near-term processing power growth

From: Eugene Leitl (eugene.leitl@lrz.uni-muenchen.de)
Date: Fri Feb 11 2000 - 15:10:31 MST


Ramez Naam writes:

> Given that today's top end processors are running 750Mhz, and that Moore's

AMD just released cryogenically unassisted 1.1 GHz processors.

> law calls for roughly a 60% performance increase each year, one would
> project 3Ghz processors in 2003 and nearly 5Ghz processors in 2004.
 
If you're going by straight clock, normal SiGe should have a hard
limit at ~150 GHz, or so. One would have to drastically simplify the
CPU core before, of course. These gate delays add up too much.

Things will also stall briefly by the time we have purely embedded
VLIW RAM/CPU architectures (essentially, just a really smart RAM with
a 4-16 kBit on-die bus) with no external RAM. Going wafer-scale
integration will offer some room for growth, by reducing costs, and
increasing parallelism (smaller dies required by higher yield result
in higher parallelism).

After that we have to move on beyond semiconductor
lithography. Photopolymerizable (first mono-, then multilayer)
Langmuir-Blodget 2d crystal molecular circuits should be available by
then, probably piggybacking on top of vanilla silicon
circuits. Afterwards you get 3d molecular crystals (from virus-sized
elementary cells, probably from engineered protein
encasing/autoassembling the molecular circuit). You can shrink that
with Drextech (at least 3 orders of magnitude denser packing), but
that's it, then. No more Moore no more, unless femtotech arrives.

> Also, as a computer scientist I find it customary and realistic to buffer
> near-term product release dates by a factor of 50%. (Though in this case it
> sounds like IBM may already have done the proper buffering.)



This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:26:45 MST