Advent of Computing - Episode 182 - Spinning Memories

Episode Date: May 18, 2026

What connects IBM, the NSA, the Third Reich, and high fidelity recordings of symphonies? The answer is: magnetic drum memory. Join me as I lose all track of scope and plot to discovery just how and why magnetic drum memory was invented. Like Advent of Computing? Then check out the after show! Adjunct of Computing is now LIVE: YouTube Spotify Apple Podcasts

the definition of a computer introduces some pretty big restraints on what a computer can actually be. And I know, I realize that's a kind of dumb sentence, but think about it for a second. To be turning complete, to be the strictest definition of a computer, a machine must support conditional execution of code. It has to be able to do an operation and then use the result of that operation to choose what to do next. that's conditional execution. Turning completeness is really cool, it's really powerful, but I'm not super in love with it.

For me, it's kind of restrictive. The largest issue here comes down to, well, it comes down to the implications. The definition implies that you're talking about a stored program computer. That puts some pretty hard limits on your design. The usual way machines handle conditional execution is to put all your operations, all your code, into some kind of storage. Each of those instructions is given some kind of identifier. That way, they can be referenced later.

Then, when the computer makes a decision, it can choose where it goes next based off those identifiers. Can you spot the implication? It's actually memory. For a computer to really fit under the definition of turning completeness, it has to have memory. It also has to be addressable. Now, there are some ways to flex the definition, but memory is always key. Our definition of a computer makes memory central to everything.

So it's no surprise that memory looms large in the history of computing. In a sense, the history of computing is the history of memory. and its uses. Our definitions make sure of that. Welcome back to advent of computing. I'm your host, Sean Hasse, and this is episode 182, Spinning Memories. Today, I'm continuing my magnetic rut by finally covering a long overdue topic, magnetic drum memory. I was actually goaded into this episode by Joe on a recent recording session for adjunct of computing, the official after show. This is also a plug for adjunct of computing. If you want a casual debrief of the main show, then please go give it a listen.

Link in the description and all that. When you look at early computers, there are really three crucial technologies. Vacuum tubes, mercury delay lines, and magnetic drums. Vacuum tubes were the core of digital logic. It was possible to make a computer using relays, but that didn't give you a very fast machine. Without the vacuum tube, you don't really have a practical computer. Mercury delay lines and magnetic drums were your only real options for memory until the middle of the 1950s. They're only gradually replaced with magnetic core memory.

Mercury is interesting because it's such a wild, nearly fantasy kind of technology. It also dies out pretty fast. Magnetic drum is a much longer-lived form of memory. That said, both of these early memory formats have similar characteristics. Both recirculate data. Both are slow. Both have this odd temporal nature. Bits literally slosh around in a mercury line, and the computer has to wait for the right slosh to arrive. Bits spin by heads on drum, and the computer has to wait for the right bit to swing by. The history of mercury lines is well attested. It's developed for radar.

specifically as a way to deal with ground clutter. These fancy radars would use mercury to remember what they've seen before and what hasn't moved. That way you could filter out trees and buildings and focus more on agile things, like birds, you know, what radars are meant to look at. But magnetic drum has always been a little more mysterious, at least to me. I know a lot about the practical application of drum because of, well, Say it with me, my beloved LGP30. Drums also come up pretty often on the show.

It really was, well, it was a very popular memory technology for about a decade. But where did it come from? I mentioned a few episodes ago, almost as a throwaway, that magnetic drum was developed in the 30s at an Austrian subsidiary of IBM. What's the deal with that? What was an Austrian doing with magnetizable drums? Was there, perhaps, another digital visionary that was whisked into obscurity by the Third Reich a la Conrad Zeus? And what exactly was IBM's stake in drum prior to the first digital computers?

This episode, I'm going to try and not even get into computing. We've already talked to death about different machines that use drums. Instead, I want to focus on the earliest days of the technology. Why was drum invented? How was it used? And what can we learn from its pre-turning complete past? The main character for the first part of today's episode is IBM. But at the same time, not the true blue IBM we're used to.

In 1933, an Austrian inventor named Gustav Tao-Touchek patented the magnetic drum members. The patent was issued while he was working at an IBM subsidiary in Germany. Now, I want to be crystal clear here. This is what you'll read if you look up what is the origin of magnetic drum memory. This may not be factual. Now, already just looking at this very short rundown, there's a complicated picture here, and the details make it more so. Tao-Chek originally worked for a company called Rheinmetal.

