Advent of Computing - Episode 128 - Cryotrons LIVE!
Episode Date: March 24, 2024Originally presented at VCF SoCal in February of 2024. The cryotron, a superconductive switch, almost revolutionized computing. It's one of those fascinating near misses. In this episode we are talkin...g about the history of the cryotron, how the NSA and supercomputing factors into the mix, and the current state of research into the topic. Did the NSA actually construct a supercomputer that ran in a vat of liquid helium? The answer is... maybe? Video of this talk: https://youtu.be/FqzSGTZ3TMU
Transcript
Discussion (0)
It feels like it's been a while since I've sat down in front of my microphone.
To you, the regular schedule should be rolling, but for me, I just got back from a vacation.
With VCF SoCal earlier this year and then going on a big family trip, my personal schedule has been a little bit off lately.
But never fear, I have the fruits of some of those labors for you today.
But never fear, I have the fruits of some of those labors for you today.
Back when I went to VCF SoCal in February, I gave a talk on the Cryotron, which was recorded,
and today I have for you the audio of that presentation.
Now, some things to mention.
There was a slide deck that went along with this presentation, and there's a video of this which I'm hosting up on my YouTube channel. I'll have a link to that in the description if you want to watch the actual
presentation with me moving around on stage, talking with my hands, and pointing to slides.
In general, the talk only references the slides at a few points. It's usually just to like point
at a picture. So in audio form, it's about the same.
Another caveat is that there was a Q&A session.
And when I got the audio back, there is a channel for my mic and a channel for the ambient audio.
That does have the questions being asked by the crowd, but it also has my voice in it.
And I wasn't able to clean that up in a nice way.
has my voice in it, and I wasn't able to clean that up in a nice way. So the Q&A section is omitted since it would kind of result in a reverberating Sean effect over everything, which
maybe you could listen to that, but in editing, I could not. So that has been redacted.
I will mention two things that came to light during the Q&A section that I need to address.
One is that for some reason, I say in this talk that deuterium is an isotope of helium.
It is an isotope of hydrogen.
In fact, very famously, an isotope of hydrogen.
I don't know why I ad-libbed that detail.
I don't know why I ad-libbed it incorrectly.
But yeah, just know when you hear that mistake, I don't know why I ad-libbed it incorrectly, but yeah, just know
when you hear that mistake, I know that's a mistake. I also mentioned the Charles Babbage
Institute, and I say it's at the wrong university. It is actually in the University of Minnesota.
That is another fact I know. I guess was just feeling the mood of being on a stage and decided to add details that were wrong.
So I profusely apologize.
With those caveats out of the way, I want to again thank the showrunners for inviting
me to speak at VCF SoCal.
I had a fabulous time.
I met a lot of good folk that I think I'll be in communication with for a long time to
come.
And also, hey, enjoy the live show. I've always
wanted to do one of these, and this is pretty exciting to get to publish this. I'll be back in
two weeks with a very normal episode of Advent of Computing. I'm going to be talking about some
programming language stuff. Until then, allow me to roll the recording of my talk at VCF SoCal 2024.
me to roll the recording of my talk at VCF SoCal 2024.
Hi everyone, my name's Sean Haas.
I'm the host of the Admin of Computing podcast, and I like to introduce myself as a dangerous freelance historian.
By day I'm a software developer, and by night I do this kind of stuff.
Today I'm going to be sharing with you the story of the Cryotron,
which is a piece of forgotten computer history
that I think is fascinating and kind of baffling.
It's this device that almost replaces the vacuum tube
before transistors become practical.
And the history of it really, besides just being unknown and fascinating,
is something that I think we're really close to a breakthrough on.
So, just to start out, full disclaimer, this is a story full of missing sources and just things we don't know yet.
Like I said, I think in the next year or two, there's going to be floodgates opening up on this topic.
But right now, the main source is
a book called The Cryotron Files, written by Ian Day and Douglas Buck, whose Douglas Buck is the
son of the man that invents the Cryotron. Besides that, we're dealing with very spooky sources.
