input esd diodes and lm358

[exe]

I have a question why in 99% the input pins of an IC accept only signal in the range of supply rails (plus 0.3-0.5V above and below rails). And the obvious answer is "obviously because of ESD diodes" . But why they are there?

I think they do more than just input protection. I think part of their function is to prevent latch-up. That's why most of ICs have them. But not all. One example I know is lm358. It seems it has only diodes to negative rail, but not positive (not shown on the schematic, but mentioned on https://www.ti.com/lit/an/sloa277/sloa277.pdf on Figure 12). I'd say in this regard lm358 is quite special.

My questions are:
1) Why lm358 doesn't have esd diodes on input to the positive rail? (perhaps, to increase common-mode range)
2) Why other ICs don't do the same?

[Zero999]

The LM358 has special input transistors which can withstand up to 32V in the reverse direction. The common mode range isn't anything special, from just a little below the negative supply to the positive supply minus 1.5V. The LM324 is the same, but includes four op-amps, in the package.

Most BJTs are damaged if the base-emitter junction is reverse biased at more than 6V. A lot of op-amps have back-to-back diodes between the inputs. The NE5532 is a classic example.

Comparators such as the LM393 and LM339 also have inputs which can withstand the full power supply votlage rating, irrespective of the actual supply voltage.

[Dabbot]

Latch up protection is generally done by surrounding each transistor with an insulating trench.

For your query on this specific chip, I quote from the PDF you linked, 2.2 Input Common Mode Range:

Note that when used as a comparator, only one input needs to be within the common mode range. The other input can be above the common mode range or above VCC and the output will be the expected VOH level (for Vin+ > Vin–) or VOL level (for Vin– > Vin+).

[exe]

The LM358 has special input transistors which can withstand up to 32V in the reverse direction. The common mode range isn't anything special, from just a little below the negative supply to the positive supply minus 1.5V.

Yeah, but below in the note it says "either or both inputs can go to 32 V without damage (26 V for LM2904), independent of the magnitude of V+", which I find unusual.

[Zero999]

The LM358 has special input transistors which can withstand up to 32V in the reverse direction. The common mode range isn't anything special, from just a little below the negative supply to the positive supply minus 1.5V.

Yeah, but below in the note it says "either or both inputs can go to 32 V without damage (26 V for LM2904), independent of the magnitude of V+", which I find unusual.

Yes, it is unusual. Regarding your query relating to latch-up: that's a property unique to CMOS ICs and the LM358 is bipolar, so it's irrelevant.

[magic]

Most inputs are connected to bases or gates of internal transistors so taking them more than a few hundred mV beyond the rails wouldn't make sense anyway.

LM358 is no exception, it doesn't even work with the inputs slightly below VCC.

[David Hess]

The LM358 is built on a junction isolated process so the individual elements are isolated from each other in "tubs" of n-type material inside the p-type substrate.  These tubs act as diodes and isolate each element as long as the tubs are reverse biased which requires the substrate to be more negative than any other node.  Usually the substrate is connected to the most negative pin but it may be brought out as a separate connection or biased negative with a charge pump.

So the diode to the negative supply is inherent to how the LM358 is constructed and this applies to all junction isolated integrated circuit processes.  If a pin is brought more negative than the substrate, then it injects carriers into the substrate which can result in all kinds of weird behavior including destructive latchup in some cases.

[magic]

That's only true because the inputs are PNP and the N tub is the base so there is indeed a parasitic diode between the input pin and ground.

NE5532 has NPN inputs and the N tub is the collector. Applying negative voltage to the input pin would only reverse bias the BC and BE junctions. Additional diodes to the substrate are therefore included to prevent BE breakdown in the input NPNs. OTOH there is no diode to VCC besides a somewhat high resistance parasitic one (in the collector load resistors), which is easily confirmed with a DMM in diode mode (wait, actually it's two diodes in series, because the current first passes through the BC junction).

LM317 has one pin which is exactly 1.25V below the substrate.