Rights for AI and Hardware Democratization
June 26, 2023

The AI era needs hardware rights: the right to tape out a custom supercomputer chip, build trusted open hardware, and run a model locally at a fraction of centralized GPU-cluster cost.

Brian Greenforest connects those rights to four years of open-FPGA work and a two-material copper-and-glass switching invention designed to move active-device fabrication closer to ordinary builders.

Open Hardware Must Reach Beyond Soft IP

Open RTL still depends on proprietary hard IP, especially SerDes blocks, closed process knowledge, and long semiconductor supply chains. Those dependencies limit security, repair, local manufacturing, and the economics of claytronics.

Brian’s FPGA fabric accepts a soft-IP bitstream and organizes reconfigurable computation, while his materials research pursues nonlinear switching without silicon, cobalt, molybdenum, or iron.

Fund the Team That Can Build the Complete Stack

Work on magnetic amplifiers and older RF mechanisms led to a nonlinear two-material system using copper and glass. That path joins device physics, circuit architecture, open tools, and local fabrication.

The invention now needs an angel investor, commercial leadership, marketing, scientists, and engineers. Join the team that can turn hardware rights into a manufacturable AI computing platform.

Build the Fabrication Path Behind Rights for AI and Hardware Democratization

The Missing Maker Fab carries this work into the tools and processes required to fabricate fast active devices locally.

The Missing Maker Fab · https://lnkd.in/g49pk3WG

Originally posted on LinkedIn

Brian Greenforest · (2023-06-26 17:22:09 UTC)

Open the original LinkedIn post · LinkedIn activity 7079146835960889344

LinkedIn status when archived: Edited · Visible to anyone on or off LinkedIn.

In the wild era of AI, ML, LLM, GPT #democratization, individuals and businesses of all sizes deserve the following #rights: 1. The right of #fabrication (#tapeout) of their own custom supercomputer chip. 2. The right to build their own secure, trusted, #open, tinkerer-friendly, #diy #hardware. 3. The right to run an ML model at a fraction of a cost at their #makerspace, home, or on premises on that custom hardware they built (comparing to what they can afford on an NVIDIA GPU cluster). For 2.5 years I've being reinventing the wheel, a special new #fpga fabric #asic IP (intellectual property) optimized to be configured from a Soft IP bitstream. Hard IP (especially SERDES) is the nastiest obstacle for #opensource #hardware in my 4-years YosysHQ ride on the surf of the #fpgas made by Xilinx, Lattice Semiconductor, Efinix, Inc., QuickLogic Corporation... Actually.... A single-crystal pure silicon wafer is so slow to grow. The "Fairchild path" stops now. We went way too far. #chipshortage shows how ridiculous the problem is... While closed-source "silicon mafia" is around, the #claytronics dream of seth goldstein, the Terminator 2 liquid metal swarm robotics won't be economically feasible. The whole year of 2022 I've being researching the 200 years of the switching circuits history, from Michael Faraday's coil to these days of Geoffrey Hinton's backprop, and Ilya Sutskever's DL and GPT. I dove into the practical aspects of magnetic amplifiers and built some fascinating prototypes; I found you don't even need diodes, nor any silicon whatsoever. I learned more and found that I don't need cobalt, molybdenum, not even iron. Iterating on top of some ancient RF ideas, I FOUND A WAY TO ENABLE NONLINEARITY IN A 2-MATERIAL SYSTEM OF COPPER AND GLASS. The problem? Nobody believes me 😂 and nobody knows what I was actually working on for the last 20 years. I always had too big of an ego. I was too much "all over the place." I always had too weak communication skills. And in America, too many cultural clashes, and my pathological inability to listen without interrupting. But! I'm a good inventor. And I'm trying to improve my personality and Western social skills 🧘‍♂️. And I'm looking for an 😇 #angelinvestor who can believe. And for a marketer. And for a CEO. And a whole team ❤️ of awesome engineers and scientists! Please, save me 🙏 Someone?.. #intellectualproperty #fundraising #venturecapital #chips #circuitdesign #asic #semiconductormanufacturing The situation with Hard IP blocks vs. Soft IP: https://lnkd.in/g49pk3WG

Comments added by Brian Greenforest on LinkedIn

These 7 comments were also preserved verbatim from Brian Greenforest’s LinkedIn data export or the public post page.

Comment 1 · (2023-06-29 00:33:32 UTC)

You are 100% right about the ego. I need a CEO. 1-3 is economically feasible due to tremendous reduction in the cost of one transistor. A fundamental breakthrough above the Moore's Law.

View the LinkedIn post

Comment 2 · (2023-06-29 00:59:43 UTC)

My pitch is: 1) I can make chips without silicon wafers. 2) I can make supercomputing data centers without Infiniband or any Bell system rack space management, using the same modular motherboard as its ultimate—literally flexible PCB—fabric. 3) I invented a post-von-Neumann's computer architecture that supercedes GPUs and FPGAs, and brings parallel distributed as first class citizen. The pitch combines all these three quite significant inventions to revolutionize how we design and deploy compute systems. This is the future. Perhaps, it's percieved as "too early", and I have nowhere to hurry up, besides the physical constraints of our frail bodies...

View the LinkedIn post

Comment 3 · (2023-06-29 01:00:49 UTC)

Oh, and I've spent 20 years nonstop inventing these.

View the LinkedIn post

Comment 4 · (2023-06-29 15:57:49 UTC)

The model is open hardware, protected by prior art. Not documented yet, what I'm working on right now is a render of a significant size dynamic parallel circuit instantiation to use as a marketing pitch. A first prototype simulator that had that core idea working I published on ShaderToy in Fall 2021. A ported version you can see here. I know it won't make sense, because there are no docs yet, and no solid developer-friendly demo. https://greenforest.io/cellular-automata-2019/cartilage3

View the LinkedIn post

Comment 5 · (2023-06-29 16:03:13 UTC)

As of siliconless switches, I worked several months to invent a magnetic amplifier that doesn't need diodes to operate. Turned out that: a) Wheatstone bridge is a must; b) The load core must have a gap to collapse the flux so it can drive a cascate of controlling coils. After finalizing that research, Corbino disc came up. Dig in, it's so simple, anyone can make it. We live in the open world nowadays, and only limited by our egos and abilities to form communities of people who will follow. As of the flex PCB fabric, it's quite clever, and I might patent it after all, because I never published it openly anywhere yet, no "risk" of prior art. Can show in a video call an example in my hands.

View the LinkedIn post

Comment 6 · (2023-07-02 00:03:25 UTC)

There are attempts to revive the hidden gems of the 1950s. Magamp wasn't the only technology that is completely feasible for DIY circuits manufacturing without silicon fabs. https://arxiv.org/pdf/2012.00304.pdf

View the LinkedIn post

Comment 7 · (2023-07-02 04:26:46 UTC)

Unless the main FPGA fabric committed is a cellular automata with switch routing instead of long-distance span wires. More sequential fused-multiply adders for Transformer model matrix multiplications all over the place, while the models can evolve for ever. No need to buy new GPUs ever again.

View the LinkedIn post