It's not an electric pulse stored in the delay line but a torsion so mechanically. Much slower then electrical pulses. This was used in the restauration of the first computer used for code breaking in the UK
Original delay line memory used sound waves through mercury. Later I saw torsion or sound wave through wire, as it is shown there. So it is "mechanical" wave through heavy medium. It needs to be slow enough (in traveling through the medium) to pack enough information into it.
@@radekhn There have been some delay lines that used either a massively long length of coaxial cable, either with a solid core, or some with a wound inner conductor. Another I saw used a lot of LC circuits, which performs almost the same function. The delay line using coax was normally used to give a few microseconds delay for signals, and the LC circuit is still a part you can get new, used as a delay line for logic signals in computers and automation, where you need to delay a digital signal for a certain amount of time so that the pulse to the latch it controls arrives at the correct time. Normally a DIP plastic package with a series of taps on it, and available with either a buffered output or unbuffered, giving delays of 5 to 50ns per tap section. Yes the longer delay lines used either acoustic delay, most common was the PAL TV system, which used a delay line made from glass, that would store the video signal for exactly one line, so that it could be used with the following line as a colour reference, allowing the extraction of the colour signal out of the phase information difference between them, with the difference being important as to determine colour, not exact phase angle as NTSC used, so cancelling out any phase shifts in the transmission path that would lead to colour shifting line by line, giving the system the backronym of Never Twice Same Color, as you had to have a tint control to keep flesh tones something approximating constant. PAL did not need that, any phase shifts were converted to small brightness differences instead, which went completely unnoticed.
You're totally right! The only delay line stuff that I'm somewhat familiar with are the LC delay lines that Tektronix and HP used in their scopes, so I was getting that type of delay line and the acoustic or torsion type of delay lines all confused in my head.
Thank you! And I think my favorite part of it is that it wouldn't be possible if I weren't running the 6AU6s at incredibly low voltages! It's a type of latch design that's only possible for this particular setup, which I think is perfectly fitting.
@@UsagiElectric I just remembered that there were some ancient eye tubes that had a low-gain triode inside-- don't think anybody has ever tried making a flip-flop out of two 6E5's. Hint Hint. The downside is they are very expensive these days. One of my Telefunken radios from 1940 even has an eye tube with a PENTODE inside.
@@georgegonzalez2476 that valve was probably an EFM1,,used by philips in some sets, now rare as honest politicians 😉😁 i managed to get hold of a used one at reasonable price , but havent tested it, the phosphor looks reasonably 'clean' and not too burnt, luckily, and heater is intact
@@andygozzo72 Yes, I think I paid like $40 for one of those for a Telefunken radio. And then I used some Windex glass cleaner on the glass dial and all the numbers came off the glass! Sigh.
This might be an absolute non-issue, but have tested how fast the VFD-based latch can switch? You're not aiming for MHz speeds, so this might be purely academical, but would still be nice to know I think
I haven't specifically tested this design for speed, but the 6977 datasheet does give us a little insight - it takes around 30 us to fully illuminate or fully extinguish. If we imagine it changing states every clock pulse, that would mean a maximum clock speed of around 30kHz, which isn't super fast. However, my current design plan is to read the instructions directly off magnetic audio tape, and using FSK the maximum speed I could attain is around 1kHz, though personally, I'm aiming for just 50 to 100Hz. Either way, the VFDs wouldn't be the bottleneck, but it's interesting to know what kind of maximum speed you could achieve with a better program design!
2:58 Neon tubes don't really have "negative resistance", since that would require that they would output energy. The English terms are a bit unspecific about this, but it has a negative *differential* resistance in the shown area whereas the overall resistance is always positive. :)
That is indeed correct! But, I've always seen it referred to as negative resistance, and when speaking to the camera, the fewer complex words I have to say in a row, the better, haha.
@@UsagiElectric is correct here. Even wikipedia calls this phenomenon "Negative Resistance": en.m.wikipedia.org/wiki/Negative_resistance The same article specifies that the term is improper and actually refers to Negative Differential Resistance (NRD), but that NR is more widely used to refer to it. So I think that while the term is improper, it's the best way to refer to it because it's what people use to refer to it.
I'll go into the entire memory map in more detail in the next video, but the current plan is to have 64-bits addressable, 48-bits of which will be the 6977/6AU6 memory combo. The remaining 16-bits will be reserved for external input and other things. Not a huge amount, but being a single bit data bus, it's going to take very long programs to even fill that tiny amount of memory up!
Very excellent as always, but note that the delay line you showed isn't electrical, but a mechanical vibration (and I assume even more painful to design). Search for "Torsion wire delay line" to learn more. For the kind of small random access memory you are looking to build, I think you going in the right direction, but for ROM (say, code) core rope memory is pretty easy.
You're totally right! The only delay line stuff that I'm somewhat familiar with are the LC delay lines that Tektronix and HP used in their scopes, so I was getting that type of delay line and the acoustic or torsion type of delay lines all confused in my head. For ROM, core rope would be perfect if I had any microcode, but being a RISC architecture, each instruction read in directly acts on all the gubbins. And being a 1-bit machine, programs are going to need to be hundreds and hundreds of lines long, which is a bit difficult with core rope. Instead, I'm thinking we'll use reel-to-reel audio tape!
