Fun_People Archive
3 May
Jujitsu for Mechanical Engineers

Content-Type: text/plain
Mime-Version: 1.0 (NeXT Mail 3.3 v118.2)
From: Peter Langston <psl>
Date: Mon,  3 May 99 12:59:45 -0700
To: Fun_People
Precedence: bulk
Subject: Jujitsu for Mechanical Engineers

X-Lib-of-Cong-ISSN: 1098-7649
Forwarded-by: david mankins
From: John L Redford
Subject: "Computational Matter" - talk by Andrew Berlin of PARC

Hello all,

We've all been hearing about micro-machining, the construction of mechanical
parts on silicon wafers, for a number of years now.  Yesterday I heard a
talk at MIT on what might be called milli-machining.  This seems to me to
have even more applications than the micro kind, with interesting
consequences for both mechanical engineering and distributed computing.

The talk was rather grandly titled "Computational Matter", and was given by
Andrew Berlin of Xerox PARC.  The room was so jammed that when Michael
Dertouzous, the director of LCS, arrived late, he had to sit on the floor.
Berlin is heading up a research group at PARC devoted to 'smart matter',
which very much ties into Dertouzous' ideas about Things that Think.

Berlin's basic work, though, was much more concrete.  He has found a way to
build a platform filled with thousands of air jets, each one under
individual electronic control.  If you put something on this platform, it
can be levitated on an air bearing, and made to move backwards and forwards
by turning on backwards and forwards-facing jets.  He showed a video where
pieces of paper were made to zoom across the platform, then suddenly stop,
then reverse, and zoom back.  He can accelerate a sheet of paper at five
gees.  By pushing the leading edge of the paper forwards, and pulling the
trailing edge back, he can keep the paper under tension and so keep it from

As you can imagine, Xerox is intensely interested in better ways to move
paper.  But this technique works for anything, even delicate biological gels
and steel plates.  Anything can be lifted up and moved around this way with
no rollers or other moving parts, except for the flaps on the air jet

He first built something like this three years ago, but had to use
individually mounted valves for each jet.  At $10 a valve, he knew this
wasn't going to be popular.

His breakthrough has been to make these valves en masse.  He does it by
building each valve as a tiny copper flap (maybe 100 um across) above a
hole.  The rim of the hole is covered by an insulator, with a conductor
underneath.  Apply about 200 volts between the flap and hole conductor, and
they close together via electrostatic attraction, thus sealing off the hole
and shutting off the air flow through it.  Turn off the voltage and air
pressure blows the flap open again. The flaps are made by cutting them out
with a laser on a thin film and then mounting that film on the board with
the holes.  He was coy about the details, understandably.  This is like
silicon micro-machining but on a much larger and cheaper scale.

The flaps can be opened and closed at about 50 Hz.  His latest ones have
terrific lifetimes - he has run them for 500 million cycles without seeing
failures.  The holes are made in a standard printed circuit board by laser
drilling.  Above the flap is another hole that directs the air stream one
way or another.  This is also made in a standard PCB, with several stacked
layers and offset holes to give the jet an angle.  The flaps are thus
protected by this upper board.

The computation part of this comes in controlling all these valves.  One
can imagine systems of millions of them, along with associated sensors to
see where the levitated object is.  It's a fine distributed real-time
control problem.

Berlin has bigger ideas, though.  He did his PhD work at MIT on the active
support of columns against buckling.  As the load on a long, thin column
increases, it tends to fail by bending and putting one side in tension
instead of compression.  If the tension stress exceeds that of the material,
it breaks and the column falls.  Berlin tried to fix this by placing sensors
along the column and piezo-electric pullers.  As the column starts to flex,
the pullers distribute the load somewhere else on it.  He was able to
increase the strength of a column by 5.6 times.

So here's where the computational part comes into the matter part.  He was
able to significantly change the mechanical properties of matter using small
forces under computer control to manage the large forces in the bulk of the
material.  He calls it a leverage point.  In just the same way, he's able
to use the small forces of a lot of air jets to move significant objects
around.  It's mech E jujitsu.  It's using something that's now very cheap,
computation power, to control macroscopic real-world forces.

Cars, planes, buildings - these could all benefit from stronger and lighter
columns.  I'm not sure that I'd want to trust the chassis of my car to the
reliability of a buckling control program written in C, but a car body
that's a fraction of the present weight would be a win for everybody except
scrap metal dealers.  It's dizzying to think of machines with piezo-electric
tendons and silicon nervous systems, but it could be coming.

/jlr (John Redford, jlr @ world . std . com,

PS I asked Berlin whether these 50 Hz flaps could be used for a video
display. He said that they had looked at it, but they didn't work that well
as light valves instead of as air valves.  It would be cute, though, to have
mechanical TVs.  It would be like those mechanical highway signs, but with
a million dots instead of a few hundred.

prev [=] prev © 1999 Peter Langston []