Reverse Engineering A Module From A Vacuum Tube Computer

It’s best to admit upfront that vacuum tubes can be baffling to some of the younger generation of engineers. Yes, we get how electron flow from cathode to anode can be controlled with a grid, and how that can be used to amplify and control current. But there are still some things that just don’t always to click when looking at a schematic for a tube circuit. Maybe we just grew up at the wrong time.

Someone who’s clearly not old enough to have ridden the first wave of electronics but still seems to have mastered the concepts of thermionic emission is [Usagi Electric], who has been doing some great work on reverse engineering modules from old vacuum tube computers. The video below focuses on a two-tube pluggable module from an IBM 650, a machine that dates clear back to 1954. The eBay find was nothing more than two tube sockets and a pair of resistors joined to a plug by a hoop of metal. With almost nothing to go on, [Usagi] was still able to figure out what tubes would have gone in the sockets — the nine-pin socket was a big clue — and determine that the module was likely a dual NAND gate. To test his theory, [Usagi] took some liberties with the original voltages used by IBM and built a breakout PCB. It’s an interesting mix of technologies, but he was able to walk through the truth table and confirm that his module is a dual NAND gate.

The video is a bit long but it’s chock full of tidbits that really help clear up how tubes work. Along with some help from this article about how triodes work, this will put you on the path to thermionic enlightenment.

32 thoughts on “Reverse Engineering A Module From A Vacuum Tube Computer

      1. I’m not entirely sure if you’re referring to the CRT’s that were used as Random Access Memory, but those were awesome. Called, the Williams Tube, they used secondary emissions from the electron beam striking the phosphor to create a charge well, which equated to a single bit of memory. Typically, a single CRT would store 1,024 bits of data. So you could easily get to the ZX80 level of memory with just a handful of Williams Tubes!

      1. Which is why they used storage tubes, magnetic drums, delay lines or core memory. All of which is smaller and less error prone than 8k vacuum tubes and won’t take a 400 volt 200 amp service to drive :)

      2. You’re totally right! We need 8-bits to make a single byte. But, there’s actually a better way to make a simple SR Flip Flop using a single dual triode setup as a multivibrator. IBM actually did this, so this dual NAND gate probably served a different purpose than being a flip flop. So, if we use a single 6201, like I used in the video, to create a single bit of memory, we need 8,184 tubes to make exactly 1,023 bytes of memory. At 300 mA at 6.3V per tube to power the heaters, that’s 15,468 Watts of power!

        Furthermore, one module has a footprint of 24mm x 24mm. If we were to make a square of modules, you would need 90 modules by 90 modules. That gives us a square that’s a little over 2 meters long, or about 7 feet.

        So, if you wanted to use vacuum tube flip flops to recreate the 1k memory of the ZX80, you’d need a 7′ x 7′ (2.1m x 2.1m) square with 8,184 modules consuming 15,648 Watts of power!

        Of course, there were much better ways to get proper memory storage, like rotating drum memory, but that’s not as fun as something needs a power sub-station to run, haha.

        1. Magnetic drum memory took quite a bit less power than flip-flops, though. There were a number of drum-based small computers in the 1950s and 60s that could be plugged into a 120V outlet.

          1. Oh, absolutely! The IBM 650 in particular was about the size of a cabinet with a little magnetic drum built in. I think it’s quite the beautiful design! Although, someday, I’d love to tinker around with magnetic core memory, which started seeing some more use near the end of the vacuum tube computer life cycle.

    1. It gets really fun when you start figuring out the power requirements. The two tubes I used in that video are the 6201 and the 6AL5. They each want about 300mA at 6.3V to power the heaters, which means 3.78 Watts of power per dual NAND gate. Multiply that by 1024, and it’s 3,870 Watts, just for the heaters!

      1. Which is why NO early computers depended on vacuum tube flip-flops for storing data. There were CRT storage tubes (Williams tubes, mentioned in another comment), mercury delay lines, magnetic drums, and electrostatic drums, all of which were developed BECAUSE it would have cost a fortune in parts to build and in power to run, using flip-flops for anything other than short-term register memory. In fact, whole computer architectures were designed around the requirements of the primary storage device. For example, many drum-based computers had encoded in each instruction, the location of the next instruction on the drum, and programming involved not only making the logic work, but also optimizing it by arranging the instructions so that as each instruction finished execution, the next instruction was just about to come around to the read heads. CRT storage could be addressed in random order, but it was very limited in capacity, and was used primarily as temporary register storage. It wasn’t until the invention of ferrite “core” memory that storage became truly random access.

