Thread by @TubeTimeUS

on this relaxing saturday afternoon i was digging through my collection of vintage 3" cathode ray tubes (as one does) and i found a *very* unique item! 🧵

it's a 3" CRT, but there are several strange things about it.

it's got the IBM logo on it, and the outside is coated in a conductive metallized layer...

the base has a somewhat unusual 11-pin arrangement. i'm not sure i have a socket that will fit.

someone wrote something on the front with a grease pen, but i can't quite make out what.

so what is it? it's a computer memory, believe it or not! this is a tube from the IBM 706 electrostatic storage unit which served as the memory for the IBM 701 computer. this tube can store 1024 bits--about half the text of this tweet.

this tube plugs into a module that looks like this. with two tubes, you get 2048 bits of memory.

https://twitter.com/lturrentine/status/161154849881337856

36 of these modules plugged into the IBM 706 memory, providing 73728 bits of storage. (9KiB!)

you can read more about the 706 here. ibm.com/ibm/history/ex…

there's also an interesting CNET article about a module found in a barn.
cnet.com/news/antique-i…

anyway, these were also known as Williams tubes after the inventor. en.wikipedia.org/wiki/Williams_…

typically there was a metal plate in front of the tube so you couldn't actually observe the data, but you could wire another CRT in parallel and view the data that way.

also, i've had this tube around for a few years but i just can't remember where i got it. most likely it was from the Electronics Flea Market.

I was going to figure out the base pinout but then I found this note inside the bubble wrap the tube was stored in. 😂

I also found my rig that I use to test CRTs.

think this is right.

the filament glows! how exciting!

not seeing anything, one of the adjustments may be off

oh here we go! the Williams tube still works! 🤩

it's not storing any bits because i need a metal plate to go in front along with the associated amplifier. stuff to figure out later, i guess.

staring at the stuff written on the faceplate, i think it says "spill+" on the second line. if true, then this tube had a rather fascinating problem

in a Williams tube memory, a dot/dash representing a bit could sometimes influence an adjacent bit. this was called "spill" because the electrons from one bit would spill over to the other, causing a bit flip. this was basically Rowhammer but in 1952.

I'm building up the test bench for this tube. the deflection amplifiers are slightly better, and the circuit drives the cathode to negative high voltage (since the faceplate must be biased close to the anode voltage)

the original deflection amplifiers were really pretty impressive. there was one set for the *entire rack* and it drove the deflection plates of all 72 CRTs in parallel!

after the differential amplifiers, there were these cathode followers (like transistor emitter followers) using 16 triodes per plate! (8 5965 tubes in parallel)

those load resistors are 1.33Kohms. rated at 150W because the voltage rail is 400V! each triode could handle 15mA or 240mA total.

why? because each deflection plate and associated wiring had something like 10pF each. with 72 CRTs, that's 720pF which needs significant current to get the voltage to move quickly. back of the napkin c*dv=i*dt -> 720e-12*100=0.24*dt -> a top speed of 0.3us to slew 100V.

in reality it's probably not being driven quite that hard, but it gives us an idea of the engineering challenge.

oh yeah that 400V rail? it's regulated and filtered to have less than 15mV of ripple. that's really hard to do. any ripple causes the electron beam to wobble and disrupt the data storage

I can't get the beam to switch off. 🤔

over 42V across the 120K grid resistor? 0.35mA is quite a bit of leakage.

hmm seems like there's 200k of resistance between the grid and focus anode

oops, left a resistor in the circuit. without that and with the filament on, the grid leak is about 100K to the focus anode. but only when Vgrid < Vfocus. I wonder if I pinned this out wrong

no I think it's damaged. the grid must have exceeded the cathode voltage at some point during its life

slightly more boring 3RP1. these are later-generation CRTs that perform better than the old 3BP1s

starting to look like bits. think I'm going to need a fancier way to generate the signals at some point, like an Arduino or something

I'm getting the feeling this isn't the best way to build a high gain low noise amplifier 😂

this is a bit better.

gain of 10,000. this isn't really the right design, I need something matched to a high input impedance.

now I need a ramp generator to make a sweep waveform. this simple current mirror circuit should be good enough.

it works! off on the right you can see the new low-noise power supplies that I added to generate +/-12V and +/-5V.

added a blanking pulse. the system is going to need a lot of adjustment to get it working well.

looking pretty good. I should start playing with the sense amplifier now.