-to-reshape-ai-compute"
    },
    {
     "title": "LinkedIn (Silicon Legal Strategy): Opticore $14M raise / Neotribe $7.5MM",
     "url": "https://www.linkedin.com/posts/silicon-legal-strategy_opticore-raises-14-million-to-advance-photonic-activity-7373390380672806912-38w8"
    },
    {
     "title": "Opticore on X",
     "url": "https://x.com/Opticore_AI"
    },
    {
     "title": "Opticore LinkedIn: CEO Zaijun Chen at AI Infra Summit 2026",
     "url": "https://www.linkedin.com/posts/opticore-inc_aiinfrastructure-photonics-aiinfrasummit2026-activity-7508964148677001216-lP5D"
    }
   ]
  },
  {
   "slug": "polyn-technology",
   "name": "POLYN Technology",
   "founded": 2019,
   "hq": "Caesarea, Israel (UK registered office)",
   "category": "analog-imc",
   "status": "active",
   "origin": "serial founders (Aleksandr Timofeev, CEO; Dmitry Godovsky, Chief Scientist)",
   "total_funding_usd_m": "conflicting: ~11 (Pulse2/EIN sourcing) vs $4.75M (Tracxn; also 2026-07 report)",
   "last_round": "Series A ($12M target) opened 2023; close not confirmed",
   "key_investors": [
    "VentureIsrael",
    "FPI Innovation Fund"
   ],
   "website": "https://polyn.ai",
   "pitch": "POLYN's \"Neuromorphic Analog Signal Processing\" (NASP) pitch: sensor data should be\nprocessed \"in its native analogue form\" at the sensor, in microseconds and at microwatt\npower, \"eliminating all overhead associated with digital operations\" — i.e., skip the\ndigitize-then-shuttle-to-a-DSP von Neumann loop entirely for always-on tiny-AI tasks.\nSources: https://polyn.ai/technology-3/, https://techtime.news/2022/06/21/polyn-technology/",
   "technical_approach": "- **Substrate:** fixed-function analog neural circuits (resistor networks + operational\n  amplifiers as neurons/axons) on standard mature CMOS; tooling supports 40-90nm nodes —\n  deliberately avoiding advanced-node cost (https://www.electronicsweekly.com/news/business/prolyn-gets-first-silicon-of-neuromorphic-processor-2025-10/).\n- **Flow:** a compiler converts a *trained* digital neural network into an\n  application-specific analog neuromorphic core (\"Neuromorphic Front-End\"); the analog\n  tier does heavy signal preprocessing/feature extraction, with any remaining digital NN\n  stages downstream. Inference-only — weights are frozen in the analog structure; no\n  on-chip learning.\n- **Products (application cores):** NeuroVoice (voice activity detection / voice\n  extraction for hearables), NeuroSense (wearable vitals, e.g. PPG heart rate),\n  VibroSense (predictive-maintenance vibration analysis)\n  (https://pulse2.com/polyn-technology-aleksandr-timofeev-profile/).\n- **First silicon:** announced 2025-10-29 — first silicon-proven NASP chip, a voice\n  activity detection (VAD) part: ~34 µW continuous operation, ~50 µs latency per\n  inference; eval kits offered, orders being taken. CEO Timofeev: \"For the first time, we\n  generated an asynchronous, fully analogue neural-network core implementation in silicon\n  directly from a digital model.\" (https://polyn.ai/polyn-technology-announces-first-silicon-implemented-nasp-chip/,\n  https://www.electronicsweekly.com/news/business/prolyn-gets-first-silicon-of-neuromorphic-processor-2025-10/)\n- **NeuroVoice VAD chip** — first NASP standard product (ASSP), debuted at CES 2026\n  (Jan 6-9, Venetian Expo): live demo of always-on voice detection at ~34 µW; eval kits\n  and early-access program announced\n  (https://polyn.ai/ultra-low-power-ai-voice-detection-demo-by-polyn-at-ces-2026/,\n  https://www.prnewswire.com/news-releases/ultra-low-power-ai-voice-detection-demo-by-polyn-at-ces-2026-302644647.html).\n- **Partnerships:** Fraunhofer Mikroelektronik (technology optimization)\n  (https://polyn.ai/polyn-technology-fraunhofer-mikroelektronik-join-forces-for-technological-optimization/).",
   "funding_rounds": [
    {
     "date": "2019-2021",
     "round": "Pre-seed + Seed",
     "amount": "$11M combined",
     "investors": "VentureIsrael, FPI Innovation Fund (Dec 2021 investment announced)",
     "source": "https://pulse2.com/polyn-technology-aleksandr-timofeev-profile/ ; https://www.einnews.com/pr_news/558480002/ventureisrael-invests-in-israeli-startup-polyn-ai-a-breakthrough-which-enables-electronics-to-sense-like-humans"
    },
    {
     "date": "2023",
     "round": "Series A (opened)",
     "amount": "$12M target",
     "investors": "not disclosed",
     "source": "https://pulse2.com/polyn-technology-aleksandr-timofeev-profile/"
    }
   ],
   "funding_notes": "Total confirmed raised: **~$11M**; the $12M Series A was reported as *open* in mid-2023\nand no close has been found in public sources as of 2026-07. **Total-raised conflict:**\nthe 2026-07 who's-who research report (_research/2026-07_whos_who_deep_research.md) puts\ntotal raised at **~$4.75M**, and Tracxn independently lists \"$4.75M over 4 rounds\"\n(https://tracxn.com/d/companies/polyn/__jJavsG78eZ1zuHtrq0xqOElUhPEFDQDWvb-dUL9jTpU) —\nirreconcilable with the ~$11M Pulse2/EIN figure; both recorded, neither replaced. Founding-year conflict:\nStartup Nation Finder says 2019; the July 2023 Pulse2 profile implies ~2018.",
   "trajectory": "Founded 2019 in Israel by Aleksandr Timofeev (CEO) with Chief Scientist Dmitry Godovsky;\nteam includes veterans of Tower Semiconductor, Marvell, Synopsys, Cadence. Fabless model:\nlicense/sell application-specific analog front-end AI chips. First production chips were\ntargeted for end-2023; actual first silicon-proven NASP chip arrived October 2025 (VAD\npart, now taking orders) — roughly a two-year slip but a real shipped-silicon milestone\nthat most tiny-analog-AI startups have not reached. Active as of mid-2026.\n\n**Founders (sweep 2026-09-22):** Aleksandr Timofeev (founder & CEO; solid-state physics; founded iGlass Technology, sold end-2020; founder/managing partner FPI VC) and Dmitry Godovsky (Chief Scientist; Dr. Habil. polymer physics; met Timofeev end-2018 after eight years on the digital-to-analog NN conversion model). Crunchbase additionally lists Boris Maslov, Ilya Rodin, Nikolay Kovshov and Val Krivenko as founders (https://semiwiki.com/ceo-interviews/319794-ceo-interview-aleksandr-timofeev-of-polyn-technology/ ; https://il.linkedin.com/in/dmitri-godovsky-44505b17 ; https://www.crunchbase.com/organization/polyn-technology).",
   "relevance": "POLYN is the \"compile-a-net-into-fixed-analog-silicon\" end of the analog spectrum:\ninference-only, weights frozen at design time — the opposite pole from Knowm's adaptive\nmemristor/AHaH thesis where the analog substrate itself learns. Their 34 µW always-on VAD\nsilicon is a concrete datapoint for what fixed analog neuromorphic delivers today.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "POLYN and **ALTER TECHNOLOGY France opened a joint chip-validation laboratory** for neuromorphic analog signal-processing chips (2026-07-08). (https://embeddedcomputing.com/technology/processing/chips-and-socs/polyn-technology-and-alter-technology-france-collaborate-on-joint-chip-validation-lab-for-neuromorphic-analog-signal-processing-chips)"
    },
    {
     "date": "2026-08-23",
     "text": "POLYN reports **VibroSense data improves ABS braking distance by 12%** in tire-sensing trials (2026-08-20). (https://www.tyrepress.com/?p=251161)"
    },
    {
     "date": "2026-10-27",
     "text": "**First VibroSense TMS engineering sensor nodes delivered for customer evaluation (2026-09-28).** (https://www.businesswire.com/news/home/20260928198992/en/POLYN-Delivers-First-VibroSense-TMS-Engineering-Sensor-Nodes-for-Customer-Evaluation)"
    }
   ],
   "sources": [
    {
     "title": "Electronics Weekly: first silicon, taking orders (2025-10)",
     "url": "https://www.electronicsweekly.com/news/business/prolyn-gets-first-silicon-of-neuromorphic-processor-2025-10/"
    },
    {
     "title": "POLYN: first silicon-implemented NASP chip",
     "url": "https://polyn.ai/polyn-technology-announces-first-silicon-implemented-nasp-chip/"
    },
    {
     "title": "Pulse2 profile: funding, products",
     "url": "https://pulse2.com/polyn-technology-aleksandr-timofeev-profile/"
    },
    {
     "title": "VentureIsrael investment (EIN Presswire, 2021-12)",
     "url": "https://www.einnews.com/pr_news/558480002/ventureisrael-invests-in-israeli-startup-polyn-ai-a-breakthrough-which-enables-electronics-to-sense-like-humans"
    },
    {
     "title": "NASP technology page",
     "url": "https://polyn.ai/technology-3/"
    },
    {
     "title": "Startup Nation Finder profile",
     "url": "https://finder.startupnationcentral.org/company_page/polyn-technology"
    },
    {
     "title": "POLYN: NeuroVoice VAD demo at CES 2026",
     "url": "https://polyn.ai/ultra-low-power-ai-voice-detection-demo-by-polyn-at-ces-2026/"
    },
    {
     "title": "Tracxn POLYN profile ($4.75M/4 rounds)",
     "url": "https://tracxn.com/d/companies/polyn/__jJavsG78eZ1zuHtrq0xqOElUhPEFDQDWvb-dUL9jTpU"
    }
   ]
  },
  {
   "slug": "prophesee",
   "name": "Prophesee (formerly Chronocam)",
   "founded": 2014,
   "hq": "Paris, France",
   "category": "event-sensing",
   "status": "active",
   "origin": "academic spinoff (Institut de la Vision, Sorbonne / UPMC Paris)",
   "total_funding_usd_m": 127,
   "last_round": "Recapitalization via judicial-recovery continuation plan, 2025 (amount undisclosed)",
   "key_investors": [
    "Intel Capital",
    "Robert Bosch Venture Capital",
    "Sony",
    "Xiaomi",
    "Sinovation Ventures",
    "Prosperity7 (Aramco)",
    "European Investment Bank",
    "iBionext",
    "360 Capital",
    "In-Q-Tel"
   ],
   "website": "https://www.prophesee.ai",
   "pitch": "Prophesee argues that conventional frame-based image sensors waste power and bandwidth by\ncapturing entire redundant frames at a fixed rate. Its \"Metavision\" sensors instead have pixels\nthat fire asynchronously and independently, only when they detect a change in the scene — mimicking\nthe human retina and eliminating redundant data at the source. (https://www.prophesee.ai/event-based-vision/)\nNote: this is event-based *sensing*, not a compute substrate — the taxonomy's \"event-sensing\"\nedge.",
   "technical_approach": "Neuromorphic event-based vision sensor (EVS / dynamic-vision sensor). Each pixel independently and\nasynchronously reports per-pixel brightness changes (\"events\") with microsecond temporal resolution\nand high dynamic range, rather than synchronous frames. Products: Gen1–Gen4 Metavision sensors;\nthe **IMX636** HD event sensor co-developed with **Sony** (stacked on Sony CIS technology, announced\n2021); **GenX320** — a compact, low-power event sensor for embedded/edge (2023), with 2025 dev kits\nfor Raspberry Pi 5 and the OpenMV platform. Sensing only; downstream processing runs on conventional\nhosts or partner processors. (https://www.prophesee.ai/2026/01/07/prophesee-recap-2025/)",
   "funding_rounds": [
    {
     "date": "2015",
     "round": "Seed/Series A",
     "amount": "€0.75M",
     "investors": "Robert Bosch Venture Capital, CEA Investissement",
     "source": "prophesee.ai / press"
    },
    {
     "date": "2016-10",
     "round": "Series B",
     "amount": "$15M",
     "investors": "Intel Capital (lead); iBionext, RBVC, 360 Capital, CEAi, Renault",
     "source": "http://www.prophesee.ai/2016/10/24/chronocam-raises-15m-in-series-b-led-by-intel-capital/"
    },
    {
     "date": "2018-02",
     "round": "Series B ext.",
     "amount": "$19M",
     "investors": "new strategic electronics investor; 360 Capital, Supernova Invest, iBionext, Intel Capital, Renault, RBVC",
     "source": "https://www.prophesee.ai/2018/02/21/prophesee-19-million-funding-round/"
    },
    {
     "date": "2019-10",
     "round": "Series C",
     "amount": "$28M",
     "investors": "European Investment Bank (lead); iBionext, 360 Capital, RBVC, Supernova",
     "source": "https://www.prophesee.ai/2019/10/28/prophesee-28-million-funding/"
    },
    {
     "date": "2022-09",
     "round": "Series C+",
     "amount": "€50M",
     "investors": "Prosperity7 (Aramco), Sinovation Ventures, Xiaomi",
     "source": "https://www.prophesee.ai/2022/09/22/prophesee-closes-50million-fundraising-round/"
    }
   ],
   "funding_notes": "Total raised reported as ~$127M / €126M. Also backed by In-Q-Tel (CIA venture arm) per later\ncoverage. Note: the 2019 release says total reached $68M after the $28M round; the 2022 round\nbrought the cumulative total to ~$127M.",
   "trajectory": "Founded 2014 in Paris as **Chronocam** by Ryad Benosman, Christoph Posch, Luca Verre and Bernard\nGilly, out of retina/neuromorphic-vision research at the Institut de la Vision (Sorbonne). Rebranded\n**Prophesee** in 2018. Became the EU's best-funded fabless semiconductor startup. In 2024 a delayed\nstrategic mobile contract in China (\"chasing the whale\") stalled a fundraising round; the Paris\nCommercial Court opened a **judicial-recovery procedure (redressement judiciaire) on 15 October 2024**,\nwith cessation of payments dated 17 September 2024. Through 2025 the company completed a\nrecapitalization with existing and new investors (iBionext, 360 Capital, Aramco/Prosperity7, Bosch\nVentures, and new backer Critical Path Ventures) and reports its financial issues resolved. In\nDecember 2025 co-founder/CEO **Luca Verre was succeeded by Jean Ferré as CEO**, refocusing on\nsecurity/defense/aerospace and industrial automation. Active as of mid-2026.\n(https://www.eetimes.com/prophesee-names-new-ceo-as-fresh-funds-target-growth/;\nhttps://www.frenchtechjournal.com/chasing-whale-nearly-sank-prophesee-new-ceo-charting-course-back-brink/)",
   "relevance": "Prophesee attacks the \"memory/data wall\" upstream of compute — by not generating redundant data in\nthe first place — rather than by rethinking the compute substrate as Knowm's memristor/AHaH approach\ndoes. No on-chip learning; the sensor is inference-agnostic and pairs with conventional or\nneuromorphic downstream processors. A complementary, sensing-side member of the non-von-Neumann race.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Post-recovery: Prophesee **raised €20M and launched Mantara**, \"the first fully integrated drone detection and tracking system built natively on event-based vision and AI\" (reported 2026-06-17) — concrete evidence of the defense/security refocus under CEO Jean Ferré. Also expanded the IDS industrial-vision partnership (2026-03-11). (https://www.yolegroup.com/industry-news/prophesee-raises-e20-million-and-launches-mantara-the-first-fully-integrated-drone-detection-system-built-onevent-based-vision-and-ai/ ; https://www.prophesee.ai/2026/03/11/ids-prophesee-expand-event-based-industrial-vision-collaboration/)"
    }
   ],
   "sources": [
    {
     "title": "Prophesee 2025 recap",
     "url": "https://www.prophesee.ai/2026/01/07/prophesee-recap-2025/"
    },
    {
     "title": "Chronocam $15M Series B",
     "url": "http://www.prophesee.ai/2016/10/24/chronocam-raises-15m-in-series-b-led-by-intel-capital/"
    },
    {
     "title": "Prophesee $19M round (rebrand)",
     "url": "https://www.prophesee.ai/2018/02/21/prophesee-19-million-funding-round/"
    },
    {
     "title": "Prophesee $28M round",
     "url": "https://www.prophesee.ai/2019/10/28/prophesee-28-million-funding/"
    },
    {
     "title": "Prophesee €50M Series C+",
     "url": "https://www.prophesee.ai/2022/09/22/prophesee-closes-50million-fundraising-round/"
    },
    {
     "title": "Prophesee files for insolvency (Image Sensors World)",
     "url": "http://image-sensors-world.blogspot.com/2025/01/prophesee-files-for-insolvency.html"
    },
    {
     "title": "Prophesee names new CEO as fresh funds target growth (EE Times)",
     "url": "https://www.eetimes.com/prophesee-names-new-ceo-as-fresh-funds-target-growth/"
    },
    {
     "title": "Chasing a whale nearly sank Prophesee (French Tech Journal)",
     "url": "https://www.frenchtechjournal.com/chasing-whale-nearly-sank-prophesee-new-ceo-charting-course-back-brink/"
    }
   ]
  },
  {
   "slug": "q-ant",
   "name": "Q.ANT",
   "founded": 2018,
   "hq": "Stuttgart, Germany",
   "category": "photonic",
   "status": "active",
   "origin": "corporate spinoff (TRUMPF; founder/CEO Michael Förtsch)",
   "total_funding_usd_m": 80,
   "last_round": "Series A extension, ~$8M (to $80M total A), 2025-10",
   "key_investors": [
    "Cherry Ventures",
    "UVC Partners",
    "imec.xpand",
    "TRUMPF",
    "Verve Ventures",
    "Onsight Ventures",
    "Duquesne Family Office"
   ],
   "website": "https://qant.com",
   "pitch": "Digital CMOS is hitting physical limits while AI's energy draw explodes; Q.ANT's photonic\n\"Native Processing Unit\" computes with light instead of electrons and is specified to\ndeliver \"at least 30 times greater energy efficiency and significant computational speed\nimprovements over traditional CMOS technology,\" including executing **nonlinear**\nmathematics natively in the optical domain.\n(https://qant.com/press-releases/q-ant-launches-first-commercial-photonic-processor-for-energy-efficient-high-performance-computing-and-real-time-ai-applications/)",
   "technical_approach": "- Substrate: proprietary **Thin-Film Lithium Niobate on Insulator (TFLNoI)** photonic\n  integrated circuits — not silicon photonics. Lithium niobate's electro-optic and nonlinear\n  properties let the chip do not just analog linear algebra but **nonlinear operations\n  \"in-flight\"** (activation-function-like math without leaving the optical domain) — a\n  differentiator vs MZI-mesh silicon approaches, and the basis of Q.ANT's claim to support\n  both inference *and* training workloads. Architecture branded **LENA** (Light-Empowered\n  Native Arithmetics). (https://abachy.com/news/qant-unveils-thin-film-lithium-niobate-photonic-processor, https://qant.com/photonic-computing/)\n- Products: first commercial **NPU** launched Nov 2024 as a PCIe card for standard x86/HPC\n  servers; **NPU 2** (second generation) showed up to 50x performance over gen-1 in\n  independent evaluation at Leibniz Supercomputing Centre. Deployed in two German HPC\n  centers — **LRZ Munich** (world's first photonic AI processor in a supercomputing center,\n  2025) and **Jülich Supercomputing Centre** — plus cloud access via an **IONOS**\n  partnership (May 2026) and a US datacenter debut.\n  (https://qant.com/press-releases/leibniz-supercomputing-centre-computes-with-light-worlds-first-photonic-ai-processor-from-q-ant-goes-into-operation/,\n  https://thequantuminsider.com/2026/05/21/qant-photonic-ai-computing-commercial-ionos/)\n- Compute-in-light for AI inference, HPC kernels, and physics simulation; control and\n  memory remain electronic. No on-chip learning in the neuromorphic sense.",
   "funding_rounds": [
    {
     "date": "2018–2024",
     "round": "Corporate backing + grants",
     "amount": "undisclosed",
     "investors": "TRUMPF (parent); German public projects",
     "source": "https://qant.com/"
    },
    {
     "date": "2025-07",
     "round": "Series A",
     "amount": "€62M (~$72M)",
     "investors": "Cherry Ventures + UVC Partners + imec.xpand (co-led); Verve Ventures, Onsight Ventures, TRUMPF, L-Bank, Grazia Equity",
     "source": "https://www.datacenterdynamics.com/en/news/photonic-chip-company-qant-announces-72m-series-a-funding-round/"
    },
    {
     "date": "2025-10",
     "round": "Series A extension",
     "amount": "to ~$80M total",
     "investors": "Duquesne Family Office",
     "source": "https://thequantuminsider.com/2025/10/30/duquesne-family-office-invests-in-q-ant-to-drive-sustainable-photonic-ai-infrastructure/"
    }
   ],
   "funding_notes": "Total disclosed external raise: **~$80M** — reported as the largest photonic-computing\nfinancing round in Europe. Pre-2025 TRUMPF investment undisclosed. No valuation reported.",
   "trajectory": "Founded 2018 by physicist Michael Förtsch as a TRUMPF spin-out (originally spanning quantum\nsensing and photonic computing; the computing line is now the headline). Unusual in this\nbatch for shipping a commercial photonic *compute* product into real HPC centers rather than\npivoting to interconnect: NPU launch Nov 2024 → LRZ/JSC installs 2025 → €62M Series A July\n2025 → $80M with Duquesne Oct 2025 → NPU 2 and IONOS cloud availability + US expansion\n2026. Active and commercializing as of mid-2026.",
   "relevance": "The strongest \"analog physics does the math, commercially, today\" datapoint on the photonic\nside — including native nonlinearity in the substrate, which parallels Knowm's argument that\nthe physics should carry more of the computation than a linear MAC array. Still\ntraining-in-the-loop electronics, though: no in-substrate synaptic adaptation à la AHaH.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Q.ANT demonstrated **generative AI (incl. xLSTM) running on its photonic NPU** at ISC 2026 (2026-06-23) — claimed as a first for photonic hardware, with ~30x efficiency on matrix ops per follow-up coverage. (https://www.hpcwire.com/off-the-wire/q-ant-runs-generative-ai-on-photonic-hardware/ ; https://quantumzeitgeist.com/q-ant-photonic-chip-qants-efficiency/)"
    },
    {
     "date": "2026-10-27",
     "text": "**Open-source SDK for its photonic NPU (2026-09-23):** C and Python APIs, examples, docs and a CPU simulation backend on GitHub; hardware access via IONOS cloud or on-prem 'in the coming months'; CEO calls it photonic computing's 'Linux moment'. Press framed it as attacking the software gap that sank Graphcore. No performance figures. (https://finance.yahoo.com/technology/ai/articles/q-ant-launches-world-first-120000709.html ; https://www.techtimes.com/articles/327988/20260924/photonic-ai-firm-qant-opens-developer-toolkit-targeting-software-gap-that-ended-graphcore.htm)"
    },
    {
     "date": "2026-11-10",
     "text": "**Q.ANT and IONOS launch a photonic-computing pilot (2026-11-09)** — Q.ANT's photonic processors offered to AI developers on IONOS bare-metal cloud infrastructure; first cloud-hosted access to its NPU. (https://www.datacenterdynamics.com/en/news/qant-partners-with-ionos-for-photonic-computing-pilot-program/ ; https://thequantuminsider.com/2026/11/09/qant-photonic-processors-ionos-cloud-infrastructure/)"
    }
   ],
   "sources": [
    {
     "title": "Q.ANT €62M press release",
     "url": "https://qant.com/press-releases/qant-raises-62-million-euro-to-transform-the-future-of-computing-with-photonic-processing/"
    },
    {
     "title": "DCD: $72M Series A",
     "url": "https://www.datacenterdynamics.com/en/news/photonic-chip-company-qant-announces-72m-series-a-funding-round/"
    },
    {
     "title": "Quantum Insider: Duquesne extension to $80M",
