forwarded message from Physics of Computation Seminar

Robin Hanson (hanson@hss.caltech.edu)
Wed, 14 May 1997 10:50:04 -0700 (PDT)


------- Start of forwarded message -------
From: Physics of Computation Seminar <poc@im.lcs.mit.edu>
Cc: phil@hplwbc.hpl.hp.com, culberts@hplwbc.hpl.hp.com
Message-Id: <199705141604.MAA05233@im.lcs.mit.edu>
Date: Wed, 14 May 1997 12:04:19 -0400 (EDT)
Subject: Physics of Computation Seminar, Tuesday, May 20

MIT PHYSICS OF COMPUTATION SEMINAR

Date: Tuesday, May 20 1997
Time: 3:30 pm (Refreshments after the talk)
Room: MIT NE43, 8th floor AI Playroom (545 Tech Sq)

"Imperfect Computers"

PHIL KUEKES and BRUCE CULBERTSON
H P Laboratories

Abstract:

We have built a reconfigurable computer with 0.22 square meters of
CMOS silicon in 864 chips, three quarters of which have known
defects. The Teramac custom computer can be thought of as a giant
programmable gate array with half a billion configuration bits.
The large number of configuration bits and the two million wires
available for routing make it very easy to place and route logic
designs onto the physical network without depending on an unbroken
symmetry in the physical network.

Teramac is big enough to let us begin to explore some of the
issues which will be faced by architects of the first nanoscale
computers. How do we make a computer with nanometer scale
components? To make a computer we need not only small components
but a method of connecting them in a reasonably arbitrary way,
chosen by the computer designer. Most self assembling strategies
at the molecular scale create very regular structures which lack
the complexity required of a computer. More elaborate assembly
strategies are likely to be very imperfect. The topology of the
Teramac interconnect is regular only in its fractal
dimensionality. The place and route software does not depend on
significant symmetry. If you have enough wires and switches you
can first build an imperfect machine and then design a perfect
one.


Hosts: Norm Margolus and Tom Knight


This talk is part of a seminar series on adapting computers and
computations to the constraints of, and opportunities afforded by,
microphysics; and on the development and application of the physical
theory of computation and information. This seminar series is
supported by the MIT AI Laboratory's Reversible Computing Project.

Please forward this notice to anyone who you think might be interested.
Anyone who wishes to be added to the distribution list for these seminar
announcements should send email to "poc@mit.edu". Copies of papers
related to recent and upcoming POC talks are available through the World
Wide Web at "http://www.im.lcs.mit.edu/cgi-bin/poc.pl".
------- End of forwarded message -------