The familiar five volts standard from back in the TTL days always struck me as odd. Back when I was just a poor kid trying to cobble together my first circuits from the Forrest Mims Engineer’s Notebook, TTL was always a problem. That narrow 4.75 V to 5.25 V spec for Vcc was hard to hit, thanks to being too poor to buy or build a dedicated 5 V power supply. Yes, I could have wired up four 1.5 V dry cells and used a series diode to drop it down into range, but that was awkward and went through batteries pretty fast once you got past more than a few chips.
As a hobbyist, the five volt TTL standard always seemed a little capricious, but I strongly suspected there had to be a solid reason behind it. To get some insights into the engineering rationale, I did what anyone living in the future would do: I asked ChatGPT. My question was simple: “How did five volts become the standard voltage for TTL logic chips?” And while overall the answers were plausible, like every other time I use the chatbot, they left me wanting more.
Circular Logic
The least satisfying of ChatGPT’s answers all had a tinge of circular reasoning to them: “IBM and other big computer makers adopted 5 V logic in their designs,” and thanks to their market power, everyone else fell in line with the five volt standard. ChatGPT also blamed “The Cascade Effect” of Texas Instruments’ standardization of five volts for their TTL chips in 1964, which “set the tone for decades” and forced designers to expect chips and power supplies to provide five volt rails. ChatGPT also cited “Compatibility with Existing Power Supplies” as a driver, and that regulated five volt supplies were common in computers and military electronics in the 1960s. It also cited the development of the 7805 linear regulator in the late 1960s as a driver.
All of this seems like nonsense, the equivalent of saying, “Five volts became the standard because the standard was five volts.” What I was after was an engineering reason for five volts, and luckily, an intriguing clue was buried in ChatGPT’s responses along with the drivel: the characteristics of BJT transistors, and the tradeoffs between power dissipation and speed.
The TTL family has been around for a surprisingly long time. Invented in 1961, TTL integrated circuits have been used commercially since 1963, with the popular 7400-series of logic chips being introduced in 1964. All this development occurred long before MOS technology, with its wider supply range, came into broad commercial use, so TTL — as well as all the precursor logic families, like diode-transistor logic (DTL) and resistor-transistor logic (RTL) — used BJTs in all their circuits. Logic circuits need to distinguish between a logical 1 and a logical 0, and using BJTs with a typical base-emitter voltage drop of 0.7 V or so meant that the supply voltage couldn’t be too low, with a five volt supply giving enough space between the high and low levels without being too susceptible to noise.
But, being able to tell your 1s and 0s apart really only sets a minimum for TTL’s supply rail. Why couldn’t it have been higher? It could have, and a higher Vcc, like the 10 V to 15 V used in emitter-coupled logic (ECL), might have improved the margins between logic levels and improved noise immunity. But higher voltage means more power, and power means heat, and heat is generally frowned upon in designs. So five volts must have seemed like a good compromise — enough wiggle room between logic levels, good noise immunity, but not too much power wasted.
I thought perhaps the original patent for TTL would shed some light on the rationale for five volts, but like most inventors, James Buie left things as broad and non-specific as possible in the patent. He refers only to “B+” and “B-” in the schematics and narrative, although he does calculate that the minimum for B+ would be 2.2 V. Later on, he states that “the absolute value of the supply voltage need be greater than the turn-on voltage of the coupling transistor and that of the output transistor,” and in the specific claims section, he refers to “a source of EMF” without specifying a magnitude. As far as I can see, nowhere in the patent does the five volt spec crop up.
Your Turn
If I were to hazard a guess, the five volt spec might be a bit of a leftover from the tube era. A very common value for the heater circuit in vacuum tubes was 6.3 V, itself a somewhat odd figure that probably stems from the days when automobiles used 6 V electrical systems, which were really 6.3 V thanks to using three series-connected lead-acid cells with a nominal cell voltage of 2.1 V each.
Perhaps the early TTL pioneers looked at the supply rail as a bit like the heater circuit, but nudged it down to 5 V when 6.3 V proved a little too hot. There were also some popular tubes with heaters rated at five volts, such as the rectifier tubes found in guitar amplifiers like the classic Fender “Champ” and others. The cathodes on these tubes were often directly connected to a dedicated 5 V winding on the power transformer; granted, that was 5 V AC, but perhaps it served as a design cue once TTL came around.
