Relay computers

[Jim Battle]

Ade Vickers wrote:
...
> I did some back-of-a-fag-packet calculations (i.e. they might be 
> completely & utterly wrong, in which case I'd appreciate corrections) on 
> a relay computer...
> 
> Assume you want a Z80-type CPU. This has ~8k gates. Typically, it seems 
> to take 1 relay per input to implement any given gate. Now, I don't know 
> how many "x-input" gates there are in a Z80, so I'll assume that - on 
> average - it will require 3 relays/gate. Thus, we need ~24,000 relays to 
> implement the Z80.
...

i've read the other replies on the subject, but there is one failure in 
your estimate that hasn't been mentioned that I noticed.

relays often have mutliple poles and multiple throws, rather than just a 
single SPST type connection.  as a result, one relay can be used on more 
than one gate.

Imagine implementing this trivial logic:

	y1 = a & b;
	y2 = a & c;

you could use four relays where the "a" is wired to energize two of the 
relays, while "b" and "c" inputs each energize one relay.

if you wire an "a" and "b" energized reley in series, there is a path 
from in to out only if "a & b".  likewise, take the other pair of relays 
  and wire them in series to implement "a & c".

Instead of four SPST relays, make one of them DPST (one electromagnet 
that triggers two parallel but otherwise independent SPST switches).  In 
this case, the "a" input energizes this DPST relay, and the "b" and "c" 
relays are as before.  Wire one pole of the "a" relay in series with the 
"b" relay and the other pole of the "a" relay in series with the "c" 
relay.  This implements the same logic but with three relays instead of 
four.

That is a simple example, but imagine a more complex case with many 
terms that include "a & (something)".  In theory, only one relay is 
needed for this part of all equations involving the "a &", limited only 
by the mechanical problem of having so many poles.

There are other ways of sharing logic as well.