Delay Lines (Was: Univac I memory tank)

[Rick Bensene]

Dwight wrote:

> The Olivetti used a piece of wire for the delay line. 

The Programma 101  indeed used a delay line.  Such delay lines use
magnetostrictive means to push a torque pulse into one end of the wire,
as well as detect a torque twist at the other end of the wire.
Magnetostrictive materials are typically a metal alloy that lengthens or
shortens depending on the polarity of an external magnetic field, and
will also generate a small magnetic field if stretched or compressed.

In a magnetostrictive delay line thin strips of magnetostrictive metal
are attached to opposite points tangential to the circumference of the
end of a nickel-alloy(typically) wire.  These strips, for whatever
reason, are typically called "tapes".

Each tape has a small coil of magnet wire surrounding it, wound
oppositely around each tape, such that when a short current pulse is
sent into the coils, one tape momentarily lengthens, and the other tape
contracts, causing a slight but sharp twisting torque to be applied to
the wire. This acts to transmit a pulse of energy into the wire.  The
torque twist mechanically travels through the wire to the other end,
where it causes one tape to lengthen slightly, and the other to compress
slightly, which induces a small current pulse into the coils around the
tapes, which can be amplified to match the electrical characteristics of
the original pulse.     Sending a current pulse through the coils in one
direction causes the twist to occur clockwise, and the pulse going the
other direction induces a counter-clockwise twist, allowing ones and
zeros to be pushed into the wire as clockwise or counter-clockwise
torque twists.

The amount of time that elapses (delay)  from the pulse being injected
to being received at the other end of the wire is based on the
metallurgy of the wire, and its length.   The wire is capable of
remembering some number of torque twists  as bits, with a clockwise
torque, for example, representing a one, and a counter-clockwise twist
representing a zero.   

The wire was typically arranged in a spiral inside a metal housing.
Silicone or rubber supports supported the wire without attenuating the
torque pulses in the wire.    In some cases, there were "taps" along the
length of the wire that used the same transducer method to pick off bits
at different delay periods. 

The use of such delay line technology in calculators arose  out of the
need to store a moderate number of bits to represent the working
registers of the calculator.  At the time, magnetic core memory was
still quite expensive, integrated circuit technology was in its infancy
and too expensive to use for mass storage in a calculator, and it was
generally cost and size prohibitive to store the bits required in
discrete transistor flip flop storage registers (though a few very early
electronic calculators did use this method).  

Given that delay line technology had been used with success on computers
(though the Univac I delay lines were very different than
magnetostrictive delay lines), they were a low cost, relatively simple
way to provide the small amount of storage required for an electronic
calculator.   A prime example of the use of magnetostrictive delay lines
in a computer was the Packard Bell 250, a low-cost "personal" computer
introduced in the early 1960's.

The bit-serial nature of the delay line was ideal for a calculator,
since a bit serial architecture is coincident with the most efficient
way to make an electronic calculator, where raw speed is not a
requirement, and minimizing the component count saves money.  The serial
nature of the delay line means that if a specific bit is needed, the
logic must wait around for the bit to arrive at the end of the delay
line.   This slows down the operation of the device, but in the case of
a calculator, where results are subject to human perception, 10s to
100's of milliseconds is well within the acceptable time for a
calculation to occur.

> I forget what the Dielh Combitron used but I know it used a two delay
lines. One was for registers and the other was for lookup tables that
loaded at turn on time from a metal tape ( as I recall ).

The Diehl Combitron did use two separate delay lines, one for the
registers(as well as learn-mode program storage) as mentioned, but the
other one wasn't really for lookup tables, but instead stored the
operating microcode that made the machine run.   The microcode was
indeed loaded from a punched metal tape at power-on time.   The
ingenious design of the Combitron was done by Dr. Stanley Frankel, a
nuclear physicist who was deeply involved in the mathematical modeling
that made the atom and hydrogen bombs possible.  After the Manhattan
project ended, he was involved in the design of quite a few computers
and calculators.  Notable computers that he designed were the
Librascope-General Precision LGP-30, the aforementioned Packard Bell
250, and some design work on early General Electric computers.   He also
designed the Smith Corona/Marchant Cogito 240 (and follow-on Cogito
240SR) electronic calculator, as well as the Diehl Combitron.   

Many calculator companies used magnetostrictive delay line technology
for storage in their earlier calculators, before the time that
integrated circuits took over the storage duties.  They included
Friden(Singer) (all of their in-house designed machines used delay
lines, e.g., 130, 132, 115x, 116x), Canon(which made machines for
Monroe), Wyle Laboratories (WS-02 and Busicom 202, 207 and 2017),
Olivetti (Programma 101 & follow-ons), Sony(early Sobax),
Victor(1400-series), Monroe(EPIC 2000/3000, 820/820A), Diehl(which made
machines for SCM and Victor), and Olympia.

-Rick
--
Rick Bensene
The Old Calculator Museum
http://oldcalculatormuseum.com