Monday, December 03, 2012

The future of Si: Into the fog

Today we had a visit at Rice from Mike Mayberry, the VP for the Technology and Manufacturing group at Intel.  He gave a very good talk about where semiconductor electronics is going (naturally with the appropriate disclaimers that we shouldn't buy or sell stocks based on anything he said).  The general rule is that there is a metaphorical fog out there about ten years off, beyond which it's not clear what the industry will be doing or how it will be doing it.  However, for the comparatively near term, the path is fairly well known.  In short: complementary metal-oxide-semiconductor (CMOS) based on Si is going to continue for quite a ways longer.  Device design will continue toward improved electrostatic control of the channel - we will likely see evolution from tri-gate finFET structures toward full wrap-around gates.  We are firmly in the materials-limited regime of device performance, and Intel has been playing games with strain to improve charge mobility.  They have even experimented with integration of III-V materials via epitaxy onto Si platforms.  It's pretty clear that there is a healthy skepticism about post-CMOS alternative technologies, particularly given the absurdly low cost and high volumes of Si.  Intel ships something like 4 trillion transistors per minute (!).  Other remarkable facts/figures:  Network traffic right now is around 7 exabytes (1018 bytes) per day, or the equivalent of 17000 HD movies every second, including around 204 million emails per minute on gmail, and these numbers are increasing all the time.  Amazing.

3 comments:

DanM said...

It is the first talk I've attended in quite some time where the word "quintillion" was used.

Charles Day said...

Intel, unlike an academic lab, ultimately has to turn research into products and services that people will buy.

Did the speaker mention what consumers might want to do with extra, beyond-silicon computing power?

Douglas Natelson said...

Charles, that's an important point. You can tell from the talk that Intel is still smarting a bit from misjudging the market's demand for mobile (low power) computing hardware - they're still playing catch-up because they were slow off the mark. Their product development people clearly worry a lot about whether they're overlooking the next big market segment. That said, I suspect that we'll need exascale computing to run Office 2023....