Have you looked into Mercury Memory *(Type of Delay memory)* ? the construction appears really easy, it consist of a tube filled with mercury and has two transducers at either end.. it was used in *EDSAC, EDVAC, UNIVAC*
In your "original" circuit, I thought of a possible solution. To keep the light in the room from being a problem, you enclose the circuit in a box that is basically lightproof. But, you say, how do you see the action? You do that by adding an L.E.D. in series with the other diode and that L.E.D. is mounted in a hole just big enough for the L.E.D. to fit through. If it allows too much light in, simply cover the bottom part of the with something to lightproof it. I like your later version, so I guess you won't be using the idea I came up with because I think you, like me, prefer it. But you might want to store the idea. Love what you are doing with all this stuff! OH, does "Professor Rabbit" have a name? ;-)
That's a solid idea! But I'm trying to avoid the use of LEDs anywhere on the machine. VFDs, tubes and germanium diodes were all things that were available in relatively decent quantities back in the 1950's, which is the theme I'm trying to go for. And the professor does indeed have a name - "Koma". It's short for "Komaru", which means troublesome in Japanese because she may be adorable, but she's also a bit of a troublemaker, haha.
They’re still manufactured to my knowledge, as they have a few niche applications. I recently got a few NE-2’s off of Digi-Key for a project, so they have some demand.
I believe there's still quite a few manufacturers around the world that are making them! As Apochromatic said, I think you can still find them on Digi-key and probably Mouser, and there's also a few sellers on eBay selling them that are US based. Just do a search for "NE-2" and you should get a bunch of results!
U could use neons and instead of using voltage use current and the computer rails act as ground or something of such Exactly how vfd and nixies work with silicon chips But a crt is the best because speed and amount of data canbe stored and put a lense over crt to expand pixles can see the data
I'm not sure I'm following you. Neons are incredibly low current devices and really need the voltage potential to get them to strike. Williams Tube / Electrostatic CRT memory is super cool, but again, that's much, much more voltage than what I'm allowing myself to use here!
@@UsagiElectric so ur valves are ground for the device and the device is the voltage drop sos 250v goes though neon and then theres like 5v to go though ur valve as hypothetical situation If u look at nixie tube with silicone chips that the chip doesn't actually have the 180v go though it just grounds the nixie.........
@@UsagiElectric th-cam.com/video/ggVu_U-CsAk/w-d-xo.html Hopefully that helps The chip runs 30v and nixie is 190v Nixie is common 190v and chip grounds all the number pins working mainly with current
if you're not using a candle, a lighter, and a snuffer arranged in some rube goldberg machine arrangement then i'm not interested >;). JKJK amazing content my dude. can't wait to see this applied to the computer!
Haha, now you've got me thinking about how to rube golberg my way to memory! Thank you, I'm quite excited to see the processor actually store some data somewhere!
Another cool thing about neon lamps is that they work as photoresistors! On the subject of photoresistors, if you were to go with the optical based memory, there are glass photoresistors with octal bases (essentially another type of tube, a diode without a heater) from the 50s that look cool! I have one in the optical sound circuit of my vintage movie projector.
Oh yeah, I totally forgot about that! If I remember right, some neons actually add a small dash of Krypton-85 to provide a little radioactive light to help the bulb strike in really dark conditions. I do actually have some of those old photo tubes though mine use an even older 4-pin base. They're a little too big for a neon based logic circuit, but still very cool pieces!
Williams tube/electrostatic CRT memory is very cool indeed, but as I said at 3:45 I'm trying to keep the entire machine at a maximum potential of just 36V (+24V B+ and -12V biasing).
@@UsagiElectric do how vfd and nixies do and do current instead of voltage sos ur low tension valves are a ground sink instead of working with high voltage or use a separate psu to drive high voltage and ur valve runs on separate low voltage sos to get the required voltage it is adding the 2x voltages in series I think a normal color crt be best make it scan screen constantly and it constantly refreshes screen and measure current of each pixel a lit one requires less current probably be able to use the colors 2 for 3x memory The electron fire gun tends to not require a high voltage I believe to do the pixel it's all the rest of the crt drive that does And make sure you use valves to run the crt because it be epic
@@UsagiElectric by looking at voltages the main one sends out the color to pixle to light seems be 30v the rest voltages are just crt drive and don't do the picture directly
I'm late for the party here.... but how about using relays to make a shift register? .... very old school... very Konrad Zuse. Love the "big blob of electrons" in the thyratron diagram.
I did actually consider relays for the memory or other various parts of the computer, but I quite like the idea of keeping the entire machine totally tube based (with the exception of program ROMs, which will be magnetic tape based). Having said that, relays are excellent for digital computing and can be made really, really compact!
If you set up your dual-triode (or pentode/"triode") design as a differential pair with hysteresis you could save yourself some passives (no diodes) and should be able to get very good control over your input logic level thresholds (due to it being essentially a kinda crappy comparator). Granted I'm not a tube engineer so there might be some downsides to doing it that way that I'm not aware of.
It was just too finicky, even with the straw or heat shrink tubing, the glass of the VFD itself let's too much light filter in. I really want the entire computer to be designed in such a way that it'll operate with massive margins of error, and the 6977/6AU6 combo is a lot more stable!
