Vacuum Tube Computer P.23 - Electrostatic, Rotating Drum and Core Memory

[@CuriousMarc]

@TubeTimeUS (Eric Schlaepfer) is currently trying to replicate a Williams tube memory from an early IBM computer. It’s a nice visual memory, but very finicky. Also you could do neon tube memory - these are both simple and very visual!

[@MAYERMAKES]

I have a suspicion that is exactly what he ends up doing!

[@trevorhaddox6884]

It would have to be VFD memory, there is not enough voltage to trigger a neon. EDIT: VFDs probably wouldn't work though, they don't have hysteresis.

[@SeekingTheLoveThatGodMeans7648]

@@trevorhaddox6884 Would it be possible to hook VFD devices in a flip flop configuration so that either one of a pair of them is on but not the other? Getting the right behavior in principle is easy; properly tuning the power, writing, and reading circuitry is harder, especially with the rough logic tolerances that his computer is operating with. I remember reading about the first vacuum tube flip flops, and that they were VERY finicky, though the modern availability of high gain vacuum tubes helps a lot.

[@CuriousMarc]

@@trevorhaddox6884 I think he will use low voltage rabbit hair memory instead ;-)

[@UsagiElectric]

I've been following TubeTime's twitter posts on the Williams Tube memory very closely! It's been a while since he updated the Twitter thread, but I found the bit about the 400V rail being smoothed to less than 15mV of ripple, which makes total sense. If the beam wobbles even a little bit, the secondary emission can leak over to the next bit and totally corrupt the data. I'm excited to see what kind of setup he builds for it in the future! I thought quite a bit about neon memory, mostly because I love neons, you can use them for just about anything! The hysteresis of the bulb makes building a 1-bit memory cell relatively easy, but with just +24V and -12V on tap, I only have about half the voltage I'd need to get it to work. Though, the bigger problem stems from the 1-bit data bus and addressing the memory. I'll go into a lot more detail on this in the next video in the series, but essentially, I have to address each bit completely individually, which means the addressing control stuff ends up much larger than the actual memory itself. Ultimately though, I'm a bit of a brute force kind of person, so I just built memory as simply as possible!

[@Dr_Mario2007]

Fun fact - magnetic core memory have essentially come back, this time much more nonvolatile. Magnetic RAM is essentially the descendant of the magnetic core memory, which both works essentially the same way, only you can read the same thing over and over without incurring the bit rot errors. Also, Magnetic RAM could be the basis of the future SSDs, as they can be shrunken down to 3 nanometers or smaller without having the charge leakage problems that sub-10nm NAND flash chips are struggling with.

[@UsagiElectric]

Whoa, that's very cool! The fact that it's non-volatile and can be read with a simple current measurement is very cool. It seems it's a little more power hungry to execute a write operation, but the added speed and density, and reduced complexity from needing a refresh circuit, sounds really promising. It'll be really interesting to see where MRAM goes in the future!

[@Dr_Mario2007]

@@UsagiElectric Indeed it is power hungry, however with the shrinking of transistors (and the replacement of Silicon with something else), it could potentially mean reduced write current. Much remain to be seen for now.

[@jeromethiel4323]

Even better, MRAM doesn't seem to have the write cycle issues that flash memory does.

[@mohinderkaur6671]

Intel Bubble ram would be fun to use

[@MadScientist267]

DRO != Bitrot

[@georgegonzalez2476]

Some other alternatives: (1) Delay-line memory: Either a long tube full of mercury, or simpler a length of steel wire. Back around 1977 the university had some CRT terminal that had delay-line memory. You could tell because the screen display would get random errors if you hit the side of the case. Also it would glitch if the room temperature got above like 85. See Wikipedia under "Delay-line_memory".

[@UsagiElectric]

I did actually consider delay line memory, but I actually shied away from it for a couple of reasons. It's not very visual. Sure the physical line itself looks cool, but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). Delay line is super cool and definitely seems like a solid alternative, but for the tube computer, I'm going to sacrifice some bit density for some blinkenlights!

[@gort59]

Caution! Mercury is poisonous in all it’s forms!

[@georgegonzalez2476]

@@gort59 Well, maybe not as bad as commonly described. In the 1950's at Oak Ridge the gummint had a plant to extract Lithium-6 isotope, for like H-Bombs. They had over one million pounds of Mercury at any one time! Somewhere between 200,000 and 2,400,000 pounds just disappeared! Most of it probably down a certain creek. And the environmental and human effects were not disastrous. You see Mercury as a pure element is mostly unreactive, almost inert-- it's compounds like methyl Mercury that are very biologically active and a huge hazard. Anyway, one study concluded that even if you ate 12 servings of fish every week from that creek, the amount of Mercury was only like 5% of a harmful dose. I guess most of the Mercury just went to the bottom and got buried in the muck and sediment. Lucky.

[@gort59]

Worked in a Pathology lab for many years and saw “firsthand” via stained brain tissue just what mercury can do. Thanks for the info though.

[@neeneko]

I am actually a bit sad there is not more hobbyist attention on delay line memory. you can find kits or howtos for building all sorts of early memory types with modern parts, but the retro interest in recreating it doesn't seem to have picked up steam yet.

[@DonnyHooterHoot]

I have an HP core memory module. It's amazing to look at it and think people had to wire it mostly by hand. And they made millions of them.

[@UsagiElectric]

That's awesome that you have a core memory module! One of these days I'll buy a module and tinker around with it. It was a very labor intensive type of memory, but the bit density you could get out of it was totally worth the extra staff you had to employ to build it!

[@davidlawton898]

Thinking of Drum memory, I used to work on repairing drum memory on an ICL Key-Edit 100 system (Key to Drum), Occasionally we would have clock reading problems (multiple spare clock tracks on the drum) and we would spare out the faulty clock track for a replacement track. Other times with read errors, we would actually use Jeweller's Rouge (a cutting / polishing compound) to re-polish the offending part of the track and bring it back into service. Another time I came across drum memory was when I visited the Alcator-C Tokamak computer room level at the Francis Bitter National Magnet Laboratory at M.I.T back in 1981 - and they used drum memory to download an enormous amount of data over a short interval (milliseconds) to store the results of each Tokamak firing. The Tokamak itself was in a large room below the computer room, accessed by a spiral staircase, with railway like tracks to dump the energy into the plasma chamber generated from a nearby power station and stored internally at the lab. Spinning drums were the fastest way to store a lot of short duration data!

[@gammaleader96]

I wanted to say that neon bulb memory might be a good idea, however Marc already suggested that. My guss on what you will propably do is somehting along the lines of RS flip flops / latches as memory? For short term storage, wire memory is also a very fascinating thing to look at, it uses a long wire that sends pulses in a loop and uses the delay of the wire as the "storage". A very similar form to this is mercury wave memory where it literally uses a long tube of mercury instead of a wire. ... which is not all that practical to have in your room.

[@UsagiElectric]

I did actually consider neon memory for a while, but I just don't have enough voltage on tap. With just +24V and -12V I only have about half the voltage I would need to get a Neon to strike. Sure I could bump up the voltage, but again, that kind of defeats the ethos of the tube computer. You're pretty much right on the money for what I'll ultimately be building though! I did actually consider delay line memory as well, but I actually shied away from it for a couple of reasons. It's not very visual. Sure the physical line itself looks cool, but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). Delay line is super cool and definitely seems like a solid alternative, but for the tube computer, I'm going to sacrifice some bit density for some blinkenlights!

