Moore's Law hitting the wall in 10 years

From: Hal Finney (hal@finney.org)
Date: Sun May 19 2002 - 13:23:43 MDT


We've been riding the Moore's Law wave for a long time, maybe even a
century by some definitions. But slashdot points today at an article
in EE Times indicating that the current technology is going to hit the
wall in about ten years. "Bell tolls for CMOS, with successor nowhere
in sight", http://eetimes.com/semi/news/OEG20020516S0024.

Current feature sizes are .13 microns, and to stay on the Moore's Law
curve they have to shrink by a factor square root of 2 (about 1.4) every
18 months (or two years, depending on the doubling time of the flavor of
Moore's Law you are using).

Next year feature sizes are expected to shrink to .09 microns, at which
time the terminology is changing to 90 nanometers. Gate sizes are 1/2
of this so they will be at 45 nm next year. By 2005-2006 feature size
should be 70 nm with gate sizes of 35 nm. At that point they have to
start making some fundamental changes. By 2012 gate lengths should
be 10 nm and that's just not going to work with the current flavor of
semiconductor technologies. We have to have something fundamentally
new by then.

The problem is that it will take an incredibly aggressive R&D schedule
to get any of the new technologies out of the lab and into commercial
production in 10 years. Chips are BIG business. Think about how many
chips come off the assembly lines every year. Consider changing every
chip manufacturing foundary into the world to instead make something
with carbon nanotubes, at commercial levels of quality, consistency,
yield and testability. We have to have that in ten years, or else
Moore's Law is going to fail (or at least hiccup).

They're also looking at benzene-based molecular devices, and at 3D silicon
technology, along with the nanotubes. But let's face it, getting any of
these into commercial production at the scale needed by our information
economy in ten years is a crazy long shot. Of course if someone succeeds,
they will be well rewarded indeed.

Hal



This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:14:12 MST