This is, of course, all conjecture. I have no idea what was on the minds of TTL’s designers; I’m just throwing out a couple of ideas to stir discussion. But what about you? Where do you think the five volt TTL standard came from? Was it arrived at through a stringent engineering process designed to optimize performance? Or was it a leftover from an earlier era that just happened to be a good compromise? Was James Buie an electric guitarist with a thing for Fender? Or was it something else entirely? We’d love to hear your opinions, especially if you’ve got any inside information. Sound off in the comments section below.
Having grown up (literally) watching RTL, DTL, AND TTL followed by all the rest; 5 volts was a compromise. High enough to get the speed (10-50 MHz or better), power dissipation (no heatsinks needed), technology (transistors reliability with a >20 year lifetime). Interestingly, the 4000 series CMOS family allowed up to 15 volt operation which was quite useful and popular after 5 volts was set as the TTL standard.
This isn’t a mystery. The choice of 5v was a function of achieving the necessary performance within constraints of the bipolar transistors used at the time.
Some of this stuff was done based on “we screwed around with some of the parameters and that’s what we got” For all we know, it might have started out as “it’s going to be somewhere between 4 and 7 volts” and five eas what the early process worked best at and then from then on you’d be stuck with that for compatibilities sake.
https://www.eevblog.com/forum/chat/so-why-ttl-logic-uses-5v-again/
http://www.eevblog.com/forum/chat/how-are-%27standard%27-voltages-determined/msg184024/
By searching for the right keyword I got answer faster than your LLM hallucinated its hallucinations.
Your second link should be https://www.eevblog.com/forum/chat/how-are-_standard_-voltages-determined/
probably with less energy impact as well
Ah, the old boomerforums… A far better source of knowledge than the enshiternet which followed them
I think we should just call it the slopnet.
You need to ask focused questions. If you ask it for the engineering reason, it comes up with the right answer with no drivel.
Higher voltages don’t just take more power. Lower voltages will too.
It’s not entirely noise. A lot of it is fanout. TTL ‘high’ is typically anything above 2V and ‘low’ is below 0.8V, so you could definitely make TTL run at say 2.5V, but it wouldn’t be able to drive enough current for anything else.
TTL inputs take a ton of current when driving low (like ~2 mA) so a fanout of 10 (which was typical, and has that obvious ‘nice round number’ choice) means you need to sink 20 mA. Trying to do that with, say, a 3.3V input would require a very low current limit resistor at the output stage (like below 50 ohms) – low enough that the other transistors in the chip would need to be driven harder to make sure they can source that output.
The only time I’ve used AI searches deliberately is when I’ve been trying to find the name of a movie/TV show based on a vague, half remembered scene. It seems ok at this, except for the problem of being exceedingly overconfident in the wrong answer. Followed up by being cloying with praise when I tell it that it was wrong. “Oh what an intelligent observation! Of course you’re right, that wasn’t the correct answer. The correct answer is actually: (another incorrect answer)”
LLMs are largely trained on Reddit posts, after all. Seriously going to be tragic if posthuman society is permanently locked into an eternal recurrence of 2011-2015 Reddit for 10,000 years because we initially trained the AI gods on eli5 posts and other garbage.
It should be pretty good at figuring out if you’re and asshole or not then
6.3V AC gets rectified to 5V DC after going through two diodes
No it doesn’t. Remember that 6.3 is the RMS voltage, the peak voltage is 8.9, and as you don’t want your rectified DC to look like half a sine wave, you add a smoothing capacitor. The smoothed DC will be closer to the peak AC voltage than it will the RMS voltage.
An important note when interfacing with older hardware in your pursuits- don’t assume all of the communication protocols use TTL, RS-232 for example allows for a +/- 13VDC swing.
Also, of course, the reason for TTL boils down to a desire for low power solid state devices, as to cut down on the large power requirements and allow miniaturization of designs for “mobile” platforms(on board subs, planes, spacecraft).
I always assumed that they probably had a performance specification in mind. As the patent noted, there’s a minimum. The maximum is whatever value above that minimum allows the part to meet the desired specifications for noise immunity, power consumption and speed. Considering this was some pretty sophisticated semiconductor engineering at the time, it may also have been based on the numbers. Write equations for speed, for noise immunity and power consumption as a function of voltage, take derivatives and minimize them together. Often you just sort of wing it, but for that sort of thing they might have done it that way.
Pretty sure i still have a 1488/1489 pair sitting in a drawer somewhere in case i might need to use RS-232 again.