[@szapcsika]

Lot of suggestion for delay line memory. Maybe the 64us acustic delay line from old tv sets' colour decoder could be used?

[@GodmanchesterGoblin]

That's a novel idea. You could store a useful number of bits in 64us without pushing the limits of the technology. I guess there are probably still parts available somewhere.

[@UsagiElectric]

I did actually consider delay line memory, but I actually shied away from it for a couple of reasons. It's not very visual. Sure the physical line itself looks cool, but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being able to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). Delay line is super cool and definitely seems like a solid alternative, but for the tube computer, I'm going to sacrifice some bit density for some blinkenlights!

[@ninefingerdeathgrip]

4:39 seems like some dreams come true!

[@kfl611]

Home built eniac computer - how totally cool !!!!!!!!!!!!!!!!!!!!!!!!! Your brain power makes my head spin! And yet, you seem so normal ! Keep up the good work.

[@DavstrWrexham]

There was another type of memory used called delay line memory as well. It was the first kind of memory created for electric computers

[@lindoran]

Delay lines can also be used as a clock, it's kind of the simplest state storage method I can think of, simply predictable time states being used to store data.

[@peterlinddk]

Yeah, I was thinking of that as well - Mercury Delay Lines or acoustic delay-lines might be the simplest choice for a 1-bit computer, since it stores everything serially as well. As I understand it mostly acts like a big shift-register, with a lot of analog circuitry though ...

[@UsagiElectric]

I did actually consider delay line memory, but I actually shied away from it for a couple of reasons. It's not very visual. Sure the physical line itself looks cool (whether it's a mercury line or a coil of wire), but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). Delay line is super cool and definitely seems like a solid alternative, but for the tube computer, I'm going to sacrifice some bit density for some blinkenlights!

[@MikePerigo]

@@UsagiElectric For visually interactive tube storage don't forget to consider the Dekatron tubes.

[@UsagiElectric]

@@MikePerigo I would love to do something with Dekatrons in the future, but they're a nightmare when it comes to voltages! And since most of the voltages needed for a Dekatron are well over 100V (or under -100V), they wouldn't quite work for the low voltage tube computer. But, Artem Kashkanov here on KZbin is building a tube computer as well, and he's using Dekatrons as one of the primary elements!

[@melkiorwiseman5234]

Rotating Drum Memory: I remember reading a story about a "Real Programmer" who worked on a machine with rotating drum memory. He'd memorised the number of clock cycles required for each instruction to process and also how long (in clock cycles) it took for the drum to rotate, so he could order the compiler to place the instructions on the drum with just enough separation so that when one instruction finished, the next instruction was just about to pass under the read head and so could be read immediately. His programs always ran faster than the computer's "optimising" compiler. Except when the place he worked obtained a printer which required a tiny delay between sending each character. Instead of programming a loop, this Real Programmer simply reversed his logic from before when writing the print routine so that when each instruction of the print routine had finished, the next instruction had just passed the read heads and so the drum needed to turn a full rotation before it could pick up the next instruction for processing. Repeated for several instructions, this produced sufficient delay for the printer without requiring the memory space for an explicitly coded delay loop. Then when he was finally fired from his job (I forget the reason, but I remember it sounded dodgy), he decided to "sabotage" an important (accounting, I think) program. Not in a "won't work" way but in an "it's annoying" way. Whenever the program was printing out its results, it would occasionally insert a random asterisk into the printout, breaking the neatness of the output. The original writer of the tale was asked to find where the asterisk was coming from and get rid of it, so he began to trace the path of the program. This persistent soul traced the program... all the way to the end of the drum. The last instruction on the drum was very obviously not the end of the program, so he was very puzzled about this for some time. Then he realised... When the program counter went past the end of the drum, where would it go? Where else? He went to the start of the drum and sure enough, the next instruction in the program was there. His admiration for the cunning of the now-absent Real Programmer caused him to give up at that point and tell his bosses that he couldn't find where in the program the asterisk was coming from. :)

[@melkiorwiseman5234]

Rotating "drum" memory would be possible to implement for the vacuum tube computer. You could even use a loop of recording tape instead (although the splice might be a problem, depending on how the join was made). There are two basic ways of recording data to magnetic media. The first way, which was used in the old TRS-80 computer, is to first record a pulse and then either record a second pulse for a 1 or no second pulse for a 0, so each bit takes up the same amount of time, but there are two pulses within that time for a 1 and only one pulse for a 0. The second method is similar but generally uses a Phase Locked Loop circuit to detect a particular frequency. Recording that frequency on the tape gives a 1 while any other frequency, or no signal at all, gives a 0. The higher the frequency you choose, the faster the bits can be recorded on the tape. The second method is essentially identical to the first if the frequency used for one bit is double the frequency used for the other bit. I still have an old kit-built computer based on the Z80 CPU which used the frequency shift keying method. Nominally, it recorded to cassette tape at 300bps, but by "poking" values into two memory locations, you could double that to 600bps or even quadruple it to 1200bps, albeit at the risk of losing what you'd saved if there was a tape "dropout" since faster speeds meant less reliable reading of the data later. In practice, I used good quality tapes and had very few problems.

[@absalomdraconis]

@@melkiorwiseman5234 : Hmm, if the splice was distinctive (e.g. conductive on a nonconductive tape) then it could function as a sync signal. Also, if you're willing to use a PLL to control the tape's motor speed, then even with tape stretch you can fairly reliably get multiple bits at a time by dividing the tape's frequency response into chunks, albeit by sacrificing one of those chunks to the tracking frequency.

[@melkiorwiseman5234]

@@absalomdraconis I'm thinking of a recording scheme which is as simple as possible. Also, it would be possible to get rid of the "dropout" of the splice by making it slanted so that the read-write head always has a recording surface under it. While recording multiple bits in a single transition is possible, it becomes progressively less reliable as you increase the number of bits stored. I can think of a scheme which uses (for example) 16 frequencies instead of 2 so that you can record 4 bits instead of 1. I can even imagine using 256 frequencies to record 8 bits at once, but with far reduced reliability.

[@brianclimbs1509]

Make a Williams tube style memory with a deflected rastering laser beam and a bunch of photocells! Okay, maybe it wouldn't be a good fit for the vacuum tube computer, but still pretty neat!

[@UsagiElectric]

I actually thought about that too at one point! It would be very cool and might actually end up being a more reliable way to do memory than by detecting tiny electrostatic charges. It's definitely a good idea!

[@otherunicorn]

The gyro effect of the drum is significant. Mine was powered by a 1/6 horsepower motor. I now use that motor to drive a small milling machine!

[@UsagiElectric]

I never thought about the gyro effect of drum memory, but you're absolutely right! That's a lot of metal spinning very quickly, it's certainly going to exhibit gyroscopic precession. Also, a 1/6 HP motor is no joke, that's a lot of power for some computer memory, haha.

[@otherunicorn]

@@UsagiElectric You should see all the heavy machinery in one of those large tape drives!

[@jeromethiel4323]

I think it's worth noting that early computers were held back mainly by a lack of memory. Memory in non-semiconductor devices tends to be large, slow, and energy intensive. Heck, even early hard drives were massive compared to what we have today, and even more expensive and energy hungry. We are spoiled today, due to extremely cheap memory that is relatively fast, and not too energy hungry. Could you imagine what Turing could have done with even 80's era memory systems? The mind boggles!

