ICCAD and Open-Source CAD

Every year in November it is ICCAD, officially the International Conference on Computer-Aided Design. This year was the 37th, presumably meaning that the first one was in 1981. This year it was held in San Diego. I wasn't able to go, but several people from Cadence were there, notably Anton Klotz who sat down with me to tell me what went on. Unlike DAC, ICCAD is an academic conference and has no tradeshow. This year there was also WOSET, the Workshop on Open-Source EDA Tools, which was just what it sounds like.

Anton's summary of the conference: "Machine learning everywhere!"

During ICCAD, there were keynotes on IoT and cloud systems from Prof Chandra Krintz of UC Santa Barbara, on DARPA's ERI from Andreas Olofsson, and on technology trends from Arm's Rob Aitken.

Cadence sponsored various activities, such as a Sunday night hackathon, the SIGDA Cadathalon. We also sponsored part of the 2018 CAD Contest, which was on name mapping in smart equivalence checking. Cadence's Wen-Hao Liu presented A New Era in Digital Routing, which was preparation for the ISPD contest (on tech files for routing the most advanced technologies).

We also sponsored a networking reception, where we presented the activites of the Cadence Academic Network.

WOSET

Basically, WOSET is trying to revitalize OS-EDA development. The key goals for the workshop were:

• To identify who is who
• Create a culture/community for open-source EDA
• Learning from past pitfalls to have a better future
• Identify the landscape of available tools, and the gaps
• Resolve common issues, such as databases, interaces, and formats

WOSET kicked off with three introductory presentations from:

• Andreas Olofsson of DARPA, who talked about Intelligent Design of Electronic Assets (IDEA) and Posh Open Source Hardware (POSH)
• Andrew Khang of the home team (UC San Diego, the city where WOSET was being held) on The State of the Digital Flow
• Sachin Sapatnekar of University of Minnesota on The State of the Analog Flow

There were 27 papers presented during WOSET. Anton's conclusion from sitting through them was that:

Most open-source EDA software is maintained by one or two engaged people, with no business model, and with little trust. It doesn't seem sustainable.

My opinion is that open-source software is clearly a wonderful development methodology provided the developers and the users are the same people. If not, then you have one lot of people (IC designers, dentists, microbiologists) who understand the problem but don't understand programming, and a lot of software engineers who know little about IC design, dentistry, or microbiology. There is also no real business model for how the developers get compensated. This is perhaps an odd thing to say just after IBM acquired Red Hat for $34B. But the VC world is also littered with the corpses of companies who were going to be "the Red Hat of <insert product here>" only to find out there is no Red Hat of <insert product here>. The best read on the topic is a16z's short 2014 read Why there Will Never Be Another Red Hat: The Economics of Open Source. Small projects can get by with grad students whose ultimate goal is a PhD not a viable business. Large projects get by selling something else (typically hardware) and wanting to have a cheap (a few engineers) level playing field on the other aspects (such as compilers and operating systems), making sure that they can't lose on that dimension to their competition.

Two of the more interesting presentations were apparently Qflow, who are taking the existing tools and combining them together in a flow, and Cloud V (not, as far as I know, anything directly to do with RISC-V) who are taking a (different) flow and moving it into the cloud.

Qflow is a digital design flow from RTL to GDSII, stitching together tools across seven separate project repositories. See the table on the right for a list of the FOSS (free open-source software) EDA tools.

Cloud V says:

A main component of the project is a fully featured SoC design workflow. This workflow incorporates an SoC designer (SoC Editor), an SoC compiler (Pollen), and a Cosimulation framework (Beekeeper).

Another interesting project, the last paper of the day, is Hammer from Borivoje Nicolić's group at UC Berkeley, which is driven by separating the concerns of tools, technology, logical design, and physical design. You can see an earlier version of this topic from when Nicolić presented at CDNLive in Munich earlier this year in my post Agile Development of Custom Hardware. One of the impressive things about the teams at UC Berkeley is that they tape out a lot of chips, and don't just do designs on paper. As every EDA developer (including me) discovers, there is a big difference between running test designs through a tool and actually trying to build silicon that may be a bet-the-company proposition, or, at the very least, a visible embarrassment if it fails (in academia or a research lab).

Summary

All the papers (or most of them anyway) from WOSET are available on Github.

If you want my opinion on these tools, I'm afraid I don't have one. Not only do I know little about them, in most cases I've never heard of them (outside of Magic, which in one form or another has been around since the early 1980s, and Graywolf which is the open source version of Timberwolf, I believe).

I think the key question is whether any of these ambitious projects to pull a whole flow together (such as OpenRoad, Qflow, Cloud V) will be successful, or whether they are the Gallium Arsenide of EDA, looking attractive in little niches, but always a few years away from widespread adoption.

Or, as Frank Liu of IBM said in his presentation:

Although there have been many discussions among EDA researchers and practitioners on open-source EDA, the open-source model has yet to become mainstream in EDA.

The big question is whether it ever will be.

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