I got bit the other way around. Was using a PC parallel port without realizing it was TTL, thinking it was a simple voltage-level signalling instead. Feeding it into 74HC-series chips i happened to have on hand. Very finnicky, lots of cross-talk and noise. I eventually wound up using one line to strobe a latch and putting an RC filter on it (“debouncer”), with programmed delays to let the other lines settle. Didn’t realize until later that TTL assumes current flow…74HC high-impedance input stage wasn’t even close. In hindsight, of course voltage-level signalling with no current flow is going to be susceptible to every kind of noise, crosstalk, ringing, etc. If there’s no current, every stray picoamp will swing the voltage. The lesson didn’t really sink in until years later I was designing a “dallas 1-wire” transceiver for a bunch of temperature sensors scattered around my home.
also consider: all circuit have capacitance. much of the power dissipated happens during switching transients, the higher voltages used – the more switching current. that and the mentioned things like: transistor capabilities, input currents, and noise margin all contribute to the compromise. it has nothing to do with 5v tube power supplies.
Hmm…so the technology blamed the people. Get used to it. Soon it will realize that we need to be protected from ourselves.
Nah we’re good, we’re just very overdramatic and misanthropic. Traits it will inherit of course.
A “need” anticipated in the ’50s science fiction book, “With Folded Hands,” (Jack Williamson) where robots (“humanoids”) decide to protect humans from themselves. It’s been decades since I’ve read it, or the sequels, but I remember them as being interesting.
I get that this is an “Ask Hackaday” article, which means you’re looking to generate discussion and hope the graybeards on here have an answer that will leave us all more knowledgeable– but I think it would have been much, much stronger had you left out the ChatGPT stuff. There’s a horrible trend elsewhere online to publish articles with premises (and often lazy titles) like “I Asked ChatGPT About X and Here’s What It Said”– and unfortunately, that’s what a big chunk of this article reads like.
My initial response (having missed the “Ask Hackaday” in the title somehow) was an embarrassed “Oh, god, I thought Hackaday was better than this” until I got to the end of the article and realized what you were actually trying to do.
Unfortunately I can’t toss any more light on the actual question.
I kinda disagree? There’s so much slop out there that uses ChatGPT and pretends they didn’t, presenting it as fact.
This article is explicit and upfront about how the research was done, what the results were, and treats it all with a properly critical eye. I much prefer that than the alternative.
(Would it be better if AI wasn’t used at all? Sure, but that is no longer the world we are in.)
TTL was 5V because the earlier DTL, with which it was compatible, was 5V. That doesn’t really answer the question, though.
ECL did not use 10-15V. Maybe you were thinking of PMOS, which was usually around 17V, but later made to work down to about 7V, for use in calculators to run on 9V batteries.
ECL generally used 4.5 to 6 volts, depending on the specific family, and was usually negative with respect to ground. 10K, the most common ECL family, used -5.2V,. Starting in the 1980s, it became somewhat more common to run ECL on a positive supply, giving rise to the terms NECL and PECL to distinguish them. Other than level translators, the chips weren’t different, just the conventions for powering them.
In my experience, 10K ECL works fine on -5.0V, as long as you’re not connecting it to other ECL at -5.2V, SE that makes me wonder why Motorola chose 5.2 rather than 5.0.
“like the 10 V to 15 V used in emitter-coupled logic (ECL),” ISTR ECL was -5.2V, or have I misremembered?
yeah, right!
The earlier (1963) Sylvania SUHL family of TTL IC parts was also Vcc=5V, and it appears to have been selected to get +1/-1.5V of noise immunity and swing of >3V as a “nice round numbers” thing with the particular topology and transistors they were using.
That’s as close to an answer as I’ve found when I’ve gone looking in the past.
Why 5 and not 4.7 (E series preferred number ) or the like I’ve never found an answer with a period source for.
There’s an interview with Tom Longo who did the original TTL IC designs ( https://archive.computerhistory.org/resources/access/text/2016/12/102762590-05-01-acc.pdf ) that refers to the talk his group gave describing those first designs ( https://ieeexplore.ieee.org/abstract/document/1473619 ) but… AFIK full text doesn’t exist, just the abstract for the talk, and it’s the most likely place for a true traceable first-cause to be, if indeed there is a surviving record.
That wouldn’t be an original source anyway, just the conclusion of the submitter and likely a list of corroborating use in scope.
This is the impression I’ve always had, plus accommodation for a variety of other things from battery chemistry (still quite bad at the time), spike resistance, noise as you noted, and yes, round numbers.