[@donmoore7785]

I had a 1024 bit storage tube from my dad that I finally sold on Ebay - it was not Williams. It did not have visible phosphor. It had a fine mesh. It could be used for RADAR (which dad was involved in). It was going to be used in a computer in the late 1950's but I don't believe that computer was ever built - as technology changed quickly. I worked on the E2C RADAR for GE in the 1980s, and my colleagues said that it had magnetic drum memory in the 1960s.

[@okb6436]

Interesting video and information! Finally someone that doesn't give up at static ram...

[@UsagiElectric]

Thank you! Ultimately, it will be static RAM, but it won't be non-volatile. Essentially, I'm just building flip flops to act as memory, but I managed to shoehorn a VFD in the middle of all of it so we can actually see exactly what's being stored in memory!

[@nynexman4464]

I wonder if a tape loop could be used as a memory. It should be similar to a delay line memory, though maybe a bit slower. Would be an interesting experiment.

[@UsagiElectric]

That's actually a really interesting idea, and something I've thought about! The hardest part is keying the tape loop so that it never gets out of sync. Ultimately, I plan to use magnetic audio tape for ROM (programs), but to use it for RAM would require some really well engineered syncing hardware. Still, it's a very cool idea!

[@absalomdraconis]

@@UsagiElectric : For tape loop sync, you just pick some sufficiently distinctive signal, such as some particular combination of distinct tones for some minimum time. Tape stretch can raise concerns about frequency drift, but if you're willing to build your own mechanism then that can be fixed by using the simplist PLL that you can build to control the motor speed.

[@j616s]

Hello. If you're still interested in williams tube memory, I used to volunteer on the replica of the Manchester SSEM thats in the Science and Industry Museum in manchester. I have circuit diagrams I might be able to send over. It can be pretty reliable once tuned. But the biggest issue we had was the tubes we used had a carbon anode coting which would flake off in miniscule amounts over time. That would cause dead spots on the screen which wouldn't support a charge well. This, and just general drift in calibration, meant the tube memory would work fine for months then randomly be unreliable for months while we messed with settings. The original team in the 40s/50s commissioned tubes with metalic anode coatings to avoid this. Also worth noting that they initially used 12" tubes, moved on to 6" after initial experimentation, and then I think the IBM implementation used even smaller tubes. I gather the main reason for that choice was purely what was available and that the 6" tubes were better quality build-wise than the 12" tubes. That said, having fairly large tubes left spare space to offset the image to avoid the dots I mentioned above. Also, another weird quirk of note. The pickup plate doesn't actually need to line up with the image. Oddly, it works with the plate on a different part of the screen...

[@j616s]

FWIW, we'd always refer to it as Williams-Kilburn tube memory. It was co-developed and an unfortunate quirk of history that only one of them ended up with their name against the invention.

[@georgegonzalez2476]

Way back in 1963 my father brought home a drum that looks exactly like the picture of the IBM 650 drum! Almost 2 feet across and about 150 fixed head read-write heads that you could screw in and out to adjust the gap. Must have weighed at least 70 pounds. Hard to imagine that spinning at 12,500 RPM. The only other thing I remember about it is that the heads had like six wires coming out to sort-of a 6-pin dip-like plug. And the wires were very slippery, could they have been using Teflon in those early years? Sadly I don't remember what happened to the drum.

[@UsagiElectric]

That's awesome! I would love to see and play with one of those old drum memories in person someday. They were certainly not a small thing though, there was some proper heft to them. Even the smallest drums were still massive and heavy things. I don't know what kind of wiring they were using on those, most of the wiring I see in vintage equipment from the 60s is either the standard plastic type coated stuff or uses a fabric like material.

[@GodmanchesterGoblin]

I would love to see a ground up core memory build, but I also know that is far from simple. You often need a few hundred mA of drive current for the X and Y lines, and capturing the sense-line output (a few hundred mV on a good day) is really hard, never mind the complexities of writing back data after a read. it reminds me of the early DRAM chips in the 1970's such as the 1103 (1k x 1 bit with weird supply voltages) - they also had destructive readout. I am looking forward to see how this develops further. I am guessing conventional double triodes for latches, or something clever with pentodes or heptodes? Off topic, but a video on Bubble Memory technology from the 1970s and 80s might be interesting - it wasn't around for long as commercial product as far as I can recall, since it was rapidly superseded by Flash memory, but like core memory, it had used magnetic structures and had non-volatility as a key attribute.

[@UsagiElectric]

Oooh, yeah, I didn't get far enough into my core memory research to start thinking about X-Y drive current, but you're right, that adds one more layer of complexity to it! The only way I've been able to kind of mentally approach core memory is to treat it like a computer all on its own. It receives an address from the CPU, then it starts its own sequencer to step through several different steps for retrieving the requested bit and offering it back to the CPU. It'll definitely take a bit of effort to figure out a decent way to handle it. Your thinking ios actually really close to how the memory is going to built on the tube computer though! Although, I'll be using 6AU6s instead of dual triodes. As it sits, one bit of memory will use four 6AU6s and one VFD, including read/write and selection control. Not exactly compact, but I think it'll look cool enough to make up for it!

[@GodmanchesterGoblin]

@@UsagiElectric Yeah... people look at core and only see the matrix. As you say, it requires a lot of hardware around it. Even with TTL in the 70s, the support hardware was always much larger than the basic core memory array and it took a lot of power, but the non-volatile aspect was always impressive. I will watch with interest to see how the tube based memory comes together.

[@taichitortoise4859]

The replica of the Manchester SSEM at the Manchester [UK] Science Museum has a functioning Williams/Kilburn tube memory. As others have noted it is 'finnicky' as the secondary electron emission is measurable but minute. IBM licenses the technology and some 701 and 702's shipped with this memory type.

[@Blacklab412294]

While I was drawing out my 16 bit Karnaugh Maps on Saturday, I was thinking about your Low Voltage use for your Vacuum Tube Computer. I was thinking about the complaints about ENIAC Computer. One of the complaints about ENIAC was, that it ate Vacuum Tubes like crazy (burnt out the filaments). And if I remember right it used High Voltage for it's tube. Q: It makes me wonder because you are using Low Voltage that means your Tubes are going to last longer? And the other Question I had, sense you are running your filaments, like Christmas Tree lights, in series: How easy is it for you to track down a Burnt Out Filament with all the Vacuum Tubes you are using ?????? When I played with 24 Volt Zone Valves in a sprinkler system, I used fuses for protection from shorts in series of each Zone Valve. One of the things I did was use a resistor in series with an LED that parallel the fuse to indicate a blown Fuse. Now I know you don't want to use a LED in your system, but what could you use as an indicator ? I know they make 24V filament Lights, but you would have to have a resistor in series with it that would not let the lamp glow unless the Tube Filament was burnt out. And if you decide not to go with a light indicator, have you thought of putting two of your vacuum tube peg cups next to your tubes (in a line) that would let you quickly measure the voltage across each filament [AKA a test point(s)] so you would not have to take a tube out to check each time something goes wrong???

[@agranero6]

Core memories do not use iron, they use ferrite and core rope memories used Permalloy. With soft iron it would be very difficult demagnetize the core without using high currents. You can still find ferrite cores for those memories that are less than a millimeter wide on e-bay. Incredibly as core memories survived until 70's and some controllers used integrated analog circuits: (I found I saw a circuit of one about 40 years ago that used 709s).

[@ibm3480]

You gotta make a card punch and reader for that computer! It can't be a tube computer without cards. LOL

[@GodmanchesterGoblin]

That would be fun... I worked on card readers in the 70s. Reading one card is easy. But reading the next hundred over about ten to twenty seconds without any feed errors or data errors is quite a challenge!