     "url": "https://thequantuminsider.com/2025/10/30/duquesne-family-office-invests-in-q-ant-to-drive-sustainable-photonic-ai-infrastructure/"
    },
    {
     "title": "Q.ANT NPU commercial launch (Nov 2024)",
     "url": "https://qant.com/press-releases/q-ant-launches-first-commercial-photonic-processor-for-energy-efficient-high-performance-computing-and-real-time-ai-applications/"
    },
    {
     "title": "LRZ deployment",
     "url": "https://qant.com/press-releases/leibniz-supercomputing-centre-computes-with-light-worlds-first-photonic-ai-processor-from-q-ant-goes-into-operation/"
    },
    {
     "title": "Quantum Insider: IONOS partnership, NPU2 50x, JSC",
     "url": "https://thequantuminsider.com/2026/05/21/qant-photonic-ai-computing-commercial-ionos/"
    },
    {
     "title": "TFLN platform explainer",
     "url": "https://abachy.com/news/qant-unveils-thin-film-lithium-niobate-photonic-processor"
    }
   ]
  },
  {
   "slug": "rain-ai",
   "name": "Rain AI (fka Rain Neuromorphics)",
   "founded": 2017,
   "hq": "San Francisco, USA",
   "category": "emerging-device-imc",
   "status": "effectively wound down — OpenAI passed on acquiring the company, then bought Rain patents (2026-08); near-total layoffs reported (was: exploring sale since May 2025)",
   "origin": "academic spinoff (University of Florida — Juan Claudio Nino's lab; Jack Kendall's memristive nanowire work)",
   "total_funding_usd_m": 38,
   "last_round": "attempted Series B, $150M at ~$600M valuation, 2024 — failed to close; $3M bridge loans 2024 and 2025",
   "key_investors": [
    "Sam Altman (personal)",
    "Prosperity7 Ventures (divested 2024)",
    "Y Combinator",
    "Buckley Ventures",
    "Gaingels",
    "Loup Ventures",
    "Metaplanet",
    "Pioneer Fund"
   ],
   "website": "https://rain.ai",
   "pitch": "Current homepage (minimal, copyright 2024): \"We're building the most energy efficient\nhardware for AI\" (https://rain.ai). In its memristor era the pitch was a \"brain-inspired\nanalog architecture that employs a 3D array of randomly-connected memristors to compute\nneural network training and inference at extremely low power\"\n(https://www.eetimes.com/rain-neuromorphics-tapes-out-demo-chip-for-analog-ai/).",
   "technical_approach": "- Original approach (2017–~2022): analog neuromorphic processor built on randomly connected\n  memristor arrays (\"Memristive Nanowire Neural Networks\" lineage from Kendall's UF work).\n  Taped out an analog demo chip and demonstrated analog AI training (EE Times, 2021–2022:\n  https://www.eetimes.com/rain-demonstrates-ai-training-on-analog-chip/).\n- Pivot: quietly moved off analog memristors to a **digital in-memory compute (D-IMC) tile\n  plus software stack**, offered for licensing for on-device AI workloads\n  (https://www.ai-market-watch.com/company/rain-ai; survey confirmation:\n  https://arxiv.org/pdf/2401.14428). June 2024: named Andes Technology as RISC-V CPU partner\n  (https://www.globenewswire.com/news-release/2024/06/04/2892703/0/en/Rain-AI-Unveils-Andes-Technology-as-Its-RISC-V-Partner.html).\n  Hired ex-Apple silicon executive Jean-Didier Allegrucci to lead hardware (2023).\n- Investor documents reportedly targeted first customer hardware for October 2024\n  (https://www.eetimes.com/rain-neuromorphics-tapes-out-demo-chip-for-analog-ai/ era coverage;\n  later reporting via https://startupwired.com/2025/05/16/rain-ai-the-rise-and-fall-of-a-chipmaking-challenger/).\n- No on-chip learning in the shipped/pivoted product; the analog-era research did target\n  training on-device.",
   "funding_rounds": [
    {
     "date": "2018",
     "round": "Angel/seed",
     "amount": "~$1M personal + $5M seed",
     "investors": "Sam Altman; Y Combinator (S18 batch)",
     "source": "https://startupwired.com/2025/05/16/rain-ai-the-rise-and-fall-of-a-chipmaking-challenger/ ; https://www.ycombinator.com/companies/rain-neuromorphics"
    },
    {
     "date": "2019",
     "round": "— (commercial LOI)",
     "amount": "$51M nonbinding purchase LOI from OpenAI",
     "investors": "OpenAI",
     "source": "https://finance.yahoo.com/news/sam-altmans-150m-ai-chip-123106283.html"
    },
    {
     "date": "2022-04",
     "round": "Series A",
     "amount": "$25M",
     "investors": "Prosperity7 Ventures (lead); Buckley Ventures, Gaingels, Loup Ventures, Metaplanet, Pioneer Fund",
     "source": "https://www.design-reuse.com/news/51363/rain-neuromorphics-funding.html ; https://rain.ai/blog/rain-ai-raises-25-million-series-a"
    },
    {
     "date": "~2023",
     "round": "Series A extension",
     "amount": "$8.1M",
     "investors": "undisclosed",
     "source": "https://startuprise.io/rain-ai-secures-8-1million-in-series-a-funding/"
    },
    {
     "date": "2024",
     "round": "attempted Series B",
     "amount": "$150M at ~$600M valuation — did not close",
     "investors": "promoted by Sam Altman",
     "source": "https://www.benzinga.com/24/11/42080933/openai-ceo-sam-altman-reportedly-leading-150m-funding-for-rain-ai-a-startup-top-analyst-calls-one-of-the-legitimate-players-at-the-table-to-challenge-nvidias-ai-chip-dominance"
    },
    {
     "date": "2024",
     "round": "bridge loan",
     "amount": "$3M",
     "investors": "—",
     "source": "https://finance.yahoo.com/news/sam-altmans-150m-ai-chip-123106283.html"
    },
    {
     "date": "2025-05",
     "round": "bridge loan",
     "amount": "$3M (to cover payroll during sale talks)",
     "investors": "—",
     "source": "https://finance.yahoo.com/news/sam-altmans-150m-ai-chip-123106283.html"
    }
   ],
   "funding_notes": "Total disclosed equity ≈ $38M (+~$6M bridge debt). Note: 2024 Prosperity7 (Aramco-linked)\nwas forced to divest its stake on US national-security grounds\n(https://finance.yahoo.com/news/sam-altmans-150m-ai-chip-123106283.html). One startupwired\naccount calls the seed \"$25M\" — conflating it with the Series A; the $5M seed + $25M\nSeries A version is better sourced.",
   "trajectory": "Founded 2017 as Rain Neuromorphics by Gordon Wilson (CEO), Jack Kendall (CTO), and UF\nprofessor Juan Claudio Nino; the three met at the University of Florida\n(https://www.crunchbase.com/organization/rain-neuromorphics). Y Combinator 2018; moved to\nBay Area. Leadership churned as the pivot unfolded: Wilson stepped down as CEO in late 2023\n(https://www.linkedin.com/posts/gordonhirschwilson_hello-world-i-have-some-big-news-to-share-activity-7135990941244411904-I0vQ);\nCOO William Passo took over as CEO (https://canvasbusinessmodel.com/blogs/owners/rain-ai-who-owns),\nand some 2025 profiles list co-founder Jack Kendall as CEO — reports conflict.\nRain reportedly began shipping evaluation NPUs in October 2024\n(https://neuromorphiccore.ai/insights/rain-ai/; also per the 2026-07 who's-who research\nreport, _research/2026-07_whos_who_deep_research.md).\nAfter the $150M Series B failed, Rain put itself up for sale (May 2025), with OpenAI,\nNvidia, and Microsoft reported as potential acquirers; OpenAI interviewed Rain engineers to\nassess an acqui-hire (https://nypost.com/2025/05/15/business/sam-altman-backed-chipmaking-startup-rain-ai-explores-sale-after-150m-fundraising-flops-sources/;\nhttps://finance.yahoo.com/news/sam-altmans-150m-ai-chip-123106283.html).\n**As of 2026-07-23 no completed acquisition, asset sale, or shutdown has been publicly\nconfirmed**; Tracxn listed ~48 employees as of March 2026 and rain.ai remains live but\nminimal (https://tracxn.com/d/companies/rain/__sKp9sqv2Nc690RoqfetuWtJafOBJ1FxqqhVp9VM2WtM).\n**Single-sourced late-2025 claims** — per the 2026-07 who's-who research report\n(_research/2026-07_whos_who_deep_research.md), unverified (the report itself flags them as\nsingle-sourced): Rain could not fund a leading-node tape-out, saw significant talent\nattrition (engineers to OpenAI's internal chip effort and to Microsoft), and was pivoting\ntoward licensing its Digital In-Memory Compute IP. Corroborate before high-stakes use.",
   "relevance": "The wave's most-hyped memristor-native startup: it raised on analog memristor crossbars,\nfound them commercially premature, pivoted to digital IMC, and still failed to reach market\n— a factual cautionary datapoint on the gap between memristor lab results and product,\nindependent of any one company's device quality.",
   "news": [],
   "sources": [
    {
     "title": "Yahoo/Benzinga — Sam Altman's $150M AI Chip Bet Crashes",
     "url": "https://finance.yahoo.com/news/sam-altmans-150m-ai-chip-123106283.html"
    },
    {
     "title": "NY Post — Rain AI explores sale",
     "url": "https://nypost.com/2025/05/15/business/sam-altman-backed-chipmaking-startup-rain-ai-explores-sale-after-150m-fundraising-flops-sources/"
    },
    {
     "title": "EE Times — Rain Neuromorphics tapes out demo chip for analog AI",
     "url": "https://www.eetimes.com/rain-neuromorphics-tapes-out-demo-chip-for-analog-ai/"
    },
    {
     "title": "design-reuse — $25M Series A",
     "url": "https://www.design-reuse.com/news/51363/rain-neuromorphics-funding.html"
    },
    {
     "title": "startupwired — Rise and Fall of a Chipmaking Challenger",
     "url": "https://startupwired.com/2025/05/16/rain-ai-the-rise-and-fall-of-a-chipmaking-challenger/"
    },
    {
     "title": "GlobeNewswire — Andes RISC-V partner",
     "url": "https://www.globenewswire.com/news-release/2024/06/04/2892703/0/en/Rain-AI-Unveils-Andes-Technology-as-Its-RISC-V-Partner.html"
    },
    {
     "title": "rain.ai",
     "url": "https://rain.ai"
    },
    {
     "title": "NeuromorphicCore.ai Rain AI profile (Oct 2024 chip shipping)",
     "url": "https://neuromorphiccore.ai/insights/rain-ai/"
    }
   ]
  },
  {
   "slug": "reexen",
   "name": "Reexen Technology (九天睿芯)",
   "founded": 2018,
   "hq": "Shenzhen, China (+ R&D in Zurich, Chengdu, Shanghai)",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (ETH Zurich / Institute of Neuroinformatics, neuromorphic engineering)",
   "total_funding_usd_m": 200,
   "last_round": "~CNY 1B across B/C tranches closed 2026-10-22/23 (Meituan, Xpeng/星航资本, miHoYo, OmniVision et al.; labels vary by outlet)",
   "key_investors": [
    "MiraclePlus (奇绩创坛)",
    "Will Semi/OmniVision's 韦豪创芯",
    "Pudong Innoventure (浦东科创)",
    "Yangtze Memory (长江存储) affiliates",
    "Yuanhe Puhua (元禾璞华)",
    "Futian Guidance Fund"
   ],
   "website": "https://www.reexen.com",
   "pitch": "Reexen attacks the bottleneck one step earlier than most CIM companies: not just\nmemory-vs-compute but **sensor-vs-compute**. Most always-on sensor data is redundant, and\ndigitizing everything before computing wastes power; its 感存算一体\n(\"sensing-memory-compute integration\") chips process signals in analog/mixed-signal form\nat the sensor and inside memory, aiming to be \"the leader in sensing-memory-compute\narchitecture chips for the intelligent era\" and a \"post-Moore ADI\"\n(https://www.reexen.com/; https://www.inno-chip.cn/329.html).",
   "technical_approach": "- **Substrate:** ultra-low-power **analog/mixed-signal (数模混合) compute-in-memory plus\n  neuromorphic front-ends**, descended from ETH Zurich's Institute of Neuroinformatics —\n  the founding lab of mixed-signal neuromorphic electronics\n  (https://startup.aliyun.com/info/103330.html).\n- **ADA100**: voice/audio \"inference sensing\" chip, ~1 GOPS equivalent at µW-class\n  always-on power; **ADA200**: vision/video processing CIM chip — both commercialized by\n  2022 (https://www.eet-china.com/mp/a151258.html).\n- Second product line: high-performance ADC/converter IP (the \"China's ADI\" ambition),\n  which funds the neuromorphic roadmap (https://www.eet-china.com/mp/a151258.html).\n- **Inference-only** (sensor-side inference); no on-chip learning claim. EU project\n  funding during the ETH period seeded the technology\n  (https://jiweipreview.laoyaoba.com/html/share/news/785427?source=h5&news_id=785427).",
   "funding_rounds": [
    {
     "date": "2020",
     "round": "Pre-A",
     "amount": "undisclosed",
     "investors": "MiraclePlus / 奇绩创坛 (Lu Qi)",
     "source": "https://reexen.com/newsinfo/785774.html"
    },
    {
     "date": "2021-06",
     "round": "Series A",
     "amount": "CNY 100M+ (~$15M; company English PR: \">10M CHF\")",
     "investors": "韦豪创芯 (Will Semi/OmniVision affiliate) + 浦东科创 co-lead; 华义创投, 磐缠投资, 上海物枢",
     "source": "https://m.cyzone.cn/article/651514; https://reexen.com/newsinfo/785817.html"
    },
    {
     "date": "2025-09",
     "round": "Series B",
     "amount": "CNY 100M+ (~$14M)",
     "investors": "元禾璞华, 韦豪创芯, 奇绩创坛, 盈浩资本 et al.",
     "source": "https://pitchhub.36kr.com/project/1678260467840003"
    },
    {
     "date": "2026-01",
     "round": "Series B+",
     "amount": "undisclosed",
     "investors": "Futian (Shenzhen) Guidance Fund, 民银国际 et al.",
     "source": "https://pitchhub.36kr.com/project/1678260467840003"
    },
    {
     "date": "2026-04",
     "round": "Series B++",
     "amount": "~CNY 200M (~$28M)",
     "investors": "incl. Yangtze Memory (长江存储) and OmniVision (豪威)-linked investors, 盈浩资本",
     "source": "https://pitchhub.36kr.com/project/1678260467840003"
    },
    {
     "date": "2026-05",
     "round": "Series C",
     "amount": "CNY 100M+ (~$14M)",
     "investors": "星航资本",
     "source": "https://pitchhub.36kr.com/project/1678260467840003"
    }
   ],
   "funding_notes": "Disclosed amounts sum to roughly **CNY 500–600M (~$70–85M) — partial**: Pre-A/B+ amounts\nundisclosed, and the 2025–2026 rounds rest on 36kr PitchHub tracker data (investor-name\nromanizations approximate). A Foxconn Industrial Internet (FII) news item also links FII\nto Reexen (https://www.fii-foxconn.com/NewsDetail?id=4638) — unverified round placement.",
   "trajectory": "Founded June 2018 in Shenzhen by **Liu Hongjie (刘洪杰)**, founder-chairwoman-CEO: Jilin\nUniv. undergrad, NTU Singapore MSc in IC design, PhD in neuromorphic engineering at ETH\nZurich (mixed-signal/neuromorphic computing lab); began mixed-signal AI processor research\nat ETH in 2013; chief scientist \"Dr. Yang\" is also ex-ETH\n(https://fgw.sz.gov.cn/ztzl/qtztzl/szscjmyjjfzzhfwpt/mqfc/myqyjdxsj/content/post_12174969.html;\nhttps://www.reexen.com/newsinfo/785768.html). R&D centers in Shenzhen, Chengdu, Shanghai\nand Zurich. Won a Deloitte Shenzhen \"Star of Tomorrow\" award 2023; reported ~CNY 50M sales\nin its first three years. Status mid-2026: active, with an unusually rapid 2025–2026\nfunding cadence (B → C in nine months) and strategic memory/sensor investors (YMTC-,\nOmniVision-linked), pointing at sensor-supply-chain integration.",
   "relevance": "Reexen is the direct commercial descendant of the Mead-lineage mixed-signal neuromorphic\nschool (ETH INI) — the same intellectual tradition Knowm draws on — but it monetizes the\nanalog front-end (sensing + inference) rather than adaptive synaptic learning; weights are\nfixed at deployment, unlike AHaH's on-chip learning thesis.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "**OmniVision Group (豪威集团, 603501.SH) disclosed a new direct investment in Reexen** (2026-08-17) — deepening the sensor-supply-chain tie beyond the earlier 韦豪创芯 rounds. Reexen is also pitching an **HBF/tiered-memory (\"存算融合\") architecture** under OCP's new HBF spec, claiming single-card trillion-parameter model inference (2026-08-05/21). (https://4g.stockstar.com/detail/RB2026081800003815 ; https://m.sohu.com/a/1065633393_121948415)"
    },
    {
     "date": "2026-10-27",
     "text": "**~CNY 1B (~US$140M) raised inside six months, closed ~2026-10-22/23.** Coverage labels the rounds inconsistently (Sina: 'B round totaling 1B'; Sohu: 'C round, 100M+'; Dealroom: '$150M'). Named investors: Meituan strategic investment, Xpeng (via 星航资本), 长存产业基金, 元禾璞华, OmniVision (豪威集团) early; 智微资本, miHoYo strategic, 嘉宾资本, 博将资本 later; existing holders 英豪资本 and 韦豪创芯 topped up. Money goes to the **SRAM-CIM + HBF (high-bandwidth flash) 3D fused chip** tape-out, the MPM (model-processor-memory) software stack and cloud/edge/device products; SRAM-CIM products are described as already in mass production. Claims: DeepSeek-V4 at ~20 decode TPS on one card + four SSDs on an engineering platform; HBF chip targets >10x per-card bandwidth/TPS. Edge partners named: vivo, Honor, Lenovo. **Working total ≈ US$200M (partial; CNY under-reported)**, up from ~70–85. (https://tech.sina.cn/2026-10-23/detail-iniweiyu8145993.d.html ; https://m.sohu.com/a/1091894649_115035 ; https://app.dealroom.co/news/note/reexen-technology-raises-150-million-for-compute-in-memory-ai-chips)"
    }
   ],
   "sources": [
    {
     "title": "Reexen site",
     "url": "https://www.reexen.com/"
    },
    {
     "title": "Reexen history page",
     "url": "https://www.reexen.com/newsinfo/785768.html"
    },
    {
     "title": "Reexen English PR: A round >10M CHF",
     "url": "https://reexen.com/newsinfo/785817.html"
    },
    {
     "title": "Cyzone: A round",
     "url": "https://m.cyzone.cn/article/651514"
    },
    {
     "title": "Aliyun/36kr: A round, founder background",
     "url": "https://startup.aliyun.com/info/103330.html"
    },
    {
     "title": "EET-China: \"China's ADI\", ADA100/ADA200",
     "url": "https://www.eet-china.com/mp/a151258.html"
    },
    {
     "title": "36kr PitchHub tracker: B/B+/B++/C rounds",
     "url": "https://pitchhub.36kr.com/project/1678260467840003"
    },
    {
     "title": "Shenzhen FGW founder profile",
     "url": "https://fgw.sz.gov.cn/ztzl/qtztzl/szscjmyjjfzzhfwpt/mqfc/myqyjdxsj/content/post_12174969.html"
    },
    {
     "title": "JW Insights founder profile",
     "url": "https://jiweipreview.laoyaoba.com/html/share/news/785427?source=h5&news_id=785427"
    }
   ]
  },
  {
   "slug": "sagence-ai",
   "name": "Sagence AI (fka Analog Inference)",
   "founded": 2018,
   "hq": "Santa Clara, CA, USA",
   "category": "analog-imc",
   "status": "active (quiet since stealth exit; product timelines unconfirmed)",
   "origin": "serial founder / industry veteran (Vishal Sarin, analog & flash-memory chip career)",
   "total_funding_usd_m": 58,
   "last_round": "Series B (details undisclosed), pre-2024-11",
   "key_investors": [
    "Vinod Khosla / Khosla Ventures",
    "Andy Bechtolsheim",
    "Atiq Raza",
    "TDK Ventures",
    "P7 Ventures (Aramco)",
    "Alumni Ventures",
    "Cambium Capital",
    "New Science Ventures",
    "MFV Partners"
   ],
   "website": "https://www.sagence-ai.com",
   "pitch": "Sagence says its analog in-memory compute \"directly addresses the untenable\npower/performance/price and environmental sustainability conundrum facing AI\ninferencing,\" claiming \"multiple orders of magnitude improvement in energy efficiency\nand cost\" versus GPU/CPU systems while keeping equivalent performance.\nSource: https://www.businesswire.com/news/home/20241119609411/en/Sagence-AI-Emerges-from-Stealth-Tackling-Economic-Viability-of-Inference-Hardware-for-Generative-AI",
   "technical_approach": "- Analog **in-memory compute** in which weights are stored in nonvolatile memory cells\n  and matrix math executes in the analog domain (Sarin's background is deep in\n  multi-level flash memory design; IEEE Spectrum profiles the approach as deeply\n  scaled analog compute-in-memory: https://spectrum.ieee.org/analog-ai-2669898661).\n- Inference-only, targeting generative-AI/datacenter inference *and* edge vision.\n  Benchmark claim at stealth exit: for Llama2-70B normalized to 666K tokens/sec,\n  10x lower power, 20x lower price, 20x smaller rack space than a leading GPU\n  (BusinessWire PR above).\n- First product was slated for 2025, aimed at vision systems\n  (https://techcrunch.com/2024/11/18/sagence-is-building-analog-chips-to-run-ai/);\n  no launch announcement found as of 2026-07-23.",
   "funding_rounds": [
    {
     "date": "2018+",
     "round": "Seed",
     "amount": "undisclosed",
     "investors": "Vinod Khosla, Andy Bechtolsheim, Atiq Raza (angels/seed)",
     "source": "https://www.businesswire.com/news/home/20241119609411/en/Sagence-AI-Emerges-from-Stealth-Tackling-Economic-Viability-of-Inference-Hardware-for-Generative-AI"
    },
    {
     "date": "pre-2024-11",
     "round": "Series A & B",
     "amount": "undisclosed split",
     "investors": "TDK Ventures, P7 Ventures (Aramco), Alumni Ventures, Cambium Capital Partners, New Science Ventures",
     "source": "same BusinessWire PR"
    }
   ],
   "funding_notes": "Totals conflict: the company's stealth-exit PR says **$58M** from strategic and\nventure investors (BusinessWire, above); Tracxn/Crunchbase-derived profiles list\n**$50.5M** with Khosla Ventures and MFV Partners as investors\n(https://tracxn.com/d/companies/sagence-ai/__JvzIBpaeqMAvUg-vc9EuuunJ_2S_P1epX_xz1Adc5uc).\nNo round-by-round breakdown has been disclosed.",
   "trajectory": "Founded 2018 as **Analog Inference** by CEO **Vishal Sarin** after decades in the chip\nindustry (analog/flash memory); operated in stealth for six years, rebranded to\n**Sagence AI** and emerged publicly on 2024-11-19\n(https://techcrunch.com/2024/11/18/sagence-is-building-analog-chips-to-run-ai/).\nSince the Nov 2024 stealth exit, no funding, product-launch, or shutdown news was\nfound through 2026-07 — Google News turns up nothing newer than early-2025 commentary.\n**Status mid-2026: presumed active but quiet**; treat product timing (\"2025 vision\nproduct\") as unconfirmed.\n\n**Founders (sweep 2026-09-22):** Vishal Sarin (sole founder, President & CEO; MSEE University of Michigan, MBA UC Berkeley; 100+ patents; \"working on storing multiple bits in flash for 30 years\" per IEEE Spectrum). Immediately prior employer not named in any company, press or database source found (https://www.sagence-ai.com/about-us/ ; https://spectrum.ieee.org/analog-ai-2669898661 ; https://www.crunchbase.com/organization/analog-inference).",
   "relevance": "Sagence is the datacenter-scale version of the flash-analog bet, backed by marquee\nValley angels (Khosla, Bechtolsheim). Its 6-year stealth and post-launch silence\nillustrate how long analog IMC gestation runs — again inference-only, with weights\nprogrammed rather than learned in the device.",
   "news": [],
   "sources": [
    {
     "title": "Stealth-exit PR",
     "url": "https://www.businesswire.com/news/home/20241119609411/en/Sagence-AI-Emerges-from-Stealth-Tackling-Economic-Viability-of-Inference-Hardware-for-Generative-AI"
    },
    {
     "title": "TechCrunch",
     "url": "https://techcrunch.com/2024/11/18/sagence-is-building-analog-chips-to-run-ai/"
    },
    {
     "title": "IEEE Spectrum",
     "url": "https://spectrum.ieee.org/analog-ai-2669898661"
    },
    {
     "title": "Tracxn profile",
     "url": "https://tracxn.com/d/companies/sagence-ai/__JvzIBpaeqMAvUg-vc9EuuunJ_2S_P1epX_xz1Adc5uc"
