Breakfast Bytes Blogs

At the 2019 International Symposium of Physical Design, the conference honored Alberto Sangiovanni-Vincentelli with a lifetime achievement award. Alberto was one of the cofounders of SDA Systems, the forerunner of Cadence, so in some ways he's a founder of Cadence. He is still on the board. I got to know him last time I worked at Cadence in an era when we had a technology advisory board (TAB) that we were both part of.

The award was presented at the end of the day at the conference, and then there was a dinner in his honor. I figured I'd not get much time with him at the dinner, so I arranged to meet him at Berkeley, then drive him to the conference and have lunch with him while we chatted a bit more informally.

Alberto was brought up in Milan in northern Italy. At high school, he studied ancient classics including five years of Greek and Latin, philosophy, history, and history of art, before making the unlikely transition of going to Politecnico di Milano to study electrical engineering. After graduation in 1971, he stayed at the university as a researcher for three years, studying large systems of equations and solving large sparse matrices, which would turn out to be very fortuitous, since that's how circuit simulators work under the hood. He became an associate professor at Politecnico in 1974.

A year later, he was seconded to Berkeley to carry out research in the circuit and system group of the Department of Electrical Engineering and Computer Science in the second half of 1975. Given the good results of his research, the Department asked him to return to take a Visiting Assistant Professor to teach classes and continue his research. However, Alberto was not keen in accepting the offer as he was scared about his lack of knowledge of the US university education system and had as an excuse the fact that his department in Italy had him scheduled to teach in Italy and would not allow him a second tour. With his great surprise, the Chair of the Berkeley department reached out to his counterpart at Politecnico to ask the permission directly without him knowing anything about it. And indeed, Politecnico gave its permission for another year of leave. Thus, he could not refuse: he had to pack and return to Berkeley in August 1976. The teaching load was heavy: two undergraduate classes and three graduate classes in a quarter system and all different. Needless to say, he never taught these classes before. In addition, his English was rather poor: only three years at Junior High School. To get to a point he would feel comfortable teaching, he took night classes both in Milano and later in Berkeley while teaching.

After the classes were over, the teaching evaluations were good and research was going well. When a regular position opened up, he was asked to apply. Again, he worried, thinking that if he accepted the offer, he would give up a tenured position for an Assistant Professor position and that he would never be able to do enough to get tenure. However, he thought he was most unlikely to get the position, so he applied anyway. He never interviewed or gave a job talk since he was already known by the faculty, but he interpreted this as a confirmation that the department had better candidates. At 11pm in a stormy night, he got a call from the Chair announcing that he had been chosen for the position. And the rest is history. He accepted, not without great hesitation, and has lived since then happily in Berkeley.

Alberto started working in research on CAD with Richard Newton, who had been a student of Don Pederson, the "father of SPICE", and came to Berkeley in 1975. This was the beginning of a long and strong partnership. Their research at the time was focused on relaxation-based methods for circuit simulation that yielded the class of fast SPICE programs that are still in use today. (Richard died tragically young in 2007.)

In 1980, he went to IBM for a year to work with Bob Brayton and Gary Hachtel in the mathematics department. Bob was the lead of the team that produced the algorithms in ASTAP, a strong circuit simulator, and was starting a program in logic synthesis. Together they developed Espresso, one of the very first logic synthesis programs, and subsequently MIS. Some of the basic ideas are still used in most if not all, modern synthesis programs like Design Compiler and Genus. Alberto was invited by IBM to join the T.J. Watson Research Center with a very good offer, but he preferred the university life where he felt that students pushed him to keep on top of research and made his professional life more fulfilling than he felt an industrial environment would do. However, working with Brayton was such a pleasure for both, that they continued to collaborate for many years until eventually, Bob Brayton joined Berkeley.

At IBM, Gelatt and Kirkpatrick developed a novel approach to solving difficult combinatorial optimization problems called simulated annealing. During his stay at IBM, Alberto felt that there was a rich set of interesting problems to solve with the help of this approach. This was the beginning of his interest in automated layout that yielded TimberWolf, a placement tool, and YACR, a detailed router (YACR is yet-another-channel-router).