[@UsagiElectric]

I actually thought quite heavily about how best to read programs on the tube computer, and both card readers and paper tape readers were really high up in the running of ideas! Ultimately though, I think I'm going to end up with magnetic audio tape. It offers quite a bit more flexibility in how I get programs from the PC to the tube computer. Plus, I think it'll end up looking stellar. This is my first rough idea of how the Program Control board will ultimately look: i.postimg.cc/ZKQV9wBG/Reel-to-reel.png

[@ibm3480]

@@UsagiElectric Very cool!!!!! Great job 👍

[@metschnikowia]

Hi! Amazing channel, congrats! Could you please tell me where can I find the tube connectors you use? Thanks!!!

[@UsagiElectric]

Thank you! The tube sockets I use on my PCBs are little 1mm PCB pin headers from Harwin, part # H3161. I go into a bit more detail about them in the video I did on my PCB production process here: kzbin.info/www/bejne/d3ObZZKsl62Sm5o

[@paulawillaminachandler-ren3725]

This was really informative. I had no idea memory had such a complex history. [I did do a little more reading on it actually] Now I understand why they were so expensive. Very good information here. Thanks.

[@UsagiElectric]

Thank you very much! In everything I've read about old tube computers, memory definitely seems to be the most difficult and expensive aspect of the entire thing!

[@paulawillaminachandler-ren3725]

@@UsagiElectric You know, you could go with relays for a simple SR latch memory module? But I bet you'll choose dual triode flip-flops, in order keep the build period correct.

[@UsagiElectric]

@@paulawillaminachandler-ren3725 I had thought about relays! And I actually really like the idea of relays because comparatively, I can make it so much more compact. But I really wanted to stick with tubes for as much of the build as I could this time. And I was actually planning to use dual triode flip flops for the memory! But, I started to feel guilty because I love the 6AU6 tube so much, haha. From a logic standpoint, I'm not really doing anything super exciting, it's essentially just D Flip Flops. But the way I implement it with the 6AU6 is pretty interesting I think!

[@paulawillaminachandler-ren3725]

@@UsagiElectric Whatever you do. I am sure it would be fantastic. But would it be too much to ask, that in the late future, to add a relay memory expansion, well not expansion but... Switchover?

[@WiztotheIzzard]

Have you considered flip-dot displays as RAM? Write with a signal / reset, read with a photodiode / magnetic reed switch.

[@Colaholiker]

When you explained the hysteresis (something I was familiar with before), I suddenly realized where the symbol for the Schmitt Trigger (something I was familiar with before) comes from. Why did my "head-sponge" never connect these two dots? 🤦‍♂ This is the moment when you watch a video, learn something and at the same time feel sooooo stupid.😅

[@pantherplatform]

Nice to live in a time with the best of everything that's ever been and grow up thru all the advancements in technology like the transistor. Can you *please* make a fully functional smart phone out of tubes and core memory and use it to watch yootoob in the highest resolution?

[@kenromaine2387]

Regarding your ref. to core & Centurion in this video. True the Warrex Computer started doing 4K core to 32K DRAM upgrades The accounting minicomputers from Fetter Computer (very early 1970s) used 4K (byte) core memory being 2 slots wide. The Fetter Computer had 3 boards making up the 8 bit TTL logic based CPU, 4K or 8K of core, Friden Flexowriter for terminal I/O & paper tape I/O also had (3)Sykes cassette tape drives. Because Warrex started upgrade others computers Warrex's first products are accounting software and hardware upgrades for late 1960s - early 1970s systems base on a CD200 mini computer like: Warrex - Centurion P/N 1001 & 1002 Disk Cont. cards for the CDC 9427H 10-MB disk drive (no more paper tape or cassette). Warrex - Centurion P/N 1003 32K byte DRAM memory to replace 4K core. Now max mem. 16K with core or 60K DRAM. Warrex - Centurion P/N 1004 4-Port Serial MUX for ADDS terminals, no more Friden Flexowriter for operator I/O. Warrex - Centurion P/N 1005 8-bit parallel interface to the Centronic 101 dot matrix printer again no more Friden Flexowriter. By the mid 1970s it seemed like we had upgrade all the CD200 mini computers we could find. Warrex had written a multi user disk operating system called @OSN (no more single user cassette tape based op-system TOS) and written a high level compiler called CPL & had a JCL based on the IBM 360 system Job Control Language. At this point Warrex changed its name to Centurion Computer and started shipping it own main CPU cards for the CD200 backplane - memory mapped I/O bus computers. The first 100% complete Centurion Computer was called the CPU4 (with CPU1,CPU2, CPU3 and DMA card) and the other cards listed above to make up a complete minicomputer system (and the rest is long lost history). The Centurion Computer you have is the CPU6 which is the 3rd generation of four generations. CD200 bus models CPU-4 (4 cards TTL) 1975, CPU-5 (2 cards TTL & AMD-2901) 1979, CPU-6 1980 (1 card TTL & AMD-2901) and the last model CPU-7 (1 card TTL & AMD-2901 but now Multibus not CD200). All the CPU models had the same op-codes so they could share all the Centurion software family for the past 10 plus years. Good luck with your tube system. When ready lets get the CPU6 at least booting from my Diag. PROM test card.

[@UsagiElectric]

Hi Ken! Thank you for yet another brilliant Centurion info dump! We're going in for a final update on family stuff this week, so I think we're just about ready to start making plans again. I'll shoot you an email mid-next week once we have a more solid gameplan. I'm ready to get this Centurion up and going for sure! And, I'm ready to pick your brain and start building a Centurion timeline. While a bit ambitious, I'd love to do a short documentary on the history of Centurion. It was fascinating company with some really amazing stories that would be a shame to be lost to history. Talk to you soon!

[@sixspeeddeath]

The fact that this video isn't called "Memory Lane" is a tragedy. 😂😂

[@UsagiElectric]

Genuinely laughed out loud at this one!

[@weazeldark3983]

u could send a signal in a loop and pick it back out turn data into a multiplexed changing frequency/modulation but tv crts are the best n most common just read a bit from glowing phosphor though energy to make glow

[@KallePihlajasaari]

Acoustic delay line memory might be practical for cache if you read a paper tape for the program.

[@UsagiElectric]

That's a good idea! For program storage / ROM though, I think I'm going to end up reading and executing directly from magnetic tape. It's not the most ideal solution, but it allows programs of any length to be executed, which when combined with the rather limited RAM and data width, should give the programmer some much needed flexibility. With some work, I should be able to get something like 50 instructions per second, which isn't blazing fast, but still quite quick!

[@wa4kdc]

Just FYI: The original navigational computer used on the C-130 aircraft used used magnetic drum memory. It was located beneath the bathroom which from a view towards maintaining it was a rather unfortunate choice.

[@jeromethiel4323]

Williams tube memory IIRC was very fast compared to magnetic drum, magnetic core, or delay line memory. But it was crazy expensive, which is why i think only the feds used it for some of the secret no budget projects like designing nuclear weapons and the like. But i could be wrong, i'm referencing information i read a looong time ago. So i could be completely wrong.

[@therealchayd]

Regarding drum memory, how about using a loop of audio tape? (at least as a proof of concept), you can run it at inches per second for slow reads and writes, you have multiple tracks for multiple bits and timing signals and the parts are widely available.

