Why not Rockets for the rest of us first, if that's easier?
The difficulty is not (entirely) technical
Link to the Zero to ASIC course that they are collaborating with: https://www.zerotoasiccourse.com/digital/
I wish for free alternatives to these.
https://github.com/shailja-thakur/VGen
Earlier from the NYU (2023)
https://zenodo.org/records/7953725
Related (?) blog post (2023)
I expect this will become the norm in a number of fields. Perhaps COBOL is next?
Perhaps it's the first open one. I was an eng manager at a hyperscaler helping one of our clients, a large semiconductor design company, build models to use internally. It was trained on their extensive Verilog repos, tooling, and strict style guides. I see this being repeated across industries, at least since 2023 there are quite a few deep-pocketed S&P 500 orgs creating models from scratch or extensively finetuning to give unique advantages they require. They're rarely announced specifically, but you can often infer from the initial investment or partnership announcements that they're working on it.
What?
Basically.
A lot of my questions went away when I got to this line though:
> He’s also fully engaged in the third leg of the “democratizing chip design” stool: education.
This is a valiant effort. Chip design is a hard world to break into, and many applications that could benefit from ASICs aren't iterating or testing on it because it sucks to do. It's a lot of work to bring that skill ceiling down, but as a programmer I could see how an LLVM-style intermediate representation layer could help designers get up-and-running faster.
That's because we don't need more digital. Digital transistors are effectively free (to a first approximation).
The axes that we need more of involve analog and RF. Less power consumption, better RF speed/range, higher speed PCI, etc. all require messy analog and RF design. And those are the expensive tools. Those are also the complex tools require genuine knowledge.
Now, if your AI could deliver analog and RF, you'd make a gazillion dollars. The fact that everybody knows this and still haven't pulled it off should tell you something.
Being a software supplier to fabless semiconductor companies is a very profitable business.
In the Gold Rush, the people who came out rich were selling the shovels and denim.
I am actually astonished. Is this what happens when the NYU board of directors tells every department they have to use and create AI, or they will stop funding? What is going on?
Improving the lived experience keeping it real! Feels so much more authentic.
More people would love AI if it communicated like an emo *Nix elitist. Train it on Daria, Eeyore, and grunge lyrics! People will love it!