A lot of the funding for research into the Cryotron came from the NSA and other three-letter organizations.
And so a lot of papers about the cryotron are still classified or sitting hidden in archives.
So what we do have to work with are a handful of declassified documents that point to other documents that don't exist anywhere in the public domain.
Things will continue to develop, but this is the story as it stands today.
And the best beginning I can think of is Project Lightning. This is an NSA project that starts in
1956. Now, according to certain lore, this project actually started at a cocktail party. There were
some NSA spooks that had just seen a press conference on computing in the USSR.
They got scared and at an after party they decide, well, how many billions of dollars
do we have this year?
Maybe we can start working something up.
Now that might be apocrypha, but the brass tacks of it is by 1956 the NSA needs a better
computer.
That's the easier story to accept.
The main problem they're dealing with is a corollary to the information problem.
As cryptographic data is piling up,
things they need to encrypt or decrypt or shoot around,
they don't have the throughput to do it.
They need more computing power, and they need more bandwidth.
So the simple goal is they just need a really big, really fast computer. The specific
goal, though, is they want to make a thousand megacycle machine. That means a machine that can do
a thousand, thousand instructions per second, or at least a thousand, thousand fluctuations per
second on a wire. They also need it to be highly parallel because a fast machine is nice, but if
it's not parallelized, you're leaving money on the table, right?
You're not able to get the throughput you need to actually churn through data.
But a central issue in 1956 is there's actually no technology that can perform at that speed for computing.
There just isn't a good option.
So why is that the case?
Well, the only option in 56 is the vacuum tube.
You know it, you love it.
They just kind of suck.
A vacuum tube is...
The laughs, I think, are very well warranted.
It's a very antiquated device by 1956.
It's essentially an automated electron switch,
or in England they call them
valves, which is kind of an apt name. You have an input, an output, and in the middle you have a gate
which you can drive with a current to control how many electrons go from the input to the output.
It's very basic. It's developed in the very late 1800s for radios.
It's used for wave rectification and audio amplification.
Well, any wave amplification, but primarily you'll see these in radios and guitar amplifiers.
Now, how these work is actually really neat.
The physics behind it is this effect called the thermionic effect,
where if you take a conductor and you get very hot,
we're talking around 700 C is the operational temperature of many vacuum tubes. If you put
electrons on that conductor, it'll just spit them out. Just says, not today, and sends them away.
If you put that in a vacuum, then you can actually catch some of the electrons on another conductor
and pull that down and read a signal. The gate part, since you can drive electrons into that,
you negatively charge it and that will block electrons from flowing.
So you get this really nice little encapsulation of an effect,
but it's not very good.
There's a lot of limitations with vacuum tubes.
For one, they're physically pretty large.
A vacuum tube's maybe one or two inches tall,
maybe three-quarters of an inch or a whole inch across.
They also need to have space around them for airflow.
You can see in this photo, the vacuum tubes aren't touching.
This is taken out of an old IBM machine,
and they have space around the vacuum tubes for air channels
because they have tiny heaters in them. They have to operate at 700 degrees C
and even though they're in a vacuum things like the conductors going up into
the vacuum will heat up if it gets too hot. In extreme cases 700 C is actually
hotter than the melting point of solder so that that doesn't bode well so you
have to cool it down which means you need more space around the tubes.
They also need a lot of support circuitry.
Once again, beautiful IBM circuit.
You can see under each tube is a pile of wires, resistors, and capacitors.
You can't just plug these directly into an outlet and expect them to do something.
