University of Washington's
Timberwolves and Huskies ...
Dr. Carl Sechen teaches at the University of Washington in Seattle, one of the singularly most beautiful college campuses in North America. Although Sechen would agree with me about the beauty of the place, he may be less than enamored of the weather.
Sechen is by his own description a world-class tennis player (for someone his age, he says), but I’m guessing that the rain that defines the Pacific Northwest does little to enhance his game. Nonetheless, Sechen is as disciplined about pursuing his tennis as he is in pursuing his academic interests. We chatted in his office in the new Bill Gates Engineering Center on the UW campus in the fall of 2003.
Carl Sechen was born and raised in Wisconsin and says that many of his values can be traced back to a childhood spent in the farming community there. He attended the University of Minnesota, where he earned a BSEE, moved on to MIT for an MSEE, and finished up at U.C. Berkeley where he earned a Ph.D. in Electrical Engineering in 1986 – Alberto Sangiovanni-Vincentelli was his thesis advisor. Sechen’s graduate work centered on the development of a new place-and-route (P&R) tool called TimberWolf.
After completing his studies at Cal, Sechen joined the faculty at Yale. He taught in the Electrical Engineering Department there from 1986 to 1992, at which point he headed back west to join the faculty at the University of Washington.
While at Yale, Sechen discovered a crucial difference between public and private universities. Research done under the auspices of the University of California, for instance, a state-funded university, is considered to be in the public domain. Faculty members and their students are discouraged from commercializing technology developed in U.C. labs for personal gain. Sechen’s TimberWolf P&R tool developed while at Cal was then, and continues to be, free and widely accessible to industry. Yale, on the other hand – a very private school – was according to Sechen more commercially minded. Technology developed in Yale labs was, and is, marketed to further the financial well being of the school and the individual researcher.
Sechen had a graduate student at Yale named Bill Swartz. In 1994, after Swartz had completed his graduate studies at Yale and Sechen was ensconced at UW, the two men founded a company. It based on Swartz’s graduate work and his etxensions and refinements on Sechen’s original place-and-route work. Their company, TimberWolf Systems, is still in business today – albeit under a different name, InternetCAD.com, Inc.
Sechen told me, "TimberWolf continues to be the best place-and-route tool in use today. When we first started the company, however, so many people had access to the original code by way of U.C. Berkeley that they thought they should get subsequent generations of the tools for free. [To counter that perception], we changed the name of the company to InternetCAD in 1997. I’ve actually been offered VC money a number of times to expand the company, but Bill Swartz has always wanted to keep things under our own control. InternetCAD has been self-funded for years. Bill’s brother, Jack, is the president and we continues to grow the company out of our offices in Dallas, Texas. We’ve always operated on a subscription model, which gives our customers the right choices. They can buy as many quarters of usage as they need."
Sechen said that the beauty of the TimberWolf technology is that it’s a variable die router. Fixed-die routing is an inferior strategy, he says, although one that’s been pursued consistently by the major players in EDA for over 10 years now:
"Cadence’s original tool, back in its early SDA days, was a variable-die router, but then they bought Tangent and [thereby] acquired TanGate, a gate-array place-and-route system. Cadence knew that their own tool wasn’t that great – that TanGate was dramatically better – so they said let’s pretend that a standard cell is actually a gate array. But their users had difficulty using the tool. The users had to place the cells in just the right way, or the router failed."
"It was at that point that SDA/Cadence became the first ones to introduce a fixed-die model. They said that their Cell3 Ensemble tool was better than anything else out on the market at the time. But, if the routing failed using the fixed-die model, users had to figure out how to change the placement such that hopefully the routing woul complete in the next iteration. Still, Cell3 Ensemble was so much better than anything else commercially available, it quickly became a philosophical issue with people."
"Meanwhile, the variable-die solution from ARCsys/Avanti was very competitive, and may have even been better than the Cell3 Ensemble solution. But when Gerry Hsiu joined Avanti [from Cadence], he wanted to install the same type of solution, the fixed-die solution, that Cadence had. The ironic thing is that Avanti went away from what may have been a better solution than the one that was in place at Cadence at the time. They went from a variable-die to a fixed-die and took a step backward, and got themselves into legal trouble in the process."
Sechen served as an expert witness in the Cadence/Avanti trial in 2000 and 2001. He said the trial was unfortunate: "I knew Gerry Hsiu when he was at Cadence. When he went off to ARCSys/Avanti, he took some key developers with him.. All of that [enhanced the impression] that code was being stolen and it raised red flags all over the place."
"I had a chance to examine a great deal of the code in question when I was an expert witness on the trial. It was amazing – I even saw my own TimberWolf code in their tool, where only a single line of code had been changed. And I don’t mean the earlier, far-inferior version of TimberWolf available from Berkeley. The version I found in Avanti’s suite was a far more state-of-the-art version that had somehow been ‘acquired’ from Yale."
