Wanted to save the bookmark of this dude’s website for future reference: http://sam.zeloof.xyz/
He also has a youtube channel. The gist of it is that he does photolithography, the process by which chips are manufactured, in his garage. This demonstrates that small chip manufacture is possible. Not top of the line EUV chips, but some old chips, that are still useful, can be produced without capital investment in the order of millions of dollars.
Related, the RISV-V chip architecture is relevant here. On the one hand, ARM and NVDA architectures are proprietary and costly, at the same time as the world is inter-connected enough that “free and open” software efforts are increasingly powerful and useful. On the other hand, RISC-V is a CPU architecture that allows peer contribution, and lacks certain licensing restrictions, to finally offer the architecture and platform to, possibly, maybe, even rival NVDA’s CUDA.
To pursue the goal of developing and using RISC-V for AI, the company Tenstorrent has received $100M in strategic funding. You can take a look at what they do here: https://tenstorrent.com/
If chips are indeed the future of everything, does that mean that now is the time to attempt to seriously boost production of *tools to produce* the chips (not the chips themselves)? And if so, how complicated is that, can a garage-style small-manufacture plant actually create chips that are useful in actual appliances?
RISC-V is an open CPU architecture. NVDA is a monopoly, so everyone wants to *not* use NVDA but unfortunately that’s not possible. NVDA’s CUDA driver for AI is proprietary, and the entire AI industry runs on it.
Then, the transition from DUV to EUV, smaller light beams for lithography, means that (1) ASML is the monopoly manufacturer of these, and (2) I think you don’t need the latest and best, to make the world turn. Tesla itself is phasing out the newest chips for slightly outdated chips, and if EUV chip-printing machines are unavailable, everyone can still pretty much use the old tech. For example my two cars are 1996 and 1999, that’s 1/4 of a century old and they work just fine.
All in all, these are quite interesting developments! I look forward to observing the next steps in the AI evolution, and hopefully being a part of it somehow. It appears to be a lot of work but then again – if our collective future livelihood critically relies on this sub-segment of technology, it makes sense to pour a lot of effort into it.