You need stuff for routing around signals,
things in some cases we're dealing with pretty high voltage power. Some
vacuum tubes operate at 50, maybe 150 volts. So they're not really a grab-and-go solution. You
have to have support circuitry. And that leads to some interesting problems we're going to talk
about in a moment. Vacuum tubes are also kind of just flaky. They can burn out. Since they aren't
kind of like a light bulb, they can just stop working. They have very finite shelf lives. They also have a maximum switching
speed, which is dependent on things like the actual size of the tube, the size of the conductors,
and the grids. So there's all these hard limitations caused by vacuum tube designs that
mean you can't drive these at a thousand megacycles, not even close. So Project Lightning can't use vacuum tubes. A more
fundamental issue though is that even the wires work against the idea of a
thousand megacycle computer. This is just a quick summary of some of the lovely
properties that conductors have. The first one is resistance. Even a good
conductor, except for the very best, have some associated resistance. They will
resist the flow of electrons in the, you know, boring brass tacks since that
wastes power and that turns power into heat, which, you know, if you're not paying
the bills that doesn't matter that much, but it means that on a long enough run
of wire,
you'll put in one voltage on one end,
you'll get out a different voltage on the other.
Multiply that by going back and forth on enough spools of thread,
and you might end up getting a logic level that's wrong.
They also have this associated inductance,
which is a resistance to the change in the flow of electrons.
Normally, if it's just a power wire,
that doesn't matter, at least not very much.
But if you're trying to send down a hundred megacycles kind of signal,
that means that you have the signal changing very rapidly.
So inductance starts to matter a lot.
It will limit the speed of transmission and make it so in a normal big bulky wire,
you're gonna have a speed limitation
for how quickly you can vary that signal.
This can also lead to problems with interference
because a varying electric current
creates a magnetic field and vice versa.
So sending a rapidly varying signal down a wire
will make a nice magnetic field around it,
which if it gets too close to another wire, will induce a charge in that wire. And then suddenly you have two conductors sending the same
signal out, which isn't usually what you want in a computer. Capacitance is another thing that
normally doesn't matter with one wire, but when you have this type of construction where you have
a lot of wires near each other, they can start coupling.
And so, once again, like with inductance, you can get interference between wires.
And not only does this get worse the more wires you have, it gets worse as you have more quickly varying signals on the wire.
And as the wires get closer together, really the way to fix this is to just not have wires.
But you can't do that in 1956. We don't have the technology yet
now some of you may be thinking that
This is late enough in the game that there should be a solution well
There there is a better way
the better way that is
Possible is the transistor, but it's still very early days for transistorized technology.
There were some earlier computers in 1956,
but the big one that was developed at MIT is TX0,
or some call it TXO.
This was a very, very early transistorized machine.
While it showed a lot of promise, it's still untested technology.
Transistors in this era were also discrete,
meaning just like vacuum tubes,
a single transistor is just in one pack.
This is years and years before we reach ICs.
So that means to make a transistorized circuit,
you still have all the issues of the wires
and the support circuits.
So that means you still are limited by all the properties,
well, all the bad properties of large conductive
wires. Early transistors are also just kind of bad. They're not very reliable. They're, in some
cases, depending on how they're built, chemically unstable, so they have physical shelf life, so they
can just up and die on you. And they're still, now this is a big kind of, they're still kind of
analog technology, at least in use, because transistors were initially developed to directly replace vacuum tubes, which means they're used for, in radar and radio, for taking analog signals and rectifying them and amplifying analog signals.
So you can use them for digital, but they're initially meant for radio.
So in a way, this is just kind of an aside,
but computers in the early era just keep using radio nerd technology.
I don't know if it's just a lot of radio nerds were into computing early on,
but it's something to keep an eye on.
I don't super trust that.
There's some weird connection there.
But that leaves us kind of at an impasse. If we're just in a transistor and vacuum tube world, we kind of can't get the
thousand megacycle computer that the NSA dreams of. So we should step back and think about what
it is we really need. And it comes down to three things. Well, it comes down to four things. We
just need a switch. And the switch needs to be really small so we can minimize all the weird effects of large
wire conductors.
It needs to be fast.
That kind of goes without saying.
To make a thousand-megacycle computer, you need to be able to switch a thousand thousand
times a second.
But it also needs to be really reliable.
When you're making a computer that uses small switches,
you're not going to be able to replace them very readily.
So you have to have some kind of core switching element
that will just hang out.
