My tour guide was the fab's manager, Craig Core, who has an enthusiasm for his work that's not always found among his peers pushing the bleeding edge of submicron semiconductor production. People seem to actually be having fun there. "That's one aspect everybody notices about our place," Core smiles. "It's not chasing the linewidth curve, which has some state-of-the-art advantages and some high-pressure disadvantages. Here, people are developing new things and doing something novel all the time. Our engineering turnover rate is very low."

The fab's dual roles in production and R&D—"it's a lot more on the 'D' side than the 'R' side," says Core—may have something to do with this exciting yet relaxed work environment. On the business side, it doesn't hurt that Analog has some 65% market share in the airbag sensor market, the leading application for the devices made there. Since most of the factory's process equipment and facility layout mimic a small chip fab (they do make chips there too, after all), there's also a comfort zone for many of the engineers and technicians who started out on the semiconductor side.

"Our strategy has always been to stay as close to semiconductor manufacturing as we can," explains Core. "There's a big industry out there developing tools and techniques to do things...we don't deviate so dramatically that we can't take advantage of that." A walk around the fab reveals the usual equipment suspects. Since it's a 6-inch house running a single-level metal, 3.0-µm BiCMOS process for circuits with micromachining down to 0.9 µm on the sensors, many tools are what Core kindly calls "ancient." He says the design is "as robust as can be, it can handle any thermal cycle we throw at it...we wanted an IC process that was bulletproof."

They've come a long way in terms of manufacturing expertise and fundamental understanding of the processes involved since Analog launched its first MEMS product in 1991 and ramped the Cambridge fab in 1997. "There were many things we had to learn about materials," Core notes. "Take vertical stress gradient: when you release stress vertically through a film, there's a bending motion.... When we first started, it was out of control and we had bending all over the place [in the micromachined sensors]. We came up with a way to measure it in-line and can correlate from end-of-line results to in-line parameters, so we have tight control locally to ensure we can make stuff day in and day out. I sleep a little bit better because of that."

Two years ago, they started developing optical MEMS. I asked the fab manager about some of the key processing differences between inertial and optical devices. "The biggest [difference] is that you need an optical reflective coating, something to bounce light off, because we're making little mirrors. The best metal for that is gold, traditionally a horrible metal to bring near a wafer fab. From a microcontamination standpoint, we set up a lot of protocols that other people have done when they've dealt with copper... we're going with different-color boxes, operator training, different gowning areas and gown types, trying to segment it as best we can. We also started doing lifetime measurements on many of our tools so we can check if we've got metal contamination in our silicon."

As the demand for MEMS devices burgeons, Core says they're looking to grow capacity soon, through the utilization of additional manufacuring space on-site and designed-in device shrinks. He mentions how accelerometers are now built into game boxes and car alarms, noting the still largely untapped potential for tilt sensors and other micromachines. "Anything that needs interaction with the real world and movement, there's an opportunity for applications."