    }
   ]
  },
  {
   "slug": "salience-labs",
   "name": "Salience Labs",
   "founded": 2021,
   "hq": "Oxford, UK",
   "category": "photonic",
   "status": "active",
   "origin": "academic spinoff (University of Oxford — Harish Bhaskaran's lab — and University of Münster — Wolfram Pernice / Johannes Feldmann's photonic in-memory computing work)",
   "total_funding_usd_m": 166.5,
   "last_round": "Series B, $125M, 2026-10-14 (Bullhound Capital lead; Intel Capital, AMD Ventures)",
   "key_investors": [
    "Applied Ventures",
    "ICM HPQC Fund",
    "Braavos",
    "Oxford Science Enterprises",
    "Cambridge Innovation Capital",
    "Silicon Catalyst"
   ],
   "website": "https://www.saliencelabs.ai",
   "pitch": "AI demand is outrunning what electronics can deliver: neural-network matrix multiplication\nexecuted in light happens \"in a single pass rather than through multiple clock cycles,\"\nenabling dense chips clocking at tens of GHz — and, in the current framing, AI datacenter\nscale-out is bottlenecked by switching/network latency and power that only optics can fix.\n(https://www.businesswire.com/news/home/20220512005064/en/, https://www.saliencelabs.ai/)",
   "technical_approach": "- Origin tech: photonic in-memory computing from Oxford/Münster (Feldmann et al.'s Nature\n  2021 photonic tensor core line of work); Salience's variant is an **amplitude-based**\n  (not phase-based) photonic multiply, packaged as a multi-chip processor — photonics on top\n  of standard electronics, deliberately foundry-compatible CMOS processes.\n  (https://www.eetimes.com/british-photonic-computing-startup-raises-seed-round/, https://semiwiki.com/semiconductor-services/silicon-catalyst/313017-ceo-interview-vaysh-kewada-of-salience-labs/)\n- **Pivot (honest read):** the 2022 seed pitched a photonic AI *accelerator* (matmul engine);\n  the 2025 Series A is explicitly for **optical switches for large-scale AI connectivity** —\n  high-radix, ultra-low-latency switching for AI datacenter networks. Like Lightmatter and\n  Lightelligence, compute-with-light became move-data-with-light.\n  (https://www.datacenterdynamics.com/en/news/salience-labs-raises-30m-for-ai-data-center-optical-switch/)\n- Inference-oriented when it was an accelerator; no on-chip learning claims. No named\n  commercial chip publicly dated yet.",
   "funding_rounds": [
    {
     "date": "2022-05",
     "round": "Seed",
     "amount": "$11.5M",
     "investors": "Cambridge Innovation Capital + Oxford Science Enterprises (co-led); Silicon Catalyst Angels, Arm-based angels, Jalal Bagherli",
     "source": "https://www.businesswire.com/news/home/20220512005064/en/"
    },
    {
     "date": "2025-02",
     "round": "Series A",
     "amount": "$30M",
     "investors": "ICM HPQC Fund + Applied Ventures (Applied Materials); Braavos, OSE, CIC",
     "source": "https://www.businesswire.com/news/home/20250207290130/en/"
    }
   ],
   "funding_notes": "Total raised: **$41.5M**. No valuation reported.",
   "trajectory": "Co-founded 2021 by Vaysh Kewada (CEO, ex-Entrepreneur-in-Residence at Oxford Science\nEnterprises) and Johannes Feldmann (CTO, Münster/Oxford photonic-computing researcher), to\ncommercialize joint Oxford-Münster research. Seed May 2022; through 2023-2024 repositioned\nfrom photonic compute toward optical switching for AI clusters; Series A closed Feb 2025 led\nby Applied Materials' VC arm — a strategic-semiconductor signal. Active as of mid-2026,\npre-product, building optical switch silicon in the UK.",
   "relevance": "Directly descended from the *photonic in-memory computing* literature (phase-change cells on\nwaveguides) — the optical cousin of memristive IMC — yet commercially it too slid from\nin-memory compute to interconnect/switching. Reinforces the article's pattern: analog\nphysics-based matmul startups converge on selling bandwidth, while adaptive/learning claims\nstay in the lab.",
   "news": [
    {
     "date": "2026-10-27",
     "text": "**$125M Series B (2026-10-14), led by Bullhound Capital with Intel Capital and AMD Ventures** plus earlier investors; Bullhound's Callum Stewart joins the board. Disclosed rounds now sum to **~$166.5M** ($11.5M seed 2022 + $30M A 2025-02 + $125M B). Use of funds: volume manufacturing of its optical circuit switch (OCS), a second Silicon Valley office, and 128/256-port follow-ons. At ECOC 2026 (Málaga, ~2026-09-23) it demonstrated a **64-port all-optical silicon-photonic OCS**, claimed as an industry first; manufacturing partners Keysight and Tower Semiconductor. The pivot from photonic compute to optical switching is now the funded product. (https://optics.org/news/uks-salience-labs-adds-125m-series-b ; https://www.datacenterdynamics.com/en/news/optical-switching-startup-salience-labs-raises-125m-with-amd-and-intel-participating-in-round/ ; https://www.fibre-systems.com/article/ecoc-2026-salience-labs-demonstrates-64-port-silicon-photonic-ocs)"
    }
   ],
   "sources": [
    {
     "title": "Businesswire seed",
     "url": "https://www.businesswire.com/news/home/20220512005064/en/Salience-Labs-Raises-11.5-Million-Seed-to-Develop-Ultra-High-speed-Chips-Using-Photonics-for-AI-Applications"
    },
    {
     "title": "Businesswire Series A",
     "url": "https://www.businesswire.com/news/home/20250207290130/en/Salience-Labs-Closed-Series-A-Round-Led-by-Applied-Ventures-and-ICM-HPQC-Fund"
    },
    {
     "title": "DCD: $30M for AI datacenter optical switch",
     "url": "https://www.datacenterdynamics.com/en/news/salience-labs-raises-30m-for-ai-data-center-optical-switch/"
    },
    {
     "title": "EE Times seed coverage",
     "url": "https://www.eetimes.com/british-photonic-computing-startup-raises-seed-round/"
    },
    {
     "title": "Semiwiki CEO interview",
     "url": "https://semiwiki.com/semiconductor-services/silicon-catalyst/313017-ceo-interview-vaysh-kewada-of-salience-labs/"
    },
    {
     "title": "Tech.eu seed profile",
     "url": "https://tech.eu/2022/05/12/meet-salience-labs-oxford-startup-that-chipped-in-115-million-to-develop-high-speed-chips-for-ai-applications/"
    }
   ]
  },
  {
   "slug": "semron",
   "name": "SEMRON",
   "founded": 2020,
   "hq": "Dresden, Germany",
   "category": "analog-imc",
   "status": "active (pre-product)",
   "origin": "academic spinoff (TU Dresden founders; CTO's memcapacitor research with Max Planck Institute)",
   "total_funding_usd_m": 10.6,
   "last_round": "Seed extension, $2.5M, 2025-02",
   "key_investors": [
    "Join Capital",
    "SquareOne",
    "OTB Ventures",
    "Onsight Ventures (Hermann Hauser)",
    "Berkeley SkyDeck"
   ],
   "website": "https://www.semron.ai",
   "pitch": "SEMRON's CapRAM computes \"using a variable capacitor ('memcapacitor') instead of\nrelying on currents like resistive approaches and transistors,\" minimizing electron\nmovement to cut noise and claiming a ~20x chip-level energy-efficiency gain — the\nroute, it argues, to running large AI models inside phones, earbuds and XR devices\ninstead of the cloud.\nSource: https://tech.eu/2024/01/30/semron-secures-eur73m-seed-funding/",
   "technical_approach": "- Device: **memcapacitor** (\"CapRAM™\") — a charge/field-based analog weight element\n  rather than a resistive (current-based) one; because cells couple by electric field,\n  static power and noise are low. Weights stored capacitively; analog in-memory MAC.\n  (https://tech.eu/2024/01/30/semron-secures-eur73m-seed-funding/ ;\n  https://techcrunch.com/2024/01/30/semron-wants-to-replace-chip-transistors-with-memcapacitors/)\n- Signature claim: **3D stacking** of memcapacitor arrays (\"industry-first 3D AI\n  chips,\" with talk of hundreds of layers) to reach the weight capacity for on-device\n  LLMs at consumer cost (tech.eu, TechCrunch above).\n- CTO Kai-Uwe Demasius's memcapacitive-device research was published with the Max\n  Planck Institute (Stuart Parkin's group) before founding; Join Capital's investment\n  memo describes the device lineage\n  (https://medium.com/the-neue-industry/on-the-edge-of-the-possible-why-we-invested-in-semron-31a0ba0a0acc).\n- Inference-only; no named commercial chip yet as of mid-2026 (development/prototype\n  stage).",
   "funding_rounds": [
    {
     "date": "2023-03",
     "round": "Pre-seed",
     "amount": "~$0.2M",
     "investors": "Berkeley SkyDeck",
     "source": "https://canvasbusinessmodel.com/blogs/brief-history/semron-brief-history"
    },
    {
     "date": "2024-01",
     "round": "Seed",
     "amount": "€7.3M (~$7.9M)",
     "investors": "Join Capital (lead); SquareOne, OTB Ventures, Onsight Ventures (Hermann Hauser) + pre-seed investors",
     "source": "https://tech.eu/2024/01/30/semron-secures-eur73m-seed-funding/"
    },
    {
     "date": "2025-02",
     "round": "Seed extension",
     "amount": "$2.5M",
     "investors": "undisclosed (per Tracxn)",
     "source": "https://tracxn.com/d/companies/semron/__yayyChO2PyIChZ_1-X9t3VNABZEUlyEWVAho3jV1gJo"
    }
   ],
   "funding_notes": "Total ≈ **$10.6M**. Germany's federal SPRIND (Federal Agency for Disruptive\nInnovation) has also profiled/supported SEMRON's seed financing\n(https://www.sprind.org/en/articles/semron/).",
   "trajectory": "Founded 2020 in Dresden by **Aron Kirschen** (CEO) and **Kai-Uwe Demasius** (CTO)\n(https://tech.eu/2024/01/30/semron-secures-eur73m-seed-funding/). Went through Berkeley\nSkyDeck (2023), raised its institutional seed Jan 2024, small extension Feb 2025.\nNo product launch, large round, or shutdown news found through 2026-07; Google News\nshows nothing significant after the Jan 2024 coverage.\n**Status mid-2026: active, pre-product, early-stage** — the smallest-funded company in\nthis batch.",
   "relevance": "SEMRON is the one \"conventional-ish device\" player whose element — the memcapacitor —\nis itself an emerging device, sitting between this batch and the memristor camp. Like\nmemristor IMC it stores analog weights in a physical state, but it computes with\ncharge/fields rather than current through a resistive filament; still inference-only,\nno on-device learning claim.",
   "news": [],
   "sources": [
    {
     "title": "tech.eu seed coverage",
     "url": "https://tech.eu/2024/01/30/semron-secures-eur73m-seed-funding/"
    },
    {
     "title": "TechCrunch memcapacitor story",
     "url": "https://techcrunch.com/2024/01/30/semron-wants-to-replace-chip-transistors-with-memcapacitors/"
    },
    {
     "title": "Join Capital investment memo",
     "url": "https://medium.com/the-neue-industry/on-the-edge-of-the-possible-why-we-invested-in-semron-31a0ba0a0acc"
    },
    {
     "title": "SPRIND on SEMRON",
     "url": "https://www.sprind.org/en/articles/semron/"
    },
    {
     "title": "Tracxn profile",
     "url": "https://tracxn.com/d/companies/semron/__yayyChO2PyIChZ_1-X9t3VNABZEUlyEWVAho3jV1gJo"
    }
   ]
  },
  {
   "slug": "spinncloud",
   "name": "Sparxity (formerly SpiNNcloud Systems; renamed 2026-10-28)",
   "founded": 2021,
   "hq": "Dresden, Germany",
   "category": "digital-snn",
   "status": "active",
   "origin": "academic spinoff (TU Dresden, Chair of Highly-Parallel VLSI & Neuromorphic Circuits — SpiNNaker2 project with U. Manchester / Human Brain Project lineage)",
   "total_funding_usd_m": "undisclosed / conflicting (~€12.5M+ in EU grants+blended funding is solid; VC seed figures range $0.6M–$11M across trackers)",
   "last_round": "EIC Accelerator blended finance, €10M, 2025-06",
   "key_investors": [
    "EIC Accelerator/EIC Fund",
    "EU grants",
    "Silicon Catalyst Angels/Ventures",
    "GS Futures"
   ],
   "website": "https://www.sparxity.com",
   "pitch": "\"Revolutionizing the inference market by democratizing AI infrastructure with ultra-low energy\nsystems. SpiNNaker2: 18x higher energy efficiency than GPUs\" (https://spinncloud.com/).\nTheir claim: brain-inspired, massively parallel event-driven many-core systems beat GPU\ndatacenter economics for sparse, real-time AI workloads.",
   "technical_approach": "- **Fully digital many-core neuromorphic architecture** — commercialization of **SpiNNaker2**\n  (successor to Steve Furber's Manchester SpiNNaker): each chip carries 152 Arm cores with\n  event-based communication fabric; systems scale to a 5-million-core Dresden supercomputer\n  (https://arxiv.org/pdf/2401.04491).\n- Hybrid workloads: spiking neural networks, sparse DNNs, and \"neuromorphic supercomputing\"\n  for research; software-defined plasticity means on-chip learning is possible in software\n  (Arm cores), unlike fixed-function inference ASICs.\n- **SpiNNcloud Platform / SpiNNaker2 systems** commercially available since May 2024 — billed\n  as the \"first commercially available neuromorphic supercomputer\"\n  (https://tudag.de/entrepreneurship-en/spinncloud-systems-announces-first-commercially-available-neuromorphic-supercomputer/?lang=en).\n- Deployments: Sandia National Laboratories SpiNNaker2 system (June 2025), named\n  \"Braunfels\", exploring neuromorphic computing for nuclear-deterrence missions\n  (https://siliconangle.com/2025/06/06/snowflake-ai-game-circle-ipo-blasts-off-elon-splits-trump/;\n  https://www.sandia.gov/ccr/news/sandia-deploys-spinnaker2-neuromorphic-system/) — the\n  ~175M-neuron scale and NNSA ASC funding route are per the 2026-07 who's-who research\n  report (_research/2026-07_whos_who_deep_research.md), unverified/single-sourced.\n- Leipzig University (announced 2025-07-28): 656,640-core SpiNNaker2 system (~4,320 chips)\n  for small-molecule drug discovery / protein folding — largest SpiNNcloud system deployed\n  to date (https://www.accessnewswire.com/newsroom/en/computers-technology-and-internet/spinncloud-advances-ai-driven-drug-discovery-with-deployment-of-b-1053033).\n- UT San Antonio, NSF-funded THOR \"Neuromorphic Commons\" (announced 2025-11-12): SpiNNaker2\n  system that per company PR \"ranks among the top four largest neuromorphic computing\n  platforms globally, simulating over 393 million neurons\" — their PR language, not an\n  independent ranking (https://www.hpcwire.com/2025/11/12/spinnclouds-spinnaker2-deployed-at-ut-san-antonio-for-nsf-funded-thor-project/).",
   "funding_rounds": [
    {
     "date": "2023-02",
     "round": "EU grant — \"SpiNNode: SpiNNaker2 on the edge\"",
     "amount": "€2.5M",
     "investors": "European Union",
     "source": "https://tu-dresden.de/tu-dresden/newsportal/news/rechentechnik-inspiriert-vom-gehirn-eu-millionen-fuer-saechsisches-deep-tech-startup-spinncloud-systems?set_language=en ; https://www.eenewseurope.com/en/spinnaker2-spiking-neural-network-project-gets-e2-5-million/"
    },
    {
     "date": "2024–2025",
     "round": "Angel/seed equity (amounts conflicting)",
     "amount": "see note",
     "investors": "Silicon Catalyst Angels/Ventures, GS Futures",
     "source": "https://pitchbook.com/profiles/company/483184-72 ; https://nordic9.com/news/spinncloud-in-an-equity-deal-funded-by-silico/"
    },
    {
     "date": "2025-06",
     "round": "EIC Accelerator blended finance (grant + equity)",
     "amount": "€10M",
     "investors": "European Innovation Council",
     "source": "https://ebrains.eu/news-and-events/2025/brain-tech-in-action-german-spinncloud-computing-startup-secures-funding-of-10 ; https://www.spinlab.co/news/spinncloud-secures-10-million-blended-funding-eic"
    }
   ],
   "funding_notes": "**Conflicting seed figures — reporting all:** StartupIntros claims an $11.0M seed in Jan 2025\n(https://startupintros.com/orgs/spinncloud); AngelsRound lists \"$14M (Seed)\" total\n(https://www.angelsround.com/p/spinncloud); PitchBook shows only $595K raised (GS Futures,\nSilicon Catalyst Angels) (https://pitchbook.com/profiles/company/483184-72); CB Insights says\n$2.64M over 5 rounds (https://www.cbinsights.com/company/spinncloud/financials); Tracxn calls\nthe amounts undisclosed (https://tracxn.com/d/companies/spinncloud/__OMQwpBsgziYccyHC3sSnrXl97uDieBdrx_wDVW5KLQw).\nNo primary press release for a VC seed was found. Solidly sourced money = €12.5M EU\ngrant/blended funding above.",
   "trajectory": "Founded 2021 by Hector A. Gonzalez (CEO), Christian Eichhorn (co-founder & managing\ndirector, https://www.crunchbase.com/person/christian-eichhorn), Matthias Lohrmann, and\nChristian Mayr (TU Dresden professor) to commercialize SpiNNaker2\n(https://tracxn.com/d/companies/spinncloud/__OMQwpBsgziYccyHC3sSnrXl97uDieBdrx_wDVW5KLQw,\nhttps://www.bayern-innovativ.de/showcontact/detail/hector-a-gonzalez-diaz/). Trajectory:\nHuman Brain Project research hardware → commercial platform (May 2024) → national-lab and\nuniversity sales (Sandia 2025, Leipzig 2025, UTSA/THOR 2025) → EIC-backed scale-up; pivot in messaging toward\ndatacenter/energy-efficient inference for mainstream sparse AI. As of mid-2026: **active**,\nselling SpiNNaker2 systems.",
   "relevance": "The \"many small ARM cores + event routing\" approach is the most software-flexible of the\ndigital SNN camp — plasticity/learning is programmable rather than physical. It competes on\nsystem-level energy per inference, not on device-level synaptic adaptation, making it a\nuseful digital-CMOS contrast to Knowm's memristor-based learning substrate.",
   "news": [
    {
     "date": "2026-11-09",
     "text": "**Renamed Sparxity (2026-10-28)** — the rename doubles as the launch of a next generation 'committed fully to sparsity for AI inference'; current SpiNNaker2 systems keep serving HPC/AI. Same day: **JUBIS, a 'multi-million-dollar' sparse-native rack at Jülich Supercomputing Centre** under JSC's three-year ESAB project (Event-based Supercomputer for AI and Beyond, funded by ERDF/EFRE-JTF NRW 2021-27); claims '10x faster sparse solver execution' with no published baseline. Company tallies: three tape-outs, >70,000 chips, >12 racks, >30 customers; successor chip SpiNNext claimed at 78x GPU energy efficiency vs 18x for SpiNNaker2, 'coming soon'. Website now sparxity.com. (https://www.storagereview.com/news/sparxity-spinncloud-jubis-sparse-native-rack-julich-supercomputing-centre ; https://www.datacenterdynamics.com/en/news/spinncloud-rebrands-to-sparxity-announces-cluster-deployment-at-j%C3%BClich-center-in-germany/)"
    }
   ],
   "sources": [
    {
     "title": "SpiNNcloud homepage",
     "url": "https://spinncloud.com/"
    },
    {
     "title": "TU Dresden: €2.5M SpiNNode grant (Feb 2023)",
     "url": "https://tu-dresden.de/tu-dresden/newsportal/news/rechentechnik-inspiriert-vom-gehirn-eu-millionen-fuer-saechsisches-deep-tech-startup-spinncloud-systems?set_language=en"
    },
    {
     "title": "EBRAINS: €10M EIC Accelerator (Jul 2025)",
     "url": "https://ebrains.eu/news-and-events/2025/brain-tech-in-action-german-spinncloud-computing-startup-secures-funding-of-10"
    },
    {
     "title": "TUDAG: first commercially available neuromorphic supercomputer (May 2024)",
     "url": "https://tudag.de/entrepreneurship-en/spinncloud-systems-announces-first-commercially-available-neuromorphic-supercomputer/?lang=en"
    },
    {
     "title": "SiliconANGLE: Sandia deploys SpiNNaker2 (Jun 2025)",
     "url": "https://siliconangle.com/2025/06/06/snowflake-ai-game-circle-ipo-blasts-off-elon-splits-trump/"
    },
    {
     "title": "SpiNNaker2 arXiv paper (Jan 2024)",
     "url": "https://arxiv.org/pdf/2401.04491"
    },
    {
     "title": "PitchBook",
     "url": "https://pitchbook.com/profiles/company/483184-72"
    },
    {
     "title": "AccessNewswire: 656,640-core Leipzig system (2025-07-28)",
     "url": "https://www.accessnewswire.com/newsroom/en/computers-technology-and-internet/spinncloud-advances-ai-driven-drug-discovery-with-deployment-of-b-1053033"
    },
    {
     "title": "HPCwire: SpiNNaker2 at UTSA for NSF THOR (2025-11-12)",
     "url": "https://www.hpcwire.com/2025/11/12/spinnclouds-spinnaker2-deployed-at-ut-san-antonio-for-nsf-funded-thor-project/"
    },
    {
     "title": "Sandia CCR: SpiNNaker2 \"Braunfels\" deployment",
     "url": "https://www.sandia.gov/ccr/news/sandia-deploys-spinnaker2-neuromorphic-system/"
    }
   ]
  },
  {
   "slug": "sync-computing",
   "name": "Sync Computing",
   "founded": 2019,
   "hq": "Cambridge, Massachusetts, USA",
   "category": "probabilistic-thermodynamic",
   "status": "acquired",
   "origin": "academic spinoff (MIT Lincoln Laboratory)",
   "total_funding_usd_m": 21.6,
   "last_round": "Series A, $15.5M, 2022-08",
   "key_investors": [
    "Costanoa Ventures",
    "The Engine",
    "Moore Strategic Ventures",
    "National Grid Partners"
   ],
   "website": "https://www.synccomputing.com",
   "pitch": "Sync Computing's founding idea was to solve hard combinatorial optimization by\nletting physics settle into a low-energy state instead of grinding through it on a\ndigital CPU — building an electronic Ising machine from coupled oscillators that\n\"relaxes\" to the optimum. The framing: certain optimization problems are\nexponentially expensive on von Neumann hardware but map naturally onto a physical\nsystem whose dynamics minimize an energy function. (Original framing: MIT News,\n\"Nature can help solve optimization problems,\" 2019; Blocks & Files 2022.)",
   "technical_approach": "- Originally: an **all-electrical, fully-connected coupled-oscillator Ising machine**\n  (invention disclosure \"All Electrical Fully Connected Coupled Oscillator Ising\n  Machine,\" filed Oct 2019 at MIT) — a physics-based / probabilistic-thermodynamic\n  optimization accelerator.\n- Pivoted to: **cloud-resource optimization software**. The team applied its\n  optimization expertise not to selling Ising hardware but to a SaaS product\n  (Gradient / the \"Orchestrator\") that continuously right-sizes and tunes cloud\n  compute clusters — notably Databricks/Apache Spark jobs — to cut cost and runtime.\n  The hardware Ising machine was left behind; the durable business is pure software.",
   "funding_rounds": [
    {
     "date": "2019-11",
     "round": "Seed",
     "amount": "~$1.3M",
     "investors": "The Engine, Underscore.vc",
     "source": "Blocks & Files / Crunchbase"
    },
    {
     "date": "2021-10",
     "round": "Venture (undisclosed)",
     "amount": "—",
     "investors": "—",
     "source": "Crunchbase"
    },
    {
     "date": "2022-01",
     "round": "Seed / emergence from stealth (total to date)",
     "amount": "$6.1M",
     "investors": "The Engine, Underscore.vc",
     "source": "GlobeNewswire, 2022-01-25"
    },
    {
     "date": "2022-08",
     "round": "Series A",
     "amount": "$15.5M",
     "investors": "Costanoa Ventures (lead); The Engine, Moore Strategic Ventures, National Grid Partners",
     "source": "TechCrunch / GlobeNewswire"
    },
    {
     "date": "—",
     "round": "Total (reported)",