In 1928, Rhein Matal spun off a subsidiary that specialized in punch cards and punch card equipment. That subsidiary was then bought by DeHomag, which was a subsidiary of IBM in Germany. With the purchase came Tao Check as well as an absolute pile of his patents. When you issue a patent, you can be the inventor, but it can be assigned to a company. You can also later sell the patent to a company, so, well, let's just say patent laws a bit of a web of ownership, licensing, and rights. The point is, Tao Check came along with a tranche of patents. He was one of those pre-World War II inventors that turned out a truly staggering number of

patents. I think his count is somewhere north of 200. These ranged from modifications for firearms to optical character recognition to accounting machines, and, yes, the drum. And this is where we have to take an immediate detour. I mentioned a company right there that you may or may not have caught, DeHomag. There's a book that should be on everyone's reading list. It's called IBM and the Holocaust.

It chronicles the connection between IBM and the Nazi party. During World War II in its lead-up, there were strong controls on how U.S. companies could operate in Germany. IBM used Dihomag, their once subsidiary, as a shell company to continue doing business with the Nazi party in direct violation of U.S. controls. That continued business directly facilitated the Holocaust. The final solution was planned and documented using machines branded as DeHomag, but many of the parts had been shipped in from IBM themselves. During the war, De Hoagmog had spun off from IBM. But once hostilities ended, it was absorbed back into IBM, as was profit generated during

the Third Reich's rule. There was also a cash flow from DeHomag to IBM during the war itself. The relationship between IBM and the Nazi party is complex, but there's a clear connection. The relationship between Tao-Chek and IBM, on the other hand, is much more muddy. During the middle of the 30s, he would work for IBM directly in New York. In 1935, he moved permanently to Switzerland, but still filed patents with IBM well into the 40s. What we do know is that once Tao Check was in the IBM fold, he stayed in. His patents materially benefited IBM in this period, and he seems to have ridden out the war in a neutral country while still selling patents to IBM.

There's a connection here, but it seems that Tao Check's relationship was more on the edge of war profiteering than what IBM corporate was doing. Luckily, the magnetic drum patent is early enough that it's not outright assigned to IBM, and I haven't found any information that points to drum memory's use in Dihomag equipment. So the deeper Nazi party connection ends there. It's more that Taocheque's relationship with the Nazi party in the Third Reich was about the same as any employee of IBM in the wartime period. I bring this up because the relationship is so different than Zeus's relationship with the Third Reich.

Zeus, as a reminder, is this kind of mad genius figure that invents a computer while in complete isolation in wartime Germany. Now, Zeus was, in large part, secluded because of the regime. His direct connection to the state prevented him from playing a larger part in the earliest days of computing. He was essentially out of circulation for most of the 1940s because of the politics of the period. And as an aside, if you haven't read Conrad Zeus's autobiography, highly, highly recommended. It is a fascinating book, and the English translation is pretty good.

Now, Tao Check, on the other hand, had a much more indirect connection. He was tied to IBM and their subsidiaries. His work made it out of Germany and out of Europe. He was free to travel and seems to have avoided some of the privations of the war. Really, he didn't have any more connection to the Third Reich than any other IBM employee who was active in this period. The key difference between him and Johnny Blusuit are that Taoisek lived in Europe, and he had worked for Diomag at one point. Really, ever since reading Zeus's autobiography and IBM in the Holocaust, I've just been keeping an eye out.

The U.S.'s approach to computing and the Nazi approach were just so different. It's hard not to try and collect data points, So this is going up on the wall as another data point. And thus ends any discussion of the Third Reich or the Nazi regime in this podcast. I'm sure it's not going to come up again this episode. That wouldn't make any sense. Anyway, let's talk about magnetic drums, right? Taoisek's patent outlines something that looks very familiar.

It's a cylinder and a magnetic head. But how similar is this to the drums we know? love. After reading Taucheck's patent, I just straight up wouldn't call what he made a magnetic drum memory. It's way too different. Drum memory, just as a quick primer, is a cylinder with a magnetic coating that has these magnetic read-write heads studded along the surface. As the drum spins, those heads can write data to it in binary and read data from it also in me. binary. How does Taocheck's patent differ from that understanding? Well, let's start with material. Tao check calls for a steel drum. That, on its own, represents a very large technical difference.

It also points us in a weird direction. So, this is something I haven't seen explained in any articles about Tao Check's drum. He explicitly states he's working from a discovery of one, Valdemar Polson. That's the inventor of the magnetic audio recorder. Specifically, Paulson invented the wire recorder that used a steel wire to store data via magnetization.

Paulson's first recorder was demonstrated in 1900. A magnetic head, basically a primitive tape head, is used to magnetize regions of steel wire and then used to read those magnetic patterns back to make audio. When folk talk about Poulson, they usually just say he invented the wire recorder. Now, I've seen wire recorders before. You can find videos of them on YouTube. They use spools of wire and function almost the same as a reel-to-reel tape recorder. The wire spools past a magnetic head. But that's not what Paulson invented.

