Will microelectromechanical systems (MEMS) follow memory, mixed-signal and regulator circuits onto the SoC bandwagon? This interesting thought, which came up in an interview with Dave Monk, sensors development engineering manager at Motorola, reveals not just the enormous power latent in our research-and-development community, but also the extent to which market reality has colored everyone's thinking since 1999.
Technically, the answer is a clear yes. MEMS in some applications-accelerometers for airbags always come up here, though there are other applications-are already being fabricated with some active circuitry on the same substrate. And since the beginning researchers have assumed that MEMS-with-circuits would be the eventual goal.
Hence, there are three techniques for getting there. In MEMS-first, pioneered at Sandia, you fabricate your MEMS in a cavity on the surface of the wafer, and then fill in the hole with some kind of sacrificial material. Then you planarize the wafer and build circuitry over the top, either in areas not occupied by the MEMS devices, on a poly layer or on a bonded SOI layer. Finally, you remove the sacrificial material, releasing the moving parts.
Similarly, there are techniques for fabricating the MEMS, either concurrently with the active electronics or after.
But the real challenges are economic. System-on-chip real estate is so expensive, and MEMS are so relatively gigantic, that there has to be some compelling reason to take up die area with MEMS.
MEMS-on-SoC may only be economically feasible when there is a functional reason to combine the two. One reason might be signal integrity-some MEMS sensors have extremely low signal levels. Another has yet to be explored: It is likely that when designers can look at the entire electromechanical system as a whole, with close coupling among digital computation, analog and mechanical functions, things will be possible that can't be done in isolation. But that day is waiting, like so many other things, on adequate design tools and models.
Ron Wilson covers microprocessors, programmable/reconfigurable logic and the chip design process. He can be reached at rwilson@cmp.com.
http://www.eet.com