        1. That’s super cool!
          Interestingly, though, if you’re alright with using silicon diodes (IBM themselves use a lot of germanium and selenium diodes in their tube computers), it’s really easy to build a NOR gate out of vacuum tubes. A single dual triode with four diodes can get you two NOR gates in a relatively tiny package. And, like the NAND gate, the NOR gate is also universal (even the Apollo Guidance Computer was built entirely out of NOR gates).
          That’s actually been the basis of most of my tube circuit builds, although I do use NAND gates on the occasion where it can save me some space.

          1. From what I have gathered : in the 50s diodes were more expensive than tubes so they were used only when absolutely required.
            My theory is that it’s not just the thermionic tube that allowed the computer industry to expand : it’s the availability of reliable and cheap diodes, because they are “passive” (don’t draw energy) and can solve many problems (diode ROMs anybody ?) in a small package.
            The transistor sure was another quantum leap but DTL was a thing for a long time and you could do complex computations with diodes and use one transistor to amplify the result.

            However getting enough cheap, reliable and mass-manufactured diodes with high yield and tight specs was a whole industrial quest (with the transistor very closely following). Silicon purity and defect management took a long time to refine…

          2. In the early IBM computers, like the 604, they definitely preferred tube diodes because they were cheaper and much more resilient than silicon diodes available at the time. But, by the mid 50s, germanium and selenium diodes were cheap enough that they were becoming more common in IBM’s computers than the tube equivalent. The 650, for example, uses a lot more silicon diodes than the 604.
            That led to a lot of circuits in IBMs computers being built similar to the DTL circuits that became ubiquitous as the transistor gained popularity.

            The 40s, 50s, and 60s were absolutely mental when it came to computing. Imagine in just 30 years, the World saw the birth of the computer, the birth of the transistor and the birth of IC, which ultimately, put men on the moon!

        2. First, alxy, I don’t know how many vacuum tubes a 650 had, but it was a “compact” machine, with the card reader and punch being almost as big as the computer itself, and the power supply was about that size again. So if 352 NAND gates is “almost” doable in vacuum tubes, then the 650, having about an order of magnitude more tubes than that, itself would clearly be impossible. Which is a funny thing to say about the most popular computer of its day. But maybe by “almost possible”, you mean for a DIY project?

          And second, UNIVAC was a transistor computer, far more capable than the 650. So I’m not sure where “imagine having your own UNIVAC” fits in here. UNIVAC was BIG. I don’t think making a computer out of the least possible number of gates would get you anywhere near the same user experience as a UNIVAC.

          1. I’ll never understand the attraction of vacuum tubes, having had to deal with them for the first twenty-five years of my life. But okay. I guess it’s like vinyl records.

      2. No problem. Run it during the winter months, and save on furnace cost!

        As added bonus, share live video feed on YT and profit. Nobody will watch your furnace, but your vacuum tube heater (sorry, computer) definitely!

  1. ENIAC-on-a-Chip
    Moore School of Electrical Engineering
    University of Pennsylvania,Titi%20Alailima%2C%20James%20Tau%2C%20D.%20J.

    The full ENIAC(TM)-on-a-Chip has been fabricated in a 0.5 micron, triple metal, nwell CMOS process. The chip has a size of 7.44mm by 5.29mm and contains about 174,569 transistors.

  2. The above comments are why I read hackaday. I assume everyone who commented has decades of experience but I still wonder what books your reading to know all of the details about IBM and the tubes.

    1. Honestly, I only really started learning about vacuum tubes in earnest at the beginning of the year. My Youtube channel (Usagi Electric) is kind of my place to share the new things I learn! My holy grail(s) have been the IBM 604 and 650 Customer Engineering Manuals. They’re absolutely packed with excellent information and I’ve learned massive amounts from them.

        1. Interesting that you mention relays, as that was where I started out with learning digital logic on a fundamental level!
          I ultimately ended up building a 10-bit serial relay adder and a 4-bit hexadecimal display relay adder. I learned an absolute massive amount about the fundamentals of logic in more complex computing circuits.

          10-bit serial relay adder:
          Hexadecimal relay adder:

    1. I don’t remember what they were called (NOT Compactrons, though those were a similar development), but for a while, radios in someplace were taxed according to the number of tubes they contained, so there were tubes that contained not only multiple active components (diodes, triodes, pentodes), but also some of the passive components. So yes, there were vacuum integrated circuits. But unlike monolithic solid state integrated circuits, these were not cheaper to produce than the non-integrated versions.

      As for Compactrons, these were invented to compete with transistors by bringing down the parts cost of vacuum tube radios and TVs, but of course the cost of transistors dropped quickly enough that they were only around for a few years. It WAS intended to replace 5-tube “American 5” radio designs with a two-tube Compactron design (one tube for the LO, mixer, and IF stages, the second for audio), but I don’t know if this ever came to pass.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.