It's not going to burn out.
It's not going to chemically die on you.
It's not going to rust.
And these three things are actually really hard to attain
using late 50s technology.
That is, unless we move to more exotic technologies. are actually really hard to attain using late 50s technology.
That is, unless we move to more exotic technologies.
Let me introduce you to the hero of this saga, one Dr. Dudley Alan Buck.
Now, in the annals of computer history,
I think there's maybe a couple handfuls of people
that you could legitimately call geniuses,
and Buck is one of them.
He somehow was just way ahead of the curve on a lot of physics
and a lot of computer science research.
Just as a quick rundown of his bona fides,
in 1948, he joins the NSA, kind of.
He actually joins one of the predecessor organizations that's secret that ends up being rolled in
with a number of other predecessor organizations
and becoming the NSA in a few years.
During this period, he's doing some kind of cryptography work.
We're not entirely sure.
There's some speculation in the Cryotron files
about what it might be,
but it's still classified information.
It's probably a juicy target
for a Freedom of Information Act request, but he's still classified information. It's probably a juicy target for a Freedom of
Information Act request, but he's doing some cryptography something with the proto-NSA.
How's that for some caveats? In 1950, he goes to MIT to do grad school. You know,
it's what you do. And he starts working with Project Whirlwind. Now, for those who aren't
familiar with this era in computing, Whirlwind was the first
real-time computer. It was a machine that initially was developed to be used for flight simulation.
That doesn't really end up working out. The project grows and grows and it becomes a computer
that's focused on real-time input outputs. Buck's working on input-output circuitry, so he's right down in the middle of the cutting
edge of this new computer. And in 1952, he's been at MIT for two years, he's finishing up his
master's degree in route to a PhD, and he invents iron RAM, which is an early form of non-volatile
RAM that, with some adaptations, is still in use today. He just does that, you know, as a master's degree program.
And then he also goes back and starts consulting with the NSA.
Well, he's still in grad school, mind you.
He starts working as a consultant on the hydrogen bomb project.
His work involved with that has to do with physics around deuterium
and refining deuterium
from normal helium.
So he's working with cryogenics.
He's doing some really high-grade physics
in addition to still getting an engineering degree at MIT.
So busy couple of years.
That leads to a really big event.
Now, this might not sound big at first, but in 1953, MIT gets a new liquid helium condenser.
The grant funding comes through.
This is going to be a turning point
for Dr. Dudley Allen Buck.
But to get to that turning point,
we got to do a quick little physics corner.
So magnets, we don't really know how they work.
No, that's a lie.
We know exactly how they work.
It's just kind of complicated.
Basically, you have two kinds of magnets, right?
You got your permanent magnets,
like the kind you have on your fridge,
the kind that you're not supposed to have
on your floppy disks.
You know, they come in bars, horseshoes.
They're always on. They always have a, horseshoes. They're always on.
They always have a magnetic field just
as a property of the material.
It has something to do with dipole moments aligning
and crystalline lattices.
That's not important.
Those are lame magnets.
The cooler magnets are electromagnets.
These are simply a nice little coil of wire.
That's it.
If any of you have a paperclip in the audience,
you can wrap that around your finger
and you have a little electromagnet.
It might not do much, but you do have an electromagnet.
Now, the apparent downside to an electromagnet
is they need power.
That's the electro part.
If you just have that coil on your finger,
that doesn't do anything.
But if you zap it, something fun happens. So you take your finger, that doesn't do anything. But if you zap it, something fun happens.
So you take your coil, give it just a little bit of current, and it becomes a magnet as good as any permanent magnet. The main difference here, at least for our story today, is this is more
flexible than a permanent magnet. And when it comes to computing, there is one large trend throughout the entire arc of history
that is very important to keep your eyes on.
Flexibility means you can do some weird tricks.
Anytime you have a technology or an idea that's flexible,
that's better than an inflexible solution
since you can start squeezing some weird effects
out of that flexibility.