[@DMahalko]

* As system memory it must be accessed very quickly. The CPU can only run as fast as the memory can be sequentially accessed, over and over. * Audio tape is physically dragged across the head with a pressure pad behind the tape. * Drum memory needs a flying head that doesn't touch the media, to protect the surface from rapid wear-out.

[@winstonsmith478]

Flexible PCB with SMD caps? Regenerative capacitor memory The first regenerative capacitor memory built was the rotating capacitor drum memory of the Atanasoff-Berry Computer (1942). Each of its two drums stored thirty 50-bit binary numbers (1500 bits each), rotated at 60 rpm and was regenerated every rotation (1 Hz refresh rate).

[@UsagiElectric]

I actually did a bit of reading into using capacitors as memory. With a properly sized capacitor, the refresh aspect of it can be eliminated as well. I really like the way the TIM relay computer handled capacitor memory and thought that would work quite well for this as well ( www.northdownfarm.co.uk/rory/tim/tim-8.htm ). Ultimately though, I decided to stick with just tubes as it simplified a lot of the supporting circuitry required. Someday though, I think capacitor memory could definitely be a viable path forward for some projects!

[@Blacklab412294]

Question: How Big are you thinking of making your memory ? I sat down last Saturday and drew up a 16 bit Karnaugh Map. 16 Bits is NOT that big of a memory (not saying that is what you were planning). And it make me glad you went with the ALU for your computer. But really, 32 Bits is too small for any real work, and 64 Bits of just RAM looks plausible. And what is your plans for ROM ? I was thinking you could use a diode switch matrix for it. Again you want something along the lines of 64 Bits or more to store both Boot routine, and any Subroutine. I was also thinking you could make something along the lines of the 74154 (tube version) to decode your address lines, and if used right could be used to drive both your RAM and ROM with some extra 'Logic Glue'. Also once invented, you don't have to reinvent that wheel. Also the 74LS193 (Tube Version) could be be used as part of your program counter. What is great about that it's both an UP and Down Counter and it's also SYNCHRONOUS and it does use SR flip flops. Also I was thinking it could be cool if you could use 5-bit paper tape to load your programs for the input part of I/O. 4 Bits (1-Nibble) for program and data, and the 5th bit for start and stop reading the tape. Not sure other then your displays for your output. IF you were to expand the bits for the paper tape to 6, one set of bits could be used as your clock.

[@absalomdraconis]

If someone is doing paper tape, then they honestly want at least 7 bits so they can use Hamming codes (they were invented because Hamming was tired of paper-tape errors!), and preferable an 8th "return to one" bit to ensure that every row has at least 1 bit open for a primitive servo to keep things aligned correctly.

[@LeonirZimmermann]

Great analysis! Would it be possible perhaps a simple circuit based on neom bubls? Like HP AC-4G Decade Counter?

[@UsagiElectric]

Thank you! Neon memory is actually a really excellent idea, and something I considered! You essentially just put a resistor on either side of a neon, write impulse go into the anode and the bit is read from the cathode. Neons are awesome because they're bi-directional and exhibit hysteresis, but unfortunately, with just +24V and -12V on tap, I just don't quite have enough voltage to use neons. Still a very excellent idea!

[@LeonirZimmermann]

@@UsagiElectric Could you consider a driver module? (step-up to write, step-down to read)... A computer Motherboard usually works with different voltage levels...

[@LeonirZimmermann]

@@UsagiElectric... But... Anyway you need to use registers on 24 volt... 🤔Now I see.

[@hanksmith3628]

Not that this will help in anyway, but when I was at Georgia Tech, I knew one of the inventors of drum memory, Dr Barnwell. Yeah, I'm that old.

[@patrickradcliffe3837]

Drum memory help out here. Brushless motors can spin in the neighborhood of 12k rpm at low voltages.

[@VonZeeple]

I have one of this cathode Ray tube at my parents house! I first thought it was a scope before looking at the manual. I found it in thé trash un the lab during my Ph.D.

[@UsagiElectric]

Awesome! Hold on to it, proper Williams Tube CRTs are getting quite rare these days!

[@loukashareangas4420]

Wha, no mention of the Selectron tubes used for the Johnniac computer?!? They were custom built vacuum tubes and they were the computer's memory. If this isn't relevant I don't know what is

[@UsagiElectric]

The Selectron tubes are brilliant and amazing pieces of memory technology! However, the purpose of this video was to talk about some types of vintage memory that I considered using on my tube computer. The Selectron tube isn't exactly an easy thing to get my hands on. They're incredibly rare, and when one does show up, it's usually more than $10,000. A bit out of my price range, haha.

[@loukashareangas4420]

@@UsagiElectric Yeah, of course it was meant in joking good spirit! More like, spread the knowledge about them, although they were a one of a kind niche device, while your other examples were more "widely" adopted. Although if you were ever to come across one, do make videos about it!

[@whyindeed9937]

Can I ask why you don't chooch up the voltage on the tubes? If you need a better power supply, I think I can help with that.

[@DAVIDGREGORYKERR]

What about a program rope using magnetic cores like what was used to hold the program for the computer of the APOLLO XI Space Craft.

[@zinckensteel]

What, no acoustic delay line memory?

[@UsagiElectric]

I did actually consider delay line memory, but I actually shied away from it for a couple of reasons. It's not very visual. Sure the physical line itself looks cool, but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). Delay line is super cool and definitely seems like a solid alternative, but for the tube computer, I'm going to sacrifice some bit density for some blinkenlights!

[@zinckensteel]

@@UsagiElectric I was being facetious; I realize all the same issues with the complexity of interface apply. An interesting visually pleasing method might involve a moving sheet of semi-long persistence phosphor, allowing an array of glowing dots to hold the data, with photocells and tiny UV bulbs set up to reinforce the pattern unless new data is input. For some materials, IR can be used to "erase" the glow rapidly, iirc.

[@weazeldark3983]

and don't worry about voltage just need a valve voltage translation or resistor divider or use optocopling

[@VandalIO]

have you thought of using a tape? or punch card as a memory ?

[@agranero6]

What about the cost? I don't know there, but here in Brazil tubes (we call them valves: "válvulas") are hugely expensive. Well I didn't found a 6AU6 but found a Miniwatt EF94 that is equivalent, it costs about US$ 7 (new old stock) which is cheap, but an ALU would used dozens and dozens of them, RF and audio tubes may cost about US$ 200 or 300.

[@TMS5100]

How about mercury delay lines?

[@jj74qformerlyjailbreak3]

Sometime just talking is better than building stuff. Great discussion

[@pe1dnn]

About magnetic drum memory, look at this kzbin.info/www/bejne/oqazfmmOlrCNq7c Really hilarious is at 1:44 where they tried a "mobile" version of the Fastrand. Be also sure to read the comment on this video. Different times...

[@UsagiElectric]

That was hilarious and awesome! I never thought about gyroscopic precession being a problem, but using a steel drum that large, yeah, I can totally see how that would be an issue!

[@der.Schtefan]

Ultrasonic delay line? Compact modules from the 80s and 90s of PAL TVs are cheap to get.

[@nucleoncode]

Bubble memory would be nice

[@UsagiElectric]

Bubble memory is very cool too! Unfortunately, it introduces a lot of the same hurdles as core memory, and the control circuitry I fear would end up being much larger than I have physical space for.

[@andypughtube]

You might want to look at acoustic delay line memory, it seems quite simple, and if you haven't seen this I think you will like it: kzbin.info/www/bejne/aHOsqWWuYrZlnKc

[@DonnyHooterHoot]

I'm fairly new can I get a personal "Hey,hey?