His original patent looks nothing like that. He has a steel wire that's wrapped around a cylinder with a magnetic head on a sliding arm. The cylinder spins and the head is synchronized to travel down the cylinder. It traces the coil of the wire as it moves. That, dear listener, is a drum. It's a drum that uses magnetic flux to store audio. What I think is the funniest part here is that Pulsin's recorder looks almost the same as earlier wax cylinder recorders.

You can go find a photo for yourself. It is a shocking similarity. It looks like someone replaced a wax cylinder with a metal one. And that's about it. I think it's clear that he's borrowing the transport mechanism, if we can call it that, from earlier wax cylinder recorders. We know that Tao Check is aware of Paulson's work. His patent literally says, quote,

in the embodiment described below, the storage elements are influenced by current pulses, series by means of magnetic induction, point magnetization, according to Pulsin, end quote. It's a little obtuse. He isn't so much saying, I borrow this idea from Poulson. Rather, he's clarifying just one piece of the invention.

But if you hold the Poulson and the Tao-Chek patent up side by side, the similarities are undeniable. Tao-check's drum is physically different. It's not a wire wrapped around a cylinder. Rather, it's a solid steel cylinder. As I said, that's different from later magnetic storage media. Later we'll see coatings made from metallic oxides, the same material as tape. But in this period, steel was the medium of choice.

Taocheque's transport mechanism is also very different. Paulson's drum spun continuously. That was needed because it recorded a waveform. It had to spin at a fixed rate to record and then reproduce audio. Tao check's drum did not. So how does that work? Well, first of all, the drum was broken into discrete rings. These aren't physical rings.

Think of them as regions where data is stored. The actual storage is done as a series of pulses. Now, to be clear, I don't mean binary encoding. I mean just straight pulses. A seven, for instance, was stored as seven magnetic pulses in a row. When you write to a ring, you first use a series of gears to move the magnetic head to the proper location. Then the drum rotates, and as it rotates, pulses are sent to the magnetic head. That records a series of pulses onto the drum.

Reading works in the same way. The head is moved to the proper ring, the drum sweeps, and the moving magnetic regions impart a current in the head. You get a series of electrical pulses that come out of a wire. That is, in a fundamental way, different from how later magnetic drums work. It's a lot closer to Pulsin's drum, or contemporary cylinder recorders. Taochecks' key changes all involve making the system more discrete. Instead of a spiral, the drum is broken into these discrete rings.

Data is stored as discrete pulses instead of waves, and you can play just one ring back. at a time. But this is missing a few crucial features that later memories have. One is the head configuration. Tauchek uses one head that moves to the location it needs to be in. That's not practical for computer memory. Later drums will have one head for each ring of data. You end up with a cylinder that's studded with magnetic reed heads. There's also the whole spin situation. Tau check's drum doesn't constantly rotate. Later drums spin up and then stay spinning at a fixed speed. Tau check's drum only spins during read and write.

That wouldn't be practical for something more sophisticated or more automated, like computer memory. So why is Tau check's drum so odd? Well, in short, he didn't know about computers. Come on. I know that sounds dumb, but it's true. This isn't digital technology. It has a few features that rhyme with the digital, but this ain't it. Dowcheck's patent tells us everything we need to know. His drum isn't meant as a general-purpose data store. It's designed for tracking account data for something like a bank. The patent isn't just for a drum that stores information, it's also for the mechanism that stores and retrieves that

data. The mechanism is itself pretty antiquated to see. It uses a rotary encoder like you'd see on an old rotary telephone. You dial up a ring which corresponds to an account number. Then you can either read data off the ring or dial in a new value. This wasn't a new idea. There are actually a pile of accounting machines that pop up during the first, about third of the 20th century. The development of these machines intersects with punch cards at some points, but in general it's kind of a parallel track. We get machines with banks of registers that record and persist account balances, interfaces and panels where bank tellers can pull up accounts and handle deposits and withdrawals. Tao Check's drum is interesting because it's using magnetism to represent data.

That's unique and important, but it's very different than what I've seen reported elsewhere. Taucheck doesn't just come on the scene in 1932 with a magnetic drum memory that stores binary data. This is more like an evolution of existing technology. Saying Taucech invented the drum isn't quite correct. He took one step towards the technology. We could just as easily argue that Poulson invented drum memory. We're seeing steps towards something more binary,

something more general purpose, and something more recognizable. But this isn't there yet. So what's the next step? Does Taochex drum evolve into something else, or is there a bit of a lapse in the technology? It's at this point I should mention the drum storage used in the at-Nasov-Berry computer.

However, I don't really want to linger on it. ABC, constructed in 1942, used a spinning drum filled with capacitors to store binary data. That's a lot closer to a magnetic drum, but there are some differences. Medium, you know, is a big one here. Capacitors are a little bit different than a magnetizable surface. The bigger reason not to linger is that ABC was very, very obscure up until the 1970s. It's not exactly a big part of digital lineage.

