Open Silicon: D-Day or Market Garden?
May 24, 2023
by Steve Goldsmith
I have tried to keep my mouth shut for the past year about the state of the open silicon community, but the new contest by efabless has my blood boiling.
I tried to be subtle by letting a chatbot be the bad cop in the Emperor has No Clothes post, but I've got nothing to lose at this point.
There is a total lack of candor at the heart of the community, namely efabless and YosysHQ. So I'll say it clearly.
efabless has a demonstrated history of not being candid about the state of the Caravel platform and should have lost the trust of the community long ago.
There are no open source formal verification tools. SymbiYosys is not free and not open source. You have to be able to write concurrent SVA assertions to actually have a silicon development process. Read this book if you don't believe me.
The first phase of open silicon is over. I compare it to the Market Garden Operation of WW2. Multiple single points of failure, and moderate success at great cost.
The US, EU, and more are pouring >$100B total into Chips Acts with $100Ms going into open source and academia. This should be our D-Day moment; with the lessons learned from over-relying on single points of failure leading to multiple solutions at every level of the stack. But when I look at most HDL and open EDA projects almost all rely on yosys for synthesis.
The open sourcing of PDKs themselves is the real gift to the community. Let's stop acting like efabless can deliver on their promises.
If you think I am being unreasonable, please reach out and I'd love to have a conversation and be convinced I am wrong, but for now I am focused more on how I can take the few resources I have and do something of import rather than wasting my time with Caravel in exchange for a "free" tapeout.
And if you agree, I suggest picking a weak part of the stack and going directly after grant money to pay for a tapeout from a real MPW service provider. A tapeout at 350nm is as cheap as $6000 from Europractice.
The three I have found that work with academia are Muse, Europractice, and CMC.
How much is your time worth?