The Return Of DAC In-Person

Apart from masked faces everywhere, you could be excused for not knowing that there was a pandemic going on. Sure, the numbers were down, the show floor was smaller, and most of the parties didn’t happen, but everyone was so happy to be able to bump elbows with their colleagues.

Buttons were available for attendees to show the level of comfort they had with various types of greetings, from “Stay six feet away from me,” to “I am ready to shake hands.” Unofficial numbers show about 3,000 in-person registrations and only about 100 for online. Out of the 3,000, about 500 were vendors. This shows the level of expectations from the industry, and from some of the reports about the numbers of people on the show floor, this was perhaps a wise decision on their part.

The opening keynote was given by Jeff Dean, senior vice president at Google Research, who provided a history of the growth in AI. His talk then pointed to a number of new problems that need to be solved and the potential for AI to be an integral part of the solution.

The industry was certainly in a good mood. Both the opening presentation from Charles Shi of Needham on Sunday, and the industry overview by Joe Sawicki of Siemens EDA, focused on why the industry is probably still in the early years of a major expansion period for semiconductors, and why that probably will lead to an expansion for EDA.

One interesting insight is that all segments of EDA are growing, not just the new tools. This is certainly an indication that much of the growth is coming from growing markets.

Fig. 1: EDA is growing across the board. Source: Siemens EDA

However, to balance that, Dean’s keynote contained a subtle warning for EDA — if you do not keep up with our needs to reduce the cost of design, we will do it for you.

Fig. 2: Internally developed tools results for placement. Source: Google

The EDA industry was borne out of captive tools groups within semiconductor companies, which decided there was little value to all chip design companies spending money to create the tools when they did not add much competitive advantage. By consolidating that work into a few companies, they had better return on money invested. If it is no longer true that design tools, or possibly even design or domain data, can add proprietary advantage, that could be a dark cloud for EDA.

Serge Leef, currently a program manager within DARPA, called for more innovation within digital simulation, saying the industry has not been making enough advancement and that we remain stuck with the techniques from the ’80s. He also wants to see the cost of developing new silicon be reduced by an order of magnitude. The pressure is on.

Day 2
Bill Dally, chief scientist for Nvidia got the second day started by looking at three vectors — how GPUs can accelerate EDA, how GPUs enable deep learning, and how deep learning can accelerate or enhance EDA.

Fig. 3: Impact of GPUs on EDA and Machine Learning. Source: Nvidia

The first part of the presentation looked at the progress of GPUs over the years and how they enabled machine learning. He outlined the technology that had been incorporated to turn the GPU into an engine more suited for machine learning. Dally then talked about a couple of ways in which GPUs could accelerate EDA, such as speeding up the rendering process. Perhaps most interesting was the work Nvidia has done to accelerate simulation. By deploying both design parallelism and cycle parallelism, the company managed to speed up regression testing by a factor of 1,000X.

The third part of his talk looked at ways in which deep learning could be used within EDA. Dally examined IR drop estimation, parasitic prediction, routing congestion, automating standard cell layouts, and the use of AI for lithography modeling.

The next presentation of the day came from Jay Vleeschhouwer of Griffin Securities, who predicted EDA growth in 2021 will come in around 10% when the final numbers are tallied. This growth is driven partly by the fact that semiconductor companies are becoming systems companies, and systems companies are becoming semiconductor companies. He believes the role of EDA has become more essential in the arms races in multiple markets and applications.

EDA’s value has grown 5X over the past decade. Income over the same period is up 120%. For 2021, Cadence revenue is up 11% to $2.97B, Synopsys is up 18% to $4.0B, and Ansys up 8% to more than $340M. The expectation is that Siemens EDA is growing at a similar rate and maintaining market share. The Big 4 account for 90% of the industry.

What is encouraging is that EDA continues to spend big on research and development, pumping an average of about 36% of revenues back into its tools. This tends to be higher than most technology companies.

Fig. 4: Continued R&D spending within EDA. Source: Griffin Securities

The day continued with presentations from Cadence about how More than Moore will enable continued advances and the benefits that EDA can get from ML. Then Kailash Gopalakrishnan, an IBM fellow, talked about roadmaps for data precision and number formats in ML. I moderated a panel discussion about the changing paradigms in SoC verification. All of these events will be covered in more detail later, and some can be watched through the DAC online registration page.

Day 3
This was the last day for the industry side of the conference, followed by one full day for academics. The keynote was given by Joe Costello, one of the visionaries of the early years of EDA, when he served as the first president and CEO of Cadence. Costello is no stranger to giving keynotes at DAC, even as someone now involved mostly as a user. He is currently the executive chairman at Arrikto, Metrics, and KWIKBIT. His talk was entitled, “When the Winds of Change Blow, Some People Build Walls and Others Build Windmills.”

Fig. 5: Joe Costello at DAC 2021. Source: Semiconductor Engineering

Costello started with a very personal testimonial to the late Jim Hogan, who was an agent of change. He noted that people do not like change, either at the personal level or the business level, and it doesn’t matter if you are successful or struggling because everyone is scared that things will get worse. We all live our lives having built a model of the world — be it good or bad, and no one likes to admit they got something wrong.

To execute change you must change the model. As people get older, they resist change more, he said. However, the older you get, the less you have to protect. The same is true for people and for companies, groups, marketplaces — even countries. Groups have their own ego and psychology. To be successful, you have to know how to change other people’s models. Costello also dared to talk about a subject taboo to many — the mistakes made by U.S. politicians and the trade war with China.

His entire talk, delivered in his typical energetic manner, can be condensed to a very simple but powerful message — do not fight a macro trend. Ride it.

The day continued with Steve Roddy, vice president of product marketing for Arm‘s machine learning group. Roddy talked about how ML is becoming pervasive, even though a large amount of it is hidden.

Sam Naffziger, AMD senior vice president and corporate fellow, talked about how domain-specific computing is necessary to meet the demands being placed on compute resources, and how chiplets and 3D stacking are needed to construct the systems of the today and tomorrow.

Finally, I moderated another panel that explored the new skill set that will be required by verification engineers as machine learning becomes an important part of the verification process. Again, many of these presentations will be explored in more detail in future articles.

Conclusion
Harry Foster, the general chair for DAC 2021, and the rest of his executive team, pulled success out of a hat and brought together a fabulous program – especially given the circumstances. The overriding theme of the conference was AI and ML are now everywhere within the industry and its impact will continue to grow.

Was it DAC as normal? No. But it was more than could have been reasonably expected, and everyone was happy to meet in person again. DAC for 2022 will back to its normal timeslot of July — again in San Francisco.