So while neat, and it does show that people can come up with drum on their own, there's enough of a disconnect that I don't want to spill a lot of ink here. I mention ABC because I want to make sure I'm being exhaustive. I mean, it did have a spinning drum that did hold data, but just not quite what I'm looking for. What I'm actually on the hunt for is a magnetic drum that stored binary data, had multiple magnetic heads, and has some connection to later data. I want to establish how we actually get to magnetic drum that shows up in computers, like my

beloved LGP30. The immediate place to look is clear, right? It's IBM. Taochecks' drum patent was owned by IBM, but did Big Blue actually make use of the patent? This is a tricky question to answer. We do, however, have one tool we can use. Patents contain citations. It's pretty easy to find patents that cite Tao Check's drum.

There are actually 10 that I can find. To be clear, this doesn't mean that there were exactly 10 things that used this drum. This will only tell us who filed patents for work that either improved on or somehow used Tao Check's drum. Does the search actually give us a useful answer, though? Well, no. No, it kind of doesn't. Oddly enough, none of these 10 patents are assigned to IBM.

There is a patent assigned to Bull S.A. in this list, and that, well, that's interesting, but also a bit of a dead end. Bull was a French company. In fact, they still are a French company. Bull was founded in 1990 to produce punch card equipment. Their patent that I'm referring to is from 1944 and concerns storing more buy binary-like data on drums. In an interesting twist, this patent is even more explicit about the Poulson connection.

It goes so far as to say that if you want it to record eight bits of data at a time, you'd have to have eight steel wires coiled around the drum coaxially. While interesting, that's as far as I've been able to take that trail. From all the sourcing that I can actually verify and back up, the true history of magnetic drum actually starts in Nazi Germany, but has nothing to do with IBM. So, yeah, I lied to you. This is actually a very Third Reich heavy episode, which I did not expect at all. Now, the history of magnetic drum is tied up with another three-letter company, Aege. Founded in 1882 as an electronics company, Aege would go on to produce the first commercial magnetic tape recorders.

These were called magnetophones. The first rough models were produced in 1936, and they were rough. They did not work very well, and they sounded awful. During World War II, the magnetophone was improved. This was done as a concerted effort directed by the Nazi party specifically. IG Farben discovered a way to make plastic tape, which was more reliable than metal. Experiments eventually led to formulations settling on iron three oxide. Biasing was discovered by radio engineers, so by the later part of the war,

it was possible to make very good recordings on tape. State-controlled radio stations were using the magnetophones to send pre-recorded broadcasts. The technology was a very deliberate cornerstone of state-run propaganda. This was literal secret Nazi super technology. Magnetic audio recording outside of Germany was rudimentary. Remember, the AEG recorders that were seeing pre-war sucked. German radio stations were broadcasting in ways that seemed impossible to the rest of the world. speeches were broadcast, while speakers were somewhere else.

Symphonies played without the distinctive crackle or hiss of vinyl or early tape. Allied Powers didn't know about this super tape until Europe was liberated. As the Allies swept Germany, all kinds of technology was plundered. A number of magnetophones were swept up along the way. This wasn't just a matter of incremental improvement. The wartime magnetophones were a massive jump in technology. Jack Mullen was one GI who helped liberate this new technology. He recounts hearing, quote, beautiful night music on the radio as soon as he landed in France.

And as he entered into Germany, he would see the radio secret for himself, to quote, via the history of music production. The station had been moved into a castle there to escape the bombing of Frankfurt. and it was being operated by the Armed Forces radio service. In response to my request for a demonstration of their magnetophone, the surgeon spoke in German to an assistant, who clicked his heels and ran off for a roll of tape. When he put the tape on the machine, I really flipped. I couldn't tell whether it was live or playback.

There simply was no background noise, end quote. Mullin looted one of these magnetophones and then shipped it back to the states in pieces. he'd sell it to Bing Crosby. That's the very, very short story of how real-to-reel tape recording made it to the States. The AEG magnetophones were cloned, improved upon, and evolved. They'd become a new technology that was a cornerstone of broadcasting in America and around the world. But that's just the audio side of things.

Mullen wasn't the only American to get his hands on a looted magnetophone. Around 1946, the U.S. Navy was in need of some kind of digital data storage. They wanted something faster than the venerable punch card. That year is crucial because it's before we really have computers. The first digital tape drives don't spin to life until the end of the decade. Mercury delay lines exist, but that's a more short-term form of memory. There actually aren't good options for fast storage in 1944, This is happening just as a new hero is emerging on the scene, a new three-letter corporation.

