Energizing a vacuum-tube flip-flop module from a 1948 IBM system

Not sure if you've seen this video about how the Apollo computers were made, but it's really fascinating.

https://www.youtube.com/watch?v=ndvmFlg1WmE

(They also anachronistically refer to women as girls which feels weird nowadays.)

I want to add an explanation of the word "trigger" because nowadays few people are aware of its correct meaning.

The early authors made a clear distinction between "trigger" and "latch", but the reasons for this have been mostly forgotten today, when almost nobody designs latches and flip-flops, because these are provided as library components in all FPGA or custom ASIC designs.

The Eccles-Jordan trigger, i.e. a closed loop formed by 2 inverters, was a "trigger" because it had a permanently active positive-feedback loop. To make it switch states, one had to force one of the outputs of the inverters towards its opposite state. That was difficult in the beginning, because the output pulled in the opposite direction to the forcing signal, trying to prevent the change. But after some threshold the modified output value went through the other inverter and then back to the output and due to positive feedback it began to pull in the same direction as the forcing signal. This triggered a very fast transition of the state. After triggering the transition it did not matter any more what the external forced signal did, the transition continued by itself.

The consequence of this is that even if the external pulse that forced the transition had a very slow rising rate, the Eccles-Jordan trigger switched much faster than its input, so it deserved the name "trigger".

The same is true for the Schmitt trigger, which also has a permanently active positive feedback loop.

Instead of using an Eccles-Jordan trigger made of 2 inverters, one can use an RS latch made of 2 NAND gates or of 2 NOR gates.

The RS latch is logically equivalent with an Eccles-Jordan trigger, but its switching behavior is very different. When the inputs of an RS latch are either "Set" or "Reset", the logical gates break the feedback loop, so the RS latch behaves like a cascade of 2 inverters without feedback, which follow the input. When the "Set" or "Reset" inputs are removed, the positive feedback loop is restored, latching the current state and keeping it until another set or reset command. Because of this, the switching of an RS latch is not "triggered", its outputs follow the inputs, with only a small rising or falling edge acceleration that is caused by the amplification of the gate in its linear range, the same as for an isolated logic inverter.

The D latch behaves in the same way as an RS latch, because for one value of its clock input the positive feedback loop is broken, allowing the change of the state, while for the other value the positive feedback loop is restored, latching the state.

Among the modern flip-flops, most are composed of latches, but there exist some types that are made with triggers.

The static RAM memory cells, like in the cache memories of CPUs, are normally made of Eccles-Jordan triggers, because the 2 inverters are simpler than the 2 NAND or NOR gates required by an RS latch.

After the Eccles-Jordan trigger, as I have written above, the next electronic bistable circuit was made in 1929, with 2 thyratrons. While the thyratron circuit looks almost identical with an Eccles-Jordan trigger, it functions in a completely different way.

A thyratron has internal positive feedback, so it latches a logic state by itself. But it does not have convenient set and reset inputs. The circuit invented in 1929 is actually a ring counter with 2 thyratrons, equivalent with a ring counter made with 2 flip-flops.

A ring counter can be extended to any number of flip-flops greater than 2. A ring counter with N flip-flops will have N states.

The same is true for ring counters with thyratrons. While the original circuit had 2 thyratrons, for 2 states, by 1931 ring counters with 10 thyratrons, thus 10 states, were made and used as components for a decimal counter, as I have said above.

This is how they looked after removing the top of their package, not how they looked inside the computer.

This is how the silicon die looked like, with the ten bonded gold wires that went to the pins of the package.

If you repaired televisions in the 70s, most of the IBM 604 trigger circuit looks completely ordinary: a couple of triodes, plate resistors, grid bias, coupling capacitors, and voltages high enough that you kept one hand in your pocket. Shirriff's observation that the trigger can oscillate, latch wrong, or refuse to switch unless the voltages are just right is especially familiar. We used to call that "the set’s acting up"; computer people called it logic margins. The later diode-controlled version is revealing too. Engineers weren’t escaping analog electronics so much as steadily boxing it in until the analog behavior became somebody else's problem.

Early digital hardware makes a lot more sense if you approach it with a voltmeter instead of a truth table.