So this takes us back to 1953, to the beginning of the year,
and a device called the Bismutron.
This is actually something that, as told by the Cryotron files,
gets Dr. Dudley Allen Buck laughed at by his colleagues.
He brings it up in a couple conversations and then abruptly stops talking about it
because everyone thinks he's wasting his time.
Essentially, what Buck does is he realizes
he can use and abuse an interesting property of bismuth.
So bismuth is a pretty poor conductor,
but it has this weird property where its resistance
will change proportionally to the magnetic field
it's exposed to.
This is a somewhat direct relationship.
The stronger the magnetic field it's in, the higher its resistance.
Eventually, it will hit the point where it is essentially an insulator.
It is fully resistive. It does not like electrons anymore.
Which means you can turn it off using a magnet.
Now, you can get boring with it.
You can take a magnet from your fridge and go ruin your buddy's bismuth circuits,
or you can make a switch.
And this is essentially how the bismutron works.
It's a single bismuth wire wrapped in another wire.
If you apply a current to the electromagnet, it turns off.
This has an input and output and a gate.
It's a direct replacement for a vacuum tube, except it's cooler.
There's no moving parts.
You don't have to have any glass ampoules.
You just have two pieces of wire.
Now, the reason this gets Buck laughed at is it just kind of sucks.
Bismuth, great word.
Even without a magnetic field, it's a very poor conductor.
So you have to pump a lot of power into a bismutron to even keep a current in it.
There's also just the matter of data encoding
on the bismutron that is a problem.
Buck proposes it in his notes
that you could encode a 1 as current going down the wire
and a 0 as no current.
That's how vacuum tubes and, in a way, how transistors work,
but the problem is you always have to have a current in it.
You always have to be pumping power into the switch. While we overcome that today by having low power
circuits and nice little ICs, that was a big problem in 1953. It kind of made this an inviolable
technology. It's not that much better than a vacuum tube. It's not at least better enough
to warrant a switch.
So the bottom line here is, you know, we know Buck's smart.
He's an MIT engineer.
He invented a new type of RAM.
He works for the NSA.
He can do better.
He can come up with something really smart, right?
So like I was saying, the BizBitron has all these issues.
It's slow, use resistance, just sucks.
But he's on the right track.
This is the right idea.
How can he exploit an electromagnet and weird exotic properties of conductors
to do something cool?
The real crux comes down to helium.
Now, helium's neat.
Helium is, well, deuterium,
which Buck was working with the NSA, is an isotope of helium.
So he was familiar with this material.
Liquid helium can get very cold.
It can get near absolute zero,
which allows you to experiment with some pretty extreme physics.
There are some downsides to helium.
One is it's very small. It's just a proton and
electron. It's, well, two protons and electron. It's a really, really tiny element, so it's hard
to contain. If you have a doer of liquid helium over time, it will just seep out. It's also a
limited resource on Earth. Despite being very common in the rest of the cosmos, Earth just doesn't have that much helium, just as kind of a quirk of where we are in our
solar system and how our planet formed. So this is a very limited resource. You might have heard
of the helium crisis. As we're running out of this, it gets harder to have things like party
balloons or MRI machines. So this is a very precious substance,
but it's also one that lets you play with some really interesting physics.
Prime among those is the beautiful superconductor.
Now, I did say before that only the best conductors have zero resistance.
Well, this is your man.
A superconductor has no resistance at all if it's very, very cold.
It has to be near absolute zero or zero degrees Kelvin to function, which if you immerse it in
liquid helium, that's cold enough for it to be a superconductor. They also have this interesting
property where they can store current indefinitely. This means if you take a superconductor,
and assuming it's very cold,
we're just gonna always assume these are very cold,
otherwise they're kind of inert ceramics.
But if you have a really cold superconductor
and you load a charge on it, just dump in some electrons,
they'll just stay there.
You can store a charge essentially indefinitely.
