Resistor–transistor logic
en.wikipedia.org
en.wikipedia.org
When I was a teenager in the 1960s, my neighbor was a sales engineer at Motorola. He saw me put up an antenna on the roof and guessed correctly that I was into ham radio. So he gave me a big Motorola databook of RTL components and said, "Let me know if you ever need anything, and we can send you some free samples."
The book had had gates, flip-flops, everything. You could get an entire JK flip-flop in a little tin can the size of a TO-5 [1] (but with more leads)!
I wanted to build an iambic Morse code keyer. This has two separate switches for your thumb and forefinger. Press with your thumb and you get one or more "dits". With your forefinger, you get a stream of "dahs". Squeeze them together and you get alternating dits and dahs, thus the "iambic" name.
I designed the circuit, built it, and it worked great!
Until I actually hooked it up to my transmitter (which was of course the point of the whole exercise).
I didn't have much of a clue about RF shielding, and RTL wasn't very robust around RF energy. So the interference from my transmitter made the circuit go crazy!
I never did figure out how to fix it, so I had to go back to a straight key.
The cost of the transistors dwarfed concerns about the cost of the boards or other components. This is a flip-flop circuit from a PDP-8. [1]
It's diode-transistor logic, primarily. It's for 3 bits. There are only two transistors used per flip-flop to store the bits. A modern designer would use more transistors to route the inputs/outputs and for addressing the flip-flop. [2] The PDP-8 designers used capacitors and diode and resistor networks, to selectively pass pulses, using transistors only when truly necessary to invert or boost a signal. Several dozen diodes, caps and resistors, plus pulse logic (which complicates the whole design) were still cheaper than a few more transistors.
With early integrated circuits, the yield was terrible. Most of the devices didn't work. It's much easier to build up a working resistor than a working diode on an IC, and it's easier to build a diode than it is a transistor. Also, with a very large process size, there's little concern about the heat of resistors. When we start increasing speed and shrinking the die, the power dissipation becomes considerable. Even by the late 1960s the heat of RTL at high speed limited how densely you could pack in it an IC. This is the same reason why, much later, CMOS would displace NMOS, despite requiring 2x as many components on chip. It runs cooler so it can be packed much more densely and run at a higher clock.
(If you absolutely needed speed, from the 1950s right into the 1990s, you would use ECL, a kind of logic that doesn't switch on/off so the transistors never enter full saturation; like a differential amplifier, the current flow is constant.)
[1] https://homepage.cs.uiowa.edu/~dwjones/pdp8/UI-8/map/R211sch...
[2] https://ars.els-cdn.com/content/image/1-s2.0-S00262692163015...
Sourcing on that, for no particular reason (also, since 1965 the dollar has inflated by ~10x):
Through-hole types that would have been used for this are around .5-1 cents in 5k quantities, and ~5 cents in units of 10+ (right on the money! I wouldn't have known that offhand.): https://www.digikey.com/en/products/filter/through-hole-resi...
SMD/chip resistors, which are used for most things except relatively small scale products, are .2-.5 cents in bulk: https://www.digikey.com/en/products/filter/chip-resistor-sur...
Note that obviously all these prices are pretty rough- digikey is not the cheapest source, and the context of the industry has changed a lot.
Iirc, the capacitance is dealt with by the gap distance of transistors, resistance by the lead
One of my "shoulda preserved that" regrets is demolishing (as a teenager) a nonworking Programma 101 for the bits and pieces. The interleaved rows of these modules made for very dense circuit packing for the discrete component era.
Classroom: learn about Karnaugh maps.
Lab: build them.
That was incredibly satisfying.
DTL circuits are composed of transistors used as inverters and of diode circuits that are either minimum circuits a.k.a. AND or maximum circuits a.k.a. OR.
When considered as min/max circuits instead of and/or, they can work even with a multi-level logic, not only with binary logic.
A maximum or minimum circuit with diodes, (i.e. an OR or AND gate) has just one diode for each input and one resistor for the output. In most cases each AND or OR gate must be followed by one inverter, to provide amplification, i.e. to restore the logic levels to values compatible with an input.
The complexity of the inverter depends on what kind of transistors are available and on what switching speed is desired, but sometimes the inverter can be made just with a single transistor without any other biasing components.
