EDAC's Emerging Companies Committee has been organizing evening seminars a couple of times a year in which Jim Hogan chats with someone with successful startup experience about what they did and what lessons they learned. Jim Hogan is today the principal of his own investment company, but he was one of the early employees of Cadence and a vice president in various roles for many years. Last week, the interviewee was John Lee. John is currently the GM & VP of the Apache business unit of ANSYS. But let's go back to the beginning. John was born in the US in Baltimore. His parents had immigrated from Korea to the US in 1959. In 1975 they decided to return to Korea. John was 9 years old and didn't speak any Korean, he had only known the US, so it was a shock, to say the least. Back then, Korea was a very poor country. He went to the local public school with 100 students per class. Since it took him some time to learn Korean, he focused on math—after all, the numbers are the same. He eventually transferred to a much smaller school where his mother taught, with just 28 in his graduating class. His math teacher told him that there is this place called Carnegie-Mellon University in Pittsburgh and he should apply. Since his family had no money (Korean salaries of the time were only about 20% of US), he would go anywhere he got a scholarship and, luckily, CMU gave him one. He was there as an undergraduate and then a graduate student with the legendary Ron Rohrer (I say legendary because he graduated from Berkeley with a Ph.D at the age of 21 and then was a thesis advisor to a guy called Aart de Geus at the age of 25). The group had eight students, many of whom are well-known names in EDA, including Charlie Huang, who just recently departed Cadence. John hadn't finished his Ph.D when Ron said, "Let's go start a company." So they created PSI with four grad students and four products doing complex simulations of ground planes, etc. They were 10 years too early, though. They didn't have any way to do circuit extraction, so it was just a technology company with no business plan. Meanwhile, another Ron company was ISS in North Carolina, which was working on DRC/LVS and starting to work on extraction. So he moved to NC and joined ISS (actually, I think ISS formally acquired PSI). So, the first lesson is not to create a technology without a plan as to how to monetize it. All of a sudden, ISS was bought by Avant! and so John was working for Gerry Hsu. Soon after that, the Arques offices of Avant! (it might still have been Arcsys) were raided by the FBI, but that is not the story for today. John said that he learned a lot of good things from Gerry. Firstly, R&D is king in an EDA company, and you must have an extreme product focus. Gerry was very business-focused (in fact, he is generally credited with inventing the time-based license that is almost universally used in EDA today). He was also a great student of culture and management. He moved to California and Andrew Yang became his boss (Avant! had acquired Andrew's company, Anagram). But they were losing every benchmark to Simplex (which would end up being acquired by Cadence) and Andrew told John to take two or three of the best engineers. There were all sorts of requirements from customers, but they were told to start very simple and to focus. Ignore the analog market. Focus on full-chip. TSMC only. Say 'no' to everything. So I guess that is a second lesson. By 2001 Synopsys had acquired Avant!. After a time there, he decided to leave to start his own company. The company was Mojave, doing physical verification in a new way. They had a big focus on starting the company IP clean, making sure that nothing came over from previous jobs. Obviously not source code, but even customer lists and other fuzzier stuff. After all, as John said, "if you are a startup and you get sued, you are dead." He'd seen what happened at Avant!, and Nassda was having its problems. The company was one of the first to notice that Linux computers were really cheap and there had to be a way to use a lot of them. DRC seemed like an embarrassingly parallel problem and should be easy to speed up. This was true, but naive. Andrew Yang and Andy Becholsteim didn't wait for a product plan, they liked the team and invested. They built the team up to nine people. Charlie Huang told them something that they didn't listen to, "if you guys are not Calibre-compatible you have no chance." Their technology worked really well, but there was a big problem. There were no foundry rule sets because foundries only wrote rules for Calibre, so they had to create them all themselves. With no rule-sets, it was hard to acquire customers and, with no customers, the foundries were not going to create rule decks themselves. Magma acquired Mojave and they rolled out the product. Early results were good. But as soon as they posted their incredible speed numbers, the four-minute-mile effect kicked in. Once it was shown to be possible, the competition caught up rapidly. Design rules got more and more complex, and there were more and more of them. This meant that the incumbent DRCs could fairly easily use lots of cheap Linux PCs simply by running each rule on its own server rather than having to come up with new algorithms. So their speed advantage didn't last more than 18 months. Magma did not have enough resource for a drawn out battle, even though it might have made it. So in 2009, he left Magma. They looked at the big EDA companies and wondered how you would build technology like that if you were starting today, with today's tools, hardware, and open-source infrastructure. Chips will have 10 billion transistors, looks like a big data problem. He wondered, how come you can instantly look at any location in the world on your phone using Google Earth, but EDA needs huge slow databases and provides nothing close to the consumer experience we expect on our phones? John was also wiser, and they focused only on problems where they could get paid in six months. They started with EMIR analysis, since that was an area where customers had pain and would pay. They started in Austin, but it was too hard to recruit engineers that they could not pay for months. So they restarted in Campbell. They built everything on Amazon AWS.The company was Gear. Again, they had a big focus on setting the company up right and the IP clean with good business practices. The challenge is always to keep ahead. In EDA there are only a few groups open to evaluating new tools and so all the startups call on the same groups. There is a sort of 12-18 month rule, which is that it only takes the competition that long to come back with something "good enough." It is easy to create an initial lead, but it is really hard to sustain the lead. Gear was a big data company with the idea that you could get any information instantly, like on Google. ANSYS acquired Gear. Despite his spending nearly a decade in Pittsburgh and ANSYS being based there, he had never heard of them. They are a 45-year-old company with 48% margins and fast growth. They have $1B in revenue, but have a bigger market cap than Cadence or Synopsys. General finance rules. $30M is a good exit today, so you can't take more than about $7M. Lucio Lanza: "Take as little cash as you can and spend it only when necessary." The living dead rule: kill yourself when it becomes obvious that you will not succeed. So John's path went from CMU to PSI to ISS to Avant! to Synopsys to Mojave to Magma to Gear to ANSYS. WIld ride.![]()
![]()
↧