Get email delivery of the Cadence blog featured here
This year's Kaufman Award recipient is Andrzej Strojvas. He is the Keithley professor of Electrical and Computer Engineering at Carnegie Mellon University (CMU) and is also the CTO of PDF Solutions. Last week, I went over to PDF in downtown San Jose to meet him.
You might guess from his name that he is Polish, and you would be right. He did his master's at Warsaw University of Technology, the best technical university in Poland. He joined the faculty and started working on his PhD. Then, Steve Director invited him to CMU (which is in Pittsburg). He was finishing his PhD there when, in 1981, martial law was introduced in Poland. So that made the decision for him as to whether to stay in the US or return to Poland. At around the same time, the SRC set up a center in computer-aided design between CMU and Berkeley. Given that funding, Steve Director pressured him to stay at CMU and he has been there ever since.
His master's thesis was on statistical simulation of semiconductor processes and devices. He continued that work for his PhD on diagnosis of the semiconductor fabrication process. After he joined the faculty, he built up a large group of PhD students. At what he calls the "peak of craziness" he had 20 PhD students, countless masters students, and three programmers. The focus of their work was developing software to make more realistic SPICE models with good correlation to escape the conservative worst-case design.
Two of those PhD students were John Kibarian and Kimon Michaels. In December 1991, they co-founded PDF solutions around a statistical simulation tool called PDFab. It started in Pittsburg, working with companies like National and TI. Although there were lots of customers, a product business on its own was not enough, consulting services were required. That changed nature of PDF. In 1995, PDF moved to San Jose. There were several suitors to acquire PDF (including Cadence) but instead they took some VC funding from Cadence/Telos and a private investor. Joe Costello, who was CEO of Cadence at the time, was instrumental in changing their focus to add consulting on yield ramps, first for IDMs starting with Toshiba.
So Andrzej had two full time jobs on different coasts. Luckily CMU realized that doing that sort of work required faculty to spread, since there wasn't a large Pittsburg-based semiconductor industry that they could work with locally. A university cannot run a state-of-the-art fab for financial reasons, but work can still go on on individual devices, even if they cannot work on realistic manufacturing problems. In fact during one sabbatical, Andrzej spent time working with TI on a big contract for single wafer manufacturing, with the goal of a 3-day cycle. This required lots of custom equipment (I seem to remember reading that they used infra-red lights instead of diffusion ovens, for example). TI offered to donate the equipment to CMU but they had to turn it down since it would be too expensive to operate and impossible to keep the equipment upgraded. But Andrzej has always been successful in getting access to state-of-the-art equipment, sending students out into industry to do PhDs split between industry and academia. There continues to be a CMU connection, although as Andrzej put it, "with 425 employees now, the percentage of my students has got lower and lower."
One other thing back then was some joint work with IBM and DARPA on extremely regular layout. Together with some grad students, Andrzej con-founded Fabbrix to work on this. It was eventually acquired by PDF who won the company versus other suitors.
I asked him about PDF's customers. He said that they definitely provide a software layer for process control and yield management. They sell to both the foundries themselves and the top tier fabless companies. They are leaders in both segments but unquestionably in fabless. They used to have separate platforms for the two markets, but now they are integrated into the Excensior product. Mostly they are providing characterization to both sides for new process nodes being developed and ramped, as well as offering the extremely regular layout that PDF calls templates, that is the outgrowth from Fabbrix. Everyone has to limit layout, templates are an extreme version with regular structures with small number of patterns but not suffering in performance, area, or power. It is a very attractive offering at 14nm, 10nm, and now 7nm. The regularity means that the process can be ramped up much faster.
Andrzej tells the story of regular layout: "We started earlier at 28nm where we would eliminate some bad patterns in customer’s standard cell libraries. But now it is an absolute must. But industry is doing a good job of restricting the patterns up to a certain level, up to routing within standard cells. Then 'here come the routers and it is free for all.'. There is proliferation of patterns, and that can get into trouble in the back end. There are zillions of design rules but a better solution would be more restrictive routing, however as an industry we are not there yet. This results in yield loss from back end (at lower levels). Some issues higher up, and they are not all fixable by dummy fill. Even there would be good to be aware of manufacturing pitfalls. CMP is more important than printability at higher metal levels."
China is a strategic direction. PDF has 120 people in Shanghai. There is an increasing emphasis on Chinese customers since that is where most of the future growth is going to occur. China, as a country, sees semiconductor as strategic and is investing a lot in making the country more self sufficient in semiconductors. To bring up modern processes requires more than just money, though, it requires the expertise that PDF and Andrzej have.
Congratulations to Andrzej on being honored with this year's Kaufman Award.
Read the ESD Alliance press release announcing the Phil Kaufman Award, and don't forget to save the date for the Kaufman Award dinner, which will be held on January 26, 2017.
Next: How Virtualization Is Changing Networking
Previous: Silicon on Nothing: the Origins of FD-SOI