How To Keep Chips Cooler

When I posted this on the USENET, several folks wrote back with some good suggestions. The best ones seemed to imply that a blackbody radiator would need to be pretty hot to bleed off enough energy as light. But are LED's blackbody radiators?

In any case, I'm just posting this here so I know where to find it in the future. Feel free to explain to me whether this idea should be deleted completely.

My November 18, 1999 Posting

Path: newsmaster1.prod.itd.earthlink.net!nntp.earthlink.net!posted-from-earthlink!not-for-mail
From: Peter Wayner 
Newsgroups: comp.arch,sci.physics,sci.engr.semiconductors,sci.electronics.design
Subject: Can We Evacuate Semiconductor Heat As Light?
Date: Thu, 18 Nov 1999 10:25:00 -0500
Content-Transfer-Encoding: 7bit
X-Posted-Path-Was: not-for-mail
X-Accept-Language: en
Content-Type: text/plain; charset=us-ascii; x-mac-type="54455854"; x-mac-creator="4D4F5353"
X-ELN-Date: 18 Nov 1999 15:31:49 GMT
X-ELN-Insert-Date: Thu Nov 18 07:35:05 1999
Organization: Flyzone.com
Lines: 86
Mime-Version: 1.0
Reply-To: pcw@flyzone.com
NNTP-Posting-Host: ip175.baltimore20.md.pub-ip.psi.net
Message-ID: <38341A35.76BDB594@flyzone.com>
X-Mailer: Mozilla 4.51 (Macintosh; I; PPC)
Xref: nntp.earthlink.net comp.arch:50464 sci.physics:335322 sci.engr.semiconductors:16055 sci.electronics.design:151178





I had an interesting idea about the heat problem
with chips. It might be very dumb, but it's stuck
in my brain. Sometimes the only way to cure these
brain fevers is to post the ideas to the Net and
see if anyone can shoot them down.

One way to analyze the behavior of the transistors on a
chip is by watching for photons coming from the
transistors. This is a great way to diagnose clock
skew and other timing problems. Apparently some
transistors  discharge energy as photons when
switched. This makes sense because LED are just
diodes and diodes are just transistors.

I wonder if anyone knows what fraction of energy
is dissapated as  photons and what comes off as
heat. Can this fraction be adjusted so that most
of the energy comes off as light? I think that it
can because LEDs are so efficient that they're
replacing light bulbs in lowpower applications.

A better question is whether the dissapation can
be adjusted in transistors that make good
computers. That is, can we take the transistors on
a G4 or a Pentium and modify the construction so
they radiate a significant fraction of their waste
energy as photons?

This would have two advantages. First, the old
EPROM packaging with a window over the chip could
be revitalized. I would think that light is a
significantly more efficient way to evacuate the
waste energy from a chip. Heat sinks and fans are
a pain. There's no way that convection moves at
the speed of light. We could go back to the days
of vacuum tubes when circuitry glowed. The movie
directors who have to shoot endless beige boxes
would love us! Good bye Dell beige, hello WOPPR.
The new transparent iMacs will really light up!

The second potential advantage would be the
opportunity to recycle the energy. A mirror or
some other device could focus this light on a
solar cell so the photons could be converted back
to electrons. Perhaps a parabolic shaped reflector
on top of a chip could focus the waste photons on
a solar cell in the center of the chip.

I suspect that there are many obstacles in
creating these transistors and packing them
together on a chip. The part of the transistor
that radiates the light must be exposed.
Unfortunately, the desire to shorten the distance
between transistors means that many chip designers
are creating vertical transistors that pack
neatly. The chip skylines are looking more like
New York City and less like suburbia. Where would
you rather sunbathe? Maybe the transistors can be
built up on towers of silicon imitating the same
way that windows cover the skyscrapers?

There are also more complicated questions about
the switching speed and the potential energy of
the electrons. Perhaps other design considerations
would push the bulk of the radiation into parts of
the spectrum that annoy the FCC. Alas, my physics
is a bit rusty.

There are details, however, that I think could be
worked out.  At the very least, this should be
interesting design consideration that could solve
heat problems and perhaps make it feasible to
recycle a significant fraction of the power used
in computation.

If anyone has any ideas on the topic, I hope they
will write me. Please shoot this down if you can.

Peter Wayner pcw@flyzone.com

My Follow Up

Path: newsmaster1.prod.itd.earthlink.net!nntp.earthlink.net!posted-from-earthlink!not-for-mail
From: Peter Wayner 
Newsgroups: comp.arch,sci.physics,sci.electronics.design,sci.engr.semiconductors
Subject: Can We Evacuate SemiCon Heat as Light II
Date: Fri, 19 Nov 1999 17:08:12 -0500
Content-Transfer-Encoding: 7bit
X-Posted-Path-Was: not-for-mail
X-Accept-Language: en
Content-Type: text/plain; charset=us-ascii; x-mac-type="54455854"; x-mac-creator="4D4F5353"
X-ELN-Date: 19 Nov 1999 22:14:11 GMT
X-ELN-Insert-Date: Fri Nov 19 14:15:30 1999
Organization: Flyzone.com
Lines: 50
Mime-Version: 1.0
Reply-To: pcw@flyzone.com
NNTP-Posting-Host: ip105.baltimore20.md.pub-ip.psi.net
Message-ID: <3835CA00.7F440DD1@flyzone.com>
X-Mailer: Mozilla 4.51 (Macintosh; I; PPC)
Xref: nntp.earthlink.net comp.arch:50519 sci.physics:335644 sci.electronics.design:151425 sci.engr.semiconductors:16084




Okay, many people have sent in some very good
responses to my question of whether silicon chips
can restructured to radiate light instead of heat.
Thanks! The best I've heard include:

1) The heat flux is pretty large. A perfect black
body radiator would need to be 1300K to give off
the approximately 15W/cm^2 of a G4.

2) Silicon doesn't give off photons in the visible
range.

3) Transistors only shed the photons when they're
'overloaded' by too many electrons.

4) Recycling the radiant heat takes too much
effort because of that old Entropy. (Slogan for
Physics Depts: We make laws even Bill Clinton
can't bend.)


I considered just shutting my mouth, but I decided
to open it again because I'm curious. Anyone who
wants to give me a few more lessons in physics is
welcome to answer. Thanks!

A) (1) and (4) certainly argue against
successfully converting 100% of the waste energy
into light. But what about 20% or 10%? Would that
be economically worth it? Lowering the termperature
makes it easier to drive CMOS faster.

B) (2) suggests that most of the radiation comes
off in the IR. This might be recyclable with an IR
CCD array, but it would be more difficult to
exhaust as pure 'light'. If I remember correctly,
there aren't many air tight materials that are
also transparent to IR radiation. Or would a
quartz window do the trick?

C) (3) may be a problem with conventional
transistors, but can new designs change the
boundaries?

Any comments are always appreciated.