There are some caveats with heating up
and other physical
effects, but you can just keep electrons in there until you discharge it. There's also
some neat stuff with magnetic field ejection. That's why you see all the demos where superconductors
get them cold and they float on a magnetic rail, but that's unimportant. That's cool
and pretty. It's a neat demo, but that's not what we want to talk about today.
Here's the property we want to talk about.
Superconductors hate magnetic fields.
If you expose a superconductor to a magnetic field,
it stops conducting charge.
Now, there are parameters on the shape
and strength of the magnetic field,
just like with bismuth,
but you can turn off a superconductor
using a magnetic field.
That means that you can make a switch out of it.
Also in 1953, very busy year for Buck,
he invents the cryotron.
This is like the bismutron,
except instead of bismuth,
bismuth, it's a very hard word, you just use a
superconductor. You make a superconductive wire, you wrap a normal wire around it, and you're done.
That's the basic design of the cryotron. It has an input, an output, and a gate. It functions like a
switch, just like a vacuum tube, except it's much, much faster.
It switches nearly instantly.
As soon as the superconductor sees that magnetic field,
it's off, it's gone.
There's switch speeds that are dependent on
how a magnetic field can fluctuate
and propagate through the coil,
but if you make some good choices
with how you build your coil,
this is an almost instantaneous switch.
This can easily handle a thousand mega cycles.
It also keeps charge, which means these can use
a different type of data encoding than we use in modern computers.
Buck proposes that to encode a 1,
you just zap some electrons on it and leave it.
You have a 1 stored in your cryotron.
To encode a zero, you just have no electrons.
You can switch from a one to a zero by energizing the coil,
the electrons bleed off.
This means that a cryotron on its own
is a one bit memory element.
That compared to vacuum tube or transistorized technology where you need multiple logic elements
to make a memory element, this is a huge advantage.
This is way beyond what anyone can do with ampoules of glass or semiconductors in the
era.
There's also a fully digital technology.
So strangely enough, this is the first non-radio nerd thing
that's proposed as a logic element for a computer.
It's either on or off.
You can't really do analog with this.
And it's also wildly simple.
As long as we don't think about the fact that this has to be immersed in liquid helium,
which is a limited resource,
and we have to use relatively exotic ceramic compounds
or special type of metal alloys.
This is actually a really simple device.
As long as you have all of that,
it's just a wire with another wire wrapped around it.
At least compared to something like a vacuumed-out ampule,
that's pretty darn simple.
In 1956, it takes a few years
to get this technology up to speed, but Buck publishes
his first paper on cryotrons.
This is really a neat paper because he has fully digital circuits.
This just as an example is a one bit adding circuit, all from cryotrons with a carry input
and output. What I'd like to direct your
attention to is this looks a lot different than a digital adding circuit. One of the interesting
details is if you look at the long rods that represent the superconductors in the circuit,
you see some of them have multiple coils wrapped around them. Those are oorgates. He's using a single superconductor
with multiple coils to create an oar. That's a lot more flexibility than we
ever had with vacuum tubes or even today than we have with transistors.
You're simply able to pull more tricks with cryotrons than you can
with other digital technology, which means there's that F
word. You have a more flexible technology, so you can make more flexible circuits, at least in theory.
Now, from 1956 forward is where things start to get really, I think, kind of wacky. So Buck
publishes, and his paper starts going around. People at GE, IBM, RCA, and the NSA,
his buddies over at the three-letter organization,
start reading his paper,
and the technology starts to propagate from there.
The next big jump comes sometime in 1957.
It's either at GE or at Buck's lab in MIT.
They start doing lithography In 1957, just for the
conventional narrative, in 1959, Robert Noyce makes the first integrated circuit, but that was a
semi-conductive integrated circuit. In 57, at least two labs are creating cryotron integrated circuits. They're using, if I remember correctly,
this photo is actually a glass slide that they etch cryotrons onto.
They're also using sapphire slides.
The process is a little wacky compared to the kind of crystal lithography we use today.
They're putting an electron reactive emulsion onto these slides.
They're putting them in a vacuum chamber
and then they're blasting them with an electron beam.
