Opamp Protection Diodes - can someone confirm my thinking

What's the context in which you're trying to protect the input?

The datasheet for the OPA192 says it's good to 4kV from the usual body model as per various JEDEC standards. It doesn't say how the protection works, but you can take it that it does. That's similar to many other chips and is OK for handling an assembled board in a manufacturing environment. You could probably kill an input in a home environment by handling the board [particularly if your environment is hot and dry and you have nylon carpets and wear synthetic fabrics], but that also applies to all the SBCs you have.

A few random thoughts on your various questions

- the BAV199 diodes are very low leakage but not stunningly fast [compared to a switching diode like a 1N4148 that is]. BAV199 recovery time is 3.5us, 1N4148 is more like 2 or 3ns. They might do. How did you come to choose them?

- both diodes leak, so there's leakage current down both diodes and the resistor will only see a proportion depending on the voltage and how well the diodes match

- the 1M resistors are a waste of time - please don't do that

- with a scheme like this, the energy from the ESD event ends up in the decoupling capacitors. There's not too much of it and the voltage shouldn't move too far, but if it looks like a problem then you'll need to catch the rail with a zener or TVS or do something else to deal with it (as you point out, the regulators themselves don't stop the rail from exceeding the normal output voltage). [I'm speaking here from the point of view of a digital designer used to planes and extensive distributed decoupling to tie planes hard together. Things might be a bit different with analogue design and need more care.]

- if you were to put some resistance between the diodes and the input then you'll end up with two stages of protection, with the external diodes taking the brunt of it and the internal doing the rest, which might be worth looking at

- I think you might first want to reconsider what you're doing with the input before designing the protection. You seem to be asking for an input that has an input resistance of 10 to the power 13 [or whatever it is for a CMOS op amp], is very low noise [over what bandwidth?], works with signal voltages down in the uV area, and can stand being zapped or connected momentarily to 50V. Not sure you'll get all of those. [You've created a conflict between the requirement for low noise and what you need to do to deal with overload/protection. The route out is probably to move away from working with a very high impedance input.]

Thanks Jon.

I chose the BAV199 because it's low leakage; other potential choices had higher leakage.  The BAV199 are available in a dual diode package so I assume they would be well matched.

The overall context that I am aiming for initially is to provide a buffer into an ADC from a sense resistor used in a DC load but I was also trying to think of it in use cases that may be a bit different (the ADC could take up to 2.048V on its input.)  So sticking to that scenario, the input voltage from a sense resistor might be 0.5V but if the driving Mosfet failed, there might be 50V generated on the input and I wanted to steer that away from the input pins; I was aiming for as little impact on the signal as possible, hence the low leakage.  I'm thinking here of what happened to "the breadboard" Peter with his (and Jan's) DC load that fried when the MOSFET failed - there was quite some discussion about the inclusion of 100K resistors and subsequently the impact they could be having on the opamps and signals.  It seemed to me that protecting against over current problems was one issue and protecting against over voltage another.  In the DC load scenario I would also expect to have some upstream protection against this (fuse, diodes and so on).  Do I really need to be concerned about uV level signals, no not really because I'm never going to be working at those levels.  I had just thought about doing it with as little impact as possible.

the 1M resistors are a waste of time - please don't do that

Yes, I obviously didn't pay enough attention to the Spice output. 

The ESD circuit is intended for out-of-circuit protection and is intended to remain inactive during normal operation.  For in circuit protection, the datasheet suggests the use of TVS diodes but they have quite high leakage and capacitance.  I'm less worried about ESD as I take precautions when handling the board and it isn't something I've ever had an issue with (yet!) 

If you can it's a really good idea to put a cap from the input pin to ground - 100nF is a reasonable value if all you care about is EMC.

In real life you often don't want a huge input shunt capacity.

A series R and the two diodes is a good compromise.

I usually use a dual diode in an SC70 type package -NXP sell nice fast ones with usually acceptable leakage.

What abuse you design your inputs to stand depends a lot on the application.

General purpose test gear like DMMs and scopes needs to cope with at least 50V DC (for minutes at least) and ideally mains AC.

But many fancy HF probes for scopes will smoke with more than a few volts DC.

Anything on a car must stand 16kV sparks on all the inputs (to pass official tests - lots of cheap things aren't that tough and often get away with it.)

Something with a plug on it that is only going to mate with the right other thing (like USB) only needs to cope with fingers, as mentioned by Doug.

MK

Did you mean something like this Nexperia 1PS301 Michael?  It has a much better recovery time but worse leakage current, probably somewhat <30nA at 7.5V. 

General purpose test gear like DMMs and scopes needs to cope with at least 50V DC (for minutes at least)

That's it, although not minutes as I'd turn it off PDQ.  Would reverse recovery time matter in such a scenario?  I could easily fit a 100nF cap on the input along with the series resistor and diode pair; it's a DC path so I don't suppose a LPF would hurt?

The part I use is NUP1301UNUP1301U which has similar leakage current to the 1PS301 but lower capacitance.

If you are only measuring low frequencies then a big shunt capacitor is a good plan - never let any signal into

your circuit that you don't need !

Reverse recovery time won't matter in an LF or DC application.

MK

Thanks Michael

Thanks Michael