[@UsagiElectric]

Hey, hey Donald! What's up?

[@edgeeffect]

I saw a drum memory once.... I'm sorry, that's the best I can do........ it was very loud!

[@UsagiElectric]

That's better than me! I haven't seen one in person yet. Someday though!

[@edgeeffect]

@@UsagiElectric we had a tour of a local mainframe when I was at school in the early 80s... it was kinda deafening in there anyway and then they lifted the lid on the drum and it was so very very loud.

[@john280z]

At the SAGE Radar Sites a rotating drum memory was used in the AN/FST-2. Never saw a failure with them the two years I worked on one. en.wikipedia.org/wiki/Burroughs_AN/FST-2_Coordinate_Data_Transmitting_Set

[@UsagiElectric]

That's awesome to you got to work on those! I imagine that once the mechanical engineering aspects are figured out, it should be a fairly reliable form of memory, but getting that drum to spin at 12,000+ RPM requires some proper engineering!

[@john280z]

@@UsagiElectric Look what just popped up: kzbin.info/www/bejne/m3rToHRtftdsppo

[@stephenforrester6776]

I know you may have already considered this, but diode matrix rom can run at low voltages and its non-volatile. Of course it's also not re-programmable because the instructions are physically soldered onto the board, but it could be a good temporary solution for now just as a proof of concept. I built a simple breadboard rom using them and a raspberry pi as a controller.

[@electronicarchaeology]

There is another memory type that came to mind while watching your video is the (Torsion wire delay line memory), the only reason I remembered it, I had an early Nixie tube calculator in my possession when I as younger in the 70s it had this type of memory at the bottom, in a metal can, probably not a lot of storage capability but it didn't look all that complicated, sadly I did the criminal thing of dismantling the whole unit, but I learned something in the process.

[@UsagiElectric]

I did actually consider delay line memory, but I actually shied away from it for a couple of reasons. It's not very visual. Sure the physical line itself looks cool, but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). Delay line is super cool and definitely seems like a solid alternative, but for the tube computer, I'm going to sacrifice some bit density for some blinkenlights!

[@electronicarchaeology]

​@@UsagiElectric I totally agree, the vacuum tube memory will look absolutely awesome, I'm really looking forward to your videos of the epic build.

[@RobertResearchRadios]

I think this is now of the best esoteric computer memory description videos on youtube. Great work!

[@UsagiElectric]

Thank you so much!

[@andrewmawson6897]

Williams worked extensively in the late 1940's at Manchester University, and Ferranti was heavily involved with his computing work being a Manchester based company. Many interesting devices were developed. One I was involved with was a magnetostricktive device used as memory for early CRT displays. A coil of magnetic wire had a transducer at one end that introduced a magnetic pulse that travelled down the wire and was detected at the other end , and through some simple electronics was reintroduced at the beginning of the wire. In this way magnetic pulses could circulate in the (coiled up) wire and by encoding 6 bit characters a full display screen could be built up and modified as required. They were a bit temperature sensitive though! (We also use drum memory (Sperry J101/102) and Bryant Drums which were later replaced by Burroughs fixed head disks of a staggering 2Mbyte!) I suspect that you will find Williams work heavily represented in the Ferranti Archive) We used to love core memory, as you could load test programs at the office and take them to site still intact :)

[@m.e.8273]

Hey! A couple of years ago I have woven a set of tiny core memory modules. One was 6 by 25 and the other 8 by 8. I can remeber they took me hours of painstaking work and dozen of retries to make them without breaking the tiny wires and to get them to look decent and sturdy. I eventually lost interest in the project and both ended up in my display cabinet of failures... Would you maybe like to have one to experiment with? I'd be happy to send you one!

[@UsagiElectric]

Thank you so much for the offer! At the moment, my experimentation with core memory is still quite a long ways away, but in the future, who knows, I may take you up on the offer!

[@chuckinwyoming8526]

Just a few others: * punched paper tape program (you teased a rotating punched drum) * punched card reader (don't forget the card sorters, once a very big usage) * rotating capacitor drum (modern SMT caps could make this an interesting option) * delay line (I still have some L-C delays in a drawer here) * magnetic tape * magnetic strip * latching relay (with latch and clear control) * acoustic (really a form of delay line) * phase change

[@UsagiElectric]

Those are all excellent suggestions! * For paper tape or punched card stuff, I should have mentioned in the video, but at the moment, I'm only working RAM. For program storage / ROM, I'll be getting into that when I get into program control. * The rotating capacitor drum is something I actually thought about, but I just couldn't come up with a good way to keep it simple enough to prevent feature creep. * For delay line memory, I did actually think about it too. Unfortunately, it's not very visual. Sure the physical line itself looks cool, but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). * For magnetic memory, we've actually talked about this a bit over on the Discord server too. Trying to think of ways to potentially use magnetic tape as RAM, but ultimately it felt like my design skills wouldn't have been up to the task. But, we will get into magnetic tape in the future! * For relay memory, that actually makes the most sense honestly. Modern mini relays are really tiny and can be had in whichever voltage you need, and with some simple VFDs, it can be very visual too! But, and this is purely a personal thing, it felt a bit like cheating, haha. I really wanted to see what I could do with just tubes! * Phase change is an interesting one, I hadn't heard of that one before! It definitely is way, way beyond my skill set, but it's fascinating nonetheless!

[@BryceSchroeder]

Rotating drums could go much slower, especially as your clock speed ambitions for the 1-bit processor are relatively modest. After all, common cassette tape only travels at a few cm per second and can be used to store data. You'd presumably want to use some kind of FM encoding for a relatively slowly rotating drum.

[@KeritechElectronics]

You've gotten yourself into a project that's as awesome as complex. It's gonna take a lot of thought to engineer something sensible within the space and voltage constraints. Fingers crossed!

[@UsagiElectric]

Thank you! And ultimately, it wasn't exactly thought and engineering ability, it was just about making enough compromises to make it fit, haha.

[@akkudakkupl]

Sewing core memory together would be infinitely painful. There is a reason why when DRAM came core memory got ditched very fast.

[@UsagiElectric]

Oh I can imagine! Using core memory on a project today would very much so be a labor of love!

[@akkudakkupl]

@@UsagiElectric I'd say it would be more the labor of a maniac :D I watched some videos about it, basicaly without a rig to hold the cores at 45 dergee in perfectly straight lines you can forget about building anything fairly dense. Saving grace for you is that you only need one plane because you have only 1 bit word width (you could say your CPU is a serial computer) :D There is a video how a Polish version of ECL 1904 is built (Elwro Odra 1304), you can see in it how the women sewn core memory - it had to take a lot of patience.

[@akkudakkupl]

​@@UsagiElectric sorry it was not 1304 but one of the RIAD series (R-32, an IBM 360 clone), here is the excerpt with core memory production, fascinating stuff. kzbin.info/www/bejne/mYrZl6yHfMetfJo EDIT: Lector said that sewing one plane takes around 160 hours of labor.

[@krallja]

Wikipedia says Williams tubes didn’t need phosphors - the charge well effect you described actually works with just the glass on the front of the tube “The presence or absence of this coating had no effect on the operation of the tube, and was of no importance to the operators, since the face of the tube was covered by the pickup plate”

[@UsagiElectric]

Interesting! I totally didn't catch that part on the wiki article, haha. It makes sense though, the goal is to have the electrons generate secondary emission, which seems like a phenomenon independent of the phosphor. I suppose the reason many of them still had phosphors is that they were designed around existing CRTs, often for scopes or radars. That makes me wonder now if there are any CRTs hiding out that don't have any phosphor on them, that would be wild to see!

