The CHIPS Alliance

On September 17, there was a meeting of the CHIPS Alliance. It was online, of course. In three hours, there were ten presentations. CHIPS is actually an acronym, Common Hardware for Interfaces, Processors, and Systems. The presentations, and The CHIPS Alliance itself, were focused on open-source hardware and on open-source EDA tools.

To give you a better idea of the range of topics covered, I'll list all of them, but then I'll focus on just a couple of the presentations.

• Open Design Verification, by Tao Liu of Google
• Enabling Fully Open Source and Continuous Integration-Drivern Flow for ASIC and FPGA Development, by Michael Gielda of Antmicro
• The Emergence of the Open-Source AIB Chiplet Ecosystem, by David Kehlet of Intel
• Chipyard: Design of Customized Open-Source RISC-V SoCs, by Borlvoje Nicolic of UC Berkeley
• SweRV and OmniXten Milestones, by Zvonimir Bandic of Western Digital
• Chisel and FIRTL for Next-Generation SoC Designs, by Jack Koenig of SiFive
• Open ML Accelerator, by Anoop Saha of Mentor
• Cloud-Based Verification of RISC-V Processors, by Dan Ganousis of Metrics
• OpenROAD Open RTL-to-GDS Update, by Andrew Kahng of UCSD and Mohamed Kassem of Efabless
• Open Source FPGA Tooling, Our Journey from Resistance to Adoption, by Brian Faith of Quicklogic

I felt that the workshop was like the first time I heard about RISC-V or Linux, very early in their life cycle, when many people were dismissive that they would ever amount to much outside of academia. I've been skeptical about open-source EDA projects myself for a couple of reasons.

Firstly, open source seems to work best when the software engineers are working on tools for their own use, meaning that they don't require one group of engineers to specify what is needed and another set of engineers to do the implementation. The skills of a good design engineer are very different from the highly specialized skills and knowledge required to develop an EDA tool like a static timing engine or high-level synthesis.

The second challenge with open source is that the engineers need to get paid. In practice, outside of academia, that means that there needs to be some way to make money other than selling the software. For example, users of Linux generally don't pay a lot for it (and if they are big users they set up their own development group). The cloud and supercomputers all run on Linux, so that is where the money is made. Having a small team of engineers contributing to Linux is a rounding error in the business.

Many of the presentations above have a business model something like that. For example, Western Digital makes most of its money selling disk drives, both SSD and rotating media. It has created and open-sourced its RISC-V processors since it is not a competitive advantage worth keeping secret. If other organizations contribute, its costs go down and time to market may be accelerated. Intel is interested in having a chiplet ecosystem leveraging their advanced packaging and their FPGA product line that uses chiplet-based interfaces (that is formerly Altera). Quicklogic makes its money selling FPGAs, and if it doesn't need to develop and maintain a complete software ecosystem, that drives down its costs and derisks the business (many many FPGA companies over the years have failed due to their software stacks being non-competitive).

Quicklogic

I'm going to focus on Quicklogic's presentation, which happened to be the last of the day. It was by Brian Faith (CEO) and titled Open Source FPGA Tooling: Our Journey from Resistance to Adoption. The reason for picking this presentation is that most of the other presentations were further from real deployment.

Brian opened with a bit of the history of Quicklogic. They actually were founded in 1989. At the time, their foundry was VLSI Technology, where I worked at the time. He also pointed out that over the last 30+ years, over 60 programmable logic companies have come and gone. Almost all of them failed due to software. And those that didn't fail, including the obvious big names, ended up with a walled garden, software that could only be used for that particular product line. In the meantime, various groups built open-source FPGA software, even though they lacked access to the detailed underlying architecture.

Eventually, Brian was persuaded to take a look at this software and discovered that it was good. and it would be even better if the programmable logic companies made the underlying architecture available to the programmers. This led to QORC, the QuickLogic Open Reconfigurable Computing.

Brian says that Quicklogic is the first programmable logic company to actively collaborate to create a fully open-source suite of development tools for their FPGA and eFPGA technology, all the way from RTL to bitstream with complete architecture and timing support. They are now fully committed to open-source FPGA tools, and more companies are getting involved.

Brian's conclusion:

• The open-source community's persistent and passion is remarkable
• Open-source FPGA tools are ready for prime time
• There will be a time in the not-so-distant future when this will be the norm not the exception

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.