About

Hi. I'm Alex Nugent, inventor, co-founder and CEO of Knowm Inc. I started working on physical neural networks before I (and most of the world) knew the word memristor, and I have spent a significant fraction of my adult life working on the problem of memristor machine intelligence. I have a story to tell, and this is where I will tell it.

knowm.ai is a long-form blog series of Knowm's work on memristor machine learning: the theory, the mistakes, the patents, the lab results, the politics, the software, the hardware, and the stubborn thread of persistence that connects all of it. It has not been an easy journey, and its still ongoing.

Knowm has been early for a long time. When I first started down this path, in 2002, machine learning could not reliably tell a dog from a cat and Intel had just gone to two cores for the first time. Machine learning had not yet become the center of the technology world and AI was entirely science fiction. The energy wall was visible and obvious if you were looking far enough ahead, but it was not yet painful enough for most people to care. Memristor synaptic accelerators sounded like a solution in search of a problem, and in all honestly, they were.

That has finally changed.

Today we have enormous models that cost more to train than it costs to build a custom chip. We have exploding inference demand, memory bottlenecks (and chip shortages), energy constraints, and a world trying to push intelligence into every device it can reach. The energy wall has been hit. The question now is whether we keep hauling learned weights through memory hierarchies forever, or whether we build hardware that can hold adaptive physical state where computation actually happens. Physics is not ambiguous, machine learning has demonstrated its value, and nature has a reference implementation. What remains is the part I've mostly lost patience for: capital defending itself while it eats the world, institutions confusing status with truth, egos mistaking ownership for understanding, and the VC and government funding hustle that turns the joy of creative technology development into a performance act for rich people controlling richer people's money.

Knowm Inc. was founded in 2017, after more than a decade of prior work, to help move computing toward memristive processors: machines where memory, adaptation, and computation collapse into the same physical substrate. Along the way, Knowm became the first company, and as far as I know still the only company, commercially providing memristors to researchers, educators, and students around the world. That may not continue. The remaining supply of Knowm SDC memristors is finite, and the path from research device to broad manufacturing requires resources far beyond what a small company in Santa Fe can carry alone. I don't want to spend my remaining days on earth pitching the government for money in exchange for bringing value to the war-fighter. Nor do I want to find new and creative ways to pitch the ultra wealthy on how they can get even wealthier. I am not a good buffer between the technology and human bullshit. Before this chapter closes, I want to leave a clear record of what we built, what we learned, what worked, what failed, and where I believe the technology needs to go next.

Knowm memristors have helped catalyze a surprisingly large body of academic work. As of May 2026, a Google Scholar search for "Knowm memristor" returns roughly 589 papers and citations across chaotic circuits, memristor logic, neuromorphic signal processing, seizure detection, threshold logic, reservoir computing, structural identification, benchmarking, and related devices. Those papers appear across IEEE venues, Electronics Letters, the International Journal of Bifurcation and Chaos, Nature-affiliated journals, and many conference proceedings. That work matters to me. I overcame significant hurdles and personally packaged and shipped every single chip so that others could have access to a device that may otherwise have been locked away. The devices left the slide deck, hit labs, and helped move people past what they wanted to be true and closer to what is true. They were tested, abused, modeled, measured, and built into things I never would have thought to build myself. My own contribution to the literature ended with our PLOS One AHaH Computing paper way back in 2014. I have kept my work mostly private since then even though I have had a number of breakthroughs. That ends today.

This site will move between personal history and technical detail because, in this case, they are not separable. The ideas came out of real devices, accidents and failures, government programs, small-business survival, academic collaborations and betrayals, and years of stubbornly refusing to give up.

The work also moved through a world shaped by money, power, politics, fear, status, and access. I have had my work denigrated in public and taken in private, sometimes by the same people at the same time. I have heard from intelligence agencies, missile-defense contractors, religious strangers convinced the work would end the world, and investors who promised millions, obtained technical information and then ghosted. Heck, at one point memristors even ended up in a high-speed police chase and shootout. I've got a lot of stories to tell.

knowm.ai exists because I want to tell the story and because I want Knowm IP to find a new home with an organization capable of carrying it faster and farther than I can. This work needs serious resources. It needs people who can span the stack from materials and fabrication to architecture, ML systems, software, infrastructure, and product deployment. I have learned that I do not want the version of the CEO job this climate rewards. I am happiest in my labs in Santa Fe, building, writing, and thinking with the last of Knowm's SDC memristors, a few wirebonders, and whatever AI and cloud resources I can pull together. Until the right home appears, I am going to develop in the open, one article at a time, and tell the untold story of Knowm.

Anyone looking for a polished corporate narrative and the stamp of name-brand approval will probably hate this. This is for people who suspect the next phase of AI will require a different relationship between memory, computation, and physical adaptation. This is for people who can still read and think for themselves.

The short version is this: I am going to build the damn thing and tell you how I got here. I hope you follow along.

Blue Skies,

Alex