4
\$\begingroup\$

The Ethereum Foundation (an open-source project) will build an open-source ASIC to support its decentralised randomness beacon.

To date, has the RTL of any ASIC been open-sourced or will the Ethereum Foundation set a precedent?

\$\endgroup\$

5 Answers 5

Reset to default
10
+100
\$\begingroup\$

Short answer: No. It is not a precedent.

  • There are plenty of Open Source RTLs out there. OpenCores and OpenSPARC are examples, but I know of international academic projects that at the very least were attempting this since the 1990's, almost since the invention of Verilog and before the concept of OpenSource was widespread. The project you pointed to seems no different.
  • Unless you are actually talking of making your own hobby foundry and you actually consider this foundry (and Arduino) "OpenSource" (instead of OpenHardware) you cannot have open source ASICs. ASICs are physical things that are quite expensive to make.
  • A group of people can organize together get an existing OpenSource RTL and combine it with an Open Cell Library (like this "fictional" one, or this one that people actually use) and use a fab (nearly all fabs like XFAB or TSMC offer this service) to contract a Multi-Project wafer run for $5000 or so that would be fully "OpenSource" free of all proprietary elements (assuming that you can actually avoid the specific requirements of that Foundry's process and that you did not leave any mistakes in your design).
  • But guess what. An academic collaboration called MOSIS that is now a stand-alone company has been doing this for more than 30 years. Together with NCSU you can actually have a full PDK that would allow you to design and verify your design before it goes to the foundry. Although initially cell libraries and PDKs would have to be made by hand, enough graduate students came together across the world and we got an OpenSource PDK. But this is not truly OpenSource as it requires Cadence, about $5000/seat/month program to run (which is free to universities, of course).
  • Due to that cost, a hobbyist group would probably be better off putting together the OpenSource RTL with a proprietary Foundry's cell library and PDK with some foundry-supported cheaper software. So much for the OpenSource PDKs.

So in short, OpenAnything is not new. As many other ideas, like the Internet or the web, it actually starts from academic collaborations that later became inclusive of the whole public and took a life of their own.

So no. They will not be the first. And no. You cannot have "OpenSource ASICs." That simply does not make sense.

The Ethereum Foundation can simply do what all of us mere mortals do and grab and OpenSource RTL and load it into an FPGA. But that will not be an "ASIC."

The Ethereum Foundation can contact one of the many fabless houses or fabs that offers the full service of RTL-to-ASIC to do that part of the work for them, but that would not be "OpenSource."

But if that is not enough, here is some people that went all the way to ASIC, released what was basically a single hobbyist labor of love as a product, and then went back to FPGA so that it could actually be released as OpenSource. So no, they will not be the first.

\$\endgroup\$
2
  • \$\begingroup\$ Wow man! This is a heck of complete and elaborate answer! Thanks, +10 if I could. \$\endgroup\$
    – Ale..chenski
    Nov 5, 2018 at 18:26
  • \$\begingroup\$ How does an "open source ASIC" not make sense? \$\endgroup\$
    – Melab
    Dec 9, 2022 at 18:40
5
\$\begingroup\$

"manufactured ASIC" is a compilation of RTL (which can be open) and particular FAB-supplied libraries of basic elements, and usually third-party I/O pad libraries, for which you need to pay big dollars. These are properties of FAB houses, and they differ in time and evolve with process node. The RTL can be "open sourced", but you need to work hard to assemble the rest (translate, place, route, generate photomasks, etc) . And again you have to pay big dollars for software tools to do all this stuff. And pay a lot to seasoned engineers who manage the whole process. The whole premise of "open-source ASIC" doesn't make sense.

\$\endgroup\$
1
  • \$\begingroup\$ Which part makes "open source" an inapplicable term? \$\endgroup\$
    – Melab
    Dec 9, 2022 at 18:55
1
\$\begingroup\$

Well, there's all the stuff at OpenCores.

However, in my opinion there's no direct equivalent to "Free Software" for hardware, because the freedom to modify it yourself as an end-user doesn't really exist. You can't sensibly get them manufactured in single quantities. Much of the tooling is closed-source, and the technology libraries are usually proprietary to the foundry.

\$\endgroup\$
2
  • 2
    \$\begingroup\$ I'm looking for manufactured ASICs (as opposed to cores). \$\endgroup\$
    – Randomblue
    Oct 26, 2018 at 14:41
  • 1
    \$\begingroup\$ There is very little of any value whatsoever on OpenCores. Lots of abandoned half-functioning, undocumented, untested and unsupported logic. I've wasted so much time trying to get anything from that site to work correctly that I'd have saved a lot of time just reimplementing it from scratch on my own. Even basic SPI or I2C cores are downright wrong on that site. \$\endgroup\$
    – akohlsmith
    Nov 3, 2018 at 1:38
1
\$\begingroup\$

Sun open-sourced the code for the UltraSPARC T1 in 2006 and the code for the UltraSPARC T2 in 2007. Both were released under the GNU General public license v2.

Source: https://en.wikipedia.org/wiki/OpenSPARC

\$\endgroup\$
4
  • 1
    \$\begingroup\$ As the other answer alludes to, the RTL is only part of the story. It's not the case that being in possession of the RTL gives you, the (wo)man on the street, direct access to the ASIC. \$\endgroup\$
    – awjlogan
    Oct 26, 2018 at 14:43
  • \$\begingroup\$ @awjlogan Your point is valid, of course. (There are brokers to FABs and you can buy "just one wafer" as part of that. Last I called about it, I found US$5k provided the basics for getting an 8" wafer done. That was 2006; so a while ago.) But I think the OP did answer one of the two precise questions the OP asked. The OP wasn't asking about whether or not the concept was good, but was asking a much more specific pair of questions. Still, I suppose the OP should update their answer to give us the final dope on the second part of the question: "will the Ethereum Foundation set a precedent?" \$\endgroup\$
    – jonk
    Oct 26, 2018 at 14:58
  • \$\begingroup\$ @awjlogan Why is that not the case? \$\endgroup\$
    – Melab
    Dec 9, 2022 at 18:56
  • \$\begingroup\$ @Melab - this is pretty well covered in the answers; essentially RTL describes logical function, but doesn't involve the physical cells, hard IP (particularly PHYs) etc. If you have core A, you'll need to implement it in a physical form compatible with the fab, which is usually under NDA. There are some initiatives to give open Product Development Kits (PDKs), but even that will tie you to a particular fab. For a given piece of RTL, there will be a lot of surrounding IP and physical IP required. \$\endgroup\$
    – awjlogan
    Dec 10, 2022 at 20:37
1
\$\begingroup\$

The OpenCelerity project manufactured a TSMC 16nm "Open-Source RISC-V Tiered Accelerator Fabric SoC".

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.