So by doing that, they're able to lay down very thin traces
of superconductive and just normal conductive
and insulative circuit elements.
So this is full stop, no caveats,
the first integrated circuits in 1957,
and they're not semiconductors.
Now, on the surface, that's cool.
It's a little bit of hidden history, right?
They're making very little cryotrons.
But this actually breaks all of the limits
that we saw with earlier technologies.
This is going to allow a thousand-megacycle computer This actually breaks all of the limits that we saw with earlier technologies.
This is going to allow a thousand-megacycle computer because look at those little wires.
You can't even see them.
They're small enough that resistance, capacitance, and inductance, they still matter, but not
nearly as much as a big, bulky vacuum tube circuit with big, thick gauge wires.
It's so small that you can pack as much circuitry as you want in there
and not worry about all those weird effects.
You can drive this basically as fast as you want.
By 1957 standards, you can drive this as fast as you can dream.
This is where we get back to lightning.
So by the late 50s,
cryotron technology looks like it's going to be the future.
This is way ahead of where we are with silicon,
and luckily, Buck has an in with the NSA.
A lot of the people he was contracting with and personal friends with
were working on Project Lightning directly.
So by the end of the decade,
lightning is standardized around the cryotron.
The only issue, though, the big hang-up is, of course, helium.
Like I mentioned, and this is a bit of a digression,
but I think it's interesting to consider,
helium's really hard to store.
That's why even a nice mylar balloon will deflate pretty
quickly. Helium is just small enough that it gets through seals. So there's this weird point in time
at the NSA where they're developing these plans for machines that would literally burn through
a scarce limited resource. Just think about how sci-fi that is, right? You have your big
supercomputer center, and every day someone comes in with a cryogenic truck and pumps your computer
full of liquid helium. You might even need to go mine an asteroid so there's enough helium
reserved to run your supercomputers. That's the kind of wacky sci-fi stuff that, I don't know,
computers. That's the kind of wacky sci-fi stuff that, I don't know, it just tickles my fancy.
But this leads to, I think, the million-dollar question, or rather maybe the $10 question.
I'd give someone $10 if they could answer this. How far did lightning get? Was there a secret NSA installation that did have the helium man showing up every day with his cryogenic tank?
Were they decrypting Russian communique Ks using superconductors? Well,
that's something of an open question. We do have the beginnings of an answer, though.
There are numerous designs for cryogenic circles, or circles, cryogenic circuits in the literature.
You can find a lot of papers that show adding circuits, ring circuits.
There's these neat associative memories
that were being developed using cryotrons.
But most of those are just schematics
that don't really go anywhere.
Now, this is one that's an interesting avenue
of research, possibly.
There may have been an entire cryogenic computer built at GE.
Supposedly, it worked.
Supposedly, it used the lithography methods that I mentioned earlier.
And supposedly, it only needed 2,000 cryotrons.
The caveat being, the evidence for it is in a GE alumni magazine.
Some ex-GE employee wrote this article saying, well, when my grandson asks me what I did
during the Cold War, I tell him I built a cryogenic computer.
And he explains the computer, I haven't found any internal documents that have been released
explaining the system's existence.
So take that with a grain of salt. There might have been some engineer at GE with one of these machines.
We also have well attested
memories that were developed using cryotrons. Now, this was actually donated, this photo
I have here, to the Computer History Museum a few years ago by an NSA engineer who was
working on Project Lightning. This is a, I forget if it's a 4 or 16-bit associative memory that's
made out of cryotrons. It's fully integrated, and as you can see, it looks pretty similar to
a semiconductor circuit. Main difference, of course being, as it stands, that's inert.
It has to be in a doer to work, but it's there.
So we have evidence that they were making
some computing elements, some pieces of the larger puzzle
out of cryotrons, and they were very sophisticated
for the era.
But this leads to the real big question, the big enchilada, is what about direct contractors
that the NSA worked with?
We know from the, well, not even a handful, from the three declassified documents that
the NSA was contracting with at least IBM and RCA. So how far did they get?
