The news release says a joint venture between Rice University and Nanyang Technical University of Singapore will launch a joint venture in very low power embedded chips, under Rice’s Krishna Palem (right).
In fact, we’ve got something much bigger, something which demonstrates just how powerful Rice is becoming.
Palem, who co-authored a report on probabilistic CMOS chips last year, which will be at the heart of the effort, is actually a very new recruit to Rice.
He had been, until this semester, at Georgia Tech, where he had founded its Center for Research on Embedded Systems and Technology. He’s now a joint fellow of both Rice and CalTech, where he is listed as a Most Distinguished Scholar.
The CalTech biography describes something of Palem’s journey, which
includes stops at IBM, Hebrew University in Jerusalem, as well as
Nanyang and National Universities in Singapore. He’s not just a
scientist or an engineer. He’s what could best be termed an
entrepreneurial researcher, for whom assembling money and alliances is
as crucial as the lab work.
Regardless of what happened between Palem and CREST or Georgia Tech, Rice is very lucky to have him in its orbit. I am certain that, back
when I was there, we never could have gotten such a heavy hitter. Rice
people can learn more, straight from the man himself, alongside two
other Rice professors, Vivek Sarkar and Moshe Vardi, next Tuesday.
Here’s a taste of Dr. Palem’s talk:
First, as device feature sizes approach the low nanometer scale,
limits on the power consumed and the heat dissipated pose hurdles,
since noise and heat are impediments to the reliable operation of
transistors. In this talk, we first see that noise-induced "unreliable"
CMOS devices can be harnessed to yield useful computing fabrics. These
probabilistic fabrics yield more than two orders of magnitude of gains
in efficiency over competing designs based on conventional
(deterministic) CMOS technology. Applications from embedded computing
are all shown to benefit from this approach.
Second, with
increasing transistor densities, non-recurring engineering cost and
time-to-market pose an ever-increasing impediment to the proliferation
of customized computing systems. With an optimizing compiler at the
heart, a range of technologies through which we have successfully
overcome these impediments will be described next.
In English, you can cut power use by 99% or more in a CMOS transistor by
forgetting about defining specific paths for the electrons. A number of different
techniques are being developed to turn this finding into useful
product, where getting it out by Friday is the only way you’re going to
make money.
Y’know, I don’t care that Georgia Tech’s football team beat Notre Dame last week while ours lost to Nicholls State.
The score, where it counts, is Sammy 1, Buzz 0.
