Monday, November 6, 2006
Interview with Jason Cong, Chairman, AutoESL Design Technologies
Our interview today is with Jason Cong, Chairman and Chief Technology Advisor of AutoESL Design Technologies (www.autoesl.com), an electronic design automation startup out of the University of California, Los Angeles. Jason is also a Professor and Chairman of the UCLA Computer Science Department, and a serial entrepreneur, having headed Aplus Design Technologies, a firm that was successfully acquired by Magma Design Automation in 2003. We caught up with Professor Cong to understand his new, funded, startup, and where it fits into the EDA industry:
Ben Kuo: What is AutoESL Design Technologies?
Jason Cong: AutoESL is in a space called electronic system level design automation (ESL), which is projected by Dataquest to be the fastest growing market segment in the Electronic Design Automation (EDA) industry, which is a four billion dollar market. We provide software tools to help integrated circuit designers to design their circuits. As you can imagine, the chips from Intel, Broadcom, Xilinx, and IBM are now fairly complex, many with over 1 billion transistors. They aren't designed by hand, they now use automated tools. Electronic Design Automation provides tools to those integrated circuit designers. Electronic System Level design focuses on system level design automation. The reason that it is an exciting area, is as Moore's law progresses, we can put exponentially more devices on a single integrated circuit. A single IC is no longer a component--you can implement an entire system on that IC. That requires new design methodology and tools to help people design the system--Systems-on-a-Chip. Specifically, we have a research project which has been looking at system level synthesis and behavior at UCLA for about five or six years, started in the 2000/2001 timeframe. We developed a numbers interesting research results, including a prototype ESL synthesis system called xPilot, which can take a high level description--almost like a program description of a circuit, written in C, C++ or a language called SystemC, where you just describe the functionality of the circtuit. Our tool will compile the RT level of that circuit so you can synthesize it in silicon. It's a significant productivity gain. For example, we showed that you can take around 5000 lines of code to describe a MPEG 4 decoder, which synthesizes into VHDL or Verilog into 50,000 lines. The big reason you can get such a huge productivity gain is in conventional methodology, you have to describe everything at a RTL, Register Transfer Level--at every clock cycle, which is a very tedious description, even though it gives human designers lots of control. What we provide, is a way to liberate human designers. You don't need to design cycle-by-cycle, but by behavior and functionality of the circuit, and have our software synthesize automatically to an RTL specification.
Ben Kuo: How do you handle things like timing, which chip designers are often concerned about?
Jason Cong: Timing is a loaded word. Some people mean throughput, or frequency, or total latency. For example, when you buy a microprocessor, when you go from 1Ghz to 2Ghz it doesn't mean your performance will double. Your clock cycle will be shorter, but some operations may go through more cycles to complete. Given a RTL design, it can be very challenging to meet a given frequency target. What is nice is our tool can actually allow you to specify multiple frequency targets, and automatically generate different RTL designs, on for each target, so that you can explore a much larger design space. You can't do that with manual RTL coding. With our automated tool, you can write your specification once, and try to run it at different speeds. It also allows you to specify your throughput or expected latency.
Ben Kuo: Can you talk about the decision to spin out?
Jason Cong: Students in my group saw this as a very viable approach to take the technology into commercialization, and to make an impact on the industry. There were several triggering events. When we started the research five years ago, ESL was not quite there. We did this research for very strong technical reasons. Two year ago, the market became very hot, and several startups in the area appeared in ESL synthesis and ESL verification. The report from Dataquest in November and December of 2005 predicated about a 36-37 percent annual growth in the next five years, with the fastest growing segment in EDA, so we saw there was market need. There was also the maturity of the technology. Our xPilot system was the third try after five or six years of research--we had written two other prototype systems, and didn't like them. We saw we had lots of potential to give commercial quality results. Another trigger event, was a few students were finishing up their Ph.D.'s, and thinking on where to go next, and it made sense to create a startup to take this technology to commercialization.
Ben Kuo: Can you talk about the company's funding and support?
Jason Cong: Our Series A funding closed earlier this year, with a combination of corporate funding, Venture Capital, and private investment.
Ben Kuo: Let's talk about your prior startup--can you tell us about your experience there?
Jason Cong: The prior startup was a different experience. I was in the early part of my teaching career. I joined UCLA in 1990, and have been there for 16 years. Since early 1990's I have been studying field programmable field arrays, in the area of architecture studies, and physical synthesis. Around 1997/1998, several companies became very interested in what we were doing. In the beginning, we just started research collaboration. Very soon, it went beyond casual interest--the companies we were working on went beyond wanting to have some software to play with to building a multi-million dollar product line (10-100M) based on our products. They needed support, customization, and enhancements, which are not very suitable to a university environment. I had a couple of students graduating, and we had customers banging on the door to use the technology. We put the company and team together, and within the first month after the R&D team was in place, we got our first multi-million dollar order from a key customer. The company was totally bootstrapped, and was in full motion since 1999. We were acquired by Magma Design Automation in 2003, for a fairly successful exit. We had about 25 people in the company at the time, mostly R&D in the U.S. and at a design center in Beijing.
Ben Kuo: How's the startup activity there at UCLA?
Jason Cong: Obviously, the best known startup from UCLA is Broadcom, which was started by Henry Samueli, who was a UCLA faculty member. His partner was also a UCLA graduate. They were tremendously successful, and have also been very generous and supportive of UCLA. Our school was renamed to the Henry Samueli School of Engineering after Henry Samueli donated $30M to the school. There also has been various startup activities within the school of engineering. The University has in general been very supportive, as they like to see technology transfer out to society. For Xpilot/AutoESL, we have a formal license agreement with UCLA, and are the exclusive licensee of the technology for commercialization.
Ben Kuo: Thanks!