     "amount": "~$21.6M",
     "investors": "—",
     "source": "Crunchbase / Costanoa"
    }
   ],
   "funding_notes": "Note: the briefed \"~$15M\" matches the single largest round (Series A); cumulative\nreported funding is ~$21.6M.",
   "trajectory": "Spun out of MIT Lincoln Laboratory in 2019 by Jeff Chou (CEO) and Suraj Bramhavar\n(CTO), supported by DARPA's Activate Fellowship and The Engine. The company began\nwith oscillator-network Ising-machine hardware for optimization, but pivoted fully\nto cloud-infrastructure optimization software, emerging from stealth in Jan 2022\n($6.1M to date) and raising a $15.5M Series A in Aug 2022 led by Costanoa. Its\nproduct optimizes Databricks/Spark and ML data-pipeline resources. Status mid-2026:\nactive as a cloud-optimization software company; the Ising/oscillator hardware is\norigin-story only.",
   "relevance": "A textbook physics-based-compute pivot: a genuine probabilistic/thermodynamic\noptimization idea (coupled-oscillator Ising machine) that never shipped as hardware\nbecause the founders found a faster path to revenue in software. For the article it\nanchors the \"thermodynamic / Ising\" branch of non-von-Neumann computing and shows\nhow commercially unforgiving custom analog-optimization silicon was even for a\ncapable MIT team — a contrast to Knowm's persistence with a physical device.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "**Status correction — Sync Computing was acquired/absorbed: the team joined Capital One Software (April 2025)** to build out the Slingshot cloud-cost product (CEO Jeff Chou's LinkedIn lists \"Joined via Sync Computing acquisition\", Apr 2025; MIT Lincoln Laboratory's spinout page: \"In 2025, the Sync Computing team joined Capital One Software\"). Co-founder/CTO Suraj Bramhavar had earlier left for UK's ARIA. Chou himself left Capital One for Databricks ~Aug 2026. The Ising/oscillator hardware lineage is fully dead; even the cloud-optimization software business no longer exists independently. (https://www.ll.mit.edu/partner-us/working-lincoln-laboratory/spinout-companies ; https://capitalonesoftware.com/news/capital-one-software-strengthens-partnership-with-databricks ; https://www.linkedin.com/in/jeffchoumit)"
    }
   ],
   "sources": [
    {
     "title": "Sync Computing nabs $15.5M — TechCrunch, 2022-08-16",
     "url": "https://techcrunch.com/2022/08/16/sync-computing-rakes-in-15-5m-to-automatically-optimize-cloud-resources/"
    },
    {
     "title": "Sync Computing Emerges from Stealth, $6.1M — GlobeNewswire, 2022-01-25",
     "url": "https://www.globenewswire.com/news-release/2022/01/25/2372653/0/en/Distributed-Cloud-Infrastructure-Innovator-Sync-Computing-Emerges-from-Stealth-Brings-in-6-1-Million.html"
    },
    {
     "title": "Ising on the cake: Sync Computing spots opportunity for cloud resource optimisation — Blocks & Files",
     "url": "https://www.blocksandfiles.com/public-cloud/2022/01/13/ising-on-the-cake-sync-computing-spots-opportunity-for-cloud-resource-optimisation/1611249"
    },
    {
     "title": "Nature can help solve optimization problems — MIT News, 2019-10-28",
     "url": "https://news.mit.edu/2019/nature-can-help-solve-optimization-problems-1028"
    },
    {
     "title": "All Electrical Fully Connected Coupled Oscillator Ising Machine — MIT TLO",
     "url": "https://tlo.mit.edu/industry-entrepreneurs/available-technologies/all-electrical-fully-connected-coupled-oscillator"
    }
   ]
  },
  {
   "slug": "synsense",
   "name": "SynSense (fka aiCTX)",
   "founded": 2017,
   "hq": "Chengdu, China + Zurich, Switzerland",
   "category": "digital-snn",
   "status": "active",
   "origin": "academic spinoff (Institute of Neuroinformatics, University of Zurich & ETH Zurich — Ning Qiao, Giacomo Indiveri)",
   "total_funding_usd_m": "undisclosed in aggregate — Tracxn says $12.7M; Chinese-market reports add CNY rounds worth far more (conflict noted)",
   "last_round": "Series B, 'several hundred million yuan', reported 2026-07",
   "key_investors": [
    "Ausvic Capital",
    "CTC Capital (Series A lead)",
    "CAS Star",
    "M Ventures (Merck)",
    "Baidu Ventures (per 2026-07 report",
    "unverified)",
    "Chinese strategic/VC investors"
   ],
   "website": "https://www.synsense.ai",
   "pitch": "\"SynSense is a world-leading neuromorphic intelligence company, dedicated to developing core\nchips for next-generation human-machine and brain-computer [interfaces]\" — building on \"two\ndecades of pioneering research\" at the Institute of Neuroinformatics\n(https://www.synsense.ai/). Their consistent framing: sensor data should be processed in\nsparse, event-driven spikes at milliwatt budgets instead of frame-based pipelines feeding\nconventional processors.",
   "technical_approach": "- **Digital event-driven SNN processors** (plus earlier mixed-signal DYNAP research lineage\n  from INI). Inference-focused; training off-chip via their Sinabs/Rockpool toolchains.\n- **DYNAP-CNN / DynapCNN**: event-driven spiking convolutional processor for dynamic-vision\n  data (dev kits ~2019–2021).\n- **Speck**: single-SoC event camera + SNN processor (\"world's first\" fully integrated\n  neuromorphic vision SoC, per company; developed with iniVation-style DVS sensing)\n  (https://www.linkedin.com/company/synsense-neuromorphic/).\n- **Xylo**: ultra-low-power digital SNN chip for audio/vital-sign/IMU sensing.\n- Feb 2024: **acquired 100% of iniVation** (Zurich event-camera maker), forming a combined\n  neuromorphic sensing+processing group\n  (https://www.synsense.ai/synsense-and-inivation-join-forces-to-form-leading-neuromorphic-technology-provider/,\n  https://www.s-ge.com/invest/en/articles/news/synsense-acquires-inivation).",
   "funding_rounds": [
    {
     "date": "2017–",
     "round": "~10 early rounds (angel/strategic, mostly CNY, largely undisclosed)",
     "amount": "undisclosed",
     "investors": "incl. M Ventures (Merck)",
     "source": "https://tracxn.com/d/companies/synsense/__JSKRuzpG-21HQ3_RrriR9LV9UU7znKY0_AEjLrZAN1c ; https://www.m-ventures.com/portfolio/synsense"
    },
    {
     "date": "~2020",
     "round": "Series A (amount undisclosed); expansion to China announced",
     "amount": "undisclosed",
     "investors": "CTC Capital (lead); M Ventures, Ecovacs, Yunding; CAS-Star, Archer Investment",
     "source": "https://www.design-reuse.com/news/8244-synsense-formerly-aictx-closes-series-a-round-and-announces-the-opening-of-offices-in-china/ ; https://www.startupticker.ch/en/news/may-2020/aictx-changes-name-to-synsense-and-closes-series-a-round"
    },
    {
     "date": "2023-03",
     "round": "Pre-Series B+",
     "amount": "US$10M",
     "investors": "led by Ausvic Capital",
     "source": "https://www.venturelab.swiss/SynSense-secures-additional-capital-from-strategic-investors ; https://www.greaterzuricharea.com/en/news/synsense-banks-investment-10-million-us-dollars-neuromorphic-computing"
    },
    {
     "date": "2026-07 (reported)",
     "round": "Series B",
     "amount": "\"several hundred million yuan\" (>$40M equivalent if literal)",
     "investors": "undisclosed",
     "source": "https://app.dealroom.co/news/note/china-s-synsense-raises-hundreds-of-millions-of-yuan-in-series-b-for-neuromorphic-chips"
    }
   ],
   "funding_notes": "**Totals conflict:** Tracxn states \"$12.7M over 10 rounds\"\n(https://tracxn.com/d/companies/synsense/__JSKRuzpG-21HQ3_RrriR9LV9UU7znKY0_AEjLrZAN1c), which\ncannot be reconciled with the $10M 2023 round plus multi-hundred-million-CNY Series B; Chinese\nrounds are systematically under-reported in Western trackers. No reliable aggregate total.",
   "trajectory": "Founded Feb–Mar 2017 in Zurich as aiCTX by Dr. Ning Qiao (CEO) and Prof. Giacomo Indiveri,\nspinning out INI (UZH/ETH) neuromorphic hardware research\n(https://www.synsense.ai/about-us-2/, https://alibabatech.medium.com/discover-the-rising-stars-006-interview-with-synsenses-founder-qiao-ning-d8dfd296436b).\nThe 2026-07 who's-who research report (_research/2026-07_whos_who_deep_research.md) names\nKynan Eng (iniVation co-founder) as a third co-founder — not confirmed in the primary\nSeries A/company sources reviewed. The report also lists CTC Capital, CAS Star, and Baidu\nVentures among backers; CTC Capital (lead) and CAS-Star are confirmed for the Series A\n(startupticker.ch above), Baidu Ventures remains report-attributed, unverified.\nRenamed SynSense ~2020 and shifted center of gravity to Chengdu while keeping Zurich R&D.\nAcquired iniVation Feb 2024 (Group CEO: Ning Qiao). As of mid-2026: **active**; fresh Series B\nreported July 2026 for mass production of neuromorphic chips (dealroom note above).",
   "relevance": "SynSense commercializes the academic SNN lineage (Indiveri/INI) in digital silicon —\ndemonstrating that the \"brain-inspired\" market wedge today is event-driven sensing at\nmilliwatts, with learning kept off-chip; adaptive-synapse hardware like Knowm's memristors\nplays no role in their shipping products.",
   "news": [],
   "sources": [
    {
     "title": "SynSense homepage",
     "url": "https://www.synsense.ai/"
    },
    {
     "title": "SynSense about us",
     "url": "https://www.synsense.ai/about-us-2/"
    },
    {
     "title": "SynSense–iniVation merger PR (Feb 2024)",
     "url": "https://www.synsense.ai/synsense-and-inivation-join-forces-to-form-leading-neuromorphic-technology-provider/"
    },
    {
     "title": "Venturelab: pre-B+ $10M led by Ausvic (2023)",
     "url": "https://www.venturelab.swiss/SynSense-secures-additional-capital-from-strategic-investors"
    },
    {
     "title": "Dealroom note: Series B hundreds of millions of yuan (Jul 2026)",
     "url": "https://app.dealroom.co/news/note/china-s-synsense-raises-hundreds-of-millions-of-yuan-in-series-b-for-neuromorphic-chips"
    },
    {
     "title": "Tracxn SynSense profile",
     "url": "https://tracxn.com/d/companies/synsense/__JSKRuzpG-21HQ3_RrriR9LV9UU7znKY0_AEjLrZAN1c"
    },
    {
     "title": "Startupticker: aiCTX→SynSense rename + Series A (CTC Capital lead, CAS-Star)",
     "url": "https://www.startupticker.ch/en/news/may-2020/aictx-changes-name-to-synsense-and-closes-series-a-round"
    }
   ]
  },
  {
   "slug": "syntiant",
   "name": "Syntiant",
   "founded": 2017,
   "hq": "Irvine, CA, USA",
   "category": "analog-imc",
   "status": "active (S-1 filed for Nasdaq IPO 2026-07-06; S-1/A 2026-07-13 and 2026-08-31; not priced as of 2026-09-22)",
   "origin": "corporate alumni (Broadcom; plus Lantronix, academia)",
   "total_funding_usd_m": 311,
   "last_round": "Private round, amount undisclosed, 2024-12 ($646.4M valuation)",
   "key_investors": [
    "Intel Capital",
    "M12 (Microsoft)",
    "Amazon Alexa Fund",
    "Bosch VC",
    "Applied Ventures",
    "Motorola Solutions VC",
    "Renesas",
    "Millennium Technology Value Partners",
    "Mirae Asset",
    "Knowles"
   ],
   "website": "https://www.syntiant.com",
   "pitch": "Syntiant's founding premise was \"moving machine learning processing from the cloud to\nthe edge\": its Neural Decision Processors (NDPs) use \"patented analog neural network\ntechnology in flash memory\" so that neural-network computation happens in/at the\nmemory that stores the weights, cutting the power of always-on inference by orders of\nmagnitude versus MCU/DSP approaches.\nSources: https://www.syntiant.com/news/ai-chip-company-syntiant-raises-25m-in-series-b-funding-round-led-by-microsofts-m12 ;\nhttps://www.syntiant.com/press-release/syntiant-demonstrates-analog-neural-network-closes-a-round-funding-led-by-intel-capital/",
   "technical_approach": "- Device: **flash memory** — early silicon demonstrated an analog neural network\n  computing in flash (2018 A-round PR above). Production **NDP** family (NDP100/101,\n  NDP120, NDP200/250) are ultra-low-power at-memory neural inference chips for keyword\n  spotting, audio and sensor workloads, typically sub-mW for always-on wake words.\n  Inference-only; training is off-chip via Syntiant's ML pipeline. (https://www.syntiant.com)\n- Scale: >20M NDPs shipped by March 2022\n  (https://www.syntiant.com/press-release/ai-chip-company-syntiant-raises-55-million-to-accelerate-growth/).\n- Dec 2024: acquired **Knowles' consumer MEMS microphone/sensor business for $114.4M**,\n  vertically integrating the microphone that sits next to nearly every NDP\n  (https://siliconangle.com/2026/07/06/syntiant-makes-low-power-chips-device-ai-files-initial-public-offering/).",
   "funding_rounds": [
    {
     "date": "2018-03",
     "round": "Series A",
     "amount": "undisclosed",
     "investors": "Intel Capital (lead); Seraph Group, Danhua, Embark",
     "source": "https://www.syntiant.com/press-release/syntiant-demonstrates-analog-neural-network-closes-a-round-funding-led-by-intel-capital/"
    },
    {
     "date": "2018-10",
     "round": "Series B",
     "amount": "$25M",
     "investors": "M12/Microsoft (lead); Amazon Alexa Fund, Applied Ventures, Intel Capital, Motorola Solutions VC, Bosch VC",
     "source": "https://www.syntiant.com/news/ai-chip-company-syntiant-raises-25m-in-series-b-funding-round-led-by-microsofts-m12"
    },
    {
     "date": "2020-08",
     "round": "Series C",
     "amount": "$35M",
     "investors": "incl. Sunstone Venture Capital, Microsoft, Applied Materials",
     "source": "https://www.businesswire.com/news/home/20200810005570/en/Syntiant-Completes-$35M-Series-C-Raise-with-Support-of-Sunstone-Venture-Capital"
    },
    {
     "date": "2022-03",
     "round": "Series D",
     "amount": "$55M",
     "investors": "Renesas, Millennium Technology Value Partners, Mirae Asset Capital; total >$100M at the time",
     "source": "https://www.syntiant.com/press-release/ai-chip-company-syntiant-raises-55-million-to-accelerate-growth/"
    },
    {
     "date": "2024-12",
     "round": "Private round",
     "amount": "undisclosed",
     "investors": "round valued company at $646.4M",
     "source": "https://siliconangle.com/2026/07/06/syntiant-makes-low-power-chips-device-ai-files-initial-public-offering/"
    }
   ],
   "funding_notes": "Totals conflict by source: S-1 coverage says **$311M raised** (SiliconANGLE, above;\nalso https://finance.yahoo.com/technology/ai/articles/intel-backed-ai-chip-software-222745613.html),\nwhile Tracxn counts $121M over 7 equity rounds\n(https://tracxn.com/d/companies/syntiant/__fpH6mCJXQ532IW1oI3j95ZxhAGxwgryghEZYGMmgYYc/funding-and-investors).\nThe gap likely reflects the 2024 financing around the Knowles deal.",
   "trajectory": "Founded 2017 in Irvine, CA by Kurt Busch (CEO, ex-Lantronix), Jeremy Holleman (chief\nscientist, professor), Stephen Bailey (ex-Broadcom Switch chief technologist) and\nPieter Vorenkamp (ex-Broadcom SVP)\n(https://canvasbusinessmodel.com/blogs/brief-history/syntiant-brief-history ; https://www.syntiant.com/bios).\nGrew from wake-word chips to an edge-AI silicon + software + sensors company via the\nKnowles consumer-sensors acquisition (Dec 2024). **Filed S-1 in July 2026** for a\nNasdaq IPO under ticker **SYTN**; ~$270M revenue for the 12 months ended 2026-03-31,\nof which ~96-97% came from the acquired sensors business\n(https://valueaddvc.com/blog/syntiant-ipo-2026-sytn-edge-ai-chip-s1-valuation ;\nhttps://siliconangle.com/2026/07/06/syntiant-makes-low-power-chips-device-ai-files-initial-public-offering/).\n**Status mid-2026: active; IPO pending** — the first of the analog/at-memory edge-AI\nstartups to attempt public markets.",
   "relevance": "Syntiant shows the commercially successful end of the flash-analog spectrum: tens of\nmillions of units shipped, but for a narrow always-on-audio wedge, inference-only, and\nwith most current revenue actually coming from acquired MEMS microphones rather than\nthe neural silicon itself — a useful data point on how \"neuromorphic\" economics play\nout in practice.",
   "news": [
    {
     "date": "2026-09-22",
     "text": "**IPO still pending, not priced.** EDGAR (CIK 0001718728) shows a further **S-1/A filed 2026-08-31** (after the 2026-07-06 S-1 and 2026-07-13 S-1/A) and, as of 2026-09-22, still **no 424B prospectus, no 8-A, no pricing** — the offering has been on file for 11 weeks without pricing. Continue to write \"S-1 filed, pending\", never \"IPO'd\". (https://data.sec.gov/submissions/CIK0001718728.json)"
    },
    {
     "date": "2026-10-27",
     "text": "**IPO still pending, not priced.** EDGAR (CIK 0001718728) shows nothing after the 2026-08-31 S-1/A: no 424B, no 8-A, no RW as of 2026-10-27 — 16 weeks on file. Only news item in the window is a conference appearance. Keep 'S-1 filed, pending'. (https://data.sec.gov/submissions/CIK0001718728.json)"
    },
    {
     "date": "2026-11-09",
     "text": "**IPO still pending, not priced (checked 2026-11-09):** EDGAR shows nothing after the 2026-08-31 S-1/A — no 424B, no 8-A, no RW. Eighteen weeks on file. (https://data.sec.gov/submissions/CIK0001718728.json)"
    }
   ],
   "sources": [
    {
     "title": "Syntiant Series B PR",
     "url": "https://www.syntiant.com/news/ai-chip-company-syntiant-raises-25m-in-series-b-funding-round-led-by-microsofts-m12"
    },
    {
     "title": "Series C PR",
     "url": "https://www.businesswire.com/news/home/20200810005570/en/Syntiant-Completes-$35M-Series-C-Raise-with-Support-of-Sunstone-Venture-Capital"
    },
    {
     "title": "$55M raise PR",
     "url": "https://www.syntiant.com/press-release/ai-chip-company-syntiant-raises-55-million-to-accelerate-growth/"
    },
    {
     "title": "SiliconANGLE IPO filing 2026-07-06",
     "url": "https://siliconangle.com/2026/07/06/syntiant-makes-low-power-chips-device-ai-files-initial-public-offering/"
    },
    {
     "title": "Value Add VC S-1 breakdown",
     "url": "https://valueaddvc.com/blog/syntiant-ipo-2026-sytn-edge-ai-chip-s1-valuation"
    }
   ]
  },
  {
   "slug": "tetramem",
   "name": "TetraMem Inc.",
   "founded": 2018,
   "hq": "Fremont/Newark area, California, USA (Silicon Valley; Tracxn lists Newark, CA)",
   "category": "emerging-device-imc",
   "status": "active",
   "origin": "corporate alumni (HP Labs memristor team) + academic (USC / UMass Amherst)",
   "total_funding_usd_m": 0,
   "last_round": "undisclosed strategic round(s); latest logged deal 2025-04; new round expected 2026 at >2x the reported $450M valuation",
   "key_investors": [
    "SK Square / SK hynix (strategic)",
    "FutureX Capital",
    "NYX Ventures",
    "Brown Venture Associates",
    "Foothill Ventures",
    "Galaxy Interactive"
   ],
   "website": "https://tetramem.com",
   "pitch": "\"We realized that we could eliminate the memory wall by using RRAM crossbar arrays for\nanalog in-memory computing, enabling real-time, low-power AI inference\"\n(https://tetramem.com/about-tetramem/). The SK hynix joint release frames analog IMC as\n\"performing matrix operations directly where the model weights reside, dramatically\nreducing data movement\" to attack AI's energy and thermal limits\n(https://www.businesswire.com/news/home/20260708244132/en/TetraMem-and-SK-hynix-Showcase-Successful-Technology-Collaboration-Advancing-Memory-Centric-AI-Computing).",
   "technical_approach": "- Device: proprietary **multi-level RRAM (memristor)** cells in crossbar arrays; the\n  founding team's academic work demonstrated thousands of analog conductance levels\n  (Nature 2023, Xia/Yang groups — pointer only; this corpus is web-sourced).\n- Compute: **analog in-memory computing (A-IMC)** — analog MVM inside the array; digital\n  periphery/SoC integration. Inference-only positioning (edge AI first, then enterprise /\n  data center); no on-chip learning claims in product material.\n- Products: MX100 evaluation chip (first multi-level-RRAM analog IMC accelerator,\n  announced ~2023); **MLX200** analog-IMC SoC with SDK (launched 2025; production samples\n  announced 2025) (https://tetramem.com/about-tetramem/). May 2026: announced MLX200-class\n  multi-level RRAM analog-IMC SoC milestone fabbed on a **commercial TSMC 22nm process**,\n  eval kits targeted 2H 2026\n  (https://www.businesswire.com/news/home/20260516556464/en/TetraMem-Announces-22nm-Multi-Level-RRAM-Analog-In-Memory-Computing-SoC-Milestone).\n- SK hynix: research partnership announced Nov 2024; July 2026 the two showcased completed\n  joint work — a memristor-based IMC SoC with efficient depthwise convolution, cover\n  feature in *Advanced Intelligent Systems*\n  (https://www.businesswire.com/news/home/20260708244132/en/).",
   "funding_rounds": [
    {
     "date": "various (2018–2025)",
     "round": "undisclosed rounds",
     "amount": "undisclosed",
     "investors": "FutureX Capital, NYX Ventures, Brown Venture Associates, Foothill Ventures, Galaxy Interactive",
     "source": "https://tracxn.com/d/companies/tetramem/__f1S-JiTZUVwgUX5OQr_mZC69JGPhDd7HhIXZQZ7IemY ; https://www.cbinsights.com/company/tetramem-hk"
    },
    {
     "date": "undisclosed",
     "round": "strategic",
     "amount": "undisclosed",
     "investors": "SK Square (SK hynix group)",
     "source": "https://www.digitaltoday.co.kr/en/view/4334/sk-square-ai-investment-results-become-visible-company-value-up-to-sevenfold"
    },
    {
     "date": "2025-04-29",
     "round": "latest logged deal",
     "amount": "undisclosed",
     "investors": "—",
     "source": "https://pitchbook.com/profiles/company/435431-08 (via search snippet)"
    }
   ],
   "funding_notes": "No round amounts are publicly disclosed — the funding league table should list TetraMem as\n\"undisclosed.\" Korean press (Feb 2026) reports a **$450M valuation** and a new round\nexpected in 2026 at more than double that\n(https://www.digitaltoday.co.kr/en/view/4334/sk-square-ai-investment-results-become-visible-company-value-up-to-sevenfold).",
   "trajectory": "Founded 2018 by four co-founders: Glenn Ge (CEO), Miao Hu (CTO), J. Joshua Yang and\nQiangfei Xia (co-founders, now scientific/process advisors)\n(https://tetramem.com/about-tetramem/). Yang, Xia, and Hu are HP Labs memristor-program\nalumni (the Stan Williams-era team); Yang and Xia went on to professorships (USC and UMass\nAmherst) and remain the leading academic memristor-crossbar groups. Trajectory: research\nchips → MX100 eval chip (~2023) → MLX200 SoC + SDK (2025) → TSMC 22nm production-process\nmilestone (May 2026) with eval kits due 2H 2026 → SK hynix joint publication (July 2026).\nStatus mid-2026: active, arguably the leading pure-play analog memristor-IMC company, with\na foundry-process proof point but no disclosed volume product revenue.",
