Alberto Sangiovanni-Vincentelli on the future of EDA

During a slack moment at DAC I had the rare pleasure of a conversation with Alberto Sangiovanni-Vincentelli, professor at the University of California Berkeley and chief technology advisor at Cadence, which titles only hint at the gentleman’s central role in the EDA industry. Our conversation quickly focused in on what Sangiovanni-Vincentelli believes to be the most pressing issue facing EDA today.

"The challenge is growth," Sangiovanni-Vincentelli said. "In today’s environment, growth for the EDA industry implies moving beyond the IC. We have to look into lateral domains that share fundamental characteristics with the process of chip design.

"For example, look at synthetic biology," Sangiovanni-Vincentelli continued. "That discipline is creating a set of tools within a platform methodology just as EDA has. I think there are many of these areas. Managing complexity is the same challenge in all of them. Assembling components and analyzing systems are the same tasks, even though the components differ from domain to domain."

Another example Sangiovanni-Vincentelli offered was the building industry, all the way from architects to construction workers and installers. He pointed out that a building comprises many interacting subsystems, including air-flow, traffic, and security to name a few.

Looking just at heating, ventilation and air conditioning (HVAC,) Sangiovanni-Vincentelli noted that controlling the temperature in a room was a problem in hierarchical control theory. Today we tend to use simple control loops in which a room or sector thermostat directly controls a valve in an air duct. But this fragmented approach leaves most of the system operating at sub-optimal air flow most of the time. A correct solution would recognize the hierarchical nature of the problem. The input from an individual thermostat would be a request to a global optimization algorithm which would readjust the flow throughout the hierarchy to keep the room within its temperature range while minimizing energy consumption.

"One of the problems," Sangiovanni-Vincentelli warned, "is that in other system domains there is not much of a tradition of spending on tools. In IC design, people are amenable to help from computers. In other areas, this is a huge psychological problem. How do you get HVAC installers to use computers so that they actually get the equipment in correctly?"

The technical problem is essentially congruent to placement and routing on an IC. But the psychological problem is very different from convincing a handful of engineers to give up their ruby-lith and let a computer program take the first cut at the layout.

While there are obvious new challenges in non-electronic areas such as synthetic biology and construction, the familiar world of IC design is changing as well as geometries continue to shrink, Sangiovanni-Vincentelli explained.

"We are just starting to look at electronic systems composed of nanostructures," he said. "We will need new methods, new architectures, and new tools to deal with the uncertainties inherent in devices this small. The challenge is to compose complex, reliable entities from unreliable components. Design comes to mean starting with a swarm of components and estimating the behavior of the aggregate."

Sangiovanni-Vincentelli suggested as a possible avenue toward solutions that we might reexplore the world of feedback systems. "We are already seeing new analog designs in which analog components are controlled by digital feedback," he said. "We are looking at how to use feedback in digital systems as well. But at what level we want the feedback—that is unclear. There is no single answer to that."

He said that there is also exciting work going on in systems with massive redundancy. "You are already starting to see ideas not unlike massive redundancy in advanced ADC design," he pointed out.

Sangiovanni-Vincentelli believes that underlying all these areas of inquiry is a single formalism that has yet to be fully developed. He calls this metamodeling. For instance, finite state machines, data-flow machines, and Petri nets are formalisms with very different properties. But are there common properties that underlie all three of these ways of describing a digital system?

Clearly this is not the thinking of a depressed individual. One quickly gets the feeling that if the EDA industry is facing serious challenges, that only means it has come to an interesting part. The future, properly investigated, should be far broader and more fascinating than the work we have done so far. The implications may be far broader than the impact that microelectronics has had on society so far. And we are just peeling the onion down to the layers of the really interesting abstract problems. Alberto Sangiovanni-Vincentelli remains one of the industry luminaries from whose conversation one leaves refreshed and anxious for tomorrow.