[@algorithminc.8850]

Been enjoying your tube-computer series --- I sent a link to you earlier today of some source code for a calculator (Busicom 141-PF) using an Intel 4004. The comment seems to have disappeared. I found someone discussing one of the algorithms in that code --- the video "Take Any Square Root by Hand - Easy to Learn!" on the channel " Tetration" I figured the idea of a simple problem like that would give you an idea of something cool to do on your machine, as you build it out. The 4004 was a 4-bit processor ... but there are many cool problems to solve with a 1-bit too. Other thoughts are to look at two or four bits (8-bit, yikes!) -- Still trying to reach some of the old game-developer associates (Jarocki) about Toru Iwatani --- Springtime here in central Florida ... Usagi everywhere ... Christopher

[@UsagiElectric]

Thank you! Yeah, KZbin is real finicky with links in comments sometimes. I suppose that makes sense because some of the larger channels get spam like crazy. I actually have a few ultimate goals that I'd like to do with the tube computer! Square roots might be pushing it (or at least my ability to program it, haha), but I think these could be quite cool: - First goal is to be able to calculate the Fibonacci sequence. Ultimately, the I/O board will be able to output an 8-bit value, so I would love to see the Fibonacci sequence printed up on a teletype. - The second big goal is to play the Game of Life in memory. I haven't shown it just yet, but each memory cell is going to have a VFD on it. And with how it'll be arranged, you'll actually be able to see a grid of memory. I would like to build a program loop that takes a 4 x 4 block of memory and plays Conway's Game of Life in it. - The kind of ultimate test of the machine though will be programming in a text adventure game. The Program Control will be able to halt the tape as a pause loop, and then on the Input/Output section there will be 8 switches to input a value and a button to resume tape reading. So, the computer can punch out something on the Teletype, and then based on the result the user punches in, continue on different paths conditionally. The program is going to have to be massive, but I think it could be quite interesting! - Also, while not entirely computer related, I really want to build a DIY tube synthesizer this year and then build a DAC that will take the 8-bit output of the computer and play it on the synthesizer. I think some really interesting stuff could be done with that as well!

[@bluerizlagirl]

All Seeburg jukeboxes from 1955 onwards contained something called the TORMAT memory unit, which was basically core memory with 1 bit per side of each record. Making a selection passed half the write current through all cores with the letter and half the write current through all cores with the number. This stored a 1 in the core with the letter and number, leaving everything else alone. (If there were no credits, the write supply was disconnected, so you could not get a free selection!) Readout was performed by passing full write current through just one core at a time at the carriage position to store a 0. A single sense wire through all the cores would pick up a pulse if that core had been storing a 1, and this would be amplified to energise the trip solenoid and start the play cycle.

[@ropersonline]

9:28: I wonder if there ever was any drum memory that also moved the read/write heads along the axle, for greater storage density (at the cost of access speed). The reason I'm wondering about this now is because I just learned about comb printers (see Wikipedia), and I'm thinking a similar principle ought to allow a great increase in drum memory capacity. But I'm not sure if something like that was ever actually implemented. Does anyone know?

[@AJB2K3]

Rotating drum looks like an evolution of the Wax cylinder record player.

[@lindoran]

I learned so much thanks for doing this video! I like the idea of making static ram cells using logic I can't wait to see where this goes :)

[@UsagiElectric]

Thank you! In the next video in the series (which I think will be the week after next), we'll look at a single bit of memory and see what I came up with. It's nothing too fancy, but it should look pretty cool when all in place!

[@bronka42]

Delay-line memory!!! 1 bit proessor compatible!

[@davidmcdonald3314]

I definitely think this is worth considering Dave Jones / EEVBlog did a tear-down, I think, of a HP (???) delay-line memory a while back You still have to have circuitry to re-introduce the stored data back into the wire - it's volatile - that's the "loop" in the name of the device

[@UsagiElectric]

I did actually consider delay line memory, but I actually shied away from it for a couple of reasons. It's not very visual. Sure the physical line itself looks cool, but you can't actually see what's being stored on it. I really want the memory to be as visual and interactive as possible on this particular build. Also, the circuitry involved in creating a bit-train that runs in a loop around the line and then being to pick out the specific bit I want was a bit beyond my capabilities at the moment (though that circuitry is ultimately very similar to Williams Tube control circuitry). Delay line is super cool and definitely seems like a solid alternative, but for the tube computer, I'm going to sacrifice some bit density for some blinkenlights!

[@davidmcdonald3314]

Visual. Hmmm That rules out magnetic memory types And so we come back to CRT memory which is too high a voltage How historically authentic does this need to be? Could you simulate tube memory with LED's each with a capacitor (to keep them lit longer)? It might end up being larger, but it would certainly be visible

[@frankowalker4662]

Build an 8 bit version ? You're a glutten for punishment. Ha ha.

[@UsagiElectric]

I've actually been playing around with different architecture ideas in Logisim for potential future computers. This 8-bit design is one I'm quite happy with: i.postimg.cc/Y9YywZpN/UE8K.png i.postimg.cc/dVVXT5Bv/UE8-K-Instr.png The design is heavily inspired by the AM2901 bit-slice ALU and the TD-4 homebrew 4-bit CPU. I think ti would be absolutely brilliant to build this out as a proper fully fledged computer and see just what it's capable of!

[@frankowalker4662]

@@UsagiElectric Cool. It does'nt look too big of a task, realy. Plus most of the PCB planning and layout has already been done for the other computer.

[@davidmcdonald3314]

Ummm. Would 300V be too high to consider for the flyback voltage on a Williams-Kilburn CRT memory? Still not child-safe, I know, but it might be just a very small thing that could be surrounded in Perspex. Google for "tiny Russian CRT" or "5lo38i". You'd have to miniaturise the sensors (or reduce the capacity) I guess. Anyway, just a thought

[@UsagiElectric]

Thank you for the idea! For this tube computer, I'm really trying to stay at just a maximum potential of 36 (+24V to -12V) at any point on the machine. In the future, I definitely want to build some stuff at proper voltages, but for this one, I quite like the idea of forcing myself to build within this quite wild limitation. It's actually really forced me to come up with weird solutions for some things that I never would have otherwise!

[@tekvax01]

(12:08) Speaking of centurion minicomputers; what happened to yours? I have been waiting to see it start up and boot. I take it, things are not going well, on that front.

[@UsagiElectric]

Things got put on a bit of hold for the Centurion. I actually got in touch with a gentleman, Ken, who used to be an employee at Centurion, and he has some much needed skills and equipment to really get the Centurion working correctly. However, some family stuff cropped up and we had to go into pretty hard lockdown for about 6 months just to be safe, so any projects that couldn't be completed by just me on the ranch by myself got put on hold. But, it's looking like we're coming out the other side and about ready to get to work on some of the bigger projects again! This spring I'm really hoping to not only get the Centurion at least booted, but also get started on the drives and maybe even run some original Centurion software. I also really want to get Ken's entire story since he was with Centurion from the very beginning and maybe make a mini-documentary on the company history and the computers they were building. So, plenty of Centurion stuff to come!