In 1983, Intel was working on a new generation of microprocessor, the 386. The design was difficult and there was a concern that it could not be completed on time without some changes to their methodology. Albert Yu, GM of the heart of Intel’s business: microprocessors, chipsets, and software, and Pat Gelsinger, Chief Architect for the 386 project came to visit Alberto at Berkeley and asked to help with design methodology and tools. Espresso, TimberWolf, YACR and MIS made it in the design flow at Intel and contributed to the success of the project, even if MIS for example, was in its very early days and not even released. The Berkeley group worked with a few engineers coming from Israel that became the core of the Design Technology Group there.

In parallel to this, Alberto and Richard Newton were being pushed by IC companies to create an EDA company to support and sell tools, derived by the research work at Berkeley. Richard and Alberto were not interested in acting as managers of the company, so they had to find someone. The selection process was taking forever and they were never quite happy with the people they interviewed. Finally, Jim Solomon at National Semiconductors cut things short and said he would do it. They created Solomon Design Automation (SDA) in 1983, with the financial support from four companies (National, Harris, Ericsson, and GE)  who each put in $1M and with the contribution of $1M by a group of VCs that included Don Lucas. By 1986, the company was in trouble since the products were delayed and the market was not developing as hoped. SDA was nearly out of business. Harris supported a further round of financing and the Board decided to look for a new CEO and, in Alberto's words:

On Jim Solomon’s recommendation, SDA decided to try a really young and inexperienced guy...Joe Costello...best choice we ever made!

From then on, SDA was profitable. In 1987, the company was ready to go public but unfortunately, the chosen day was Black Monday, October 19, 1987, when stock markets around the world crashed. Of course, the IPO was withdrawn but much work had gone into preparing the company to this event, a large waste of time and effort unless another way to be on the stock market was found. ECAD, another EDA company, had gone public a few months earlier. The idea came to merge with ECAD and by doing so, become a public company. Few people know that SDA and ECAD were supposed to start their operation together. Paul Huang, the technical leader of ECAD, was part of the original team but became impatient when the choice of a CEO for the company took too long. Hence, together with Glen Antle decided to start ECAD waiting for the original company to start. When finally SDA was formed, ECAD was already established with the DRACULA DRC program. The interest in being one company was no longer there. However, the teams remained in friendly terms and the merger was relatively easy to carry out. The merged companies got a new name...Cadence.

Among Alberto’s projects was the logic synthesis. Alberto and Richard saw the industrial potential and wanted to convince SDA to develop this technology, too, but it was at the time SDA had other problems that warranted all the management attention. Hence, they decided to help found another EDA company that would focus exclusively on logic synthesis. Aart de Geus and his team at GE were also interested and they got together with the same funding scheme used in SDA: corporate sponsorship (GE and Harris) and VCs. The original company was called Optimal Solutions Inc (OSI) and was established in North Carolina. It moved to Silicon Valley in 1988 under the name of Synopsys.

At that point, Alberto and Richard were consulting for both companies, Alberto was the Chair of the Technology Advisory Board of both companies and Richard was on the board of Synopsys since they didn't really compete in that era. In 1991, it became clear that the two companies were encroaching in each other domain and that Alberto had to make a choice. Aart offered him a position on the Board of Directors on a Monday of 1992 and the very same day, Joe Costello made the same offer, knowing nothing of the Synopsys offer. The following months were spent in a continuous nightmare, the decision was too difficult but in the end, it had to be made and Alberto chose Cadence where he still is on its Board of Directors.

Alberto, after Synopsys was formed, became interested in system design methodology and tools. He started working with automotive companies, first with Magneti Marelli, then with Mercedes-Benz, BMW and finally with General Motors. At that time, Alberto recommended Cadence start assembling a team for system-level design by acquiring tools such as SPW and Bones from Comdisco, and Redwood Design Automation. That combination was structured as Alta Group, a totally owned entity by Cadence, but with independent R&D and sales. Alta group had the largest market share in a domain that was just starting. It is now much more visible with some of the companies that Cadence is collaborating with such as Green Hills and MathWorks. It also developed new tools such as VCC. However, Alta was eventually fully absorbed into Cadence losing market share and focus until its remnants were sold to CoWare. Alberto is very excited that Cadence is coming back to this domain with determination and does hope it will become once more number 1 in the field.

So that's a perspective on the very early days of EDA design tools, the EDA industry. Or how Alberto came for six months to California, and like so many of us, never really went back.

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