I know you started researching and going to a particular topology... but, insisting with delay lines, (I do not say that you go there, it is more something that maybe I try)... How about a (slow and easier to read/write) magnetic tape loop? maybe stereo, one channel is data, another channel is clock
We actually talked a bit about doing something like that on the Discord, kind of like the audio tape loop Look Mum No Computer showed on his channel a little while back. It totally could work, but there's a certain timing problem that arises. For example, I need some markers in the tape that identify the tapes exact position and someway to track that. There also needs to be specific control for reading and writing, which can be quite complex with transformers, amplifiers and precise timing. This isn't all that different from delay memory, but we rapidly approach a point at which the number of tubes involved is about the same as doing it with flip flops but then we lose the cool blinkenlights that the flip flops give us. Also, I'm totally planning on using magnetic tape for the program ROM, so there will be some good magnetic tape use going on!
@@UsagiElectric just rambling... how about taking advantage of a stereo tape; channel 1 with pulses as a marker, and channel 2 with data. if channel 1 and channel 2 have a pulse, it is a one, if only channel 1 has a pulse it is a zero. channel 1 would increase an address counter, and at the end of the tape loop, which the counter instead of resetting, would simply return to zero by overflow. you can use a binary comparator with only diodes to determine if the address read is equal to the address requested, although the functioning is somewhat "analogic"
nice! another thing against 12AU7s is if you need to buy more, audiophools covet the things causing prices to be silly for some brands,, 6AU6s must be cheap and easily available where you are? not so much here in the uk , known more as EF94s,, EF91/6AM6s were more common here , used to be loads of MOD surplus ones in the 50's , they were everywhere!
You might try to find some information on 1960's Raython KDU's (keyboard display units), they used 2 types of memory you haven't mentioned, a sonic R/W fifo and cathode read only memory. The sonic memory was simply a piezo transducer to produces ultrasonic vibrations on a length of wire about 30 feet and another piezo receiver at the end this was neatly enclosed in a 12 x 12 x 1 inch metal box the purpose was to store the screen display characters of 40 characters per line 20 displayed lines on the screen. When a key was pressed the character would replace the character in the circulating store, constantly refreshed probably at 25 screens /sec. The cathode memory was a fixed target within the crt, as the character was displayed the appropriate character in the crt would be scanned, the change in crt current would represent 1/7 of the character. Each line was scanned 7 times in sync with the display CRT. The KDU's used DTL logic chips. So many innovative solutions to memory when memory was precious and programs were much more efficient and tighter written.
One design I remember seeing, but cannot find online quickly, is a rotating capacitor memory. Brushes connect the R/W circuit to different caps as the drum contains the caps rotates. The caps only need to be large enough to hold a charge for a little over a rotation, using a regenerative circuit in R/W control. using larger caps, and LEDs, the entire memory would be completely visible. The design allows many more bits of storage per active circuit element.
whatever you are doing here can be translated into a homemade chip with mosfets ... Maybe team up with sam Zeelof to translate to MOS Silicon tubes/mosfets! Would be a lot of un!
Before any technical limitations you may have, delay line memory was already out of scope due to your self imposed voltage limitations. One of the main effects of using such low voltages is the huge switching time of the tubes. In order to get the circuitry to operate fast enough you would need to drive it with the high voltages you have been avoiding..
I guess you already checked this but I had the thought that your first idea with the LDR might work, if you find an optical filter that has a very narrow response to the greenish blue light of the VFD? This might bring your ambient threshold down enough to work, however it would still be an insane amount of work to assemble and you might have "bleeding" issues inbetween two neighboring dots of the VFD tubes. I like how you sneakily used two diodes as an OR gate but it makes sense, since early solid state diodes and vacuum tubes were definately used at the same time. Oh well, you might have a slight stock of 6AU6 tubes to use, so I definately get why you want to use them everywhere. The VFD as a triode is an ingenious idea, I can only add to what the other commenters said, very nice for visualisation. I know that vacuum tubes are very high impedance anyway and it makes sense to get everything done with rather high resistances but did you run into any problems of noise or metastability due to high impedances and potentially high currents on the heaters? I guess it should only be an issue with changing currents like when you turn on the heaters, where the tubes won`t conduct anyway. How many RAM do you plan to use, something like 32 bits or more?
Interestingly, I actually mentioned the bleeding issue between neighboring dots in my early edit of the video, but ultimately it got cut to try to keep the video runtime reasonable. I haven't done any experimenting with filters, which could work quite well, but I did do quite a lot of experimenting with three dots setup. Because the anodes are in the center of the glass, no matter how well I try to shield the LDRs, there's always some light from neighboring dots. I actually had to skip a dot to even get it to work, and even then if the neighboring dots on both sides are lit up, it can cause issues. It's a shame because it's such a cool looking piece of memory! The primary logic gate I built the entire processor out of was the NOR gate and I used two little diodes to make those NOR gates. It's cheating in that they're little schottky diodes, which are a little too modern, but germanium diodes of the era were indeed quite common with vacuum tube computers. The LGP-30 is a wonderful example of this! Thanks, I'm really happy with how the VFD and triode combo turned out! Regarding the heaters, everything is super unstable until all the heaters warm up and stabilize, which takes about 30 seconds on average. Flip Flops and latches almost always seem to initialize into the same state (for example, the instruction register always warms up into the same state, which if I remember is 1101). But, so far, once everything is fully warmed up, it all seems to work rather well! Granted, I'm not aiming for high clock speeds though. The ultimate goal is around 100 Hz, and at that speed, everything should be really stable. I'll go into the entire memory map in more detail in the next video, but the current plan is to have 64-bits addressable, 48-bits of which will be the 6977/6AU6 memory combo. The remaining 16-bits will be reserved for external input and other things.
