Talk Of Building A Computer...

From: Tony Duell <ard_at_odin.phy.bris.ac.uk>
Date: Thu Nov 20 04:51:20 1997

On Thu, 20 Nov 1997 Philip.Belben_at_powertech.co.uk wrote:

[Can we build a modern version of a valved computer]

> I think there is something even more fundamental here. Valves
> (thermionic, in tubes) have quite different behaviour to trannies.

Hmm... Whether that's a problem depends on how close-to-original you want
to be.

Remember, the standard bistable will work with _any_ type of switching
component (relays, bipolar transistors, jfets, mosfets, valves, etc).
Similarly, you can make multi-input gates, which, while not original,
perform the same functions.

Thus you could easily make a machine with the same architecture. It's not
quite as simple as replacing the valved AND gate with the transistorised
(or IC) equivalent (I suspect timing considerations are going to be a
problem, and a lot of the earlier machines used AC coupling between
stages, etc), but it should be possible.

>
> A JFET behaves fairly like a triode, but designs that use pentodes and
> nonodes and things as multi-input gates are going to be very difficult
> to translate.

Well, there were pentodes that could be cut off by 2 of their grids (the
6F33 springs to mind) and were thus NAND gates, but I don't think they
were that common in computers. Anyway, there are work-arounds

There are certainly dual-gate MOSFETs made (used as RF mixers, etc) that
could probably be used for gates as well.

> Or if you want to be way out, what about tubes with several valves in?

Double triodes were common in computers. Somewhere I have 1 module from a
Ferranti Pegasus computer, and that has 3 double triodes (essentially
selected 12AX7/12AU7/12AT7/ECC8x devices) on it.

> Things like double diode triodes are quite common, and someone even put
> most of a radio set into one tube (passive components and all).

You mean the Loewe 1-valve? That thing had a single glass envelope with a
tuned circuit connected to the input and a pair of headphones connected to
the output. The valve contained (I think) 3 triodes and all the R's and
C's for the radio. Came out in the 1920's, and never caught on.

>
> So how about making our own. A tube containing, say, a 4-bit D-type
> latch? Make a few in that range and a valve computer becomes almost
> manageable! Besides, the smaller it is, the faster you can make it...

Hmm.. There are/were a few special-purpose valves already, like the
Trochotron (10 state high-vacuum counter tube - I guess an 8-state or 16
state one could have been made), Selectron (OK, so it never worked, but it
would have been a 256 bit memory), etc. If you don't need that high speed,
then I suspect you could make a shift register using the nixie tube
principe - Borroughs made a thing called a 'self-scan display' that was
exactly that, although most of the electrodes were not available
externally.

Also remember that the Klystron proved that the transit time of the
electrons was not the limiting factor of the speed of a valve (The
_spread_ in transit times is).

I wonder if you could use the glass delay lines from PAL colour TVs to
store (say) 64 bits. It would make a neat register for an EDSAC-a-like :-)

>
> Philip.
>
>

-tony
Received on Thu Nov 20 1997 - 04:51:20 GMT