Blazing a Trail for Chip Designers
July 2006 -- He’s been called ‘one of the best-known academic researchers in design automation.’ Indeed, Jacobs School of Engineering professor Andrew Kahng has led the definition of the technology roadmap for design of integrated circuits since the mid-1990s.
Now the computer engineering professor can also be called an entrepreneur.When he returns full-time to UCSD in the fall after a two-year leave of absence, Kahng will be able to offer students first-hand insights into the process of taking a technology from the lab to the marketplace—and doing so successfully.
Kahng formed Blaze DFM, Inc. with former Ph.D. student Puneet Gupta and industry veteran David Reed.The Sunnyvale, CA-based company, which now counts 20 employees, has hit all of its benchmarks since closing a $6 million initial round of venture-capital financing in late 2004. Alpha testing of its first product, Blaze MO, began in March 2005, and the product was installed at a beta customer site last October.This spring, a customer obtained its first product silicon optimized with Blaze’s software tools.
“Results so far with our first product have exceeded our customers’ most optimistic expectations,” says Kahng.“Now we want as many chip designers as possible to benefit from it.”
The “DFM” in the company’s name stands for “design for manufacturing,” the concept of designing integrated circuits in a way that improves the percentage of chips on a wafer that meet product specifications and therefore can be shipped. But as chip sizes drop and densities increase, improving yields on sub-100- nanometer (nm) processes becomes an even bigger problem because of power leakage and manufacturing variability.
“DFM is the semiconductor industry’s only hope for a costeffective continuation of Moore’s Law,” says Kahng.“It’s been my research focus since 1997. I felt I had to do something to help solve critical challenges of variability, power and cost—if only as a safety net for the industry.”
Kahng’s new technology does DFM with a twist. He calls it ‘electrical’ DFM. Most of the DFM tools used today have a geometric mindset: they focus on whether the shapes of circuit features in silicon exactly match the layout as designed. But, argues Kahng, shape fidelity is not the same as maximizing the number of sellable chips per wafer (and therefore revenue per wafer).To optimize the yield of chips on a wafer, he says, designers must take into account electrical requirements, notably power and timing.
“Leakage power is a big part of the problem at the sub-100nm level,” notes Kahng.“Because of leakage currents, a fast chip may burn too much power and be unusable in, say, a mobile application, and a slow chip may not be usable either. Electrical DFM looks at these electrical requirements and optimizes the chip design so that manufactured silicon meets parametric specifications while burning as little power as possible. Our products also directly mitigate the manufacturing variation that is so challenging at leading-edge process nodes.”
When he decided to start a company, Kahng recused himself from negotiations with UCSD’s Technology Transfer and Intellectual Property Services (TechTIPS) office to avoid any conflict of interest. “Blaze DFM has been a canonical example of what technology transfer at UCSD and TechTIPS is all about,” says Kahng.“UCSD and the Jacobs School enable faculty to take their innovations into the marketplace, and to return to the university with a clearer understanding of how to innovate with real-world impact.”
Blaze obtained an exclusive license to Kahng’s technology, and UC received equity in the company and royalty interest from future revenues. Adds Kahng: “If the company is a success, the university, my home departments of ECE and CSE, my lab— as well as I and other inventors of the IP—will benefit.”
His co-founder Puneet Gupta (Ph.D., ’04), says of his experience so far,“Working in Andrew’s lab kept me much closer to industrial requirements than most other academic research groups, so the transition to a pure industry setting was not as tough as I had anticipated. Moreover, DFM being a nascent field, there has been ample research activity inside Blaze itself.We hope to bring to the semiconductor industry several radically new ideas over the next few years.”
Later this year, Blaze DFM’s second product is slated to hit the market, but Kahng will be back at UCSD. He is looking forward to returning to teaching and spending more time at home in San Diego after commuting during the work week to northern California.“Every week and every day has a cost of being away from home and family and students,” explains Kahng.“So I never lose sight of the timeline and the mission, and of my scheduled return to UCSD when my leave expires.”