Well, if you asked me that question two years ago, I would have said,
maybe they got a little ways, but you know, maybe one day we'll know more. That would be nice.
But standing here today, I do have some information I can share.
today, I do have some information I can share.
Now, this information comes by way of a pile of identified documents, some of which I've scanned, and a current ongoing Freedom of Information Act request.
The identified documents are a bunch of progress reports that are held at the Charles Babbage
Institute at the University of Wisconsin at Madison.
I had one of my buddies
in the Midwest go out and scan a bunch for me. Luckily, they were in the area. They were not
very pleased about it. I've bought them many beers since. So I have about a dozen progress reports
that show extensive development of a cryogenic computer up to a point. The main contractor that these progress reports cover
was IBM.
IBM was receiving, I don't have the budgets on hand,
a whole lot of money from the NSA
to develop one of these things.
They got so far as they have a full,
well, at least by the early 60s,
they have a full simulation model of cryogenic circuit
elements that they can use to simulate components of a cryogenic supercomputer. They were refining
manufacturing techniques to reliably have yields of integrated cryotron circuits. And they have
designs of very large-scale cryotron circuits. The kind of piece de resistance that I wish I could show,
but I am bad at planning
and didn't sort out the copyright details,
is I have a giant schematic
for a very large associative cryotron memory.
It's really neat.
It would have been even bigger than the IC I showed you before,
but with how archives work, I need to find the original person that wrote that report and who donated it to the archive and get their
permission for copyright materials, but
So it's out there. There were very large-scale
designs.
But none of these progress reports yet point to a full computer.
The latest one I have, the 10th progress report,
which closes up the series, basically says that they're getting close. They need maybe a year or
two, and then they might have a full computer. The big hurdle that they were trying to overcome
is the actual fast switching domain with Cryotrons that has, there's some weird effects that
go on once you hit that regime.
Nothing deal breaking, but it's hard to simulate.
And so by the 10th progress report, they were working on getting their simulation down well
enough that they could ramp up to a computer.
The big hope though is in 2021, I filed a Freedom of Information Act request with the NSA.
If you're not familiar with FOIA, it is, I believe, the greatest institution in these United States.
Basically, if you say some magic words, fill out the paperwork properly and send it to the right address,
then any federally funded agency has to reply and say, yes, we will get you those papers.
Thank you for your magic words.
They just don't have to be fast about it.
So as of right now, I'm waiting for,
well, one of two outcomes.
I'll either disappear into a black van
or I'll get a...
Oh, you laugh.
Imagine the tax implications for that ride.
The better outcome is I'll get a big pile of redacted documents
for a nominal fee,
which I always joke with my friends,
once that happens,
I'm going to go off grid for a while
because that's going to be
hopefully a lot of data
and will hopefully settle things.
My best guess,
my gut feeling from everything
I've read in progress reports and the Cryotron files is they might have gotten close.
If that GE alumni paper is anything to go off, then it shouldn't have been too hard to make a cryogenic computer.
We have all the elements pointing to it.
We have memories.
We have adding circuits.
We have things for busing.
We have things for latching.
There should be all the elements that were needed
to make a cryogenic computer.
We just don't have that smoking gun yet.
I think once we get it, that's going to open up
a whole new pile of secret history in computing.
Just imagine how wild it would be if we do have evidence
that the NSA did have one of these,
and there was some friendly neighborhood helium merchant
that came by every day to top off their tanks.
But until we learn more,
I'm just going to stay in my corner
waiting for my FOIA request to come back.
Thank you so much for listening to my talk.
If you want to follow me,
I have a podcast, Advent of Computing.
This is, I saw the QR codes earlier
and so I added one.
It seems like a nice touch.
My podcast covers this kind of history.
I mainly focus on the more obscure
and esoteric parts of computer history
that aren't covered elsewhere.
I'm hoping to be releasing more as I learn more,
but until then, enjoy the show.
I have some nice episodes up.