Mechanical memory (and logic gates)

[fjkraan]

Based om some publications, I tried to recreate 3D-printable mechanical components as they could have been implemented in the Zuse Z1, Z2 and Z4 mechanical programmable calculators (computers according to some). The logic gates were only used in the Z1, but the memory cell was also employed in the Z2 and Z4. The latter even had 1024 * 32 bits!

See https://fjkraan.home.xs4all.nl/mechanics/zuseGates/

 

[Chuck(G)]

Your mention of a mechanical memory brought to mind a device that was developed around the WWII era. It was a typewriter-to-morse-code machine, but with a memory--that is, one could type ahead of the code being sent. Each keystroke set a pawl on a rotating drum, which pawl positions would be read while sending. I don't recall what the thing was called, but I'm sure there's a patent out there somewhere. This probably was an early implementation of keyboard buffering.

Of course, read-write mechanical memory is very old. Consider the adjustable combination stops on a pipe organ.

 

[Dwight Elvey]

I don't think Zuse ever did get his mechanical memory to work well. I do like his relay based carry mechanism. It was much faster than a relay based carry that depended on the actuation of each bit's relay to pass the carry.
Dwight

 

[fjkraan]

The mechanical memory worked well enough to use it in the relay based Z4. It were the mechanical logic gates that weren't reliable. The Z2, Z3 and Z4 used relay based logic gates.

 

[g4ugm]

Based om some publications, I tried to recreate 3D-printable mechanical components as they could have been implemented in the Zuse Z1, Z2 and Z4 mechanical programmable calculators (computers according to some). The logic gates were only used in the Z1, but the memory cell was also employed in the Z2 and Z4. The latter even had 1024 * 32 bits!

See https://fjkraan.home.xs4all.nl/mechanics/zuseGates/

It appears from Wikipedia that the Z series machines could not perform conditional branches, and so had to evaluate all possible outcomes of a problem. I believe that the essence of being a "Turing Complete Computer" is that you don't have to evaluate all possible outcomes. However Zuse was most certainly a pioneer in Computing

I wonder how reliable this would be in practice. The first "electronic" computer was probably the ENIAC which ran a program from ROM (banks of switches) in April 1948. Operating ENIAC in "stored program" mode slowed its operation considerably, but increase the speed with which it could be programmed. The Manchester Baby was, as far as we know the first computer to run a program from "RAM" in July 1948. However this machine was not a true "digital" computer as it uses analogue techniques in many places, including the "subtractor". However the "williams tube" memory was licenced by IBM and used in the IBM701...

Creating sufficient memory was a real problem for the early computing pioneers. All sorts of devices were used, and it wasn't until 1953 when core memory became available that compact reliable RAM was available.

 

[Dwight Elvey]

The change from things like analog adders and subtractors mixed in with digital machines didn't go away until the cost of using diodes with transistors instead of resistors with tubes became cheaper. With tubes it was easier to make an AND function with two resistors creating a sum that was greater than the threshold of a tube ( an analog function ). It was all about minimizing the number of tubes.
Dwight

 

[Chuck(G)]

Well, yes and no. One of the problems with tubes was reliability (ask any of the old SAGE people). Transistors were initially suspect, so a lot of diodes were used instead (e.g. PB250)--and alternate technologies stepped in to fill the gap. Parametrons (NEC systems) and core logic (Univac Solid State) were considered state-of-the-art at one time because their extreme reliability.

The big advantages that silicon transistors had over tubes that really closed the books was speed, space and consistency in manufacture. When that materialized, the alternate technologies were dropped.

I used to have some old logic boards with the little wire-leaded "peanut" tubes on them. PCB in the middle, with aluminum "wings" on either side. 6 tubes; 3 on each side, held in clips; every board that I bought had double triodes for all the tubes. They were very cheap on the surplus market; I was told that the only cost-effective way to troubleshoot these was to replace the whole assembly. I used several of the little triodes in some high-frequency experimentation.

 

[Dwight Elvey]

The Russians used tubes on their Venus lander so they can't be all that bad ( just kidding ). The NASA guys have been talking about making a lander using silicon carbide transistors for a Venus lander.
I can't imagine using a tube computer for anything other than op-amps. Still, towards the end of tubes, they were making tubes that were really quite reliable. The smaller ones were capable of high frequencies as well. Filament power per compute was still an issue. They have been talking about making vacuum tubes with carbon nano tubes as cathodes. The idea is to use no heaters, just the strength of the fields will create free electrons.
Dwight

 

[daver2]

I also recollect Russia using valves/tubes on some of their fighter planes - EMP resiliant?