These days, Sechen’s more occupied with growing and maintaining the VLSI program at UW, than he is with picking through Avanti code. He said that when he first arrived on campus in 1992, the program was ranked about 30th in the nation among engineering programs in VLSI design and CAD. He’s visibly proud that today UW is ranked 14th in the nation.
"When I first came here in 1992, there was only a minimal VLSI program. In fact, at the undergraduate level, there wasn’t anyone teaching VLSI at all. Andrew Yang and Mani Soma were here, and a couple of CAD faculty members, but Mani only taught a graduate class on testing. I started out by teaching layout and also some analog circuit design. I didn’t actually want to compete with industry in place-and-route research, because I felt the important work needed to be done in the industry. At this point, however, I’m not really sure there’s been any significant improvement in place-and-route technology in over 10 years. There have been some changes, but they’re only minor."
"Anyway, I decided to direct my research, and that of my students, into IC design, keeping the work close to layout and synthesis in the process. [Accordingly], I started a full 3-quarter sequence to teach design. [The program was] fully functionally by 1997; the first two quarters covered standard cell libraries, circuit design, and memory and data path design, and the third quarter covered contemporary topics in VLSI design. At the peak [of the boom] in 2000/2001, we had 130 students enrolled in the series."
"The downturn has changed that, however. Students aren’t seeing a lot of work opportunities in this area, so we’re down to about 100 enrolled in the sequence this year. But that’s not bad; our program certainly remains one of the largest VLSI design programs for undergraduates in the U.S."
Sechen is the one responsible for purchasing the EDA tools used by both undergraduates and graduate students at UW. He says he has a fairly large budget and a reasonable amount of cooperation from most of the large EDA vendors. He has less than positive things to say about the one company who’s technology he would most like to have available for teaching, however – the simulation tools from Nassda.
"Most EDA vendors are happy to have us teaching with their tools; they have nice programs with us. They know that students often come out of school as fully committed users. We’re training 100 to 150 students a year here. I know that when I go out and see my former students at companies, they’ll be using the tools they learned to use here at UW."
"Nassda, on the other hand, seems to be one of those companies that takes the exact opposite approach [to making tools available to universities]. From what I hear, they’ve got the best simulator out there for larger circuits, but they don’t seem to understand the value of working with universities. HSIM [from Nassda] is a great tool, one that appears to work quickly on large blocks in the design and one that my students really want to use. The comparable simulators from other vendors are nowhere near as fast, but Nassda charges $1000 or more per copy."
"[In comparison], Cadence and Synopsys, through their university programs, give us virtually unlimited licenses for their tools for a flat fee of $5000 per year. And, their tools don’t really need much support. It’s true that the students do sometimes have problems, but we usually find a work-around. In general, we start out using the tools in the research groups and only then do we move tool usage down to the classroom. Again, Nassda is out of sync here – they make us sign an agreement not to move their tools down into the classroom. The fact of the matter is, we do a lot of benchmarking of tools here, and Nassda would only benefit by allowing us to benchmark their tools – we publish our results a lot and present frequently at technical conferences."
Meanwhile, Sechen was outspoken about some of the disparities and harsh realities in industry: "We train students here that go out and create technical success for the companies. But engineers continue to earn less money and less respect than people who enter industry with a degree in business or an MBA. Someone with an MBA out of Harvard, for instance, can earn $150,000 a year right out of school. New engineers earn significantly less than that. But, hello – what does that MBA really know, especially compared with the engineer? I think that there’s little in school – even in business school – that makes a good manager. If there were and considering all of the MBAs we’re producing, we’d see real improvements in management. But nobody really believes that management is getting any better. In fact, it’s just the opposite."
"In any case, the technical industries should be providing serious funding to the universities. I like to say, ‘We’re training your people for you and in most cases, we’re getting next to nothing in return. And, when we want to use your products and can provide visibility for those products, we don’t get a lot of cooperation.’"
Sechen also commented on the vagaries of funding for research in universities. "The SRC funding is going down, and the size of NSF awards hasn’t changed in 20 years. Meanwhile, the cost to universities to train students has gone up threefold. DARPA has alsoi moved away from funding universities, so that makes it all even more challenging. We did have to build up a lot of faculty to teach VLSI and CAD at the height of the boom, but even today it’s hard to say that there are too many professors teaching VLSI design today, even if the enrollment is down."
"Meanwhile, I don’t think that the universities can really be on the bleeding edge any more. We can’t compete with what’s going on in R&D in industry. The problems are too intense and the costs of solving those problems are too high. We can continue to educate, however. And we shouldn’t have to defend teaching these technologies at the university level. VLSI design and CAD tools continue to be scientific disciplines. We should remember that anything that stimulates people’s minds is always highly suitable for universities. And don’t forget that these students are able to step into a company after graduation and make contributions almost immediately."
Editor's Note: This article was first posted in May 2004 in in Play in EDA.
Peggy Aycinena owns and operates EDA Confidential. She can be reached at email@example.com