A group of codebreakers from the Navy's communication supplementary activity Washington, aka Seesaw, had just left federal life. Three of the Seesaw researchers had broken off to start an engineering firm. This firm would come to be known as Engineering Research Associates, or ERA. You may remember them from my coverage of analog and hybrid computers. The actual formation of ERA is a little interesting. They basically took over this other company, Northwest Aeronautical Corporation, that was closing up shop. ERA came in, took their physical space, and because of that, there was a little bit of a time lag between NAC winding dense, and ERA winding up. So ERA's earliest contracts were actually signed to NAC. The story of how

ERA formed is recounted in computers and commerce by Arthur Norberg, and surprisingly, it does go back to Nazis again, at least in large part. So ask yourself this question. What exactly was C-Saw doing during the war? What kind of? What kind of? of projects would have set up Seesaw engineers to create a computer company. Well, in a word, it's all code-breaking. Seesaw was the American equivalent of Bletchley Park, at least it was one of the American equivalents. Seesaw constructed and operated bomb machines, co-breaking devices that verged on the complexity of computer. Bombs could brute-force messages encrypted with Nazi Enigma machines. This was the real start of ERA, researchers and their associates trying to extend and

improve the bomb. The war was over, but there was still a lot of money to be had in cryptography. Norbury explains how this led directly to storage research. In short, decryption has to be fast to be useful. Encrypted messages often contain time-sensitive information, so it's best if you can crack the code fast. When you get down to it, that means you need a way to churn through data quickly. You need some type of high-speed data storage. Bombs work, but they're very manual machines. Problems were wired up using patch cables, much like ENIAC. There is an immediate desire to speed up the process, especially when it came to data entry. There are also other machines that worked better than bombs for more general decryption tasks.

Seesaw worked through attempts with punch cards and paper tape, but those are very limited media. When ERA forms, they take up this challenge. Some of their first contracts involve finding better and faster forms of data storage. The large project they were working on this period is called Project Goldberg. And the tale of that contract comes to us via an oral history with SIEC, Sidney Rubens. He was an early ERA employee who worked directly on this contract. Quote, they weren't sure how they were going to build this device. The original specifications called for a digital store using photographic film, which could not

be altered, of course, but would be useful for one run. They had a captured German magnetophone tape recorder here with lots of good tape, a couple good heads on the machine. The electronics was a mess. We never did use it as a recorder, and they asked me to look into ways of proceeding." The magnetophone looted from some Nazi-controlled radio station is the key connection here. Rubens and his coworkers didn't read Taoisek's patent and think, ah, gee, we could spice that up. Rather, the first real digital magnetic drum memory was built from the best recording media at the time. Nazi rust. But that wasn't immediately apparent.

Project Goldberg investigated a number of ways to store data magnetically. One attempt involved a disc, but not in the way you'd think. The team used a spinning steel disc and attempted to record magnetic pulses on its edge. That showed promise, but didn't pan out. They had issues attempting to store multiple tracks of data. They would also try tape itself, but that didn't really lead anywhere. This is where Rubens entered the picture. He, unlike the rest of Goldberg's team, knew a thing or two about magnetism. He knew that steel wouldn't work well for holding pulses. It also helps that he was able to actually do some literature research. Rubens went back to the source. He translated a paper by Heinz Lubbock called Magnetic Sound Recording with Films and Ringheads.

This paper describes the state of the magnetophone in 1937. It sounds as if no one in ERA had actually bothered to go back and read the theory behind how these new recorders worked. With better theoretical knowledge, and armed with some very nice tape, Rubens was able to enact a plan. That plan involved a drum. Where did Rubens get the idea to use one? Well, we aren't sure. I've read everything I can find about him, and I can't find a specific reason. The best I can find is from his oral histories.

Quote, I thought, well, the first thing I'll try is a five-inch aluminum drum, end quote. And maybe that's all there is to it. Faced with a few options in the fact that disc and tape didn't really show problems, He went with a drum. But note the material. Aluminum is what's called paramagnetic. It doesn't really react to magnetic fields. So then how was he planning to record data onto this new drum?

The answer is simple. Rubens had access to all this nice magnetophone tape. Why not use that? He used this trick where you can soak scotch tape and tallywine. That lets you remove the... adhesive from the plastic backing. With some careful maneuvering, Ruben was able to build up a layer of adhesive on his drum. Then he stuck strips of magnetophoam tape to the aluminum drum. I want to be sure you get the right picture here. The drum's five inches in diameter,

but it was only about an inch long. This is barely removed from a disc. The difference, however, is all in the length. Ruben fit two strips of magnetophone tape on the surface of the drum. The geometry of the drum ends up just being a better fit for this problem. It also let him do the whole adhesive trick, so that's kind of nifty. Each ring of tape functioned as an independent track. One track contained a timing pulse, the other contained data. The timing track was needed so you could actually tell

