Return to On Line Documents

The Strange World of Memory Tubes

October, 2004
pages 5 through 14

Williams tube, Selectron, Radechon, Graphecon, Charactron

On April 1, 2011, Ludwell Sibley - tubelore at jeffnet dot org - gave me the "Go ahead!" to post the above article from "Tube Collectors". This is the website of the organization.

Click on The Strange World of Memory Tubes
This 4.7 MegaByte .pdf file was scanned at 400 dpi to milk as much detail from the printing process as practical.

An even earlier dynamic memory (needs refresh) was invented by Atanosoff and used in the ABC Atanosoff-Berry-Computer


from Rick Dill - rdill at cyburban dot com
In the summer of 1954, I was a summer hire at IBM, working for Joe Logue. My assignment (to come up with integrated circuits using germanium) is irrelevant to this discussion. One of the things I saw in Poughkeepsie's South Road Lab were "drawers" of the Williams tube storage used by IBM for the 701 computer. Joe Logue had a major role in designing those memories, which I believe were only 32 bit. The tubes used were simply surplus 3 inch radar tubes, then available in large numbers and very cheap. A tin oxide transparent electrode was used to sense the charge changes in the phosphor and actually allowed the bits to be seen.

These were packaged two tubes per drawer, slipping into a rack mount. I am a little fuzzy on the cycle, but I think that in something like 12 microseconds, a bit location could be read or written and about a dozen refreshed. At this point they were not experimental, but rather the production memories used in the 701 and immediately following computers.

I probably have some documentation on these in notes from a three week summer course they gave new employees and summer hires while the company took its annual vacation all at the same time.


from Grant Saviers
Hi Ed,

I can confirm one part of the article. Sperry-Univac did offer the Charactron as a peripheral to the Univac 1107, we had one at Case. However, it was not a console display, that was relegated to a Type 28 TTY printer with separate electronic keyboard.

My memory is probably faulty, but I seem to remember te Charactron was made by Stromberg-Carlson. It was definitely a bitch to keep in any sort of alignment. Characters came via the beam mask and it could also draw vectors. I'm not sure if it had any internal memory, but the 1107 had an excellent DMA system for its 64k x 36 bit words.

A good story about its only use. A fellow grad student of mine was doing a MS Math/CS and his thesis was about some novel fast algorithms for calculating orbits of satellites. So he got the NASA space catalog of the time, I think about 1000 objects (1967) and using Cleveland as the display center could show the real time overhead horizon to horizon satellite trajectories. Then he read everything he could find about ABM work and dropped MIRVs (real and decoy) on Cleveland and had a few cycles left over (the CPU was 100% dedicated) to try to sort spoofs from real nukes. He thought this demo was really cool, as did I and the rest of the Computing Center staff, until some G-men showed up and took him away - semi-permanently!

Next I heard he was at Hughes in Redondo Beach. I guess they made him an offer he couldn't refuse.

Re Scan conversion tubes - did the Nike sites have links to off site radars? I remember in 1965 visiting in Oberlin what I think is now called Cleveland Center, an en-route of aircraft, and I think the busiest in the nation (wikipedia says it is busiest in the world). The Univac ?? computers were prone to printing the route ticket for the AT controllers sometime after the flights landed, so the great networked radars (Raytheon I think) from several sites were the only means to keep track of everybody. Somehow those radar signals were shipped some distance, I think as TV signals, scan converted from the source PPI.


additional from Rick Dill
In thinking about tubes for memory, I'm not sure that they ever saw the light of day, but IBM had a follow-up to the surplus 3 inch CRT storage they used on the 701 and similar machines. It was called the "barrier grid" tube. It was a CRT which had a mica target with a metal electrode on the back-side (aka William's tube). Just in front of the target was a grid which served to catch the electrons ejected from the target .. or force them back .. based upon bias level. There have to be reports on this project, but it got eclipsed by magnetic cores.

BTW, I think it would be really easy to create a Williams Tube storage today. All you need is a small CRT with electrostatic deflection. You would drive the deflection with a few bits out of an A/D converter to place the beam at a storage location. The front surface could be either coated with a transparent SnO coating or with a simple metalization of paint-on conductor. When the beam was directed to a bit location, there would be a pulse if the site was charged (electrons driven out) or no pulse if it wasn't. That would give the signal for read and refresh. A very small AC deflection ring pattern could be used to refresh interrogated spot and erase the content of the spot. It doesn't seem like a very big project to me. The tin oxide is a nice touch because you can see the bit glow when it is written. It is sure nice to have digital electronics to do chores that took a lot of hard tube circuit design. Yes, I have seen the "bits" on an IBM storage tube.

The barrier grid tube was one of the projects in the "pickle factory" .. IBM's original site in Poughkeepsie .. on the river in an old pickle factory. It was there that IBM manufactured rifles in WWII under the guise of Munitions Manufacturing Corp. I had a toolmaker/lab tech who worked for me for some years who started there. His hands and those of another ex-navy tech were the ones who built IBM's first alloy transistors and who hand made them smaller and smaller until about 1958 there was a 2 ghz oscillator to demonstrate their frequency capability. The tech who worked for me was Earle Harden, who later was the first to lap magnetic thin film heads and brought in the process still used today. His partner was Ralph MiGibbon. Their work on transistors resulted in the ones used for the 701 and also for the Stretch machine. The former were simple alloy transistors with a ring base around the emitter. The latter had a diffused base region which express shipped minority carriers across to the collector.

In 1954 I got some partial prototypes for a germanium decimal counter (copying the neon ring counters) built for me in the pickle factory. Those never achieved any usage, but led to my graduate school thesis doing numerical modeling in one dimensions of a semiconductor device .. the double-base diode .. later called the unijnction transistor. That turned out to be one of the earliest uses of numerical computations to model semiconductor devices .. but it got published very obscurely, so it is a never-referenced publication.

In 1955 at RCA Labs, I learned to make silicon diodes .. and was able to generalize that to lots of other devices over the next decade. There I met Herb Kroemer, who is credited as "father" of the drift transistor .. and also winner of a Nobel Prize for other work on physics of semiconductors. On the patent side, some years later, I found the patent for the diffused base (drift) transistor which was invented by Lloyd Hunter. While Joe Logue may disparage others, particularly Hunter for their attention to point contact transistors, Hunter perhaps has a bigger role in junction transistors than Logue's simple "sense of the way to go". In fairness, both are/were very good people and technology often seems to move with a "first responder" when it is in reality ready for the next thing to happen.

In my career, what I worked on and achieved varied from things which if our team hadn't scored the point, another was ready and technology would have advanced without a moment's delay, to the opposite end to where my friends and I had found a corner of our own and were able to work deliberately without anyone coming into our space. The semiconductor laser, which I was involved in, would have happened whether the IBM team invented it or not. On the other hand understanding and modeling photoresist imaging was something where we were able to work for an extended period and finally publish a complete story all at once.

Return to On Line Documents