Delay line memory... Where's Veritasium? LOL Makes me want to play with thyratrons - got two European side-contact and a few Soviet heptal ones. Crazy PCB art, as always :) Looking at those curvy traces (not to be confused with curve tracers) makes me really happy. I love your charisma, uniqueness, nerve and talent. UE is one of the funniest channels to watch on the electronics part of YT. I'm so glad I found it. Definitely needs more recognition!
2:25 you can technically make sram using tubes but, would need a minimum of about 6-8 tubes per bit of sram, while you can make a one bit worth of dram with one tube and a capacitor with some feedback to act as dram refresh, so two tubes per bit of dram and a capacitor, well you can try using a vfd display to act as dram cells , and also use another vfd to provide dram refresh and addressing... and also see what the bits look like, since teh can be used as makeshift triodes.........
Hi! I like the approach with vacuum tubes. What about using capacitors as memory? Maybe, one cap can hold more than 2 states, like 4 voltage levels (24 divided to 4 voltage ranges) which means 2 bits/cap. I think you can fit 64 x 64 capacitors to a single board, and One vacuum tube circuit that is refreshing all the values. Basically demonstrating DRAM, and using charge as in modern SSD-s. This would give 1KB of memory! It would be enough to drive a monochrome Display, and write some games for this computer :D
Also I just remembered that valve computers they used special valves because they didn't like being powered up and long waits between signal changes it lessens life of tube etc I was thinking if you sent a modulating signal though it that doesn't effect the hi n low states can keep the valve life longer and bonus u could make a circuit to pick up the signal that tell you if valve is working and could even live see life of valve by how signal is
Thank you! Megabytes? About 0.000006 MB, haha. I'll go into the whole memory map in much more detail in the next video, but I'm aiming for 64-bits addressable, of which 48-bits will be this 6977/6AU6 combo.
If I were using LEDs, it totally would! But, I'm using VFDs here, which are designed to have a plate voltage of 20V to 30V. Even still, having resistors on the buttons is good practice, I just didn't draw them in the schematic for simplicity sake.
I am indeed using little diodes in there. The primary logic gate I built the entire processor out of was the NOR gate and I used two little diodes to make those NOR gates as well. It's kind of cheating in that they're little schottky diodes, which are a little too modern, but germanium diodes of the era were indeed quite common with vacuum tube computers and could be put in place here with no modifications necessary at all. The LGP-30 is a wonderful example of germanium diodes and tubes working together, but even IBM regularly used semiconductor diodes on there tube computers too!
@@UsagiElectric I didn't know they used the germanium diodes. when I was in school the Si transistors were replacing germanium. As I remember the Germanium devices had stability problems and were hard to bias. But that was long ago. Keep up the good work.
The 6977 uses 1V at 30mA for the filament, which means when it is fully warmed up, it has a resistance of 33 ohms. I'm using an 820 ohm resistor from the 24V rail, which should drop 23.1V across it to supply 0.9V into the 6977 filament and allow 28mA of current to pass through, which is prefect.
@@UsagiElectric mohinder's observation is that the filament appears to be (camera or it appears that bright by eye?) glowing brightly, which is normally a no-no with VFD's.
![[Full Episode] The Restaurant War Thailand ศึกพ่อค้าซ่าแม่ค้าแซ่บ Episode 2 | 29 ก.ย. 67](http://i.ytimg.com/vi/r3lfLNH6Ke4/mqdefault.jpg)
It's not an electric pulse stored in the delay line but a torsion so mechanically. Much slower then electrical pulses. This was used in the restauration of the first computer used for code breaking in the UK
Original delay line memory used sound waves through mercury.
Later I saw torsion or sound wave through wire, as it is shown there.
So it is "mechanical" wave through heavy medium. It needs to be slow enough (in traveling through the medium) to pack enough information into it.
@@radekhn There have been some delay lines that used either a massively long length of coaxial cable, either with a solid core, or some with a wound inner conductor. Another I saw used a lot of LC circuits, which performs almost the same function. The delay line using coax was normally used to give a few microseconds delay for signals, and the LC circuit is still a part you can get new, used as a delay line for logic signals in computers and automation, where you need to delay a digital signal for a certain amount of time so that the pulse to the latch it controls arrives at the correct time.
Normally a DIP plastic package with a series of taps on it, and available with either a buffered output or unbuffered, giving delays of 5 to 50ns per tap section. Yes the longer delay lines used either acoustic delay, most common was the PAL TV system, which used a delay line made from glass, that would store the video signal for exactly one line, so that it could be used with the following line as a colour reference, allowing the extraction of the colour signal out of the phase information difference between them, with the difference being important as to determine colour, not exact phase angle as NTSC used, so cancelling out any phase shifts in the transmission path that would lead to colour shifting line by line, giving the system the backronym of Never Twice Same Color, as you had to have a tint control to keep flesh tones something approximating constant. PAL did not need that, any phase shifts were converted to small brightness differences instead, which went completely unnoticed.
Yeah, the delay line memory he showed a picture of isn't too dissimilar from a spring reverb, as found in old guitar amps.