when the data track flashed a zero. It also simplifies the overall design if the drum can make its own timing. That means you don't have to do anything fiddly with timing or synchronizing speed. The timing is the drum itself. The squat drum was mounted to, of all things, a bench grinder. One of Ruben's colleagues worked up an adapter to make that happen. This had the benefit of spinning the drum at a moment. more or less constant speed and was easy to set up. Toolposts on the grinder were used to mount

tape heads. Crucially, these heads were never touching the drum. They sat just above it, which limited wear on the delicate tape. In fact, no contact meant really nowhere at all. What Rubens had worked up was a very rudimentary device. I mean, using a bench grinder is a bit of a tip-off. This first drum didn't spin all the time. It was actually written to while the drum was stationary. A set of gears were used to rotate the drum to the next bit, and then data was fed to the tapehead. The disc was stepped this way until it filled up with information. Reeds were carried out while it was spinning. This was rough, but it showed that a drum could be used for data storage. From there, things scaled and improved.

there was an immediate sprint towards a deliverable drum for the Navy contract, and there was the longer project towards a better drum in general. In 1946, the result of Project Goldberg was done, a practical and usable magnetic drum. The first Goldberg device was scaled up from the bench grinder. It had 25 magnetic heads, one for timing, the rest for binary data. It was 36 inches across and a solid foot long. The surface was still magnetic tape pasted onto metal.

It was still written too statically in red well in motion, but that was workable. The main advancement came down to construction and scale. The naval drum was shrouded. The drum itself was housed in an aluminum tube, that tube had a series of holes drilled in it, and each hole was threaded to accept a magnetic head module. This allowed heads to be replaced if they burnt out. It also made the drum more safe and more stable. An open drum is asking for problems.

Dust could basically destroy the thing, and at 600 RPM, this could mangle your hand pretty badly. Once we reach this drum, we're basically there, right? We have a drum that uses ferric oxide to store magnetic flux. As it spins, it sends data out a wire. That's much closer to the drum of the LGP30 than Tao Checks device. It's just that the Goldberg drum is still pretty rudimentary. Part of that is coding, but that would change pretty quickly.

By 1948, 3M was producing a lacquer with iron-3 oxide particles in suspension. That made it possible to spray on the magnetic coating instead of taping it on. Experience showed it was actually possible to write to a drum in motion. So by the time spray-on drums were developed, the awkward step-and-write cycle was dropped. That's when we get the first drums that appear fully modern, so to speak. So what's this new wonder device used for, exactly? Why on earth would the Navy even want a magnetic drum in the first place? Going into this episode, I really imagined that drum memory must have been used on something prior to computers.

That a tabulator or a calculator, maybe some military device used drums. But that's not entirely the case. This is one of those things where we have very blurry lines. I was hung up on this idea because of the whole borrowed aspect of early computing. Mercury delay lines come from radar dishes, tubes from radio, all that jazz. Drum doesn't follow that path. In fact, maybe we should just say that magnetic media doesn't really follow that borrowed path at all. We saw a few episodes back how digital tape was developed almost from scratch for use in digital computers.

The tape deck used by machines like Radak, Univac, and the IBM 701 share very little with audio decks. This is largely due to timing. When the Eckert Mouchley computer companies start developing digital tape, the state of audio tape is very, very rough. They also don't have access to a magnetophone, actually, but it's... a separate thing. The same goes for drum. When ERA starts developing digital drum storage, there isn't really any prior art to borrow from. Taucheck's drum exists in theory, but that doesn't really help. The patent's too different, and the ERA staff probably didn't even know about the patent. They had to create a drum from scratch in the most literal sense of the word. So we don't see tabulators

in the 40s that used drum because drum didn't exist. We don't see a tabulator in 1947 that uses drum because drums were being developed for computers, or at least for some kind of very sophisticated digital logic machine. So what's using these? I have to clarify something here. I said that the drums at ERA were developed as part of Project Goldberg. They were part of the larger project.

That's true. But the overarching project was to create a computer. That's part of the secret that Rubens didn't originally know. This is where we get into some very, very muddy territory. ERA produced three unique machines at the tail end of the 1940s. One of those enters mass production. Those machines are sometimes called Goldberg-Demond, and Atlas. All these machines used drum. The Atlas is then produced as the ERA 1101, which in turn

becomes the Univac 1101. That machine is well known. Goldberg and Demon, less so, and they may not have even been fully fledged computers. Goldberg and Demon were special-purpose crypto-analytical machines. The break for me in this was finding an oral history written by John Hill, an employee at ERA, who was a little higher up the food chain than Rubens. Hill was assigned to Goldberg after Rubens figured out how to get a drum to sing. And that's when he learned the truth about the project, quote, well, at that point with my clearance, I was now privy to quite a bit of what was going on there. And I found out then, for the first time, how these streams are characters were being passed by one another in an attempt to find what they called hits,

which would be the coincidence of a character sequence among two tapes or among two streams of data. He continues, from the first day on there, and perhaps with Rubens having developed something I wasn't aware of up to that time, we started cementing tape onto cylindrical surfaces and doing all our work from thereon with those kinds of surfaces, end quote. the drum was to be the heart of the Goldberg machine, and the drum was created explicitly for Goldberg. So therefore, Goldberg is the first machine to use drum memory. But it was really being used for fast data recall.

