Seeking to empower the future of EDA, around 30 EDA researchers from academia and industry met at the CCC/SIGDA Workshop on Extreme Scale Design Automation in Tampa, Florida in late February 2014. Participants were invited to articulate a vision for the future of design automation, and to prepare action items for a report that will help identify research and funding priorities.
Attendees were presented with kind of a good-news, bad-news situation. The good news is that EDA algorithms and methodologies may be useful in many other fields, ranging from synthetic biology and spintronics to something as mundane as intelligent traffic lights. The bad news is that funding for academic EDA research is lagging in the U.S., and that coordination between industry and academia is often lacking when it comes to research.
I was privileged to attend this workshop, but before going further, let me explain some background and decode a few acronyms. The CCC is the Computing Community Consortium, a "catalyst and enabler for the computing research community," and it is funded by the National Science Foundation (NSF) through a "cooperative agreement" with the Computing Research Association (CRA). SIGDA is the Special Interest Group for Design Automation of the Association for Computing Machinery (ACM).
The CCC/SIGDA workshop provided a small group setting for exploring the future of EDA
With support from the CRA/CCC and ACM SIGDA, several key academic researchers organized a series of three workshops to chart the future of EDA in the "extreme scale" era. These organizers are Iris Bahar (Brown University), Alex Jones (Univ. of Pittsburgh), Srinivas Katkorri (Univ. of South Florida), Patrick Madden (SUNY Binghamton), Diana Marculescu (Carnegie-Mellon), and Igor Markov (Univ. of Michigan).
The first workshop focused on issues related to workforce, markets, and education, and was held in Pittsburgh, Pennsylvania in March 2013. The second workshop focused on EDA for emerging technologies and extreme scales, and was held in conjunction with the Design Automation Conference in June 2013 in Austin, Texas. What I attended in Florida in February was the third and final in this series of workshops.
Moving EDA Forward
Alex Jones (Univ. of Pittsburgh) opened the Florida workshop by thanking attendees for participating in this "effort to shape the vision of design automation as we move forward over the next decade." Originally, he said, the six organizers of the workshop series went to the NSF out of concern that the "funding situation doesn't seem to be commensurate with the needs of the industry and the community."
Jones said that tools for extreme scale design are "already in an under-achieving state, and are constantly running into new problems as technology scaling continues." At this point, he said, extreme scale EDA tools should be able to:
- Efficiently harness systems integrating 1015 devices
- Effectively model emerging technologies
- Provide trustworthy validation of industry strength systems
- Address current and emerging design metrics
- Extend design tools beyond ICs to entire systems
- Model interactions with users and the environment around them
If EDA tools cannot meet these requirements, Jones said, "from a U.S.-centric perspective, we may be positioned to lose market share in this design automation area." He noted that research investment in Asia is significant and growing, while in the U.S. it is relatively flat.
The first CCC/SIGDA workshop, held in March 2013, pointed to two things EDA needs to do to survive, Jones said. One is to look at how core EDA can permeate to other areas, using the "toolbox" it has developed over the past three decades and applying it to "big data" problems. Another is to "regenerate the excitement" that EDA once held, perhaps with something similar in spirit to the "Intel inside" advertising campaign.
The second workshop, held at DAC 2013, came away with two suggested directions. One is to increase investment in core EDA at extreme scales. Jones noted that there are still many unsolved core EDA issues - system-level design, power and thermal, reliability, and analog/mixed-signal - that still need research. The second direction is to "enable the mission to Mars with design automation." The idea is to find a high-visibility problem that requires the design automation toolkit for success.
In this final workshop, Jones said, the goal is to "craft a vision of the future and be able to sell this vision" so EDA research can get the investment it needs. With a pre-workshop attendee survey as a starting point, the workshop was tasked with developing action items for a report that will help guide future research and funding.
The workshop included two keynote speeches. In the first speech Bill Joyner, director of CAD and test at the Semiconductor Research Corp. (SRC), offered a good overview of that organization and how and why it funds EDA research. In the second speech Rob Rutenbar (University of Illinois) presented his experiences with an EDA MOOC (massive open online course).
For much of the workshop, however, we split into four breakout groups - Traditional EDA, Education, New Markets, and Emerging Technologies. I was in the Traditional EDA group. Following are some of the more interesting points that emerged when breakout group leaders reported back to the attendees.
Much of the CCC/SIGDA workshop consisted of discussions in breakout groups
Group 1 -- Traditional EDA
- "Traditional EDA" refers to a design flow that produces a single chip or board. It includes synthesis, placement and routing; RTL verification; and analog design.
- Industry is often unaware of high-quality tools that are shown at academic conferences (for example - gate sizing tools that were benchmarked at the International Symposium on Physical Design surpass many commercial tools)
- EDA is not as attractive to students as it once was, and it's harder than more lucrative options (such as social media)
- U.S. government funding for semiconductor research has decreased substantially
- There is often a "surprisingly low level of technical interaction" between academia and the EDA industry
- "If we don't do research in traditional EDA, it will be difficult to retain confidence in the ability to attack related and emerging problems"
Group 2 - Education
- How can we inspire students to get into EDA? Reach out early with competitions at the high school or college freshman level that introduce EDA algorithms in different ways
- Massive Open Online Courses (MOOCs) can be a good recruiting tool to reach a wider audience
- Students who want to be designers should take coursework in EDA. They will be better designers if they know how tools work.
Group 3 - New Markets
- New markets EDA can potentially address include medical electronics, drug discovery, synthetic biology, smart grid, renewable energy, cloud computing, and more
- However, there is no "community" for applying EDA algorithms and methods outside EDA. There's a need for conferences, workshops, and sessions that address these challenges.
- Another area of challenge is pulling together the models that are needed for new areas. And, both optimization and analysis tools are needed.
Group 4 - Emerging Technologies
- The A in EDA should stand for "aided." Emerging technologies are just developing new models and will not be fully automated.
- What new EDA tools will be needed for emerging technologies like MEMS, spintronics, and carbon-based electronics? These new possibilities demand a fundamentally different approach. We must think outside of centralized clocks, handle variability, and develop new computing models with better statistical methods.
- We don't need high accuracy for early physical analysis tools - we first need to know if the task is even worth doing using EDA methods. "The notion of physical layout and extraction will need a fundamental rethinking."
As I write this post, the report is still under development. I'll have more to say about that at a later time.
Richard Goering
Related Blog Posts
BSIM-CMG FinFET Model - How Academia and Industry Empowered the Next Transistor
Preparing the Next Generation of IC Designers