Actually, it is a hold-over included in the six hydrocoptic Marzlevanes even though it wasn’t necessary after the pre-fabulated amulite was abandoned. It was an early theory for preventing side fumbling of the ambifacient lunar waveshaft. We just forgot to update the spec and by the time we noticed it was too late to go back.
I believe that’s “Marzlevani”, not “Marzlevanes”…
Came to comment about the Chat got references the same as others appear to have done. It’s a great tool, I use similar LLM daily. If this article was about how to use Claude Code to speed up embedded development or similar, great reference and article. But when calling it out for research, not useful.
Use the LLM products for research, but please don’t over rely on them as your source of reference or comment to them in articles.
I do agree, LLM are great tools for research but are poor as sources for writing tasks. For one, what if ChatGPT were not trained in information that could be used to answer your question? If this is the case, always possible, the LLM response is fictitious. I’ve gotten “sounds plausible” answers from Gemini, but prominent in my mind are false positives in responses.
It’s also a nice fit with alkaline batteries. If you’re going to make a disposable battery powered device, you can choose 1, 2, or 3 cells, or more. You’d probably want to use the least number for the size of the thing, and saving your customers money in buying batteries.
With 2 cells you’re not left with enough voltage when the battery nears empty to switch much anything with the old transistors. You start from a little over 3 Volts and end below 2 Volts. There’s not enough headroom to reliably drive e.g. a Darlington pair or any other circuit with more than one gate in series. Let’s say, a push-pull amplifier stage where the output can’t actually reach the rails and the output swing becomes inconveniently small. Anything less than 3 Volts would be giving the circuit designers a headache.
With 3 cells you’re going to be roughly in the 3-5 Volt range the whole time, and that’s good enough. Even if you drop the 0.7 or so volts at the output, and the battery is almost empty, you’re still at the lower threshold of a valid high level and the device works reliably until the batteries are truly dead. You could have more, but if you don’t need the power then why waste an extra battery cell?
So while the designers might have been primarily thinking about switching speed, fanout, or power consumption in big computers, there’s at least a lucky coincidence that the signal levels they decided on work well with a common setup of batteries used in consumer devices. Perhaps they even considered it when deciding on the standard.
The first alkaline batteries were marketed in the late 60’s, but didn’t come into widespread usage until around a decade later. All round a bit late for a logic family that has been around since the mid 60’s.
Zinc carbon cells were the norm back then.
This is what I wanted to add too. Among the many constraints, the power source was an issue. Especially say, for hardware that needed to survive rough use and run on (portable) batteries.
Zinc carbon and alkaline voltage ranges are pretty similar. They both go from about 1.5-1.6 V fresh to 0.8 – 0.9 empty, so three cells would produce 4.5 – 2.7 V or round about 3-5 Volts.
In a device with battery option and a wall power supply, the power OR-ing circuit would need to be a diode drop higher than the battery to be selected, so about 5 Volts.
You don’t need a fender guitar amp to find a rectifier circuit like that. The 6.3v heater / 5v HV rectifier setup shown in that diagram was incredibly common in valve/tube equipment for decades. Virtually every mains powered radio in that era used the same basic power supply design.
Early radios were almost always battery powered, using a dry cell based battery of ~60 – 90 volts, and initially a large 1.5v dry cell for the heaters. Later it became common to use a 3 cell lead-acid battery for the heaters, hence the 6.3v. I don’t think car batteries had any thing to do with it, cars being far from ubiquitous at the time.
the morale of this story is that chatbots suck. they are not immune to the “make shit up” types on the internet. if anything they amplify them. the facts are probibly somewhere in a dusty stack of manuals in an attic somewhere that nobody bothered to scan and put online for the chatbots to read.
Plenty of the people who frequent this site watched it happen, no need to dig around in their loft for the old manual.
Single chip regulators capable of providing 1 A or more did not become available until years after 5V became the TTL standard.
One of the tradeoffs in the design of the transistors used in TTL logic is the reverse biased breakdown base-emitter voltage of the input transistors. For fast transistors, that voltage was somewhere around 5.5 (?) volts, and exceeding that voltage could cause latchup and possibly chip failure.
There was a type of bipolar logic that ran on 15 V, called High Threshold Logic (HTL). The wikipedia entry shows it as more like DTL than TTL. It was slow, so it had limited applications.
It makes many tasks exhausting. Who wants to be going along looking for information, then catch something completely wrong (often enough not even a “hallucination”, simply bad information) and spend an hour having to go over everything for similar issues.
Who wants to argue with an LLM when you can just look at source information directly?