You're totally right!
The only delay line stuff that I'm somewhat familiar with are the LC delay lines that Tektronix and HP used in their scopes, so I was getting that type of delay line and the acoustic or torsion type of delay lines all confused in my head.
The part of the latch that stores the information actually illuminates .... BRILLIANT!
Thank you!
And I think my favorite part of it is that it wouldn't be possible if I weren't running the 6AU6s at incredibly low voltages! It's a type of latch design that's only possible for this particular setup, which I think is perfectly fitting.
@@UsagiElectric I just remembered that there were some ancient eye tubes that had a low-gain triode inside-- don't think anybody has ever tried making a flip-flop out of two 6E5's. Hint Hint. The downside is they are very expensive these days. One of my Telefunken radios from 1940 even has an eye tube with a PENTODE inside.
@@UsagiElectric any diagrams for analog processing?
@@georgegonzalez2476 that valve was probably an EFM1,,used by philips in some sets, now rare as honest politicians 😉😁 i managed to get hold of a used one at reasonable price , but havent tested it, the phosphor looks reasonably 'clean' and not too burnt, luckily, and heater is intact
@@andygozzo72 Yes, I think I paid like $40 for one of those for a Telefunken radio. And then I used some Windex glass cleaner on the glass dial and all the numbers came off the glass! Sigh.
This might be an absolute non-issue, but have tested how fast the VFD-based latch can switch? You're not aiming for MHz speeds, so this might be purely academical, but would still be nice to know I think
I haven't specifically tested this design for speed, but the 6977 datasheet does give us a little insight - it takes around 30 us to fully illuminate or fully extinguish. If we imagine it changing states every clock pulse, that would mean a maximum clock speed of around 30kHz, which isn't super fast. However, my current design plan is to read the instructions directly off magnetic audio tape, and using FSK the maximum speed I could attain is around 1kHz, though personally, I'm aiming for just 50 to 100Hz.
Either way, the VFDs wouldn't be the bottleneck, but it's interesting to know what kind of maximum speed you could achieve with a better program design!
2:58 Neon tubes don't really have "negative resistance", since that would require that they would output energy. The English terms are a bit unspecific about this, but it has a negative *differential* resistance in the shown area whereas the overall resistance is always positive. :)
That is indeed correct! But, I've always seen it referred to as negative resistance, and when speaking to the camera, the fewer complex words I have to say in a row, the better, haha.
@@UsagiElectric is correct here. Even wikipedia calls this phenomenon "Negative Resistance": en.m.wikipedia.org/wiki/Negative_resistance
The same article specifies that the term is improper and actually refers to Negative Differential Resistance (NRD), but that NR is more widely used to refer to it.
So I think that while the term is improper, it's the best way to refer to it because it's what people use to refer to it.
@@aliceitc CookieCrasbe's comment is spot on. He did not criticize the use of the term, rather he pointed out it is improper - which you agree with.
@@donmoore7785 it might be "improper" but given that the distinction isn't useful is there a point?
This is cool memory the visual indicator is the memory this will be really cool finished
Thank you! I'm really happy with how the combo of VFD and low voltage 6AU6 turned out!
How many bits of memory are you building ?
I'll go into the entire memory map in more detail in the next video, but the current plan is to have 64-bits addressable, 48-bits of which will be the 6977/6AU6 memory combo. The remaining 16-bits will be reserved for external input and other things.
Not a huge amount, but being a single bit data bus, it's going to take very long programs to even fill that tiny amount of memory up!
@@UsagiElectric That's a lot of tubes 😀
Clickbait title for the VFD/LDR solution: This computer is afraid of the light!
Haha, I gotta work on my clickbait titles!
Very excellent as always, but note that the delay line you showed isn't electrical, but a mechanical vibration (and I assume even more painful to design). Search for "Torsion wire delay line" to learn more. For the kind of small random access memory you are looking to build, I think you going in the right direction, but for ROM (say, code) core rope memory is pretty easy.
You're totally right!
The only delay line stuff that I'm somewhat familiar with are the LC delay lines that Tektronix and HP used in their scopes, so I was getting that type of delay line and the acoustic or torsion type of delay lines all confused in my head.
For ROM, core rope would be perfect if I had any microcode, but being a RISC architecture, each instruction read in directly acts on all the gubbins. And being a 1-bit machine, programs are going to need to be hundreds and hundreds of lines long, which is a bit difficult with core rope. Instead, I'm thinking we'll use reel-to-reel audio tape!
Semiconductor diodes? Isn't that cheating?
Have you looked into Mercury Memory *(Type of Delay memory)* ? the construction appears really easy, it consist of a tube filled with mercury and has two transducers at either end.. it was used in *EDSAC, EDVAC, UNIVAC*
Did you try using two 6977s as a fliplop?
In your "original" circuit, I thought of a possible solution. To keep the light in the room from being a problem, you enclose the circuit in a box that is basically lightproof. But, you say, how do you see the action? You do that by adding an L.E.D. in series with the other diode and that L.E.D. is mounted in a hole just big enough for the L.E.D. to fit through. If it allows too much light in, simply cover the bottom part of the with something to lightproof it. I like your later version, so I guess you won't be using the idea I came up with because I think you, like me, prefer it. But you might want to store the idea.
Love what you are doing with all this stuff!