The economy of components vs. RTL appears in any complex circuits, because it is very easy to make DTL gates with a very large number of inputs, while RTL gates are restricted to a much smaller number of inputs, due to insufficient amplification from the transistors, so you need many RTL gates to replace one DTL gate with many inputs.
Using a diode matrix (and a decoder circuit, which can be made with another diode matrix), you can make a PLA or a ROM memory (the difference between PLA and ROM is that in the former both diode matrices have arbitrary diode patterns, while in the latter the diode matrix of the decoder has a fixed pattern, for selecting 1 of 2^N outputs with an N-bit input), and such PLAs or ROMs made with discrete diodes were still used in computers (e.g. for storing microprograms) many years after the integrated circuits had replaced simpler gates and flip-flops, because the ICs still had cost and size disadvantages.
The diode matrices have disappeared completely only after the introduction of integrated PROMs and PLAs (during the early seventies) that could be programmed by the end-users, by burning internal NiCr fuses.
This is a reason why DTL circuits can be much faster than the RTL circuits, because in the latter the resistances slow down the charging and discharging of the various parasitic capacitances of the circuit.
The current spikes are not a problem (because they are very short in time), they just require the presence of good decoupling capacitors close to the gates, to keep the current spikes localized, and various other shielding or decoupling measures may be needed, to avoid creating radio interference above what is accepted by standards. This is unavoidable for fast logic circuits, regardless of their schematic.
There are only two ways to make a logic circuit faster, either making it smaller so that its capacitances become smaller too, or increasing the current spikes that can be delivered by the transistors. The problem in logic circuits is that the transistors always limit the current spikes to lower values than desired, which limits the attainable clock frequency, and never that the current spikes are too high.
ttl was the default choice for building discrete logic circuits until about 01980, after which point it was obsolete because cmos (74hcxxx, not cd4xxx) was better in every way except esd
if you're running an educational computer lab on a tight budget, esd is still the dominant consideration, because students will burn out all your cmos chips with static after only a few dozen uses, while ttl chips will survive most of their mistakes
but outside a circuit lab or possibly repair of 40-year-old devices there's no reason to use ttl
When making DTL circuits with bipolar transistors, especially at low supply voltages, in order to make the logic levels at the output equal to the logic levels at the input, a simple solution was to add a diode at the output of an AND gate made with diodes, to shift down the logic levels by the voltage drop over that diode.
When the diodes used for gates were bipolar diodes, not Schottky diodes, they stored a big electric charge when on and they could be turned off quickly only if there was a path to evacuate the stored charge. Also the bipolar transistor of the inverter stored a big charge in the base, which had to be evacuated quickly for turning it off.
The current needed to evacuate the stored charge was required to pass in reverse direction through the level-shifting diode, which is not possible, so the DTL gates with level-shifting diodes were slow.
In bipolar IC technology, all diodes are made from transistor junctions, in order to not have separate process steps for making diodes. The junctions suitable for fast diodes are the emitter-base junctions. So all the diodes of a DTL gate were made inside an IC as multiple emitters of a transistor, with a short over the base-collector junction, to disable the transistor effect and make it work as a bunch of diodes.
At this point in the history of integrated DTL circuits, someone made the observation that seems trivial in hindsight, that removing the short over the base-collector junction allows to use it as the level-shifting diode, saving a diode.
Moreover, this not only saved a diode, but the bipolar transistor effect results in passing current through the reverse-biased level-shifting diode, allowing the fast turn off of the inverting transistor.
So in those early times, when DTL circuits could be made only with bipolar diodes and with bipolar transistors, TTL was the best variant.
Later, when the bipolar diodes were replaced with Schottky diodes, which store negligible charge when on, and when diode clamps were used over the inverting transistor, so that it no longer reached deep saturation and it no longer stored a big charge, TTL was no longer the optimal implementation and some of the so-called Schottky TTL families were actually DTL circuits, not TTL, but they had retained TTL as a marketing term, as this had become almost synonymous with bipolar logic integrated circuit.
TTL could never be used in discrete circuits, because it is based on bipolar transistors with multiple emitters and/or multiple bases, which have never been available as discrete parts.
And I'm sure that circuit designers who'd invested the majority of their career in that async paradigm, only to see it facing obsolescence due to 'inexorable' tooling (digital EDA being that era's Copilot/GPT) advances, would've also downvoted an analogous comment :D