   "relevance": "The closest thing in the wave to a direct commercial descendant of the HP Labs memristor\nprogram: multi-level RRAM crossbars for analog MVM, inference-only — contrast with Knowm's\ndiscrete SDC devices and on-chip-learning (AHaH) focus. TetraMem's 22nm foundry milestone\nis the strongest current evidence that analog memristor IMC can reach commercial process\nnodes; its funding remains undisclosed, so money comparisons are not possible.",
   "news": [
    {
     "date": "2026-10-27",
     "text": "**MLX300 taped out — claimed as the first 12 nm analog-IMC SoC on multi-level RRAM (PR 2026-10-27).** Second full-mask tape-out of 2026 after the 22 nm MLX200/MLX200M; designed on Synopsys Cloud FlexEDA across US and Singapore teams. Target: single-chip accelerator for audio, vision and small language models, **up to 64M INT4 parameters on-chip with no external DRAM for weights**, targeting **>50–80 TOPS/W** (company figures, pre-silicon). First A-IMC move to a FinFET node in the cohort. (https://finance.yahoo.com/technology/ai/articles/tetramem-advances-analog-memory-computing-084100770.html)"
    }
   ],
   "sources": [
    {
     "title": "About TetraMem",
     "url": "https://tetramem.com/about-tetramem/"
    },
    {
     "title": "Businesswire — 22nm multi-level RRAM analog IMC SoC milestone (2026-05-16)",
     "url": "https://www.businesswire.com/news/home/20260516556464/en/TetraMem-Announces-22nm-Multi-Level-RRAM-Analog-In-Memory-Computing-SoC-Milestone"
    },
    {
     "title": "Businesswire — TetraMem and SK hynix collaboration (2026-07-08)",
     "url": "https://www.businesswire.com/news/home/20260708244132/en/TetraMem-and-SK-hynix-Showcase-Successful-Technology-Collaboration-Advancing-Memory-Centric-AI-Computing"
    },
    {
     "title": "DigitalToday (KR) — SK Square AI investments; TetraMem $450M valuation",
     "url": "https://www.digitaltoday.co.kr/en/view/4334/sk-square-ai-investment-results-become-visible-company-value-up-to-sevenfold"
    },
    {
     "title": "Tracxn — TetraMem profile",
     "url": "https://tracxn.com/d/companies/tetramem/__f1S-JiTZUVwgUX5OQr_mZC69JGPhDd7HhIXZQZ7IemY"
    },
    {
     "title": "CB Insights — TetraMem",
     "url": "https://www.cbinsights.com/company/tetramem-hk"
    }
   ]
  },
  {
   "slug": "unconventional-ai",
   "name": "Unconventional AI (Unconventional, Inc.)",
   "founded": 2025,
   "hq": "San Diego, CA, USA",
   "category": "probabilistic-thermodynamic",
   "status": "active",
   "origin": "serial founders (Naveen Rao — Nervana, MosaicML/Databricks; + MIT/Stanford faculty)",
   "total_funding_usd_m": 475,
   "last_round": "Seed, $475M at $4.5B valuation, 2025-12",
   "key_investors": [
    "Andreessen Horowitz",
    "Lightspeed Venture Partners",
    "Lux Capital",
    "DCVC",
    "Databricks",
    "Jeff Bezos"
   ],
   "website": "https://unconventional.ai",
   "pitch": "Naveen Rao's stated goal is to build a computer \"as efficient as biology.\" The a16z\ninvestment thesis spells out the mismatch: GPUs excel at deterministic arithmetic over\nlarge numerical arrays, but AI models are fundamentally probabilistic — so Unconventional\nwill \"store exact probability distributions in the underlying physical substrate\" instead\nof representing them as floating-point arrays, with a theoretical target of \"O(1,000x)\nlower power consumption than digital computers.\"\nSources: https://a16z.com/announcement/investing-in-unconventional/,\nhttps://techcrunch.com/2025/12/09/unconventional-ai-confirms-its-massive-475m-seed-round",
   "technical_approach": "- **Substrate:** analog and mixed-signal chips designed for probabilistic AI workloads;\n  the company says it is exploring **oscillators, thermodynamics, and spiking neurons** as\n  candidate physical foundations (https://a16z.com/announcement/investing-in-unconventional/).\n  DCD headlines it as a \"neuromorphic compute startup\"; Rao has also publicly framed it as\n  brain-inspired/analog computing (https://www.datacenterdynamics.com/en/news/neuromorphic-compute-startup-unconventional-ai-raises-475m-in-seed-funding/).\n- **Stage:** pre-product research lab as of mid-2026 — no announced chip, prototype, or\n  benchmark. MeeLan Lee leads mixed-signal chip design engineering.\n- **Training vs inference:** not yet specified publicly.\n- **Team pedigree:** cofounders Mike Carbin (MIT, programming systems/novel computation)\n  and Sara Achour (Stanford, analog/unconventional compilation) alongside Rao. The 2026-07\n  who's-who research report (_research/2026-07_whos_who_deep_research.md) also lists\n  MeeLan Lee (mixed-signal chip lead, above) as a fourth co-founder — unverified.",
   "funding_rounds": [
    {
     "date": "2025-12-08/09",
     "round": "Seed (first installment toward a targeted ~$1B round)",
     "amount": "$475M at $4.5B valuation",
     "investors": "Co-led by Andreessen Horowitz and Lightspeed; Lux Capital, DCVC, Databricks, Jeff Bezos; Rao personally invested $10M",
     "source": "https://finance.yahoo.com/news/unconventional-ai-confirms-massive-475m-002416708.html ; https://www.bloomberg.com/news/articles/2025-12-08/ai-computer-startup-hits-4-5-billion-valuation-in-seed-round"
    }
   ],
   "funding_notes": "Total disclosed: **$475M**. Notes: TechCrunch first reported the raise effort in Oct 2025\nat a targeted $5B valuation; it closed at $4.5B. Rao told Bloomberg the $475M is a first\ninstallment toward up to $1B for the round — watch for a second close in 2026. The 2026-07\nwho's-who research report (_research/2026-07_whos_who_deep_research.md) independently\ncorroborates the round details: announced 2025-12-08, a16z/Lightspeed co-leads with Lux\nCapital, DCVC, Databricks, and Jeff Bezos participating, plus Rao's $10M personal stake.",
   "trajectory": "Founded ~October 2025 in San Diego by Naveen Rao immediately after leaving his post as\nVP of AI at Databricks. Rao's track record: co-founded Nervana Systems (acquired by Intel,\n2016, reported $400M+) and MosaicML (acquired by Databricks, 2023, $1.3B). The company\nclosed its seed roughly two months after founding — reported as the fastest company ever\nto reach unicorn status (https://mugglehead.com/unconventional-ai-becomes-historys-fastest-unicorn-with-us4-5b-valuation-in-just-2-months/).\nAs of mid-2026: hiring, no announced silicon. This is by far the largest single funding\nevent in the neuromorphic/physics-based-computing category.",
   "relevance": "The $475M seed is the strongest market signal to date that \"efficient as biology,\"\nprobabilistic, analog computing — the territory Knowm's AHaH/thermodynamic-computing work\nstaked out in 2014-2015 — has become a venture-scale thesis. Unconventional has published\nno substrate choice yet (oscillators/thermodynamics/spiking all on the table), so direct\ntechnical comparison to memristor-based adaptive hardware is premature.",
   "news": [
    {
     "date": "2026-09-22",
     "text": "**First substrate disclosure (secondary source):** on the All-In podcast (coverage 2026-09-21) Naveen Rao said Unconventional AI has **taped out a physical \"dynamical computer\" built from coupled oscillators**, generating images at ~500 nW each, targeting a 1,000x power-efficiency gain over GPUs and a rack-scale datacenter product within ~2 years. This resolves the profile's \"oscillators / thermodynamics / spiking — no substrate chosen\" note toward the oscillator path. Single podcast-derived write-up, no company PR or technical paper yet — quote as Rao's statement, not a verified benchmark. (https://finance.biggo.com/news/9911762380f2e23b)"
    }
   ],
   "sources": [
    {
     "title": "a16z: Investing in Unconventional",
     "url": "https://a16z.com/announcement/investing-in-unconventional/"
    },
    {
     "title": "TechCrunch: Unconventional AI confirms $475M seed",
     "url": "https://techcrunch.com/2025/12/09/unconventional-ai-confirms-its-massive-475m-seed-round"
    },
    {
     "title": "Bloomberg: $4.5B valuation in seed round",
     "url": "https://www.bloomberg.com/news/articles/2025-12-08/ai-computer-startup-hits-4-5-billion-valuation-in-seed-round"
    },
    {
     "title": "DCD: Neuromorphic compute startup raises $475M",
     "url": "https://www.datacenterdynamics.com/en/news/neuromorphic-compute-startup-unconventional-ai-raises-475m-in-seed-funding/"
    },
    {
     "title": "Mugglehead: fastest unicorn",
     "url": "https://mugglehead.com/unconventional-ai-becomes-historys-fastest-unicorn-with-us4-5b-valuation-in-just-2-months/"
    }
   ]
  },
  {
   "slug": "vaire-computing",
   "name": "Vaire Computing",
   "founded": 2021,
   "hq": "London, UK + Seattle, WA, USA",
   "category": "probabilistic-thermodynamic",
   "status": "active",
   "origin": "serial founders (Rodolfo Rosini) + academic (Hannah Earley, Univ. of Cambridge)",
   "total_funding_usd_m": 10,
   "last_round": "Seed (cumulative ~$10M reported), 2025-05",
   "key_investors": [
    "7percent Ventures",
    "Seedcamp",
    "Lifeline Ventures",
    "Jude Gomila"
   ],
   "website": "https://vaire.co",
   "pitch": "Vaire attacks the heat-dissipation floor of irreversible logic: conventional chips destroy\ninformation at every gate and pay the Landauer energy cost, while reversible computing\n\"performs calculations while generating a relatively negligible amount of heat,\" recycling\ncharge instead of dumping it — aimed at \"near zero-energy chips.\"\nSources: https://techcrunch.com/2024/07/01/vaire-computing-raises-4-5m-for-reversible-computing-moonshot-which-could-drastically-reduce-energy-needs/,\nhttps://seedcamp.com/views/vaire-computing-secures-4-5m-to-create-near-zero-energy-silicon-chips-using-reversible-computing/",
   "technical_approach": "- **Substrate:** standard CMOS (22nm for the first test chip) running adiabatic/reversible\n  logic with on-chip resonators that recover and re-use charge each cycle — no exotic\n  devices required.\n- **What the physics does:** energy stored in circuit capacitances is resonantly recycled\n  rather than dissipated; logic is (near-)reversible so information — and thus energy —\n  is not destroyed at each step.\n- **Ice River test chip:** announced 2025-03-09; fabricated in 22nm CMOS; demonstrated net\n  energy recovery — energy-recovery factor 1.77 for a capacitor array and 1.41 for an adder\n  (>1.0 = net recovery). Vaire calls it the first foundry-fabbed full-stack demonstration\n  of reversible computing. Sources: https://www.eetimes.com/vaire-demos-energy-recovery-with-reversible-computing-test-chip/,\n  https://www.datacenterdynamics.com/en/news/vaire-computing-tapes-out-demo-chip-capable-of-recycling-50-of-its-energy-intensity/\n- **Claims:** resonator component showed ~50% energy savings in initial tests; press\n  summaries cite ~30% AI energy reduction for Ice River-class designs — early, not yet\n  independently verified (https://fortune.com/2025/05/20/uk-startup-vaire-reversible-computing-chip-gpu-alternative-energy-savings-ai/).\n- **Training vs inference:** general-purpose logic/arithmetic direction (AI inference and\n  always-on edge first); no learning-in-physics claim.\n- **Team:** Mike Frank (ex-MIT/Sandia, the field's best-known reversible-computing\n  researcher) leads chip design; Andrew Sloss (ex-Arm) also aboard. ~20 staff across\n  London, Cambridge, Sunnyvale, Portland (Fortune, May 2025).",
   "funding_rounds": [
    {
     "date": "~2022-23",
     "round": "Pre-seed",
     "amount": "$0.5M",
     "investors": "7percent Ventures + angels",
     "source": "https://fortune.com/2025/05/20/uk-startup-vaire-reversible-computing-chip-gpu-alternative-energy-savings-ai/"
    },
    {
     "date": "2024-07",
     "round": "Seed",
     "amount": "$4.0M (total then $4.5M)",
     "investors": "7percent Ventures lead; Jude Gomila; Seedcamp",
     "source": "https://techcrunch.com/2024/07/01/vaire-computing-raises-4-5m-for-reversible-computing-moonshot-which-could-drastically-reduce-energy-needs/"
    },
    {
     "date": "by 2025-05",
     "round": "Seed extension(s)",
     "amount": "to ~$10M cumulative",
     "investors": "7percent Ventures, Lifeline Ventures, Seedcamp, Jude Gomila",
     "source": "https://fortune.com/2025/05/20/uk-startup-vaire-reversible-computing-chip-gpu-alternative-energy-savings-ai/"
    }
   ],
   "funding_notes": "Total reported: **~$10M** as of May 2025. Conflict note: DCD/Fierce reported the 2024\nround as \"$4M\"; TechCrunch/Seedcamp as \"$4.5M total\" — the $0.5M pre-seed accounts for\nthe difference.",
   "trajectory": "Founded 2021 by Rodolfo Rosini (CEO, serial entrepreneur) and Hannah Earley (CTO,\nCambridge unconventional-computing researcher). Taped out and validated Ice River\n(announced March 2025); as of May 2025 test units were shipping to prospective customers\nand academic labs. Roadmap: further tape-outs improving performance/energy re-use, with\nfull-scale production targeted ~2027. Active as of mid-2026; no 2026 round disclosed in\nsources checked.\n\n**Founders (sweep 2026-09-22):** Rodolfo Rosini (CEO; serial founder, \"raised 5x, exited 2x\") and Dr Hannah Earley (CTO; PhD Applied Mathematics & Theoretical Physics, University of Cambridge — came to reversible computing in 2016 during a PhD on molecular programming, via a Florida-group thesis her supervisor forwarded). No lab spinout or technology licence documented; senior scientist Mike Frank (ex-Sandia) joined later (https://vaire.co/about ; https://techcrunch.com/2024/07/01/vaire-computing-raises-4-5m-for-reversible-computing-moonshot-which-could-drastically-reduce-energy-needs/).",
   "relevance": "Vaire attacks the same thermodynamic cost-of-computation problem Knowm framed with\nkT-RAM, but from the opposite direction: eliminate dissipation via reversibility rather\nthan exploit fluctuations for adaptive/stochastic computation. Purely deterministic, no\non-chip learning — a useful contrast column in the physics-based taxonomy.",
   "news": [],
   "sources": [
    {
     "title": "TechCrunch: $4.5M seed",
     "url": "https://techcrunch.com/2024/07/01/vaire-computing-raises-4-5m-for-reversible-computing-moonshot-which-could-drastically-reduce-energy-needs/"
    },
    {
     "title": "EE Times: Ice River energy recovery",
     "url": "https://www.eetimes.com/vaire-demos-energy-recovery-with-reversible-computing-test-chip/"
    },
    {
     "title": "DCD: demo chip recycling 50%",
     "url": "https://www.datacenterdynamics.com/en/news/vaire-computing-tapes-out-demo-chip-capable-of-recycling-50-of-its-energy-intensity/"
    },
    {
     "title": "Fortune: Vaire profile, $10M total (2025-05-20)",
     "url": "https://fortune.com/2025/05/20/uk-startup-vaire-reversible-computing-chip-gpu-alternative-energy-savings-ai/"
    },
    {
     "title": "Seedcamp announcement",
     "url": "https://seedcamp.com/views/vaire-computing-secures-4-5m-to-create-near-zero-energy-silicon-chips-using-reversible-computing/"
    }
   ]
  },
  {
   "slug": "vivum-computing",
   "name": "Vivum AI (Vivum Computing)",
   "founded": 2023,
   "hq": "South San Francisco, CA + Bloomington, IN, USA",
   "category": "digital-snn",
   "status": "active",
   "origin": "serial founders (Aldo Carrascoso) + academic (Derek Whitley, complex systems / cognitive science)",
   "total_funding_usd_m": 0,
   "last_round": "undisclosed (government R&D funding reported; SBIR listing exists)",
   "key_investors": [],
   "website": "https://vivum.ai",
   "pitch": "Vivum pitches \"Evolutionary AI\" (E-AI): instead of ever-larger backprop-trained networks\non power-hungry hardware, evolve compact *dynamic neural models* that \"harness the power\nof biological intelligence,\" claiming neural-network power reductions of \"up to 6000\ntimes\" for edge autonomy — with explainability for high-stakes missions as a co-equal\nselling point. Sources: https://vivum.ai/about, https://vivum.ai/",
   "technical_approach": "- **Substrate:** digital — evolved dynamic/spiking neural models (described in third-party\n  profiles as liquid time-constant-style continuous-time networks) deployed on FPGAs and\n  embedded processors on drones and other uncrewed systems\n  (https://www.neuromorphiccore.ai/insights/vivum-ai/).\n- **Learning locus:** models are produced by an offline evolutionary-search platform\n  (\"accelerates the evolution of highly efficient Dynamic Neural Models\"); deployed\n  models are adaptive/dynamic at runtime, but this is algorithmic adaptation on digital\n  hardware, not learning in device physics.\n- **Products:** E-AI platform + deployed autonomy stacks; company advertises real-world\n  deployments on drone hardware and FPGA systems (2025-2026 per NeuromorphicCore profile).\n  An active Senior FPGA Engineer posting (Bloomington, IN) corroborates FPGA focus.\n- **Claims caveat:** the \"6000x\" energy figure is a company claim without a published\n  benchmark methodology in the sources reviewed.",
   "funding_rounds": [
    {
     "date": "—",
     "round": "none disclosed",
     "amount": "—",
     "investors": "—",
     "source": "https://www.crunchbase.com/organization/vivum (no disclosed rounds surfaced in search)"
    }
   ],
   "funding_notes": "No venture rounds disclosed. Non-dilutive signals: an SBIR award listing exists\n(https://www.sbir.gov/awards/214636); CTO Whitley's underlying research was funded by\nNASA, the U.S. Air Force, and the Office of Naval Research (https://vivum.ai/about).\nTotal raised: **undisclosed**.",
   "trajectory": "Founded 2023 by Aldo Carrascoso (CEO; previously InterVenn Biosciences, Veem, Jukin\nMedia) and Derek Whitley, Ph.D. (CTO; Navy veteran, dual Ph.D. in Complex Systems and\nCognitive Science, Indiana University orbit). 11-50 employees. Positioning leans heavily\non defense/aerospace (team pedigree cites NASA, SpaceX, Lockheed Martin, U.S. Navy/Marine\nCorps) and \"Ethics First\" explainable autonomy. Active as of mid-2026, in a\ncommercial/scaling phase per third-party profiling; funding transparency is low.\n\n**Founders (sweep 2026-09-22):** Aldo Carrascoso (founder & CEO; MBA; previously InterVenn Biosciences, Veem, Jukin Media) and Derek Whitley (co-founder, CTO/CSO; dual PhD Complex Systems and Cognitive Science, Indiana University; evolvable-hardware researcher). Company site now names only Carrascoso in leadership; no lab or corporate parent (https://vivum.ai/about ; https://insideunmannedsystems.com/unlocking-true-autonomy-vivum-ais-dynamic-neural-networks-for-unmanned-systems/ ; https://bootstraplabs.com/community/derek-whitley/).",
   "relevance": "Vivum represents the \"neuromorphic as algorithm, not device\" corner of the taxonomy:\nbiologically-inspired dynamics realized on conventional FPGAs. Contrast with Knowm's\ndevice-physics approach — Vivum's efficiency comes from small evolved models, not from a\nnon-von-Neumann substrate, so it tests how much of the neuromorphic promise is capturable\nwithout new hardware.",
   "news": [],
   "sources": [
    {
     "title": "Vivum AI — About",
     "url": "https://vivum.ai/about"
    },
    {
     "title": "Vivum AI — home",
     "url": "https://vivum.ai/"
    },
    {
     "title": "NeuromorphicCore.ai profile",
     "url": "https://www.neuromorphiccore.ai/insights/vivum-ai/"
    },
    {
     "title": "SBIR award listing",
     "url": "https://www.sbir.gov/awards/214636"
    },
    {
     "title": "Crunchbase profile",
     "url": "https://www.crunchbase.com/organization/vivum"
    }
   ]
  },
  {
   "slug": "weebit-nano",
   "name": "Weebit Nano Ltd",
   "founded": 2015,
   "hq": "Hod Hasharon, Israel (ASX-listed, Australia)",
   "category": "emerging-device-imc (ReRAM memory IP; neuromorphic is R&D-only)",
   "status": "public (ASX:WBT)",
   "origin": "academic spinoff (Rice University — Prof. James Tour's SiOx ReRAM patents, licensed 2014/2015)",
   "total_funding_usd_m": 150,
   "last_round": "ASX placement + SPP, ~A$102M total (A$80M institutional placement + oversubscribed SPP), 2026-05/06",
   "key_investors": [
    "ASX public market investors",
    "Krongold family (early)"
   ],
   "website": "https://www.weebit-nano.com",
   "pitch": "Weebit sells ReRAM as the next non-volatile memory, not a compute paradigm — but on\nneuromorphics its CEO Coby Hanoch has said: \"using Neuromorphic techniques based on ReRAM\ntechnology has the potential to emulate the way the brain works, and making the computing\nprocess significantly more efficient\"\n(https://www.weebit-nano.com/news/press-releases/weebit-nano-and-cea-leti-to-demonstrate-brain-inspired-neuromorphic-demo/, 2019-07-19).",
   "technical_approach": "- Device: **SiOx (oxygen-vacancy / OxRAM) ReRAM**, invented in Prof. James Tour's lab at\n  Rice University; Weebit engaged Tour in 2014 and licensed the patents\n  (https://www.weebit-nano.com/weebit-nano-turns-10only-the-persistent-survive/;\n  https://en.wikipedia.org/wiki/Weebit_Nano). Long-running joint development with CEA-Leti\n  since 2016 (https://www.eenewseurope.com/en/leti-to-develop-siox-memory-with-weebit-nano/).\n- Business model: **embedded NVM IP licensing** (digital memory replacing embedded flash at\n  advanced nodes) — not an AI-compute company. Inference/learning: N/A for products.\n- Neuromorphic angle (R&D only): July 2019 joint demo with CEA-Leti at Flash Memory Summit\n  2019 — SiOx ReRAM as artificial synapses running CEA-Leti spiking-neural-network\n  inference for object recognition\n  (https://www.weebit-nano.com/news/press-releases/weebit-nano-and-cea-leti-to-demonstrate-brain-inspired-neuromorphic-demo/).\n- Product milestones: SkyWater volume-production agreement Sep 2021; first production\n  wafers Nov 2022; first commercially available ReRAM IP Mar 2023; DB HiTek license\n  Oct 2023 (130nm BCD); Efabless collaboration May 2024; onsemi Treo platform integration\n  Jan 2025; **Texas Instruments license Dec 2025**; selected for a Korean government AI\n  computing initiative Mar 2026 (https://en.wikipedia.org/wiki/Weebit_Nano).",
   "funding_rounds": [
    {
     "date": "2016-08",
     "round": "ASX listing (reverse takeover of Radar Iron)",
     "amount": "A$5.04M",
     "investors": "ticker WBT",
     "source": "https://www.weebit-nano.com/weebit-nano-turns-10only-the-persistent-survive/ ; https://announcements.asx.com.au/asxpdf/20151111/pdf/432wpj7r6ns4ts.pdf"
    },
    {
     "date": "2023-04",
     "round": "Placement + SPP",
     "amount": "~A$60M (~US$40M)",
     "investors": "A$45M underwritten placement + oversubscribed SPP",
     "source": "https://www.globenewswire.com/en/news-release/2023/04/24/2652816/0/en/Weebit-Nano-successfully-secures-approximately-US-40-million-to-accelerate-development-and-commercial-roll-out-of-its-ReRAM.html"
    },
    {
     "date": "2024-12",
     "round": "Underwritten placement",
     "amount": "A$50M",
     "investors": "institutional",