OH, does "Professor Rabbit" have a name? ;-)
Limo tint window box so you keep it dark
That's a solid idea! But I'm trying to avoid the use of LEDs anywhere on the machine. VFDs, tubes and germanium diodes were all things that were available in relatively decent quantities back in the 1950's, which is the theme I'm trying to go for.
And the professor does indeed have a name - "Koma". It's short for "Komaru", which means troublesome in Japanese because she may be adorable, but she's also a bit of a troublemaker, haha.
Love the glow lamps. Are they still made anywhere today? I don't think I'll be able to get any old ones from Russia any time soon...
They’re still manufactured to my knowledge, as they have a few niche applications. I recently got a few NE-2’s off of Digi-Key for a project, so they have some demand.
I believe there's still quite a few manufacturers around the world that are making them! As Apochromatic said, I think you can still find them on Digi-key and probably Mouser, and there's also a few sellers on eBay selling them that are US based. Just do a search for "NE-2" and you should get a bunch of results!
U could use neons and instead of using voltage use current and the computer rails act as ground or something of such
Exactly how vfd and nixies work with silicon chips
But a crt is the best because speed and amount of data canbe stored and put a lense over crt to expand pixles can see the data
I'm not sure I'm following you. Neons are incredibly low current devices and really need the voltage potential to get them to strike.
Williams Tube / Electrostatic CRT memory is super cool, but again, that's much, much more voltage than what I'm allowing myself to use here!
@@UsagiElectric so ur valves are ground for the device and the device is the voltage drop sos 250v goes though neon and then theres like 5v to go though ur valve as hypothetical situation
If u look at nixie tube with silicone chips that the chip doesn't actually have the 180v go though it just grounds the nixie.........
@@UsagiElectric th-cam.com/video/ggVu_U-CsAk/w-d-xo.html
Hopefully that helps
The chip runs 30v and nixie is 190v
Nixie is common 190v and chip grounds all the number pins working mainly with current
if you're not using a candle, a lighter, and a snuffer arranged in some rube goldberg machine arrangement then i'm not interested >;).
JKJK amazing content my dude. can't wait to see this applied to the computer!
Haha, now you've got me thinking about how to rube golberg my way to memory!
Thank you, I'm quite excited to see the processor actually store some data somewhere!
I love your idea of a flip flop with light (LDRs...), this was really thinking outside of the box!
Another cool thing about neon lamps is that they work as photoresistors!
On the subject of photoresistors, if you were to go with the optical based memory, there are glass photoresistors with octal bases (essentially another type of tube, a diode without a heater) from the 50s that look cool! I have one in the optical sound circuit of my vintage movie projector.
Oh yeah, I totally forgot about that! If I remember right, some neons actually add a small dash of Krypton-85 to provide a little radioactive light to help the bulb strike in really dark conditions.
I do actually have some of those old photo tubes though mine use an even older 4-pin base. They're a little too big for a neon based logic circuit, but still very cool pieces!
I'd use a crt and pixles
Williams tube/electrostatic CRT memory is very cool indeed, but as I said at 3:45 I'm trying to keep the entire machine at a maximum potential of just 36V (+24V B+ and -12V biasing).
@@UsagiElectric do how vfd and nixies do and do current instead of voltage sos ur low tension valves are a ground sink instead of working with high voltage or use a separate psu to drive high voltage and ur valve runs on separate low voltage sos to get the required voltage it is adding the 2x voltages in series
I think a normal color crt be best make it scan screen constantly and it constantly refreshes screen and measure current of each pixel a lit one requires less current probably be able to use the colors 2 for 3x memory
The electron fire gun tends to not require a high voltage I believe to do the pixel it's all the rest of the crt drive that does
And make sure you use valves to run the crt because it be epic
@@UsagiElectric by looking at voltages the main one sends out the color to pixle to light seems be 30v the rest voltages are just crt drive and don't do the picture directly
I'm late for the party here.... but how about using relays to make a shift register? .... very old school... very Konrad Zuse.
Love the "big blob of electrons" in the thyratron diagram.
I did actually consider relays for the memory or other various parts of the computer, but I quite like the idea of keeping the entire machine totally tube based (with the exception of program ROMs, which will be magnetic tape based).
Having said that, relays are excellent for digital computing and can be made really, really compact!
Pretty neat idea to use the VFD triode this way!
Thank you!
I really like how it turned out and it should look absolutely epic once programs are reading and changing the memory in real time!
If you set up your dual-triode (or pentode/"triode") design as a differential pair with hysteresis you could save yourself some passives (no diodes) and should be able to get very good control over your input logic level thresholds (due to it being essentially a kinda crappy comparator).
Granted I'm not a tube engineer so there might be some downsides to doing it that way that I'm not aware of.
ALL HAIL THE VFRAM
also the straw over the ldr would make it more directional and sensitive like I told you bevore XD
It was just too finicky, even with the straw or heat shrink tubing, the glass of the VFD itself let's too much light filter in.
I really want the entire computer to be designed in such a way that it'll operate with massive margins of error, and the 6977/6AU6 combo is a lot more stable!
