I'm having difficulties finding information on how repeatable the output of an opamp in this simple comparator configuration is. Give the following schematic:


simulate this circuit – Schematic created using CircuitLab

I do know that a real opamp has a non zero input offset voltage, so the actual differential voltage required to make the output 0V might be a couple of mV for a jellybean opamp. So in the example above, the actual voltage might be 2.503V and might also drift with temperature, but assumed the temperature stays constant, how repeatable is it? Repeatable meaning at two points in time, how different can the input voltages be so the output is the same?

Unfortunately, not being an EE, I cannot figure out how to read an opamp datasheet properly. How do I find out how repeatable the output an opamp really is?

Bonus question: Is the output from an opamp more repeatable than the output from a comparator or a schmitt-trigger? How can I find this out from looking at the datasheet?

PS: I've run some tests with an LM393 and a 74hc14 and with the voltage references and multimeters that I have access to, I pretty much don't see a difference. Changing the input by 10µV reliably changes the Output between 5V and 0V (at a different absolute voltage of course).

To elaborate on the purpose: The input is an analog signal with a relatively high resolution. I'm not interested in the actual value of the analog signal, but I need to know exactly when it crosses a certain threshold. The absolute value of the threshold doesn't matter as long as it stays as constant as possible.

  • \$\begingroup\$ 1. Your schematic doesn't show the feedback resistor you say your actual circuit has. It won't cost you any extra to include it so we know exactly what circuit you're asking about. 2. Have you considered using a chip designed to be a comparator, instead of an op-amp or a schmitt-trigger gate? 3. Just exactly which op-amp are you considering using, if you must use an op-amp? \$\endgroup\$
    – The Photon
    Nov 8, 2016 at 5:09
  • \$\begingroup\$ Repeteability is never specified. What is specified is accuracy. But it gives you an idea of what you can expect (and something like 74hc14 is, indeed, much worse than a comparator and a reference voltage). Anyway, what is your target? How much mV/µV of deviation can you tolerate? \$\endgroup\$
    – dim
    Nov 8, 2016 at 5:13
  • \$\begingroup\$ The input is a signal between 0V and 5V and I could tolerate a deviation of about 100µV. Is that 5V/100µV = 50 mV/µV or 100µV/5V = 2*10^8 mV/µV? \$\endgroup\$ Nov 8, 2016 at 5:21
  • \$\begingroup\$ When you say "from moment to moment", it implies that your input signal is changing in time. How fast it is changing? \$\endgroup\$ Nov 8, 2016 at 6:17
  • \$\begingroup\$ Moment to moment means there's a certain input signal on one day and another one on another day. Let's say, I've got an input of 2.50000000000V on Monday and 2.49999999999V on Tuesday. Will I get the same output? (Assuming the reference on the non-inverting input is always exactly 2.50000000000V) \$\endgroup\$ Nov 8, 2016 at 6:23

2 Answers 2


The main specifications you are looking for are 'long term stability' and 'temperature coefficient of offset voltage'.

Where you are looking to make a very precise comparison against a reference, it is common to use a precision amplifier with a modest gain, like 100x, ahead of a more conventional comparator.

For example, look at the OP177 with its max specification of 0.3uV/C for input offset versus temperature, and a typical input offset drift of 0.3uV per month. Those are an order of magnitude or more better than cheap 'cooking grade' opamps and comparators. This is my 'go-to' device for low cost (digikey $6 in ones for highest F grade) high precision work. If an amplifier or comparator doesn't specify drift/time, then you can assume it's worse than that!

Indeed given a stability of the order of uV in 2.5v, your difficulty is not going to be the comparison system, but the reference voltage. The trick to getting good reference stability is to a) choose a device well specified for drift (buried zener is the type to look out for) b) run it continuously 24/7/365 c) from a regulated voltage d) at a constant temperature. Users often use two cascaded stages of voltage and temperature regulation, as it eases the stability requirements for those.

How do you check that you have a stable voltage? Build two or three systems, and compare them.

Mind you, one you have a decent low drift amplifier, and a low drift reference, is the stability of the process that you're measuring meaningful at this level?


This is just a guess, but I suspect that when your engineer friend says that an op-amp or comparitor configured as a Schmitt Trigger is more precise than an ordinary Schmitt Trigger, he or she is talking about a Logic Gate (IC) Schmitt Trigger such as 74HC14.

He or she is absolutely correct IF that is what they are comparing to. The absolute voltage thresholds of a typical Logic Gate Schmitt Trigger have a wide variance.

It really depends on what you need your Schmitt Trigger for. If you need precise voltage thresholds, use comparitors connected as Schmitt Triggers. If the goal is to clean up a noisy logic signal, the rough-and-ready Logic Gate Schmitt Trigger is probably sufficient.

  • \$\begingroup\$ Hmm... But I don't care about the precision. I care about the repeatability. I do need switching at the same voltage, but I don't care about the actual value of that voltage. It this case: Is a comparator or an opamp significantly better than a 74HC14? How do I find this out from reading the datasheet? \$\endgroup\$ Nov 8, 2016 at 5:13
  • \$\begingroup\$ When you say repeatability, do you mean from moment to moment, or are you going to build 100 of these and you need them all to have the same threshold? \$\endgroup\$
    – The Photon
    Nov 8, 2016 at 5:15
  • \$\begingroup\$ Moment to moment. \$\endgroup\$ Nov 8, 2016 at 5:23
  • \$\begingroup\$ It is easier to specify the required accuracy for comparator error than analyze all bad and good choices. This depends on signal spectrum, range and SNR from the source and Vio(T,Vcc) worst case. How many parts per million or mV? \$\endgroup\$ Nov 8, 2016 at 5:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.