     "source": "https://www.listcorp.com/asx/wbt/weebit-nano-limited/news/weebit-nano-completes-a-50-million-underwritten-placement-3115557.html ; https://www.yolegroup.com/industry-news/weebit-nano-raises-50-million-to-advance-its-reram-memory-tech/"
    },
    {
     "date": "2026-05/06",
     "round": "Placement + SPP",
     "amount": "~A$102M total",
     "investors": "A$80M institutional placement + A$15M+ SPP; ~A$170M+ cash post-raise",
     "source": "https://www.weebit-nano.com/news/press-releases/successful-completion-of-a80-million-institutional-placement/ ; https://finance.yahoo.com/markets/stocks/articles/weebit-nano-valuation-102-million-131418779.html"
    }
   ],
   "funding_notes": "Public company — capital raised via ASX equity, not VC rounds. Major disclosed raises:\nCaveat: Weebit also made multiple smaller placements/SPPs 2017–2022 not enumerated here;\nthe frontmatter total (~US$150M equivalent) is an approximation of the major raises above\n(A$220M+ cumulative), not an audited total. FY2025 revenue A$4.41M, +333% YoY\n(https://stockanalysis.com/quote/asx/WBT/).",
   "trajectory": "Incorporated in Israel 2015 around the Rice/Tour SiOx patents; listed on ASX Aug 2016 via\nreverse takeover arranged with early investors the Krongold family; first CEO Yossi Keret,\nreplaced by Coby Hanoch in 2017; chairman is former Intel EVP David \"Dadi\" Perlmutter\n(https://www.weebit-nano.com/weebit-nano-turns-10only-the-persistent-survive/). Trajectory:\na decade from lab license to commercial embedded-ReRAM IP with tier-1 licensees (TI,\nonsemi, DB HiTek). Status mid-2026: active, public, well-capitalized; neuromorphic remains\na demo/roadmap item rather than a product line.",
   "relevance": "Shows the \"memory-first\" commercialization path for emerging devices: license ReRAM as\nembedded NVM to incumbents and treat synaptic/neuromorphic use as future upside — the\nopposite sequencing from Knowm's compute/learning-first (AHaH) approach, and currently the\nbest-funded pure ReRAM play in the cohort.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Weebit PR (2026-07-31): **expanded licensing agreements with key customers; three customer chip designs taped out to date** — first product-customer silicon milestone beyond demo chips. FY26 results due late Aug 2026 (guidance ≥A$13.5M revenue). (https://www.weebit-nano.com/news/press-releases/weebit-nano-expands-licensing-agreements-with-key-customers-three-customer-chip-designs-taped-out-to-date/ ; https://kalkine.com.au/news/technology/weebit-nano-slips-today-despite-record-cash-pile-and-texas-instruments-tape-outs-is-this-just-a-breather-after-a-monster-run)"
    },
    {
     "date": "2026-09-22",
     "text": "**FY26 results (2026-08-28):** revenue **A$15.3M, up 246%** from A$4.4M (above the ≥A$13.5M guidance); statutory loss A$54.9M (FY25: A$38.4M); **cash A$168.3M** at 30 June 2026 after raising A$102M in the year. Texas Instruments licensed as the largest customer, DB HiTek qualification achieved, working chips with onsemi, a US subsidiary established, and a role in Korea's national In-Memory Compute program. Guidance: ≥A$7.1M revenue in 1H FY27, \"multiple\" new license agreements expected in FY27. (https://www.fool.com.au/2026/08/28/weebit-nano-fy26-record-revenue-new-customer-wins/ ; https://stocksdownunder.com/weebit-nano-asxwbt-fy26-result/)"
    },
    {
     "date": "2026-09-22",
     "text": "**Added to the S&P/ASX 200 index** in the September 2026 rebalance (effective 2026-09-21) — index inclusion for a pre-mass-production ReRAM IP company. (https://www.marketscreener.com/news/weebit-nano-limited-asx-wbt-added-to-s-p-asx-200-index-ce785adbd981f727 ; https://kalkine.com.au/news/technology/why-is-weebit-nano-asxwbt-stock-rising-today)"
    },
    {
     "date": "2026-10-27",
     "text": "**Q1 FY27 update (2026-10-16/19):** cash **A$164.2M** (from A$168.3M), receipts A$6.1M, net cash used A$3.2M; FY26 burn <A$20M. **DB HiTek completed full ReRAM qualification on 130 nm BCD** and opened customer designs; onsemi has functional test chips and is in qualification; TI transfer on track. Three product-customer tape-outs done in 2026 vs a target of one; customers 1–2 have samples and target 2027 production, customer 3 samples before end-2026; Overlord Labs' Genesis chip (DB HiTek) first silicon July 2026. Two new AI/IMC exploration customers; management says a fully commercial IMC product is **≥2 years out**. Royalties expected from calendar 2027. Shares fell ~8% on the day and are ~70% below the 52-week high. AGM 2026-11-23. (https://www.investing.com/news/company-news/weebit-nano-q1-fy27-slides-reram-qualified-targets-hit-stock-falls-93CH-4956263 ; https://kalkine.com.au/news/announcements/weebit-nano-reports-second-customer-begins-silicon-validation-1642-million-cash-balance-in-q1-fy27)"
    },
    {
     "date": "2026-11-09",
     "text": "**Weebit Matrix IMC research program launched (2026-11-04):** gives academic and industry researchers direct access to ReRAM crossbar hardware (evaluation board + package) for in-memory compute experiments; the company's projection for one inference workload is ~80x lower power and ~100x lower latency than a GPU (own figures, pre-product; management said in October a commercial IMC product is ≥2 years out). Demo at electronica 2026 (Munich, 2026-11-10/13). Keep the 'memory IP, not compute' framing — this is a research platform. (https://embeddedcomputing.com/technology/ai-machine-learning/ai-logic-devices-worload-acceleration/weebit-nano-launches-weebit-matrix-in-memory-compute-research-program ; https://www.weebit-nano.com/products/embedded-reram-ip/matrix-imc-in-memory-compute-reram-rram/)"
    },
    {
     "date": "2026-11-10",
     "text": "**Weebit Matrix shown prototyping IMC algorithms at Electronica (2026-11-10)** (Electronics Weekly); separately, a **second customer is testing ReRAM silicon** as qualification progresses (Kalkine, 2026-11-10, headline-level). (https://www.electronicsweekly.com/news/business/weebit-matrix-2026-11/ ; https://kalkine.com.au/news/small-cap-stocks/weebit-nano-asxwbt-progresses-reram-qualification-as-second-customer-tests-silicon)"
    }
   ],
   "sources": [
    {
     "title": "Weebit Nano Turns 10 — company history",
     "url": "https://www.weebit-nano.com/weebit-nano-turns-10only-the-persistent-survive/"
    },
    {
     "title": "Weebit/CEA-Leti neuromorphic demo PR (2019-07-19)",
     "url": "https://www.weebit-nano.com/news/press-releases/weebit-nano-and-cea-leti-to-demonstrate-brain-inspired-neuromorphic-demo/"
    },
    {
     "title": "Wikipedia — Weebit Nano (milestones, licensees)",
     "url": "https://en.wikipedia.org/wiki/Weebit_Nano"
    },
    {
     "title": "GlobeNewswire — US$40M raise (2023-04)",
     "url": "https://www.globenewswire.com/en/news-release/2023/04/24/2652816/0/en/Weebit-Nano-successfully-secures-approximately-US-40-million-to-accelerate-development-and-commercial-roll-out-of-its-ReRAM.html"
    },
    {
     "title": "Listcorp — A$50M placement (2024-12)",
     "url": "https://www.listcorp.com/asx/wbt/weebit-nano-limited/news/weebit-nano-completes-a-50-million-underwritten-placement-3115557.html"
    },
    {
     "title": "Weebit PR — A$80M placement completion (2026)",
     "url": "https://www.weebit-nano.com/news/press-releases/successful-completion-of-a80-million-institutional-placement/"
    },
    {
     "title": "Yahoo Finance — valuation after A$102M raise",
     "url": "https://finance.yahoo.com/markets/stocks/articles/weebit-nano-valuation-102-million-131418779.html"
    }
   ]
  },
  {
   "slug": "weina-hexin",
   "name": "Weina Hexin (微纳核芯; Hangzhou Weina Hexin Electronic Technology Co.)",
   "founded": 2021,
   "hq": "Hangzhou (Xiaoshan), Zhejiang, China",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (incubated at the Zhejiang Peking University Advanced Institute of Information Technology, 浙江省北大信息技术高等研究院)",
   "total_funding_usd_m": 420,
   "last_round": "C2, >CNY 1B (~US$140M+), 2026-10 (建投投资 lead)",
   "key_investors": [
    "建投投资 (JIC Investment)",
    "蓝驰创投 (Lanchi Ventures)",
    "毅达资本 (Yida Capital)",
    "中芯聚源 (SMIC Capital)",
    "锦秋基金",
    "优势资本",
    "元禾璞华",
    "德联资本",
    "华勤技术",
    "美格智能"
   ],
   "website": "n/d (no English site found; brands 3D-CIM™, LP-CIM™)",
   "pitch": "\"全球首创三维存算一体 3D-CIM™ 架构\" — the company bills 3D-CIM™ as the world's first\nthree-dimensional compute-in-memory architecture, fusing \"3D 近存计算 + SRAM 存算一体 +\nRISC-V 异构存算\" (3D near-memory compute on DRAM, SRAM compute-in-memory, and RISC-V\nheterogeneous compute), aimed at both edge AI (LP-CIM™ series, co-developed with memory\nvendors) and cloud inference (PCIe-CIM series)\n([新浪财经 2026-09-09](https://finance.sina.com.cn/roll/2026-09-09/doc-inirfcpw5732663.shtml);\n[雷达财经 via 新浪 2026-10-30](https://finance.sina.cn/stock/jdts/2026-10-30/detail-iniwzazt8174544.d.html)).",
   "technical_approach": "- **Where the compute happens:** two places. The SRAM part is true compute-in-memory\n  (\"其 SRAM 存算一体技术将计算单元深度嵌入存储阵列\" — compute units embedded in the SRAM\n  array); the 3D part is near-memory compute next to stacked DRAM, not inside the DRAM\n  array. A granted patent (per Tianyancha, relayed by 雷达财经) covers a\n  \"三维集成异构存储介质的存算一体神经网络处理器\" — a CIM neural-network processor on 3D-\n  integrated heterogeneous memory, stacking volatile memory, logic, and non-volatile memory.\n- **Digital vs analog:** not stated. Treated here as digital/mixed SRAM CIM, the same\n  bucket as Houmo (SRAM) and Yizhu (3D DRAM). Inference-only; no on-chip learning claim.\n- **Claims (company-reported, baseline unspecified):** measured tape-out data show\n  \"4-6 倍算力密度提升、5-10 倍能效比提升\" versus a von Neumann design; \"十余项 ISSCC 世界纪录\"\n  over six years; ~20 papers in ISSCC/JSSC/TCAS-I (citmt.cn).\n- **Products / status:** first-generation LP-CIM and PCIe-CIM series, three chips in total,\n  \"已全部投片\" (all taped out) as of 2026-10; customer sampling \"预计2027年初\" (early 2027)\n  per the C2 announcement, although the C1 announcement a month earlier said sampling\n  \"年内\" (within 2026). Boards/servers and a software stack are described as complete.",
   "funding_rounds": [
    {
     "date": "2025-11",
     "round": "Series B (strategic)",
     "amount": ">CNY 100M (~$14M+)",
     "investors": "蓝驰创投 (lead); 中芯聚源, 锦秋基金, 君科丹木, 劭恒投资",
     "source": "https://k.sina.cn/article_3307324252_c521c35c020029mp2.html"
    },
    {
     "date": "2026-06",
     "round": "B3 + B4",
     "amount": "B-series total >CNY 1B (~$140M+; per-round split undisclosed)",
     "investors": "n/d",
     "source": "https://www.178448.com/thread-2159286-1-1.html (界面新闻 repost)"
    },
    {
     "date": "2026-09",
     "round": "C1",
     "amount": "CNY 1B (~$140M)",
     "investors": "no lead named; 优势资本, 元昊投资, 金浦投资, 阳光人寿, 元禾璞华, 德联资本, 宽桥恒松, 厚雪资本; strategic: 美格智能, 恒旭资本, 华勤技术; existing holders added",
     "source": "https://finance.sina.com.cn/roll/2026-09-09/doc-inirfcpw5732663.shtml ; https://www.163.com/dy/article/L6I9STBR05568W0A.html"
    },
    {
     "date": "2026-10",
     "round": "C2",
     "amount": ">CNY 1B (~$140M+)",
     "investors": "建投投资 (lead); 杭州资本, 凯泰资本, 晓池资本 et al. (participant list partly garbled in source)",
     "source": "https://www.citmt.cn/news/202610/127033.html ; https://finance.sina.cn/stock/jdts/2026-10-30/detail-iniwzazt8174544.d.html"
    }
   ],
   "funding_notes": "**Total:** disclosed floor **>CNY 3B ≈ US$420M** (B-series >1B + C1 1B + C2 >1B; earlier\nangel/A rounds exist per the round lettering but no figures were found). One garbled source\n(citmt.cn) appears to say \"超30亿元\" cumulative, consistent with the floor. No valuation\ndisclosed. Flag **†** — floor, CNY-converted at 0.14.",
   "trajectory": "Founded 2021 in Hangzhou, incubated at the Zhejiang Peking University Advanced Institute of\nInformation Technology; the founder/chief scientist is a Peking University School of\nIntegrated Circuits professor (citmt.cn names the affiliation; the name is garbled in that\ncopy — 叶乐 (Ye Le), PKU professor and vice dean of the same institute whose stated research\narea is CIM AI chips, is the likely match but is **not confirmed** by any readable source).\nTeam of \"100+\" PhDs, postdocs and engineers (C1 release). Four rounds in twelve months\n(2025-11 → 2026-10), each ≥CNY 1B from mid-2026, make it one of the fastest-funded\ncompanies in the census. Status: active, pre-revenue; first silicon sampling early 2027.",
   "relevance": "A conventional-memory (SRAM + DRAM) compute-in-memory play at datacenter scale, inference\nonly, with no adaptive/learning hardware. It belongs with Houmo and Yizhu in category 2 and\nis further evidence that Chinese CIM companies raise CNY-1B-class rounds that English-\nlanguage coverage misses entirely: this company surfaced only through a zh-cn discovery\nquery. Its total alone adds ~6% to the census money.",
   "news": [],
   "sources": [
    {
     "title": "新浪财经: 微纳核芯完成10亿元C1轮融资 (2026-09-09)",
     "url": "https://finance.sina.com.cn/roll/2026-09-09/doc-inirfcpw5732663.shtml"
    },
    {
     "title": "网易/猎云网: 微纳核芯完成10亿元C1轮融资 (2026-09-11)",
     "url": "https://www.163.com/dy/article/L6I9STBR05568W0A.html"
    },
    {
     "title": "网易: C1 round with 财联社 data (2026-09-09)",
     "url": "https://www.163.com/dy/article/L6E0Q8210550B1DU.html"
    },
    {
     "title": "新浪财经 (合肥市投资基金协会 member news): C1 round (2026-09-15)",
     "url": "https://finance.sina.com.cn/roll/2026-09-15/doc-inirwwmy3702661.shtml"
    },
    {
     "title": "雷达财经 via 新浪: 微纳核芯完成超10亿元C2轮融资 (2026-10-30)",
     "url": "https://finance.sina.cn/stock/jdts/2026-10-30/detail-iniwzazt8174544.d.html"
    },
    {
     "title": "citmt.cn: C2 round, 建投投资 lead (2026-10-29)",
     "url": "https://www.citmt.cn/news/202610/127033.html"
    },
    {
     "title": "新浪 (大众财经网): 超亿元B轮战略融资 (2025-11-03)",
     "url": "https://k.sina.cn/article_3307324252_c521c35c020029mp2.html"
    },
    {
     "title": "178448.com (界面新闻 repost): B3/B4, B-series >10亿元 (2026-06-15)",
     "url": "https://www.178448.com/thread-2159286-1-1.html"
    },
    {
     "title": "CCF YEF 2025 speaker bio: 叶乐 (background only; not a company source)",
     "url": "https://ccf.org.cn/yef2025/speaker_d_2032"
    }
   ]
  },
  {
   "slug": "witmem",
   "name": "Witmem / Zhicun Technology (知存科技)",
   "founded": 2017,
   "hq": "Beijing, China",
   "category": "analog-imc",
   "status": "active",
   "origin": "academic spinoff (UCLA flash in-memory-computing research)",
   "total_funding_usd_m": 115,
   "last_round": "Series B2, CNY 200M (~$29M), 2023-01",
   "key_investors": [
    "SDIC Venture Capital (国投创业)",
    "Shenzhen Capital Group (深创投)",
    "Huawei Hubble (哈勃投资)",
    "China Fortune-Tech / 中芯聚源",
    "iFlytek Ventures",
    "Qingkong/China Merchants VC"
   ],
   "website": "https://www.witmem.com",
   "pitch": "Witmem frames AI's bottleneck as the \"memory wall\": shuttling neural-network weights\nbetween separate memory and processor dominates energy and limits edge AI. Its answer is\n存算一体 (compute-in-memory) — store the weights in flash cells and do the matrix math\ninside the memory array itself, claiming order-of-magnitude power reduction and \"first in\nthe world to mass-produce commercial CIM chips\" (https://www.witmem.com/;\nhttp://www.xinhuanet.com/tech/20230106/ec33e33e368740729d06a4dc093ab768/c.html).",
   "technical_approach": "- **Device/substrate:** analog compute-in-memory in mature **NOR flash** — flash cell\n  conductances hold NN weights; analog vector-matrix multiply in the array\n  (https://www.eenewseurope.com/en/china-startup-witinmem-uses-analog-flash-for-compute-in-memory/).\n- **WTM2101** (mass production announced Feb 2022): ultra-low-power CIM SoC for always-on\n  voice/audio/health AI in wearables and hearables; shipping at \"KK-level\" (millions of\n  units) per the company site (eeNews Europe above; https://www.witmem.com/).\n- **WTM-8 series** (2nd generation): 3D heterogeneous-bonded CIM vision/\"spatial compute\"\n  chip for phones and edge devices; company cited 4–32 TOPS (later 6–48 TOPS) targets and\n  ~10% of the energy of comparable solutions; **trial mass production reached Dec 2024**\n  (https://www.cls.cn/detail/1903646; https://pitchhub.36kr.com/project/1679660495360776).\n- A third-generation WTM-C series (~100 TOPS) was announced as roadmap in Nov 2022\n  (https://www.eet-china.com/news/202211115175.html).\n- **Inference-only**: training happens off-chip; no on-chip learning claim.",
   "funding_rounds": [
    {
     "date": "2019-08",
     "round": "Series A (A1)",
     "amount": "~CNY 100M (~$14M)",
     "investors": "中芯聚源 (SMIC-affiliated) lead; 普华资本, 招商局创投, 三峡鑫泰, 科讯创投",
     "source": "https://www.asmag.com.cn/news/201908/100174.html"
    },
    {
     "date": "~2020",
     "round": "A2",
     "amount": "undisclosed",
     "investors": "国投创业 (SDIC VC) lead",
     "source": "https://www.forbeschina.com/investment/55770"
    },
    {
     "date": "2021-06",
     "round": "A3",
     "amount": "~CNY 100M; A-series total ~CNY 300M",
     "investors": "飞图创投 lead; 万魔声学 (1MORE), 科宇盛达, 仁馨资本",
     "source": "https://startup.aliyun.com/info/105184.html"
    },
    {
     "date": "2021-09",
     "round": "Pre-B",
     "amount": "undisclosed",
     "investors": "Huawei's Hubble Investment (哈勃) became shareholder",
     "source": "https://www.taibo.cn/p/75626"
    },
    {
     "date": "2022-01",
     "round": "B1",
     "amount": "CNY 200M (~$31M)",
     "investors": "领航新界 lead; 天堂硅谷, 瑞芯投资; returning 讯飞创投, 招商局创投/清控招商, 普华资本",
     "source": "https://www.52audio.com/archives/106421.html"
    },
    {
     "date": "2022-09",
     "round": "B1+",
     "amount": "CNY 100M (~$14M)",
     "investors": "深创投 (Shenzhen Capital Group) lead; 国开科创",
     "source": "http://www.news.cn/tech/20220928/6d13a5c7d4584c8a901eba4b225aff33/c.html"
    },
    {
     "date": "2023-01",
     "round": "B2",
     "amount": "CNY 200M (~$29M)",
     "investors": "国投创业 lead; 水木春锦资本, 领航新界 (add-on)",
     "source": "http://www.xinhuanet.com/tech/20230106/ec33e33e368740729d06a4dc093ab768/c.html"
    }
   ],
   "funding_notes": "Disclosed rounds sum to roughly **CNY 800M (~$115M at ~7 CNY/USD), partial** (angel and\nPre-B amounts undisclosed). Conflict: 36kr PitchHub states cumulative \"CNY 600M\"\n(https://pitchhub.36kr.com/project/1679660495360776) — report both. A wave-2 gap-sweep\nnote of \"CNY 200M Series B at CNY 3.8B valuation\" could **not** be confirmed in any\nprimary source found; no valuation figure is citable here. No post-2023 round surfaced in\npublic reporting — Chinese trackers under-report, treat as unknown rather than none.",
   "trajectory": "Founded Oct 2017 in Beijing by **Wang Shaodi (王绍迪)**, founder-CEO, whose PhD work at\nUCLA was on flash-based in-memory computing (bio per Chinese press, e.g.\nhttps://ccf.org.cn/cftc2025/speaker_d_1151). Steady product cadence: WTM2101 volume\nproduction 2022 → WTM-8 trial production late 2024. As of mid-2026 the company is active,\nmarkets itself as the global first mover in mass-produced CIM chips, and its CEO is\npublicly pushing in-memory compute for LLM inference (CCF 2025 talks). Named a CLS\n\"most investable AI company\" Dec 2024 (https://www.cls.cn/detail/1903646).",
   "relevance": "Same core idea as a memristor crossbar — analog conductances as synaptic weights doing\nin-array VMM — but implemented on mature NOR flash instead of emerging devices, which is\nhow they got to million-unit shipments first. Inference-only: weights are programmed, not\nlearned on-chip, in contrast to Knowm's AHaH on-device-learning thesis.",
   "news": [
    {
     "date": "2026-08-23",
     "text": "Major commercial win: **Anker's Thus™ A1 — a compute-in-memory AI audio chip co-developed with Witmem/知存科技 over three years** — is shipping in Anker noise-cancelling earbuds, which earned a Guinness \"clearest call\" certification (2026-05-22). Reported ~5 GFLOPS vs ~30 MFLOPS for conventional BT audio chips (~150x). 36Kr ran a joint interview with Anker's and Witmem's CEOs (2026-06-05). (https://www.leiphone.com/category/weiwu/SE0UCzo94OXxs9aG.html ; https://m.36kr.com/p/3838642718607622 ; https://www.52audio.com/archives/277272.html)"
    }
   ],
   "sources": [
    {
     "title": "Witmem site",
     "url": "https://www.witmem.com/"
    },
    {
     "title": "eeNews Europe: Witinmem analog flash CIM",
     "url": "https://www.eenewseurope.com/en/china-startup-witinmem-uses-analog-flash-for-compute-in-memory/"
    },
    {
     "title": "Xinhua: B2 round",
     "url": "http://www.xinhuanet.com/tech/20230106/ec33e33e368740729d06a4dc093ab768/c.html"
    },
    {
     "title": "Xinhua: B1+ round",
     "url": "http://www.news.cn/tech/20220928/6d13a5c7d4584c8a901eba4b225aff33/c.html"
    },
    {
     "title": "52audio: B1 round",
     "url": "https://www.52audio.com/archives/106421.html"
    },
    {
     "title": "a&s: 2019 A round",
     "url": "https://www.asmag.com.cn/news/201908/100174.html"
    },
    {
     "title": "Aliyun/Cyzone: A3 round",
     "url": "https://startup.aliyun.com/info/105184.html"
    },
    {
     "title": "Taibo: Huawei Hubble stake",
     "url": "https://www.taibo.cn/p/75626"
    },
    {
     "title": "36kr PitchHub profile",
     "url": "https://pitchhub.36kr.com/project/1679660495360776"
    },
    {
     "title": "CLS: WTM-8 trial production",
     "url": "https://www.cls.cn/detail/1903646"
    }
   ]
  },
  {
   "slug": "yizhu",
   "name": "Yizhu Technology (亿铸科技)",
   "founded": 2020,
   "hq": "Suzhou / Shanghai, China",
   "category": "analog-imc",
   "status": "active",
   "origin": "serial founders / corporate alumni (Wave Computing China GM; UT Austin PhD)",
   "total_funding_usd_m": 280,
   "last_round": ">CNY 1B (~US$140M), round letter/investors undisclosed, 2026-10-23; before that Series D, amount undisclosed, 2026-08-12 (上创信德投资, 上海纳米创投, 中科英智投资 et al.); total CNY 1B+ confirmed 2026-07 excludes this round",
   "key_investors": [
    "CAS Star (中科创星)",
    "Legend Star (联想之星)",
    "Longqiu Capital (隆湫资本)",
    "Prosperity7 Ventures (per PEdaily)",
    "Shenzhen SED/盛视科技",
    "ABC International (农银国际)"
   ],
   "website": "https://yizhu-tech.com",
   "pitch": "Yizhu says large-model compute growth slams into the \"power wall and memory wall\" (功耗墙,\n内存墙) of processor-centric von Neumann chips; founder Xiong Dapeng argues future AI\nchips must be \"memory-centric,\" and that ReRAM-based fully digital compute-in-memory can\ndeliver ~10x energy-efficiency on a mature 28nm process — architecture innovation as the\n\"second growth curve\" replacing Moore's law scaling\n(https://cn.chinadaily.com.cn/a/202209/13/WS632042eba310817f312edd5d.html;\nhttps://zhuanlan.zhihu.com/p/737616205).",