I know you started researching and going to a particular topology... but, insisting with delay lines, (I do not say that you go there, it is more something that maybe I try)... How about a (slow and easier to read/write) magnetic tape loop? maybe stereo, one channel is data, another channel is clock
We actually talked a bit about doing something like that on the Discord, kind of like the audio tape loop Look Mum No Computer showed on his channel a little while back. It totally could work, but there's a certain timing problem that arises. For example, I need some markers in the tape that identify the tapes exact position and someway to track that. There also needs to be specific control for reading and writing, which can be quite complex with transformers, amplifiers and precise timing. This isn't all that different from delay memory, but we rapidly approach a point at which the number of tubes involved is about the same as doing it with flip flops but then we lose the cool blinkenlights that the flip flops give us.
Also, I'm totally planning on using magnetic tape for the program ROM, so there will be some good magnetic tape use going on!
@@UsagiElectric just rambling... how about taking advantage of a stereo tape; channel 1 with pulses as a marker, and channel 2 with data. if channel 1 and channel 2 have a pulse, it is a one, if only channel 1 has a pulse it is a zero. channel 1 would increase an address counter, and at the end of the tape loop, which the counter instead of resetting, would simply return to zero by overflow.
you can use a binary comparator with only diodes to determine if the address read is equal to the address requested, although the functioning is somewhat "analogic"
nice! another thing against 12AU7s is if you need to buy more, audiophools covet the things causing prices to be silly for some brands,, 6AU6s must be cheap and easily available where you are? not so much here in the uk , known more as EF94s,, EF91/6AM6s were more common here , used to be loads of MOD surplus ones in the 50's , they were everywhere!
You might try to find some information on 1960's Raython KDU's (keyboard display units), they used 2 types of memory you haven't mentioned, a sonic R/W fifo and cathode read only memory. The sonic memory was simply a piezo transducer to produces ultrasonic vibrations on a length of wire about 30 feet and another piezo receiver at the end this was neatly enclosed in a 12 x 12 x 1 inch metal box the purpose was to store the screen display characters of 40 characters per line 20 displayed lines on the screen. When a key was pressed the character would replace the character in the circulating store, constantly refreshed probably at 25 screens /sec. The cathode memory was a fixed target within the crt, as the character was displayed the appropriate character in the crt would be scanned, the change in crt current would represent 1/7 of the character. Each line was scanned 7 times in sync with the display CRT. The KDU's used DTL logic chips. So many innovative solutions to memory when memory was precious and programs were much more efficient and tighter written.
One design I remember seeing, but cannot find online quickly, is a rotating capacitor memory. Brushes connect the R/W circuit to different caps as the drum contains the caps rotates. The caps only need to be large enough to hold a charge for a little over a rotation, using a regenerative circuit in R/W control. using larger caps, and LEDs, the entire memory would be completely visible. The design allows many more bits of storage per active circuit element.
whatever you are doing here can be translated into a homemade chip with mosfets ... Maybe team up with sam Zeelof to translate to MOS Silicon tubes/mosfets! Would be a lot of un!
Before any technical limitations you may have, delay line memory was already out of scope due to your self imposed voltage limitations.
One of the main effects of using such low voltages is the huge switching time of the tubes. In order to get the circuitry to operate fast enough you would need to drive it with the high voltages you have been avoiding..
That is quite elegant considering your design goal. Going back to what others perfected was a good place to start.
Super smart to use the display element as half the flip-flop. I enjoyed the heck out of that. Can't wait for the next episode.
Hey does anyone know the title of the book with the neon computer circuits in it?
Later in the project, why not using nixie tubes as a display ?
You're just brilliant. What a project.
I guess you already checked this but I had the thought that your first idea with the LDR might work, if you find an optical filter that has a very narrow response to the greenish blue light of the VFD? This might bring your ambient threshold down enough to work, however it would still be an insane amount of work to assemble and you might have "bleeding" issues inbetween two neighboring dots of the VFD tubes.
I like how you sneakily used two diodes as an OR gate but it makes sense, since early solid state diodes and vacuum tubes were definately used at the same time.
Oh well, you might have a slight stock of 6AU6 tubes to use, so I definately get why you want to use them everywhere.
The VFD as a triode is an ingenious idea, I can only add to what the other commenters said, very nice for visualisation.
I know that vacuum tubes are very high impedance anyway and it makes sense to get everything done with rather high resistances but did you run into any problems of noise or metastability due to high impedances and potentially high currents on the heaters?
I guess it should only be an issue with changing currents like when you turn on the heaters, where the tubes won`t conduct anyway.
How many RAM do you plan to use, something like 32 bits or more?
Interestingly, I actually mentioned the bleeding issue between neighboring dots in my early edit of the video, but ultimately it got cut to try to keep the video runtime reasonable. I haven't done any experimenting with filters, which could work quite well, but I did do quite a lot of experimenting with three dots setup. Because the anodes are in the center of the glass, no matter how well I try to shield the LDRs, there's always some light from neighboring dots. I actually had to skip a dot to even get it to work, and even then if the neighboring dots on both sides are lit up, it can cause issues. It's a shame because it's such a cool looking piece of memory!
The primary logic gate I built the entire processor out of was the NOR gate and I used two little diodes to make those NOR gates. It's cheating in that they're little schottky diodes, which are a little too modern, but germanium diodes of the era were indeed quite common with vacuum tube computers. The LGP-30 is a wonderful example of this!
