Kurt Petersen has been called a "founding father of MEMS" because of his pioneering research work on microelectromechanical systems at IBM Corp. in the 1970s. He has co-founded four MEMS startups, the most successful of which, NovaSensor, has delivered hundreds of millions of MEMS sensors worldwide.
Petersen also pioneered the fusion of MEMS and microfluidic technology in a biological detection system made at another of his startups, Cepheid. It's now used by the U.S. Postal Service to screen mail for possible anthrax contamination.
Petersen's most recent startup, SiTime Corp., plans to substitute MEMS and CMOS chips for all the billions of bulky quartz-crystal oscillator cans that maintain time bases for everything from watches to the electronics on printed-circuit boards. He recently told EE Times' R. Colin Johnson where he sees MEMS going in the future.
EE Times: Your work in the 1970s resulted in the seminal IEEE paper "Silicon as a Mechanical Material." What first got you started in MEMS?
Kurt Petersen: In a word, Stanford University. I had just finished my PhD in electrical engineering [from the Massachusetts Institute of Technology] and was interviewing in the [San Francisco] Bay area. I went to Xerox Parc to interview. But this was in 1975 and they didn't have any labs yet, so they were doing most of their work in Stanford's labs.
During the interview, they took me over to Stanford, where I saw a poster from Jim Angell's group. It showed a gas chromatograph-on-a-wafer done by Steve Terry. That's when it hit me--you can use silicon as a mechanical material.
EE Times:So your first industry job was at Parc?
Petersen: No, I ended up joining IBM in San Jose, and within about four months was doing micromachining. We built a little fab and began dealing with MEMS-specific issues like various applications, stiction and stress.
EE Times: And what types of devices were you working on?
Petersen: Back then we were trying to do optical MEMS. The first [silicon-dioxide] beams we made were to deflect light and were very, very flat. That made me think, "Wow, this really works." But on the second try the beams got all curled up like potato chips. So we discovered we had a lot of issues to deal with at the beginning, just to get consistent results.
EE Times: Were you building optical devices exclusively?
Petersen: No, we built some accelerometers, some switches, some inkjet nozzles, even some resonators.
EE Times: Were all of your devices silicon?
Petersen: We typically built our mechanical structures on silicon wafers. These wafers have a P+ etch-stop diffusion covered with a 10-micron epitaxial layer. We would then etch the epi layer away to release silicon-dioxide mechanical devices.
EE Times: So you were doing bulk micromachining of single-crystal silicon?
Petersen: Yes, but in the early '80s I met with Roger Howe, who was then an electrical engineering graduate student at Stanford [and is now a Stanford professor]. He introduced me to surface micromachining using polysilicon.
EE Times: Did you call your work "MEMS" back then?
Petersen: No. The MEMS acronym was actually introduced in the late '80s. I think the word "MEMS" was coined by NSF [the National Science Foundation]. The term micromachining was coined by [the late] Jim Angell. Jim was a visionary professor at Stanford whose students did some of the earliest work. But then he went on to other work and retired a few years later. His students spun off a company [Microsensor Technology] to manufacture that MEMS-based gas chromatograph, which was eventually purchased by [Hewlett-Packard]. That's how HP got its start in MEMS.
EE Times: When was that?
Petersen: HP bought Microsensor in the late '80s or so, but before then--up until my paper in 1982--most of the MEMS development work was being done at individual labs. Besides Stanford, there was work being done in Henry Guckel's lab at the University of Wisconsin, in my lab at IBM in San Jose and also at [the IBM Research center in] Yorktown Heights, N.Y. [It was also being done by] TI [Texas Instruments], Kulite in New Jersey and a few other places. But for the most part, people didn't realize that there were other labs that were doing work that was similar to theirs--MEMS hadn't formed into one field yet.
EE Times: Did your paper bring people together and define MEMS as an independent research field?
Petersen: I think so.
EE Times: What do you see as the milestones that have been passed since you presented that paper in 1982?
Petersen: First of all, there was no volume production at that time, because the big companies had not gotten heavily involved yet. But by 1985, MEMS pressure sensors were in volume production at Motorola. Then the automotive applications took off at Delco and industrial applications took off at Foxboro, then at National Semiconductor. And lots of research [was done] at IBM.
EE Times: Didn't you start your own initial MEMS company around that time?
Petersen: Yes. SiTime is my fourth MEMS startup, but the first was Transensory Devices, which I started with Jim Knutti, Henry Allen and Joe Brown, who joined me there from IBM.
EE Times: What was the biggest obstacle for your first startup?
Petersen: Definitely the fab. SiTime is fabless, but in those days there were no foundries, so the first thing we had to do was build our own fab from the ground up. It was small, but it gave us control over all aspects of our development work.
EE Times: Your own fab? That must have been incredibly expensive for a startup.
Petersen: Well, we were just doing applied research. In those days, we weren't yet taking into account the production issues or the testing issue or the packaging issues. Most of our revenue was from R&D contracts, but I wanted to make products. So when I met Janusz Bryzek and Joe Mallon, who were already very experienced in traditional pressure sensors, I left to help start Novasensor [a MEMS sensor maker that has since been acquired by GE]. We had MEMS pressure sensors in production just six months after founding.