   "technical_approach": "- **Device/substrate:** **ReRAM (memristor) compute-in-memory, fully digital** — an\n  explicitly all-digital CIM design that removes AD/DA converters and uses sparsity, to\n  avoid analog precision loss while keeping in-array compute\n  (https://cn.chinadaily.com.cn/a/202209/13/WS632042eba310817f312edd5d.html).\n- Domestic supply chain: ReRAM IP/process via partner 昕原半导体 (Xinyuan Semiconductor)\n  and SMIC, claimed fully China-sourced from IP to fab\n  (https://m.36kr.com/p/1688429264884613; https://www.cls.cn/detail/1139532).\n- First large-compute POC chip \"lit up\" (点亮) — a high-precision, low-power ReRAM CIM\n  chip — reported by PEdaily's profile; in 2023 the company proposed a \"CIM\n  super-heterogeneous\" (存算一体超异构) architecture, and by 2026 brands itself as China's\n  first \"general-purpose in-memory computing\" (GP-IMC®) large-compute AI chip company\n  (https://m.pedaily.cn/vc/company/614300.html; https://m.tech.china.com/redian/2026/0702/072026_1907537.html).\n- **Market wedge:** datacenter/large-compute AI inference (plus autonomous driving named\n  early on). **Inference-only**; no on-chip learning claim.",
   "funding_rounds": [
    {
     "date": "2021-12",
     "round": "Angel",
     "amount": "over CNY 100M (~$16M)",
     "investors": "CAS Star (中科创星), Legend Star (联想之星) and Huixin Investment jointly led",
     "source": "https://yizhu-tech.com/newsv.php?id=4"
    },
    {
     "date": "2022-10",
     "round": "Pre-A",
     "amount": "over CNY 100M (~$14M)",
     "investors": "Longqiu Capital (隆湫资本) lead",
     "source": "https://m.cyzone.cn/article/701649.html"
    },
    {
     "date": "2023-08",
     "round": "Series A",
     "amount": "undisclosed",
     "investors": "per PEdaily tracker: China Merchants-affiliated fund, returning Longqiu, others",
     "source": "https://m.pedaily.cn/vc/company/614300.html"
    },
    {
     "date": "2024-10 (PEdaily dates it 2024-05)",
     "round": "Series B",
     "amount": "several CNY 100Ms",
     "investors": "\"well-known overseas fund\" lead (PEdaily records Prosperity7 Ventures, Aramco's VC); 盛视科技, 行至资本; ex-co-founders of a famous overseas GPU company added",
     "source": "https://www.laoyaoba.com/n/964077; https://m.pedaily.cn/vc/company/614300.html"
    },
    {
     "date": "2025-10",
     "round": "New round (B+)",
     "amount": "undisclosed",
     "investors": "兴湘资本, 农银国际 (ABC International), 盈科值得, 合鼎共资本",
     "source": "https://news.pedaily.cn/202510/556561.shtml"
    },
    {
     "date": "2026-04",
     "round": "Strategic",
     "amount": "undisclosed",
     "investors": "listed in PEdaily tracker (investor name unverified)",
     "source": "https://m.pedaily.cn/vc/company/614300.html"
    }
   ],
   "funding_notes": "**Total: CNY 1B+ (~$140M at ~7 CNY/USD) completed as of 2026-07-02, with a new round\n\"about to open\"** (https://finance.sina.com.cn/stock/estate/integration/2026-07-02/doc-inifkuqa2722327.shtml;\nhttps://www.tmtpost.com/nictation/8050245.html). Per-round amounts are only partially\ndisclosed — the round table is **partial** and lead-investor names for 2023+ rounds rest\non tracker data.",
   "trajectory": "Registered June 2020 in Suzhou (operations Shanghai). Founder/chairman/CEO **Xiong Dapeng\n(熊大鹏)**: BS Xidian 1983, MS SCUT 1986, PhD UT Austin 1998 (plus three master's);\n~30 years in chips in the US and China; worked on GPU-for-AI chips from 2015 and was China\nGM of US AI-chip firm Wave Computing before founding Yizhu\n(https://static.eetrend.com/content/2023/100571710.html;\nhttps://course.zhidx.com/teacher/detail/YzE3M2UyZWNiM2YwYTc0MzQ2NTM=). Status mid-2026:\nactive, well-funded for a ReRAM player (CNY 1B+), state and strategic capital on the cap\ntable, still pre-large-scale-commercial-deployment on public evidence.",
   "relevance": "The closest of the China batch to Knowm's substrate: memristive (ReRAM) devices in the\ncompute path. But Yizhu deliberately goes **digital** CIM — no analog conductance math, no\nADCs, no on-chip learning — using the memristor purely as dense non-volatile storage\ninside the compute array, the opposite bet from analog-synapse approaches like AHaH.",
   "news": [
    {
     "date": "2026-08-23, unverified",
     "text": "A Sohu funding-tracker item \"亿铸智能科技获D轮投资\" (2026-08-13) reports a **D round** for what is most likely Yizhu's registered entity (亿铸智能科技) — consistent with the \"new round about to open\" note from 2026-07. Single low-quality source, investor/amount undisclosed: verify before citing. (https://m.sohu.com/a/1062588388_115035)"
    },
    {
     "date": "2026-09-22",
     "text": "**D round now verified** (second, named-investor source): 亿欧 data reports Suzhou Yizhu Intelligent Technology (苏州亿铸智能科技有限公司, founded 2020-06-08) **completed a Series D on 2026-08-12**, jointly invested by 上创信德投资, 上海纳米创投, 上海轩乃怡, 中科英智投资, 众物投资, 元信景合(海南), 兰溪聚垚, 大米创投, 普超资本, 树院基金 and 涌铧投资. **Amount undisclosed.** The August \"unverified\" flag is lifted; the CNY 1B+ cumulative total (as of 2026-07) therefore excludes this round. (https://view.inews.qq.com/a/20260814A02J1B00)"
    },
    {
     "date": "2026-09-22",
     "text": "**Substrate shift — 3D DRAM, not ReRAM, is the production path.** At WAIC 2026 (2026-07-21) co-founder/president 徐芳 presented Yizhu's \"three laws of general-purpose CIM\" and its **all-digital 3D-DRAM general-purpose CIM AI chip, which the company says has taped out** and completed full-chain validation; September coverage repeats it and adds claimed results vs mainstream cloud GPUs (~2x performance, ~1/5 power, ~1/6 cost, up to ~88% lower inference TCO), a CUDA/PyTorch-compatible YICA software stack, ~300 staff (80% R&D) across Suzhou/Shanghai/Hangzhou/Chengdu/Changsha, and a shortlist spot in MIIT's 算力强基揭榜行动. Yizhu now describes ReRAM/RRAM-class prototypes (its own 2023 28 nm ReRAM test chip included) as capacity-limited \"MB-class\" media and says its architecture \"is not bound to a single storage medium\". **Reclassification candidate:** the company no longer fits emerging-device-imc on its productized path — it is DRAM-based digital CIM (closest peer: Houmo's SRAM digital CIM in category 2). All performance figures are company claims. (https://news.tom.com/202607/4265382577.html ; https://www.eetrend.com/content/2026-09/a6ca010c-fc03-4acc-b632-ce8b8d3fa0fa-100604189.html ; https://m.sohu.com/a/1077890515_121207965)"
    },
    {
     "date": "2026-10-27",
     "text": "**New round of >CNY 1B (~US$140M) announced 2026-10-23**; investors and round letter not disclosed (36Kr, 智东西, 搜狐 all carry the company statement; 创业邦's headline names 经纬 and 蓝驰 as repeat backers — unconfirmed in the body text we could read). Funds go to R&D, product iteration and industrialization. Company restates: 3D-DRAM all-digital general-purpose CIM chip **taped out and 'test results basically meet design targets'**; claims vs cloud GPUs unchanged (~2x throughput vs H200 on an FPGA platform, ~1/5 power, ~1/6 cost); YICA stack CUDA/PyTorch-compatible; founder/CEO 熊大鹏 (UT Austin PhD, ex-Wave Computing China GM). Prior rounds restated: Prosperity7-led 数亿元 (2024-10); 兴湘资本/农银国际 et al. (2025-10); plus the undisclosed Series D (2026-08-12). **Working total ≈ US$280M** (CNY 1B+ through 2026-07 + this CNY 1B+; the August D still excluded as undisclosed). (https://m.zhidx.com/p/603963.html ; https://www.36kr.com/newsflashes/4039705825858436 ; https://m.sohu.com/a/1092946225_120773109)"
    }
   ],
   "sources": [
    {
     "title": "ChinaDaily: 28nm ReRAM digital CIM",
     "url": "https://cn.chinadaily.com.cn/a/202209/13/WS632042eba310817f312edd5d.html"
    },
    {
     "title": "Yizhu PR: angel round",
     "url": "https://yizhu-tech.com/newsv.php?id=4"
    },
    {
     "title": "Cyzone: Pre-A round",
     "url": "https://m.cyzone.cn/article/701649.html"
    },
    {
     "title": "Laoyaoba/JW Insights: 2024 round",
     "url": "https://www.laoyaoba.com/n/964077"
    },
    {
     "title": "PEdaily: Oct 2025 round",
     "url": "https://news.pedaily.cn/202510/556561.shtml"
    },
    {
     "title": "PEdaily company tracker",
     "url": "https://m.pedaily.cn/vc/company/614300.html"
    },
    {
     "title": "Sina Finance: CNY 1B+ total, 2026-07",
     "url": "https://finance.sina.com.cn/stock/estate/integration/2026-07-02/doc-inifkuqa2722327.shtml"
    },
    {
     "title": "China.com tech: GP-IMC® positioning",
     "url": "https://m.tech.china.com/redian/2026/0702/072026_1907537.html"
    },
    {
     "title": "CLS interview with Xiong Dapeng",
     "url": "https://www.cls.cn/detail/1139532"
    },
    {
     "title": "36kr: Xinyuan ReRAM partnership",
     "url": "https://m.36kr.com/p/1688429264884613"
    }
   ]
  },
  {
   "slug": "zhizi-guangyuan",
   "name": "Zhizi Guangyuan (智子光垣; Zhizi Guangyuan Technology (Shanghai) Co.)",
   "founded": 2026,
   "hq": "Shanghai (Minhang District), China",
   "category": "photonic",
   "status": "active",
   "origin": "n/d (founders not named in available sources; Shanghai Jiao Tong University AI fund among investors)",
   "total_funding_usd_m": 28,
   "last_round": "Seed, 数亿元 (≥CNY 200M, ~US$28M+), 2026-11",
   "key_investors": [
    "云启资本 (Yunqi Partners)",
    "国泰海通",
    "险峰K2VC",
    "原子创投",
    "同鑫力诚",
    "上海交通大学AI未来基金",
    "海桐国际创新中心"
   ],
   "website": "n/d",
   "pitch": "A general-purpose **all-optical computing chip** (\"通用光计算芯片\") built on a reconfigurable\nall-optical architecture that combines **interference and diffraction** computing, scaled by\nchiplet cascading, with two first products: the server chip 烛龙 OPU-08 (claimed 150 TOPS)\nand the edge chip 玄鉴 TOCU-07 (claimed 100 TOPS); both first versions \"已送流片\" (sent for\ntape-out) ([DRAMeXchange/TrendForce 2026-11-05](https://www.dramx.com/News/ict/20261105-41397.html)).\nThe company's stated goal is \"not a special-purpose chip that computes fast\" but a general\noptical processor ([zhcjzx.cn 2026-10-20](http://www.zhcjzx.cn/cd/2026/203419652.html), snippet;\npage refused connection).",
   "technical_approach": "- **Substrate:** free-space/integrated hybrid is not specified; \"interference + diffraction\"\n  suggests a combination of Mach-Zehnder-style interferometric meshes and diffractive\n  optical layers, the two mainstream approaches to optical matrix multiplication.\n- **Scaling:** chiplet cascading to grow matrix size.\n- **Status:** founded May 2026; two chips taped out within six months (claimed). No\n  measured results, power figures, or precision data are public. Inference-oriented;\n  no learning claim.",
   "funding_rounds": [
    {
     "date": "2026-05 → 2026-11",
     "round": "\"five seed rounds\" (company statement; individual sizes undisclosed)",
     "amount": "n/d",
     "investors": "—",
     "source": "https://www.dramx.com/News/ict/20261105-41397.html"
    },
    {
     "date": "2026-11",
     "round": "Seed (latest)",
     "amount": "数亿元 (hundreds of millions CNY; ≥CNY 200M ≈ $28M+ taken as floor)",
     "investors": "no lead named; 云启资本, 国泰海通, 险峰K2VC, 原子创投, 同鑫力诚, 上海交通大学AI未来基金, 海桐国际创新中心",
     "source": "https://www.dramx.com/News/ict/20261105-41397.html"
    }
   ],
   "funding_notes": "**Total:** floor ~US$28M (†). Whether the 数亿元 figure is the latest round or the cumulative\nseed series is ambiguous in the source. Five seed rounds in six months is unusual and is\nrecorded here as a company claim.",
   "trajectory": "Registered as 智子光垣科技（上海）有限公司, HQ in Minhang District, Shanghai; founded May 2026.\nFounders and their backgrounds are not named in the sources read; a Shanghai Jiao Tong\nUniversity AI fund is among the investors, suggesting a university link. Status: active,\npre-silicon-results.",
   "relevance": "The newest company in the census and the only 2026 founding, in the photonic category that\nalready holds the most money. It shows the photonic wave has not ended: Chinese optical-\ncomputing startups continue to form and raise CNY-100M-class seed money in late 2026.\nInference-only; nothing adaptive.",
   "news": [],
   "sources": [
    {
     "title": "DRAMeXchange (TrendForce 集邦咨询): 智子光垣完成数亿元种子轮融资 (2026-11-05)",
     "url": "https://www.dramx.com/News/ict/20261105-41397.html"
    },
    {
     "title": "zhcjzx.cn: 智子光垣打造通用光计算芯片，已推动连续多轮融资 (2026-10-20)",
     "url": "http://www.zhcjzx.cn/cd/2026/203419652.html"
    }
   ]
  }
 ],
 "patents_table_raw": [
  [
   "#",
   "Company",
   "Category",
   "Families",
   "Pubs",
   "Granted / pending",
   "Earliest priority",
   "Latest grant",
   "Top CPC",
   "Representative patents",
   "Reassigned / note"
  ],
  [
   "1",
   "Lynxi Technology (灵汐科技)",
   "digital-snn",
   "635",
   "763",
   "298 / 465",
   "2015-06-16",
   "2026-08-21",
   "G06F, G06N, Y02D",
   "[Method and apparatus for converting numerical values into spikes, elec](https://patents.google.com/patent/US11783166B2/en) (US11783166B2, 2020-07-24); [Path creation method and device for network on chip and electronic app](https://patents.google.com/patent/US11398981B2/en) (US11398981B2, 2018-12-28)",
   ""
  ],
  [
   "2",
   "Crossbar, Inc.",
   "emerging-device-imc",
   "172",
   "186",
   "151 / 35",
   "2010-06-08",
   "2026-06-30",
   "G11C, H10N, H10B",
   "[Selector device for two-terminal memory](https://patents.google.com/patent/US9761635B1/en) (US9761635B1, 2014-03-11); [Resistive memory architecture and devices](https://patents.google.com/patent/US9768234B2/en) (US9768234B2, 2014-05-20)",
   "→ Crossbar, Inc."
  ],
  [
   "3",
   "AlpsenTek (Shenzhen Ruishizhixin Technology)",
   "event-sensing",
   "146",
   "162",
   "61 / 101",
   "2019-04-25",
   "2026-08-21",
   "H04N, G06T, G06F",
   "[Delta vision sensor](https://patents.google.com/patent/US12114090B2/en) (US12114090B2, 2019-04-25); [Image sensor and image output method and application thereof](https://patents.google.com/patent/CN114422726A/en) (CN114422726A, 2022-03-15)",
   ""
  ],
  [
   "4",
   "Lightelligence",
   "photonic",
   "127",
   "136",
   "38 / 98",
   "2018-06-05",
   "2026-08-21",
   "G02B, H10W, G06F",
   "[Optical modulation for optoelectronic processing](https://patents.google.com/patent/US11719963B2/en) (US11719963B2, 2020-04-29); [Processing matrix operations for rate limited systems](https://patents.google.com/patent/US11526737B2/en) (US11526737B2, 2019-02-01)",
   ""
  ],
  [
   "5",
   "Yizhu Technology (亿铸科技)",
   "analog-imc",
   "125",
   "128",
   "37 / 91",
   "2022-11-07",
   "2026-07-21",
   "G06F, Y02D, G06N",
   "[Tensor processing unit and method, and computer-readable storage mediu](https://patents.google.com/patent/WO2025124578A1/en) (WO2025124578A1, 2023-12-13); [Computing system, method executed by computing system, and storage med](https://patents.google.com/patent/WO2025124574A1/en) (WO2025124574A1, 2023-12-13)",
   ""
  ],
  [
   "6",
   "Houmo.ai (后摩智能)",
   "analog-imc",
   "113",
   "121",
   "17 / 104",
   "2021-03-26",
   "2026-01-13",
   "G06F, G06N, G11C",
   "[Circuits based on digital domain in-memory computing](https://patents.google.com/patent/CN112711394B/en) (CN112711394B, 2021-03-26); [Circuit based on digital domain in-memory computing](https://patents.google.com/patent/US20240168718A1/en) (US20240168718A1, 2021-03-26)",
   ""
  ],
  [
   "7",
   "SynSense (fka aiCTX)",
   "digital-snn",
   "112",
   "120",
   "41 / 79",
   "2019-04-09",
   "2026-06-23",
   "G06N, G06F, G06T",
   "[Event-driven integrated circuit having interface system](https://patents.google.com/patent/US12432466B2/en) (US12432466B2, 2021-04-19); [Event-driven spiking neutral network system for detection of physiolog](https://patents.google.com/patent/US12665078B2/en) (US12665078B2, 2019-06-24)",
   ""
  ],
  [
   "8",
   "Witmem / Zhicun Technology (知存科技)",
   "analog-imc",
   "94",
   "116",
   "35 / 81",
   "2018-08-02",
   "2026-05-29",
   "G06F, G11C, H10B",
   "[Analog vector-matrix multiplication circuit](https://patents.google.com/patent/US11379673B2/en) (US11379673B2, 2018-08-02); [Flash memory chip and calibration method and apparatus therefor](https://patents.google.com/patent/US11995339B2/en) (US11995339B2, 2018-11-28)",
   ""
  ],
  [
   "9",
   "Weina Hexin (微纳核芯; Hangzhou Weina Hexin Electronic Technology Co.)",
   "analog-imc",
   "90",
   "106",
   "29 / 77",
   "2019-04-10",
   "2026-08-04",
   "G06F, G06N, Y02D",
   "[Feature extraction circuit, neural network, system, integrated circuit](https://patents.google.com/patent/CN113326918A/en) (CN113326918A, 2021-04-29); [Capacitance-to-digital conversion circuit](https://patents.google.com/patent/CN113422608A/en) (CN113422608A, 2021-05-25)",
   ""
  ],
  [
   "10",
   "Reexen Technology (九天睿芯)",
   "analog-imc",
   "82",
   "106",
   "34 / 72",
   "2017-08-31",
   "2026-08-04",
   "G06N, Y02D, G06F",
   "[Time division interleaving band-pass filter for use in voice activity ](https://patents.google.com/patent/US11894820B2/en) (US11894820B2, 2020-04-10); [Subunits, MAC arrays, bit-width reconfigurable analog-to-digital hybri](https://patents.google.com/patent/CN113627601B/en) (CN113627601B, 2020-05-08)",
   ""
  ],
  [
   "11",
   "Lightmatter",
   "photonic",
   "77",
   "82",
   "31 / 51",
   "2012-11-27",
   "2026-06-09",
   "G02B, G06N, G06F",
   "[Hybrid analog-digital matrix processors](https://patents.google.com/patent/US12153975B2/en) (US12153975B2, 2019-02-26); [Photonic communication platform](https://patents.google.com/patent/US12092867B2/en) (US12092867B2, 2019-03-06)",
   ""
  ],
  [
   "12",
   "TetraMem Inc.",
   "emerging-device-imc",
   "70",
   "81",
   "28 / 53",
   "2018-08-17",
   "2026-08-18",
   "—",
   "[Forming-free random-access memory (RRAM) devices](https://patents.google.com/patent/US12137622B2/en) (US12137622B2, 2022-07-15); [Suppressing random telegraph noise in crossbar circuits](https://patents.google.com/patent/US12406726B2/en) (US12406726B2, 2023-07-28)",
   ""
  ],
  [
   "13",
   "Q.ANT",
   "photonic",
   "52",
   "55",
   "3 / 52",
   "2018-07-27",
   "2026-01-29",
   "—",
   "[Method for joining an optical crystal to a substrate](https://patents.google.com/patent/US12338168B2/en) (US12338168B2, 2020-09-09); [Method for calibrating a particle sensor, particle sensor and device h](https://patents.google.com/patent/EP4264227A1/en) (EP4264227A1, 2020-12-18)",
   ""
  ],
  [
   "14",
   "Prophesee (formerly Chronocam)",
   "event-sensing",
   "42",
   "45",
   "23 / 22",
   "2013-09-16",
   "2026-03-03",
   "H04N, G06V, G06T",
   "[Sample and hold based temporal contrast vision sensor](https://patents.google.com/patent/US10469776B2/en) (US10469776B2, 2016-04-04); [Pixel circuit for detecting time-dependent visual data](https://patents.google.com/patent/US10721423B2/en) (US10721423B2, 2016-10-20)",
   ""
  ],
  [
   "15",
   "Weebit Nano Ltd",
   "emerging-device-imc",
   "31",
   "32",
   "10 / 22",
   "2019-06-12",
   "2025-08-13",
   "G11C, H10B, H10N",
   "[Method for manufacturing an OxRAM type resistive memory cell](https://patents.google.com/patent/US12349605B2/en) (US12349605B2, 2019-06-12); [Memory comprising a matrix of resistive memory cells, and associated m](https://patents.google.com/patent/US12052876B2/en) (US12052876B2, 2020-12-16)",
   ""
  ],
  [
   "16",
   "Rain AI (fka Rain Neuromorphics)",
   "emerging-device-imc",
   "31",
   "31",
   "3 / 28",
   "2018-04-05",
   "2025-04-08",
   "G06F, G06N, G11C",
   "[Flexible compute engine microarchitecture](https://patents.google.com/patent/US12271439B2/en) (US12271439B2, 2023-06-23); [Electrical networks using analytic loss gradients for design, analysis](https://patents.google.com/patent/US11922296B2/en) (US11922296B2, 2021-07-28)",
   "→ OpenAI Opco, LLC"
  ],
  [
   "17",
   "Extropic",
   "probabilistic-thermodynamic",
   "25",
   "31",
   "0 / 31",
   "2023-06-16",
   "—",
   "G06N, G06F, H03K",
   "[Circuits for mixing voltage distributions associated with random varia](https://patents.google.com/patent/US20260031800A1/en) (US20260031800A1, 2024-07-29); [Configuring a circuit for generating samples from a target distributio](https://patents.google.com/patent/US20260031794A1/en) (US20260031794A1, 2024-07-29)",
   ""
  ],
  [
   "18",
   "Applied Brain Research (ABR)",
   "digital-snn",
   "23",
   "33",
   "12 / 21",
   "2012-12-05",
   "2026-08-11",
   "G06N, G06F, G10L",
   "[Methods and systems for training multi-bit spiking neural networks for](https://patents.google.com/patent/US12248871B2/en) (US12248871B2, 2019-11-01); [Methods and systems for storing and querying database entries with neu](https://patents.google.com/patent/US11741098B2/en) (US11741098B2, 2019-11-01)",
   ""
  ],
  [
   "19",
   "Syntiant",
   "analog-imc",
   "21",
   "22",
   "4 / 18",
   "2017-07-18",
   "2026-06-16",
   "—",
   "[Microcontroller interface for audio signal processing](https://patents.google.com/patent/US11270198B2/en) (US11270198B2, 2017-07-31); [Neuromorphic synthesizer](https://patents.google.com/patent/US12657442B2/en) (US12657442B2, 2017-07-18)",
   ""
  ],
  [
   "20",
   "Optalysys",
   "photonic",
   "20",
   "25",
   "4 / 21",
   "1997-12-12",
   "2024-07-09",
   "G06E, G06F, G02F",
   "[Reconfigurable optical processing system](https://patents.google.com/patent/US10289151B2/en) (US10289151B2, 2012-07-04); [Optical processing systems](https://patents.google.com/patent/US12033066B2/en) (US12033066B2, 2018-04-27)",
   ""
  ],
  [
   "21",
   "BrainChip",
   "digital-snn",
   "19",
   "22",
   "11 / 11",
   "2008-09-21",
   "2026-07-21",
   "G06N, G06F, G10L",
   "[Method, digital electronic circuit and system for unsupervised detecti](https://patents.google.com/patent/US11853862B2/en) (US11853862B2, 2016-11-21); [Secure voice signature communications system using local and remote ne](https://patents.google.com/patent/US11429857B2/en) (US11429857B2, 2014-06-28)",
   ""
  ],
  [
   "22",
   "Innatera Nanosystems",
   "digital-snn",
   "18",
   "20",
   "3 / 17",
   "2018-11-18",
   "2025-08-15",
   "G06N, G01S, H03M",
   "[Resilient neural network](https://patents.google.com/patent/US12380320B2/en) (US12380320B2, 2018-11-18); [Adaptation of snns through transient synchrony](https://patents.google.com/patent/EP4226281B1/en) (EP4226281B1, 2020-10-05)",
   ""
  ],
  [
   "23",
   "Luminous Computing",
   "photonic",
   "17",
   "22",
   "9 / 13",
   "2018-11-08",
   "2025-05-13",
   "G02B, H04B, G02F",
   "[Photonic bandgap phase modulator, optical filter bank, photonic comput](https://patents.google.com/patent/US12085790B2/en) (US12085790B2, 2019-07-11); [Method for phase-based photonic computing](https://patents.google.com/patent/US10656336B1/en) (US10656336B1, 2018-11-08)",
   "→ ADVANCED MICRO DEVICES, INC."
