IP: re: world's fastest computer

[David Farber <farber () cis upenn edu>]

To: farber () cis upenn edu
Subject: Re: IP: world's fastest computer 
Date: Fri, 09 May 1997 18:12:50 -0400
From: "Jonathan S. Shapiro" <shap () eros cis upenn edu>


Dave: 


I am fascinated by John McDonald's grant announcement.  As far as I
know, the laws of physics haven't changed substantially in the past
few years, nobody has discovered a new spatial dimension that is
readily accessible, and thermodynamics hasn't undergone a radical
reformation.  Given which, I have to wonder why DARPA is funding work
in bipolar technology.


In a given lithography, the critical challenges to building processors
that go at high clock rates have to do with heat dissipation, gate
delays, and clock synchronization.  Both gate delays and clock
synchronization problems are reduced by minimizing the size of the
standard cells from which modern processors are designed.  In a given
lithography, CMOS cells are smaller than their NMOS counterparts.
CMOS has always had lower power requirements than bipolar, resulting
in lower heat.  As heat goes with the square of the frequency, a
factor of 10 increase in clock rate means a factor of 100 increase in
generated heat.  Every bit you can save matters.


Forgetting about cost issues, CMOS won on pure performance over a
decade ago.


I can accept without difficulty that hooking up a boat load of Pentium
Pro's is not the right way to build a superfast machine, but absent a
change in device physics bipolar isn't either.


Do you have any insight into what the thinking may be here?




shap