The true purpose was to try and break enciphered text. But this still leaves us with a mystery. What exactly was the Goldberg computer? and was it even a computer? This one has been quite the rabbit hole. So check it out. I was able to find a number of declassified reports from the NSA that partly answer my questions. But just to be clear, this is like teasing out a riddle.

The reports I've found are from the 1950s, so aren't perfectly contemporary. They're also heavily redacted and, reference documents that I can't find. Take this as the best picture I can put together right now. Goldberg wasn't exactly a computer, but it was close. It was known as a tape comparator. These are a specific type of cryptography machine that was used by code breakers. Now, comparators differ from the bombs of World War II. The idea here is to compare long samples of cipher time. encrypted data. If you can find similarities, then there are methods you can use to break that encryption. It all comes down to how cryptographic keys are used and the size of the key versus

the size of the message. Tape comparators were developed, I think, in the 1930s. I say I think because I have scant literature. What I get is one NSA paper called machines in the service of crypto analysis. I've been able to back up very little of this tale. So, grain of salt, I got one, maybe one and a half sources that are all the NSA. The first comparator was created by none other than Vannevere Bush in 1938. It was called the 70-MM comparator because it used 70-millimeter wide paper tape. Data was fed into the machine as reels of punched paper tape. The tapes were compared to each other, and similarities were flagged. The machine entered service in 1942 and could run out a staggering 85 characters per minute. I'm hedging my bets here a little bit, because

Van of your Bush doesn't mention this at all in his autobiography. So, again, my source, A handful of NSA documents. Now, that 85 character per minute thing might sound slow, but there are some extenuating circumstances. This isn't just a straight comparison between two tapes. The process actually entails a sliding window kind of algorithm. You're looking for chunks of tape one that match up to chunks of tape two. And you're looking for a statistical comparison, not a direct match.

something like Region A has a 5% similarity to Region B. That process for all the possible regions is slow, especially for tech in the 30s. Goldberg hits the scene in 1948, and is much more sophisticated than its predecessors. From Cryptoanalytic machines in the NSA, quote, Goldberg is a high-speed general-purpose comparator and scoring device, with magnetic drums designed to compare two or three streams of data. It does coincidence and frequency counts, cribbageing, wheel stripping, transformations, pattern, and round-robin search,

weighting, distribution matching, and calculation of certain statistics, end quote. The report gives us more detail about Goldberg. The machine is special purpose, but it can serve a lot of purposes. It contained two magnetic drums and a tape reader. When Goldberg compares two streams, it does so drum to drum. When it does three, it would have used the tape reader for the third source of data. The reader is also how data is loaded onto the drums. Goldberg is so early that it still uses the awkward, write-in-advance method to store data.

The thing is almost a computer, and I really mean almost. It has complex mathematical and logic circuits. It has memory. It has I.O. It even has a printer. What it lacks is programmability. That's the only thing preventing Goldberg from being a full computer. That said, you will see places online that call Goldberg a computer. And that's not strictly correct, but it feels very close to right.

There's a short bridge here in the history of the drum that I have to mention. That's Project Demon. It was the direct successor to Goldberg, and it was a very aggressive successor. One oral history I read even recalls that once ERA started working on Demon, the existing Goldberg team was essentially gutted. We have even less detail on Demon. What we do know was its use case. Quote, Demon 1, specifically for Redacted. The redaction there goes on for about a full page.

I believe that Demon and Goldberg were both used for breaking encrypted Soviet messages, but I don't have anything concrete to prove that. That's kind of a contextual guess. I bring up Demon for one fascinating observation. The NSA reports I've been working out of describe machines in terms of the tape they accepted. So Goldberg is sometimes said to use Goldberg tape. It's sometimes called a Goldberg tape machine. the same with 70mm.

There's a bit of a convolution going on here, and that gets really funny when we get to demon. When discussing the early use of the device, one report said this, quote, a 25-position demon-punch check was developed to check tapes, end quote.

From this, one can infer that the demon used a 25-bit tape. The drum was probably the setup that we've discussed earlier. 24 bits for data, one bit for clock or parody or some other check information. But I love the implication here. If you need a checker for tape, that also means you need something to make that tape. So the NSA must have had a device that would have just been called the 25-position demon punch.