  ],
  [
   "24",
   "POLYN Technology",
   "analog-imc",
   "16",
   "22",
   "3 / 19",
   "2020-06-25",
   "2025-06-10",
   "G06N, G06F, G06V",
   "[Optimizations for analog hardware realization of trained neural networ](https://patents.google.com/patent/US12327182B2/en) (US12327182B2, 2020-06-25); [Systems and methods for detonation control in spark ignition engines u](https://patents.google.com/patent/US11885271B2/en) (US11885271B2, 2020-06-25)",
   ""
  ],
  [
   "25",
   "GrAI Matter Labs (GML)",
   "digital-snn",
   "16",
   "17",
   "1 / 16",
   "2018-08-29",
   "2023-12-12",
   "—",
   "[Processing device and method of operating the same](https://patents.google.com/patent/US11842230B2/en) (US11842230B2, 2019-07-15); [Message based processor, message based processing method and record ca](https://patents.google.com/patent/WO2022219201A1/en) (WO2022219201A1, 2021-04-16)",
   "→ SNAP INC."
  ],
  [
   "26",
   "iniVation",
   "event-sensing",
   "14",
   "14",
   "5 / 9",
   "2018-04-27",
   "2026-03-17",
   "H04N, G06F, G06T",
   "[Switched capacitor circuit](https://patents.google.com/patent/US12206429B2/en) (US12206429B2, 2019-10-07); [Eye tracking device, eye tracking method, and computer-readable medium](https://patents.google.com/patent/US12474770B2/en) (US12474770B2, 2020-07-03)",
   "Toyota co-assigned on ~14%; acquired by SynSense 2024"
  ],
  [
   "27",
   "MemryX Inc.",
   "boundary",
   "14",
   "14",
   "6 / 8",
   "2019-05-07",
   "2026-08-04",
   "G06F, G06N, G11C",
   "[Memory processing units and methods of computing dot products](https://patents.google.com/patent/US10998037B2/en) (US10998037B2, 2019-05-07); [Memory processing units and methods of computing DOT products includin](https://patents.google.com/patent/US10853066B1/en) (US10853066B1, 2019-05-07)",
   ""
  ],
  [
   "28",
   "Koniku",
   "biological",
   "13",
   "16",
   "2 / 14",
   "2015-07-17",
   "2022-10-18",
   "G01N, C12M, B01L",
   "[Systems for detection](https://patents.google.com/patent/US11474103B2/en) (US11474103B2, 2017-02-17); [Stacked integrated platform architecture](https://patents.google.com/patent/US11336285B2/en) (US11336285B2, 2018-08-10)",
   ""
  ],
  [
   "29",
   "Mythic",
   "analog-imc",
   "12",
   "15",
   "8 / 7",
   "2014-08-14",
   "2026-01-20",
   "—",
   "[System and methods for mixed-signal computing](https://patents.google.com/patent/US11726925B2/en) (US11726925B2, 2017-09-15); [Mixed-signal circuitry for computing weighted sum computation](https://patents.google.com/patent/US10255551B2/en) (US10255551B2, 2017-02-07)",
   ""
  ],
  [
   "30",
   "CyberSwarm, Inc.",
   "emerging-device-imc",
   "12",
   "12",
   "3 / 9",
   "2017-10-31",
   "2024-08-20",
   "G06N, G11C, H10N",
   "[Synapse array](https://patents.google.com/patent/US12069870B2/en) (US12069870B2, 2018-06-11); [Method and system for data validation using memristors](https://patents.google.com/patent/US11455258B2/en) (US11455258B2, 2019-08-05)",
   ""
  ],
  [
   "31",
   "Eta Compute (now ModelCat)",
   "digital-snn",
   "12",
   "12",
   "7 / 5",
   "2014-07-16",
   "2025-08-05",
   "G06F, Y02D, G06N",
   "[Asynchronous processor that adjusts a respective operating voltage for](https://patents.google.com/patent/US9423866B2/en) (US9423866B2, 2014-07-16); [Self-timed processors implemented with multi-rail null convention logi](https://patents.google.com/patent/US10951212B2/en) (US10951212B2, 2017-04-10)",
   "→ ModelCat, Inc. (rename)"
  ],
  [
   "32",
   "Fractile",
   "boundary-digital",
   "10",
   "14",
   "0 / 14",
   "2024-07-26",
   "—",
   "G06F, G06N, G06J",
   "[Analog multiplier](https://patents.google.com/patent/US20260195550A1/en) (US20260195550A1, 2025-01-09); [Hardware architectures for efficient vector-matrix multiplication](https://patents.google.com/patent/WO2026022403A1/en) (WO2026022403A1, 2024-07-26)",
   ""
  ],
  [
   "33",
   "Aspinity",
   "analog-imc",
   "10",
   "10",
   "4 / 6",
   "2016-03-10",
   "2025-02-25",
   "G06F, B60R, G06N",
   "[Audio tone detection using analog processing](https://patents.google.com/patent/US12126969B1/en) (US12126969B1, 2024-03-18); [Analog voice activity detection](https://patents.google.com/patent/US10090005B2/en) (US10090005B2, 2016-03-10)",
   ""
  ],
  [
   "34",
   "ANAFLASH",
   "analog-imc",
   "9",
   "9",
   "5 / 4",
   "2017-11-29",
   "2026-07-15",
   "—",
   "[Neural network circuits having non-volatile synapse arrays](https://patents.google.com/patent/US11361216B2/en) (US11361216B2, 2017-11-29); [Multiply-accumulate unit](https://patents.google.com/patent/US12423056B2/en) (US12423056B2, 2019-09-17)",
   ""
  ],
  [
   "35",
   "AnotherBrain",
   "analog-imc",
   "9",
   "9",
   "2 / 7",
   "2017-04-28",
   "2024-07-10",
   "G06V, G06N, G06T",
   "[Automated method and device capable of providing dynamic perceptive in](https://patents.google.com/patent/US11164049B2/en) (US11164049B2, 2017-04-28); [Automated method and associated device for the non-volatile storage, r](https://patents.google.com/patent/EP3616131B1/en) (EP3616131B1, 2017-04-28)",
   ""
  ],
  [
   "36",
   "Mentium Technologies",
   "emerging-device-imc",
   "9",
   "9",
   "3 / 6",
   "2018-01-11",
   "2026-08-04",
   "G06F, G11C, G06N",
   "[Programming non-volatile memory arrays with automatic programming puls](https://patents.google.com/patent/US11170839B2/en) (US11170839B2, 2019-07-17); [Using data correlation to reduce the power consumption of signal proce](https://patents.google.com/patent/US11409356B1/en) (US11409356B1, 2019-11-20)",
   ""
  ],
  [
   "37",
   "Salience Labs",
   "photonic",
   "9",
   "9",
   "0 / 9",
   "2023-01-24",
   "—",
   "G02F, G06E, H04B",
   "[An optical coupling device](https://patents.google.com/patent/WO2024156984A1/en) (WO2024156984A1, 2023-01-24); [An optical coupling device with adjustable coupling coefficient](https://patents.google.com/patent/EP4655641A1/en) (EP4655641A1, 2023-01-24)",
   ""
  ],
  [
   "38",
   "SEMRON",
   "analog-imc",
   "9",
   "9",
   "1 / 8",
   "2019-11-21",
   "2025-01-21",
   "G06N, G11C, H10D",
   "[Capacitive synaptic component and method for controlling same](https://patents.google.com/patent/US12205637B2/en) (US12205637B2, 2019-11-21); [Arrangement for carrying out a discrete Fourier transformation](https://patents.google.com/patent/DE102021004453A1/en) (DE102021004453A1, 2021-09-01)",
   ""
  ],
  [
   "39",
   "AIStorm",
   "analog-imc",
   "7",
   "7",
   "3 / 4",
   "2018-03-02",
   "2025-11-25",
   "—",
   "[Charge domain mathematical engine and method](https://patents.google.com/patent/US12481868B2/en) (US12481868B2, 2018-03-02); [Charge based switched matrix and method therefor](https://patents.google.com/patent/US12061975B2/en) (US12061975B2, 2018-10-09)",
   ""
  ],
  [
   "40",
   "Ambient Scientific",
   "analog-imc",
   "7",
   "7",
   "5 / 2",
   "2018-03-16",
   "2026-06-09",
   "G06F, G11C, G01R",
   "[Dynamic adjustment of operating conditions of integrated circuits](https://patents.google.com/patent/US10503184B2/en) (US10503184B2, 2018-03-16); [Static power reduction in SRAM](https://patents.google.com/patent/US10832762B1/en) (US10832762B1, 2019-04-30)",
   ""
  ],
  [
   "41",
   "Cortical Labs",
   "biological",
   "7",
   "7",
   "3 / 4",
   "2018-11-21",
   "2026-03-25",
   "—",
   "[Biological computing platform](https://patents.google.com/patent/US11651188B1/en) (US11651188B1, 2018-11-21); [Closed-loop perfusion circuit for cell and tissue cultures](https://patents.google.com/patent/US12312575B1/en) (US12312575B1, 2019-07-01)",
   ""
  ],
  [
   "42",
   "Sync Computing",
   "probabilistic-thermodynamic",
   "7",
   "7",
   "3 / 4",
   "2020-04-13",
   "2025-01-28",
   "G06F, G06N",
   "[Optimization processing unit having subunits that are programmably and](https://patents.google.com/patent/US11604913B2/en) (US11604913B2, 2020-04-13); [Real time scheduling using expected application resource usage](https://patents.google.com/patent/US11966783B1/en) (US11966783B1, 2020-08-17)",
   "→ Capital One Services, LLC"
  ],
  [
   "43",
   "Sagence AI (fka Analog Inference)",
   "analog-imc",
   "6",
   "8",
   "5 / 3",
   "2018-06-26",
   "2026-03-17",
   "G06N, G11C, H03M",
   "[Methods and systems of operating a neural circuit in a non-volatile me](https://patents.google.com/patent/US12136028B2/en) (US12136028B2, 2018-06-26); [Methods and systems of cell-array programming for neural compute using](https://patents.google.com/patent/US11961570B2/en) (US11961570B2, 2018-06-26)",
   ""
  ],
  [
   "44",
   "MemSilicon (铭芯启睿; Beijing MemSilicon Technology Co.)",
   "emerging-device-imc",
   "6",
   "6",
   "0 / 6",
   "2025-10-23",
   "—",
   "G06F, G11C, H10B",
   "[A memory circuit structure for suppressing leakage current](https://patents.google.com/patent/CN121354607A/en) (CN121354607A, 2025-10-23); [NPU fault-tolerant computing method and system based on repeated execu](https://patents.google.com/patent/CN121901017A/en) (CN121901017A, 2025-12-15)",
   ""
  ],
  [
   "45",
   "Neurophos",
   "photonic",
   "6",
   "6",
   "1 / 5",
   "2019-08-04",
   "2023-12-12",
   "G02B, G06N, G06E",
   "[Computation with adjustable resonant optical metamaterials](https://patents.google.com/patent/US11843182B2/en) (US11843182B2, 2019-08-04); [Liquid crystal tunable single-coaxial and bicoaxial metamaterial eleme](https://patents.google.com/patent/US20230333431A1/en) (US20230333431A1, 2021-01-22)",
   ""
  ],
  [
   "46",
   "LightOn",
   "photonic",
   "5",
   "6",
   "2 / 4",
   "2015-02-04",
   "2022-04-26",
   "—",
   "[Method and system for machine learning using optical data](https://patents.google.com/patent/US11137289B1/en) (US11137289B1, 2020-03-11); [Methods for the stabilization of interferometric systems and interfero](https://patents.google.com/patent/US11313667B2/en) (US11313667B2, 2019-08-09)",
   ""
  ],
  [
   "47",
   "Lumai",
   "photonic",
   "5",
   "6",
   "0 / 6",
   "2023-07-20",
   "—",
   "—",
   "[Analog computing system and method of performing analog computing](https://patents.google.com/patent/EP4747714A1/en) (EP4747714A1, 2023-07-20); [Optical core system](https://patents.google.com/patent/GB202417552D0/en) (GB202417552D0, 2024-11-29)",
   ""
  ],
  [
   "48",
   "Blumind",
   "analog-imc",
   "5",
   "5",
   "2 / 3",
   "2021-05-07",
   "2025-06-03",
   "H04N, G06F, G06N",
   "[System, method, and computer device for transistor-based neural networ](https://patents.google.com/patent/US11636326B2/en) (US11636326B2, 2021-05-07); [Systems and methods for analog image sensing and processing](https://patents.google.com/patent/US12323724B2/en) (US12323724B2, 2022-02-17)",
   ""
  ],
  [
   "49",
   "iHW (Intelligent HW)",
   "analog-imc",
   "4",
   "5",
   "0 / 5",
   "2024-02-28",
   "—",
   "—",
   "[Memory device with switching characteristics and method of operation t](https://patents.google.com/patent/US20250275492A1/en) (US20250275492A1, 2024-02-28); [Analog computing method and apparatus for performing deep neural netwo](https://patents.google.com/patent/US20250348726A1/en) (US20250348726A1, 2024-05-10)",
   ""
  ],
  [
   "50",
   "Ceryx Medical",
   "analog-imc",
   "4",
   "4",
   "0 / 4",
   "2019-09-10",
   "—",
   "A61N, A61B, A01N",
   "[Pacemaker device](https://patents.google.com/patent/US20250018196A1/en) (US20250018196A1, 2019-09-10); [Controller for a stimulation device](https://patents.google.com/patent/ES2952383T3/en) (ES2952383T3, 2021-02-01)",
   ""
  ],
  [
   "51",
   "femtoAI (fka Femtosense)",
   "digital-snn",
   "4",
   "4",
   "1 / 3",
   "2020-07-09",
   "2025-12-16",
   "G06N, G06F, H03M",
   "[Methods and apparatus for matrix and vector storage and operations](https://patents.google.com/patent/US12499174B2/en) (US12499174B2, 2020-07-09); [Methods and apparatus for system-on-a-chip neural network processing …](https://patents.google.com/patent/US20230133088A1/en) (US20230133088A1, 2021-11-01)",
   ""
  ],
  [
   "52",
   "FinalSpark",
   "biological",
   "3",
   "3",
   "0 / 3",
   "2018-09-08",
   "—",
   "—",
   "[Microfluidic system for robust long-term electrical measurement and/or](https://patents.google.com/patent/US20240425787A1/en) (US20240425787A1, 2021-10-21); [Methods and system for remotely interacting with biological neural net](https://patents.google.com/patent/US20250124250A1/en) (US20250124250A1, 2018-09-08)",
   ""
  ],
  [
   "53",
   "Vivum AI (Vivum Computing)",
   "digital-snn",
   "3",
   "3",
   "1 / 2",
   "2006-02-22",
   "2010-11-09",
   "A61B, G01N, G21F",
   "[Method and device for analyte measurement](https://patents.google.com/patent/US7830519B2/en) (US7830519B2, 2006-02-22); [Method for the radiological characterization of the radiation inventor](https://patents.google.com/patent/DE102019115733A1/en) (DE102019115733A1, 2019-06-11)",
   ""
  ],
  [
   "54",
   "General Vision Inc.",
   "digital-snn",
   "2",
   "2",
   "0 / 2",
   "2005-06-30",
   "—",
   "—",
   "[Monolithic Image Perception Device and Method](https://patents.google.com/patent/US20110052011A1/en) (US20110052011A1, 2005-06-30); [Sensor unit for environmental observation with neural processor](https://patents.google.com/patent/DE102008052160A1/en) (DE102008052160A1, 2008-10-20)",
   ""
  ],
  [
   "55",
   "Intrinsic Semiconductor Technologies Ltd (branding also \"Intrinsic AI Solutions\")",
   "emerging-device-imc",
   "2",
   "2",
   "0 / 2",
   "2025-03-27",
   "—",
   "—",
   "[Method for resistive random access memory devices](https://patents.google.com/patent/GB202504516D0/en) (GB202504516D0, 2025-03-27); [Resistive memory elements](https://patents.google.com/patent/GB202509858D0/en) (GB202509858D0, 2025-06-20)",
   ""
  ],
  [
   "56",
   "Normal Computing",
   "probabilistic-thermodynamic",
   "2",
   "2",
   "0 / 2",
   "2022-10-14",
   "—",
   "G06F, G06N, G01R",
   "[Thermodynamic computing system for sampling high-dimensional probabili](https://patents.google.com/patent/WO2024081827A1/en) (WO2024081827A1, 2022-10-14); [Thermodynamic artificial intelligence for generative diffusion models ](https://patents.google.com/patent/WO2024118915A1/en) (WO2024118915A1, 2022-12-02)",
   ""
  ],
  [
   "57",
   "MemComputing, Inc.",
   "adjacent",
   "1",
   "2",
   "0 / 2",
   "2021-12-28",
   "—",
   "—",
   "[System on chip self-organizing gates and related self‑organizing logic](https://patents.google.com/patent/WO2023129466A1/en) (WO2023129466A1, 2021-12-28)",
   ""
  ],
  [
   "58",
   "OLIX Computing (fka Flux Computing)",
   "photonic",
   "1",
   "2",
   "0 / 2",
   "2024-08-08",
   "—",
   "G02F, G02B",
   "[Method and system for optical computing](https://patents.google.com/patent/WO2026033226A1/en) (WO2026033226A1, 2024-08-08)",
   ""
  ],
  [
   "59",
   "Aspirare Semi",
   "analog-imc",
   "1",
   "1",
   "0 / 1",
   "2024-08-22",
   "—",
   "G06J, G06N, H03M",
   "[Electronic circuit and device for computation](https://patents.google.com/patent/US20260057197A1/en) (US20260057197A1, 2024-08-22)",
   ""
  ],
  [
   "60",
   "EnCharge AI",
   "analog-imc",
   "1",
   "1",
   "0 / 1",
   "2023-12-04",
   "—",
   "—",
   "[Systems and methods for high-resolution, high-speed, analog voltage de](https://patents.google.com/patent/WO2025122596A1/en) (WO2025122596A1, 2023-12-04)",
   ""
  ],
  [
   "61",
   "GEMESYS",
   "emerging-device-imc",
   "1",
   "1",
   "0 / 1",
   "2023-08-02",
   "—",
   "G06F",
   "[Method for producing an electronic component based on phase change mat](https://patents.google.com/patent/DE102023120474A1/en) (DE102023120474A1, 2023-08-02)",
   ""
  ],
  [
   "62",
   "Opticore (Opticore AI / Opticore Inc.)",
   "photonic",
   "1",
   "1",
   "0 / 1",
   "2023-10-13",
   "—",
   "—",
   "[Hyper-multiplexed homodyne photonic circuits for high-throughput, ener](https://patents.google.com/patent/EP4792220A1/en) (EP4792220A1, 2023-10-13)",
   ""
  ],
  [
   "63",
   "Sparxity (formerly SpiNNcloud Systems; renamed 2026-10-28)",
   "digital-snn",
   "1",
   "1",
   "0 / 1",
   "2022-01-28",
   "—",
   "—",
   "[Method and hardware architecture for event-based classification of …](https://patents.google.com/patent/WO2023144396A1/en) (WO2023144396A1, 2022-01-28)",
   ""
  ],
  [
   "64",
   "Arago",
   "photonic",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "65",
   "Bio Inspired Technologies",
   "emerging-device-imc",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "66",
   "The Biological Computing Co. (TBC)",
   "biological",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "67",
   "Grayscale AI",
   "digital-snn",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "68",
   "Great Sky",
   "photonic",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "69",
   "Neurobus",
   "digital-snn",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "70",
   "Neuromorphyx",
   "digital-snn",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "71",
   "Neuronova",
   "analog-imc",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "72",
   "Unconventional AI (Unconventional, Inc.)",
   "probabilistic-thermodynamic",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "73",
   "Vaire Computing",
   "probabilistic-thermodynamic",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ],
  [
   "74",
   "Zhizi Guangyuan (智子光垣; Zhizi Guangyuan Technology (Shanghai) Co.)",
   "photonic",
   "0",
   "0",
   "0 / 0",
   "—",
   "—",
   "—",
   "—",
   "no rows under own name"
  ]
 ]
}