EE Times: How did you get a fab up and running so quickly?
Petersen: Instead of building our own at NovaSensor, to start, we would rent time on the third shift at other people's fabs and make our sensors there in the middle of the night.
EE Times: Did you actually manufacture your MEMS pressure sensors one shift at a time like that?
Petersen: Yes. The amazing thing was that we took our first order for 50,000 chips before we even had a product, then delivered them in 10 weeks. And not only did they work, but they had tighter specs and worked much, much better than the [traditional sensor] parts they were replacing.
EE Times: What types of structures were you building then?
Petersen: At NovaSensor, we were using several technologies that are commonplace today, but were revolutionary in those days. For instance, building reproducible piezoresistors on the same chip with the MEMS structures was very, very tricky, but we solved the problems by using ion implantation when everybody else was using diffusion.
We started using silicon nitride too, which was also revolutionary for use in pressure sensors at that time because of stress issues. And we used electrochemical etch stops, so we could accurately define the thickness of the diaphragm. Everybody was just timing the etch, but by using stops we gained much tighter control.
Since then, there have been hundreds of millions of pressure sensors built using our process at Novasensor. Motorola also serves that market, but I think that even today Novasensor has a bigger market share.
EE Times: So did you go directly from Novasensor to found SiTime?
Petersen: No, I stayed at Novasensor for 11 years, but I wanted to start a company that actually made end-user products, so I founded Cepheid. [That company] used MEMS in microfluidics devices to do rapid DNA analysis using the polymerase chain reaction with fluorescent detection.
EE Times: So Cepheid's microfluidic devices identify medical maladies by their DNA?
Petersen: Yes, but the post office is actually Cepheid's largest customer now. For medical tests, you can tolerate as much as a 1 percent false-positive rate, but to identify anthrax in letters for the post office, they run millions of tests. And every time you have a false positive, it's a huge deal. [It costs] them millions of dollars to shut down a facility and bring in the FBI and the CDC and test all the employees, so they demanded a false-positive rate of better than one in 500,000. Cepheid's microfluidic devices exceeded that spec. To date, they have conducted over 4 million tests without a single false positive.
EE Times: How did you go from making microfluidic devices to making MEMS resonators for SiTime?
Petersen: One day Joe [Brown] came to me and told me about this great process that [Robert] Bosch had to make resonators. And I told him, "Come on Joe, people have been trying to perfect resonators for 30 years--it's just too hard a problem."
The problem with resonators is that they sense everything--temperature, stress, water vapor, everything. And anything will change their frequency, making them unsuitable as frequency standards. I knew from my experience at IBM that you had to be stable to within parts per million or even parts per billion, which I thought was impossible.
EE Times: Even harder than the anthrax problem?
Petersen: Yes, much, much harder. So Joe convinced me to take a look at their process, and after hearing about it, I came away convinced that Robert Bosch had finally developed a process that was good enough. It is going to be hard for our competitors to match [it].
EE Times: What makes Bosch's process, for which SiTime has an exclusive licensing deal, so different?
Petersen: It's the packaging. Everybody else has to add some clunky laminated encapsulation process to seal their MEMS structures. Bosch has the only MEMS process that seals at the wafer level in the clean room at 1,100°C by depositing an epitaxial layer that is only 15 microns thick. That packaging solution eliminates all contaminating water vapor and [other] contaminants, and is now unique to SiTime's resonators. Not only does it ensure that absolutely no contaminants get inside, but it also enables our MEMS chips to look and behave just like every other silicon chip.
EE Times: So will silicon resonators replace quartz crystals in all applications?
Petersen: I always say no to that question, because it's a very big marketplace. We divide the market into three segments, each of which is about a billion-dollar market today. The low end is like the crystals in your watch--very low-cost with a precision of about 200 parts per million. The middle billion dollars is consumer electronics, with a precision of about 50 ppm. Then [you have] the top billion dollars, which [is composed of] high-precision parts such as oven-controlled crystal oscillators for cell phone basestations and TCXOs [temperature-compensated crystal oscillators] for cell phones--there the precision is about 1 ppm.
EE Times: So are you going to try to serve that whole market?
Petersen: Today our parts are aimed at the middle billion-dollar consumer electronics market. But we have two more resonators in development, one that will move upward into the top $1 billion tier and another that goes into the bottom $1 billion tier.
EE Times: Do you have any plans to make any other MEMS devices besides resonators and oscillators?
Petersen: By 2010, the resonator and timing-chip markets will total over $7.5 billion--that's enough for us. SiTime will be directly attacking about one-third to one-half of this [consumer electronics] market. n
Kurt E. Petersen
April 13, 1948, San Francisco
BSEE, UC Berkeley, 1970, MSEE, Massachusetts Institute of Technology, 1972; PhD, MIT, 1975
Chairman and CEO, SiTime Corp., 2004-present
President and CTO, Cepheid, 1995-2004
Executive vice president of technology, NovaSensor, 1985-1995
Vice president of technology, Transensory Devices Inc., 1982-1985
Research staff member, IBM Corp., 1970-1982
Winner of IEEE Simon Ramo Medal, 2001
Elected to the National Academy of Engineering, 2001
Fellow Member of IEEE