Dave

 

[Chuck(G)]

The Russians used tubes on their Venus lander so they can't be all that bad ( just kidding ). The NASA guys have been talking about making a lander using silicon carbide transistors for a Venus lander.
I can't imagine using a tube computer for anything other than op-amps. Still, towards the end of tubes, they were making tubes that were really quite reliable. The smaller ones were capable of high frequencies as well. Filament power per compute was still an issue. They have been talking about making vacuum tubes with carbon nano tubes as cathodes. The idea is to use no heaters, just the strength of the fields will create free electrons.
Dwight

RCA did introduce the idea of a "dark cathode" back in the 60s, but I don't know what became of the idea. Similarly, "tunnel cathodes" have been proposed, but in general, innovations on general-purpose tube types has been at a standstill for decades--and is likely to remain so. One friend who worked with SAGE said that the racks of tubes was a convenient place to leave your lunch to warm. Although high-frequency specialized tubes such as magnetrons, TWTs and klystrons are still used, the big enemy is the space between electrodes when it comes to making high-speed digital electronics.

 

[Hugo Holden]

It is hard to ignore the Babbage Analytical Engine. On the remake of the Time Machine movie they created a machine that was similar looking with the many gears and rotating shafts, but it had a date display of course, picture attached.

The best description I ever heard of the machine in the first movie from the 1960's, which was much more decorative, was that it looked like something Elton John would have driven through the Everglades.

 

[Chuck(G)]

Well, in truth, a hard disk is a "mechanical" memory, meaning that mechanical motion is required for it to function.

Another early machine to consider is ABC from the 1930s.

 

[g4ugm]

The change from things like analog adders and subtractors mixed in with digital machines didn't go away until the cost of using diodes with transistors instead of resistors with tubes became cheaper. With tubes it was easier to make an AND function with two resistors creating a sum that was greater than the threshold of a tube ( an analog function ). It was all about minimizing the number of tubes.
Dwight

I realize this and should have said. You can build an analog adder with fewer valves than a pure digital adder, and it will be almost as reliable. The Manchester Baby and the Cambridge used a lot of single A-Stable circuits as temporary latches rather than flip-flops, again as it saves a valves, although the Baby also used EF50 pentodes in "Dual Control" mode as AND gates.

 

[g4ugm]

Well, yes and no. One of the problems with tubes was reliability (ask any of the old SAGE people). Transistors were initially suspect, so a lot of diodes were used instead (e.g. PB250)--and alternate technologies stepped in to fill the gap. Parametrons (NEC systems) and core logic (Univac Solid State) were considered state-of-the-art at one time because their extreme reliability.

The big advantages that silicon transistors had over tubes that really closed the books was speed, space and consistency in manufacture. When that materialized, the alternate technologies were dropped.

I used to have some old logic boards with the little wire-leaded "peanut" tubes on them. PCB in the middle, with aluminum "wings" on either side. 6 tubes; 3 on each side, held in clips; every board that I bought had double triodes for all the tubes. They were very cheap on the surplus market; I was told that the only cost-effective way to troubleshoot these was to replace the whole assembly. I used several of the little triodes in some high-frequency experimentation.

Most of the problems with tubes/valves are caused by heaters popping at switch-on. On the Baby replica in Manchester we run the heaters on half voltage for a few minutes. We run the machine several days per week, and only experience a couple of valve failure a year. Its got over 500 valves/tubes, so not as many as a Sage...

http://www.cs.man.ac.uk/CCS/SSEM/volunteers/valvetypes.htm

but we only get one or two failures a year....

 

[Chuck(G)]

I recall that there was, back in the 1950s, an add-on TV and radio kit consisting of a large carbon-block NTC resistor ("Globar") to soften the inrush current. I don't know if any manufacturers at the time employed it in stock sets. Interestingly, you can still find them sold today.

Running filaments/heaters at 90% of rated voltage can result in greatly increased life, particularly in power tubes--but you have to be careful to degrade the ratings somewhat.

You can still find inrush protectors on modern equipment--mostly to save the diode bridge in the power supply.

 

[m_thompson]

You can build an analog adder with fewer valves than a pure digital adder, and it will be almost as reliable.

I am restoring a DEC-PDP-9 and found a problem with the adder. The transistorized adder circuit is far from digital and has four possible logic voltages on the summer. We made an LT-SPICE simulation of the adder because we really didn't understand how it worked, and were a little challenged to fix it.