No Time Like MEMS Time - by Dr. Ian Cutress
The next generation of high performance processors are really kicking it up a notch. We’ve moved from 300 W to 600 W and now we’re looking at 1200 W+ designs quickly coming down the pipe. TSMC published research a couple of years ago about how to cool these monsters up to 2600 W, and IBM has a lot of patents on intra-die-stacked cooling from the mid-2000s. Moving to larger silicon, or lots of chiplet silicon, at high power is driving high frequencies, and one of the things we don't often think about is how do we get those frequencies - how do we enable the power efficiency of that chip through a smart timing solution that remains flexible and consistent?
Historically, we've enabled this sort of timing through oscillating quartz crystals. These are physically manufactured things, literally with saws and polishing, to within 0.1 Hz, just to make sure that our chips run at the right speed. This industry is being innovated rapidly through the use of in-silicon devices called MEMS: Micro-Electro-Mechanical Systems. In short, instead of vibrating piezoelectric crystals, we can use silicon itself for that function.
The beauty of using MEMS over quartz is multi-faceted. We can enable much more precise silicon timing and oscillators. They consume less power, and suffer lower delay in rapidly changing frequency environments. They can be made programmable, allowing a design that can be applied to a large array of industries with a single unified solution. Of course, there is still a market for custom designs – but the whole point about programmability is that it enables that customization straight out of the fab.