Thanks, I'm really happy with how the VFD and triode combo turned out! Regarding the heaters, everything is super unstable until all the heaters warm up and stabilize, which takes about 30 seconds on average. Flip Flops and latches almost always seem to initialize into the same state (for example, the instruction register always warms up into the same state, which if I remember is 1101). But, so far, once everything is fully warmed up, it all seems to work rather well! Granted, I'm not aiming for high clock speeds though. The ultimate goal is around 100 Hz, and at that speed, everything should be really stable.
I'll go into the entire memory map in more detail in the next video, but the current plan is to have 64-bits addressable, 48-bits of which will be the 6977/6AU6 memory combo. The remaining 16-bits will be reserved for external input and other things.
Delay line memory... Where's Veritasium? LOL
Makes me want to play with thyratrons - got two European side-contact and a few Soviet heptal ones.
Crazy PCB art, as always :) Looking at those curvy traces (not to be confused with curve tracers) makes me really happy.
I love your charisma, uniqueness, nerve and talent. UE is one of the funniest channels to watch on the electronics part of YT. I'm so glad I found it. Definitely needs more recognition!
2:25 you can technically make sram using tubes but, would need a minimum of about 6-8 tubes per bit of sram, while you can make a one bit worth of dram with one tube and a capacitor with some feedback to act as dram refresh, so two tubes per bit of dram and a capacitor, well you can try using a vfd display to act as dram cells , and also use another vfd to provide dram refresh and addressing... and also see what the bits look like, since teh can be used as makeshift triodes.........
sram= Set Reset Always Memorized
dram = Don't Remember After Memorized
Cooool
Thank you!
Of course you are using a VTVM! No transistorized VOM or DMM when you are committed to tubes.
Then there are the two Mean Well switching power supplies. But let's not get picky.
Hi! I like the approach with vacuum tubes.
What about using capacitors as memory? Maybe, one cap can hold more than 2 states, like 4 voltage levels (24 divided to 4 voltage ranges) which means 2 bits/cap. I think you can fit 64 x 64 capacitors to a single board, and One vacuum tube circuit that is refreshing all the values. Basically demonstrating DRAM, and using charge as in modern SSD-s. This would give 1KB of memory! It would be enough to drive a monochrome Display, and write some games for this computer :D
two videos so far with mostly just us getting spoken at with details man come on what even was the point. I'm out, I am done.
Well, I thought that the survey of options considered was quite interesting.
So fascinating! Thanks for sharing so much detail 🙌🏼. Also it was me watching your video a bunch of times from 5 min on 😏
Also I just remembered that valve computers they used special valves because they didn't like being powered up and long waits between signal changes it lessens life of tube etc
I was thinking if you sent a modulating signal though it that doesn't effect the hi n low states can keep the valve life longer and bonus u could make a circuit to pick up the signal that tell you if valve is working and could even live see life of valve by how signal is
Cool and clever solution. How many Megabytes of RAM will you give it ? LOL.
Thank you!
Megabytes? About 0.000006 MB, haha.
I'll go into the whole memory map in much more detail in the next video, but I'm aiming for 64-bits addressable, of which 48-bits will be this 6977/6AU6 combo.
@@UsagiElectric Nice. That's plenty for this valveset. (you can't say chipset, there are no chips. LOL)
HELLO, When you press the set button won't the led get full +24 volts and burn out?
If I were using LEDs, it totally would! But, I'm using VFDs here, which are designed to have a plate voltage of 20V to 30V. Even still, having resistors on the buttons is good practice, I just didn't draw them in the schematic for simplicity sake.
@@UsagiElectric KEEP UP THE GOOD WORK
Why not use 2 VFD's?
Wow, what a original idea! Totally bonkers too.
Thank you! I'm really happy with the weird mixture of VFD and low voltage 6AU6, it's just weird enough to be a perfect fit for the tube computer.
Using solid state diodes?????
I am indeed using little diodes in there. The primary logic gate I built the entire processor out of was the NOR gate and I used two little diodes to make those NOR gates as well. It's kind of cheating in that they're little schottky diodes, which are a little too modern, but germanium diodes of the era were indeed quite common with vacuum tube computers and could be put in place here with no modifications necessary at all. The LGP-30 is a wonderful example of germanium diodes and tubes working together, but even IBM regularly used semiconductor diodes on there tube computers too!
@@UsagiElectric I didn't know they used the germanium diodes. when I was in school the Si transistors were replacing germanium. As I remember the Germanium devices had stability problems and were hard to bias. But that was long ago. Keep up the good work.
After the forthcoming nuclear armageddon, Usagi Electric has the only functioning computer remaining in the world.
several radhard microprocessors exist
use capacitors.
Is this know as 'dark memory"?
I might be tempted to call it "light" memory instead since it uses a VFD light!
Cool
Logic gate
I am impressed, thanks.
Thank you!
please reduce filament current of 6977
The 6977 uses 1V at 30mA for the filament, which means when it is fully warmed up, it has a resistance of 33 ohms. I'm using an 820 ohm resistor from the 24V rail, which should drop 23.1V across it to supply 0.9V into the 6977 filament and allow 28mA of current to pass through, which is prefect.
@@UsagiElectric The filament glow should not be visible for maximum life. Try 1k ! Will extend tube life
@@UsagiElectric Calculations are right. Experimentation with higher values is not a bad idea.
@@UsagiElectric mohinder's observation is that the filament appears to be (camera or it appears that bright by eye?) glowing brightly, which is normally a no-no with VFD's.