That's hilarious to me. It sounds like something from Fists of the North Star or some kung fu-foo movie. Beware the man marked by three scars, for he knows the secret 25-position demon-punch. I would like to know who the NSA was using this 25-position demon punch on, and also perhaps how to execute it for myself. Anyway, these early drum machines weren't exactly computers, but they were very close. In fact, this is similar to the situation we get with tabulators and calculators around the same period. These are machines that are just about as complicated as computers, but they just don't

have the programmability piece needed to make them technically computers. It's also important to note that we just get two of these machines that aren't computers but use Drum. That's what I mean when I said that Drum is really a computing technology. We just have two outliers. So then, when do computers actually enter the picture. That gets us to something that's been stuck in my craw for a while. As I said earlier, and I'll keep talking about this, I've been cruising this old 1953 survey on digital computers that the Office of Naval Research put out. That survey includes the year each computer enters operation. If you chart things out, you notice a gap, at least from that source, it appears to be a gap. The first machines come online around 1945, then we get another big batch of machines

around 1950. During that five years, we only see a few computers in our operation. Drum appears in that gap, so we get these non-computer things that use drum before we actually get the next batch of new computers. As an aside, I don't know how real this gap is. It's a, at least a gap as far as the sourcing I've seen goes. I need to do some work to see if I can compile a better chronology for these five years, because this is really interesting to me, but that's a separate task. When we finally get computers with drums, it's closer to 1950. What we end up with is probably a bit of a tie.

ERA themselves produced a computer called Atlas, which entered service in 1950. This started as a unique machine and was then adapted into the ERA 1101, which was serially produced. Atlas was a full-on von Neumann architecture stored program computer. It's exactly what you'd expect out of a machine just with drum instead of more sophisticated RAM. And at that point, it was using a spray-coated drum that could be written to while in motion. so we have the full-on finished technology here. That same year, the Navy turned on a computer called ABLE. According to reports,

able is a, quote, relay computer for low-speed parallel computation, including arithmetic and other logical processes. The same report describes it as a relay analog of Atlas, which is fascinating. Most important of all, Abel was a drum computer. These factors together actually make Abel super unique. There were not many relay-based programmable computers, probably fewer with drum,

and clone machines in this period were basically unheard of. That's one cluster of drum computers, all centered around the Navy. IBM was the other location where drum memories were under consideration. Big Blue was, in part, goaded up. on by work done at ERA. But that's a story for another time. During the latter part of the decade, IBM made attempts to get a drum to work and to make mass-produced drum-based machines. That project didn't hit its stride until 1950 or 52 with the CPC and the IBM 650. I bring this up because of the ERA connection. During the development phase, IBM entered into a contract with ERA to get

designs for magnetic drums. Those designs would ultimately have an impact on the direction of IBM's work. In IBM's early computers, it's made pretty clear that the choice to even continue investigating drum memory came because of work done in papers published by ERA. That's the ultimate lineage. Magnetic drum memories are developed at ERA and proliferate into computers from there. ERA's story is quite clear. Their technology is adapted from very early work on magnetic tape. In other words, Tauchak's patent has basically nothing to do with the story. All right, that does it for the origin of magnetic drum memory.

It's clear to see how cool drum is. As an electromechanical form of memory, it's already in rarefied air. In general, the field of early memory just feels kind of magical to me. Mercury delay lines will always hold a special spot in my heart, partly because Mercury doesn't really excrete from your body, but drum lives pretty close to it. The convoluted history of the drum makes it even more cool. I'm inclined to say that drum was invented twice

and that it was some kind of lost technology, but that's not really correct. As we've seen, Tao Check's drum bears some similarities to later magnetic drums, but it's not really there. The largest similarities are in shape and magnetism, but that's it. When it comes to magnetic storage, I've learned that the devil is really in formulation. And Tao Check just straight up didn't have that. The drum also moves and functions in ways different from memory drums.

Drums, as we know them, come out of ERA. That device is created from scratch and from some looted reel-to-reel tape machines. In fact, the magnetic drum shares a lot more DNA with tape storage in general, which is something I never expected. All later drums derive from the ERA design, and I couldn't find any connection between ERA and Tao check. The lineage appears just to not exist there. And with that, I'm finally out of my magnetic rut, I think.

But I'm going to stay in this era for a while. There's some mystery in this early period that I'm still trying to work out for myself. Until I pull the curtain back on that, thanks for listening to Admin of Computing. I'll be back in two weeks' time with the next episode, and I'll be back in a week with the next episode of Adjunct of Computing, the official Advent of Computing after show. Go give it a listen. You can find links to everything including my Patreon. Please go subscribe at advent ofcomputing.com. And as always, have a great rest of your show.

your day.