CRA/CCC Workshop on Extreme Scale Design Automation
February 21-22, 2014
Embassy Suites at University of South Florida
Tampa, FL, United States
2014 Events, 2014 Visioning Activities, Visioning Activities, Workshop
Design automation tools have been an enabling force in the computing revolution. Beginning in the 1970s, rapid advances have allowed semiconductor chips to evolve from a handful of transistors to modern processors and systems with billions devices. The Design Automation research community has been instrumental to the progress by leveraging highly collaborative research and continually exploring new aspects of electronic design.
While electronic design automation has had an astonishing run, the field appears mature. EDA vendors are offering extensive software packages that effectively support many design types. Rather than seek new, distinctive breakthroughs, ongoing work focuses on supporting new technology nodes and supersized systems, while also adding incremental improvements. Remaining research challenges require greater handson experience than a typical academic group can offer, making close collaboration with industry as critical as ever. Funding agencies have reduced support for traditional design automation research, while industry funding for EDA research is also limited. On the other hand, it is clear that the IC industry cannot remain viable without a healthy EDA ecosystem, and continuing innovation is paramount in the era of decelerating Moore scaling.
These changes motivated the CRA/CCC visioning workshop series on Design Automation. The first workshop focused on academia and funding, with a key conclusion being that things will not improve for traditional design automation areas. Funding for academic work is already scarce, and will likely dwindle further. Limited funds make it impractical to train graduate students in legacy areas and maintain core expertise, as academics are strongly encouraged to move on to new research topics. The second workshop focused on the semiconductor industry needs and comprehensive strategies to handle reliability, power management, and verification. Outside of these topics, however, most industry based participants felt current solutions were adequate. There is broad agreement that what constitutes Design Automation and Design Automation research must evolve. If left untended, the design automation community will splinter, with individual researchers moving off in their own directions. A goal of the workshop organizers is to hold the design automation community together, while still shifting into new areas. The shared expertise and experience of the community is an asset that should not be lost.
For the third workshop, the focus was on expansion into both “extreme scale” silicon, and into emerging technologies. Given the current state of emerging technologies, silicon is unlikely to disappear or be replaced. Future systems will integrate advanced forms of traditional technologies with novel technologies that are still in the experimental stage today.
An important challenge to the industry and the researchers is to identify new ways to communicate important research problems and solutions, as well as protocols for productive collaboration. To this end, the third workshop included the discussion of effective problem abstractions, representative design metrics, and actionable benchmarks. For example, lambda design rules proposed in the 1970’s as a simple abstraction made the development of design automation tools possible. Coupled with common abstractions were benchmarks and metrics, leading to the development of effective optimization techniques and steady, measurable progress. At present, the abstractions are lagging behind the increasing complexities of modern semiconductor systems, leaving a vacuum when it comes to benchmarks and metrics, for both nanometer scale silicon, and for emerging technologies. If a problem cannot be formulated clearly, it is unlikely that effective solutions can be developed. One of our goals was to make the emerging problems tractable for traditional design automation researchers so that these problems can be “brought into” design automation, rather than the current situation of design automation researchers leaving for other fields.
To stimulate discussion, we posed the following questions to participants:
- What critical problems in traditional EDA are not getting the attention they deserve? How can these problems be made tractable for academia?
- What are the most promising emerging technologies to replace or significantly augment silicon? Is there one in next 10 years?
- How can we best leverage the past few decades of EDA investment into new areas and technologies? Which are the most appropriate areas and technologies?
- What can the design automation community (individual researchers, funding agencies, professional society groups) do to jumpstart research in new areas?
Effective design and problem abstractions, representative design metrics, and actionable benchmarks? How do we evaluate if an EDA approach is appropriate, and how much impact it might have? What tangible items do we need, and who will create them?
At the end of the workshop, a report was prepared which proposed specific action items for academics, industry, funding agencies, and professional society leaders. Participants in the workshop were encouraged to propose clear, actionable steps that will help the design automation community address the challenges ahead.
February 21, 2014 (Friday)
|09:00 AM||Workshop opening and Introduction|
|09:30 AM||Keynote Talk 1|
|10:30 AM||Coffee Break|
|11:00 AM||Discussion session 1|
|01:30 PM||Discussion session 2|
|03:00 PM||Coffee Break|
|03:30 PM||Drafting visioning outcomes|
February 22, 2014 (Saturday)
|09:00 AM||Keynote 2|
|10:00 AM||Discussion Session|
|11:30 AM||Box lunch and drafting visioning outcomes|