[@duality4y]

isn't any flipflop and latch on there technically memory ? (in response to you saying it doesn't have any memory :P )

[@UsagiElectric]

Haha, you are technically correct. The best kind of correct!

[@Pulverrostmannen]

would be really fun to see a more modern version of each of these types of memories :)

[@UsagiElectric]

Each of these memories totally fascinates me and I would love to build some version of them in the future! Core memory is probably the most useful, but I can't help but be drawn to Williams tube / Electrostatic memory. Someday I really want to take the old scope CRT I have laying around and do a Williams tube project with it!

[@Pulverrostmannen]

@@UsagiElectric making the scope tube memory would indeed be epic to see but it surely sound complex to pull off and you gonna need some HV for that project too

[@Kae6502]

Interesting look into some of the various options for data storage. Thank you Doctor Memory. And thank you for the bunnies. :)

[@UsagiElectric]

Thank you! The little cottontail babies we rescued here were absolute terrors, but so, so adorable!

[@idahobob]

Have you thought of using Neon Bulbs? www.massmind.org/techref/mem/neonlamp.htm They would need a much higher voltage supply to light (70-120V) but not too outrageous of a voltage. You could use neon bulbs as logic... www.introni.it/pdf/GE%20Glow%20Lamp%20Manual.pdf (See page 45) Maybe even your next computer could be all Neon Lamps!

[@UsagiElectric]

I did indeed think about neons! And I would love to use neons but for the tube computer, I'm trying to keep the highest voltage at just 24V (or 36V total from +24V to -12V). But, I have been getting quite obsessed with neons lately! They have some excellent properties and the hysteresis of them can be exploited in a ton of different ways. They're kind of like the 555 timer of the tube era! I've been giving some serious thought to building a full neon logic computer actually. They need to be combined with an LDR in order to get amplification, but some logic elements, like multivibrators, can be made without LDRs. So, a combination of all the different types of elements could be a ton of fun to build! Thank you for the link to the GE manual on Neons! Here's another manual that I've been reading through lately as well: www.tiffe.de/roehren/neon.pdf

[@idahobob]

@@UsagiElectric I used to have a lot of tube gear growing up, as mainly "hand me downs" because nobody wanted tube stuff anymore. One day, after being "tickled by B+ on an audio amp with 6V6GT's (350V).. I had enough and stopped working on any tube equipment and got ride of them. "No, I won't work on your TV, I don't work on mine" I would tell people. It blows me away you can get solid operation at 24V! Reminds me of my experments with u741 opamps, powered with two batteries (+1.5 and -1.5) and actually doing things. (They are mighty tweekie at well under half their miniumum voltages, and had horrable slew rates,, but I actually could read my pulse holding two screwdrivers, connected to the inputs!)

[@davefrancis4970]

I'm just glad you didn't go down the mercury delay line memory route. Handy for building radar systems, by the way. ;-)

[@davids5148]

I heard that sdcards are cheap this days!

[@UsagiElectric]

Hmm, convincing an SD card to talk to my ridiculous confusing tube computer might be a bit difficult 😅

[@stevenbliss989]

Never understood why "valves" were never made much smaller, ..more efficient and much lower voltage? ...what am I missing??? BTW ..I miss using 74xxx family stuff way back when I was designing in the 80's & 90's :(:(:(

[@davidlawton898]

Another memory (sic) was working on magnetic core memory from California Data Corp. They made 8K (and possibly 16K ?) core memory modules attached to a logic and timing board. As engineers on the ICL KeyEdit 50 (Key to Disk systems), we used to try and repair these memory modules sometimes on an extender board attached to the bus. They would be extremely difficult to repair, as the timing feedback loops were disturbed by the distance of the extender board, so while we could access the logic board outside the computer and scope the lines, the timing feedback would cause certain operations to fail as a process of measuring them on a scope. Extremely difficult things to fix and the whole mem module often had to be exchanged for a replacement. The magnetic core module itself could not be repaired in the field.

[@ghostlucian12]

Also dynamic memory, requires a capacitor and a tranzistor(triode), also a refresh controller, also a buffer, for the result and refresh

[@homeopathicfossil-fuels4789]

What about the EDSAC approach? Mercury Delay Lines, you dont even need to use mercury. There are other ways to get waves propagating at a delayed rate.

[@RaymondSwanson-u9y]

Rotating drum memory was the worlds first swap file.

[@robot797]

no vf-ram? to bad I was looking forward to that one

[@UsagiElectric]

Ultimately, I don't think vf-ram is going to work out, but I will actually talk a bit about it in the next episode!

[@robot797]

@@UsagiElectric yay also I do understand

[@nanolith]

If the point is just to get enough memory to store some basic data, you could set up relay memory with a series of rotary encoders -- powered by stepper motors perhaps -- to locate a single bit. I'll let you think on the details, as you're plenty creative, but it would not be difficult to build this. I don't think that you'd want to scale larger than maybe 64 or 128 bits, but that would be enough to add, subtract, and multiply by single stepping through your computer.

[@GHILLIESARCADEANDMORE]

Cool

[@UsagiElectric]

Thank you!

[@thepostapocalyptictrio4762]

I learned about drum memory because delay/echo/vibrato music circuits in the 40's and 50's used drum delays quite a bit.

[@DamacusSquared]

I know it's a vac-tube computer but you could always use latching relay memory as an alternative.

[@gort59]

Hi, have you considered some sort of relay based memory for your computer? I built a version of the MINIVAC 601 and it uses relays for many of the experiments, including memory. P.S. I have really enjoyed watching your computer evolve and your educational videos!

[@UsagiElectric]

I actually did think a bit about using relays for memory! Some of my first homebrew computing experiments were with relays, heavily inspired by Jeroen Brinkman's relay computer ( www.relaiscomputer.nl/ ). There's a couple reasons why I ultimately decided not to use relays, but the main one is that they aren't quite visual enough. For the relays that I can afford at least. Relays with clear cases can get quite pricey, especially when you need a whole lot of them. Also, I think it would be cool to brute force my way through this with tubes, haha. I remember checking out your Minivac601, it's a really cool build! It would be interesting to see if it could be wired up in such a way that would allow it execute programs from a separate ROM like paper tape, kind of like as a final evolution of the Minivac design!

[@terrypokorny3858]

This was very interesting on explaning the differant types of memory

[@byterock]

Need some wire rope ROM., or as it was called in the day LOL programs. LOL meaning Little Old Lady ;) Though some of the original wire rope ROMs where made by hand 'ie stitched though it is a misnomer as it is more of a weave' it was largely made my machine. I still say neon light memory if the way to go

[@UsagiElectric]

I actually considered core rope ROMs as well! In this episode I only talked about RAM, but for Program Control in the future, I'll need to figure out a good way to read programs. Initially, I was planning on using core rope memory because it's just so cool. But, the 1-bit architecture means that programs are going to need to be massively long, and long core rope programs is perhaps a little... much, even for me, haha. Neon light memory is also a brilliant idea, but I just don't have the voltage for it. I really want to keep everything on this build at a max of 36 (+24V to -36V), which isn't quite enough to strike a neon. But, I have been getting quite a bit obsessed with neons lately, and really want to build something using neons as the primary logic elements!

[@__--JY-Moe--__]

how it does what it does! it's all magic!!

[@TheFinalFrontiersman]

something tells me a dekatron may get involved somewhere ;)

[@kylewilson4097]

Mercury delay lines for short term storage?

[@fieldomoss]

wonderful video! Thank you for the clear explanations!

[@UsagiElectric]

Thank you so much!