Path: utzoo!attcan!utgpu!news-server.csri.toronto.edu!mailrus!cs.utexas.edu!samsung!usc!apple!fernwood!portal!cup.portal.com!mmm
From: mmm@cup.portal.com (Mark Robert Thorson)
Newsgroups: comp.arch
Subject: Re: Re: Mercury delay lines
Message-ID: <31511@cup.portal.com>
Date: 7 Jul 90 17:29:07 GMT
References: <3040@softway.oz> <2694@wrgate.WR.TEK.COM>
  <1990Jun7.210822.5230@esegue.segue.boston.ma.us> <2701@wrgate.WR.TEK.COM>
  <1317@m1.cs.man.ac.uk> <645274007.10856@minster.york.ac.uk>
  <11001@medusa.cs.purdue.edu>
Organization: The Portal System (TM)
Lines: 82

warnock@sgi pointed out how optical computers of the future may share
a characteristic of mercury delay line computers, namely that the dynamic
state of the program is stored as one or more wavefronts in a suitable
medium.  This reminds me of an idea I had for a type of fluidic computer.

A fluidic computer consists of channels cut in a solid material, through 
which a pressurized liquid or gas flows.  Because of certain effects which
occur near the surface of a fluid-carrying channel, it is possible to
cut shapes which act as switches.  These switches have no moving parts,
except for the fluid itself.

An early backer of fluidic technology was the DoD.  In the late 50's and
early 60's, they were looking for a technology to provide small flight
computers for nuclear missiles.  Fluidic computers had four main advantages:
  1)  very rad-hard
  2)  cheap (they were made by etching plastic sheets, in a manner
      similar to making PC boards)
  3)  they could be powered by pressurized gas (this is more convenient
      than electricity on a rocket)
  4)  their control outputs are pressurized gas  (again, more convenient
      to control valves and flight surfaces with a cylinder than a
      solenoid)

Integrated circuits knocked fluidics out of the race for the flight
computers.  In so doing, they pretty much wiped out fluidic technology,
because flight computers had demands which were well-suited to fluidic's
strengths.  In other applications, fluidics was a loser.  The last company
to push fluidics for computation was Corning.  As recently as the mid-70's,
they had a line of logic modules you could use to build small all-fluidic
digital circuits.  They tried marketing them for embedded control applications
in high-noise environments, but by then transistors had left them in the dust.

My idea would be to have a fluidic computer which exists as a wavefront
in a suitable medium.  The channels and switches would be cut into this
medium.  The power would come from hot gases.  These gases would be 
products of combustion at the leading edge of the wavefront.  The medium
would be like the fuel/oxidizer mix in solid rockets, and the computer would
only pass through it once (unlike the optical computer, where it recirculates).

The computer would control where combustion is most vigorous by controlling
the back-pressure at the combustion sites.  By using fluidic switches to
control the release of the exhaust gases, the fluid computer can control
the rate of burning at each burning site, hence it can carve whatever
it wants into the raw unburned material ahead of the wavefront.

The computer would be implemented by casting the rocket fuel around a set
of thin wires.  The computer would begin execution when a large pulse
of electricity vaporizes the wires.  This would allow the initial state
of the machine to have any shape of channels and switches that might be
required.

It could proceed like the nautilus, building a new home for itself in front,
and sealing off the previous chamber behind.  I would guess the initial state
would be a large version of the computer (to simplify the process of making
the exploding wires), and that each successive generation would 
build a smaller version when
it replicates, until some limiting factor was reached.  This strategy would
allow the designer to reach a level of resolution beyond his wire-forming
technology.  To minimize the number of wires needed, the initial computer
would be sort of a bootstrap loader, which would be expecting instructions
on how to contruct the real McCoy.

Although not as fast nor as dense as electronic computers, this wavefront
fluidic computer has one property they don't have:  given a suitable
medium, it exists only as its dynamic state.  If it executed a halt
instruction, it would die.  The only record of its existence 
would be channels it had cut while it was executing.

Another unusual property is that it might exist in nature.  A suitable
computing medium need only combine solid and liquid or gas states, 
have a source of pressurized gas or liquid, an exhaust, and
some means for controlling the cutting of new channels.  Magma recently
arrived from the earth's interior contains a lot of dissolved gas,
and at times combines the solid, liquid, and gas states.  And indeed,
some examples of solidified lava contain many intricate channels, as
we might expect to find in material consumed by a wavefront fluidic
computer.  This leads to the obvious question of what the minimum 
requirements for a self-reproducing wavefront fluidic computer would be,
and whether such a structure could arise spontaneously, or through any
conceivable sequence of evolutionary events.  Perhaps some future
civilization of computers will adopt such a theory as their creation
myth, once they've forgotten about us!