LM339 Comparator input limits?

The 339, like the LM324, has some sneak paths that squirt current all around the chip if you pull any input much below ground. One such pulled-low input, even a tiny current, can freak out all four channels in weird ways. It's better to not allow any input to go negative. The severity seems to vary among vendors, with the classic National part being really, really bad.

National was the worst eh? That explains why they don't work so well in a HP 8116A function generator's reset circuit. Which manufacturer's part is the least sensitive?

I don't know. JT claimed that the Motorola parts were better, but I don't know if they can still be had... perhaps from OnSemi. There are probably drop-ins from someone else, with a different process. A honkin' schottky clamp diode will probably help.

When I last used them in production, late '80's, Motorola's 339 was best because they had split up the bias; but that same bias structure implemented in their 324 had the worst cross-over distortion. ...Jim Thompson

How about using a diode across the input to limit the negative voltage? The comparator input reads about 1000 using a DMM to check diode drops. A silicon diode reads about 800, so it seems a diode across the comparator input to ground would absorb all the reverse current?

-Bill

As I recall, a silicon PN diode isn't good enough. The chip screws up at lower voltages than you'd expect from the ESD diode drop. Both the Fairchild and National data sheets rate the abs max input voltage range as -0.3 to +36. That -0.3 number seems to be serious.

A power schottky diode might work, if it keeps the negative swing below -0.3.

The "quality" way to do it, avoiding leaky Schottky's or finding a Germanium diode, is an active clamp, as I demonstrated several years ago in a 4-20mA circuit for Spehro...

The 324/339 input stage architecture is a Darlington differential pair made up of lateral PNP's, with an NPN mirror load on the inner pair of PNP's.

This "screws up" if driven below ground by a Vbe, because a lateral device doesn't know its emitter from its collector, never mind a hole in the ground.

Thus a positive input pulled below ground causes a phase reversal and the output rises instead of continuing to fall.

Driven hard, this could inject current into the substrate and upset things, but I've never seen that... rational designers limit current ;-)

Another effect is what I think Larkin confuses with the front-end below ground... take inputs too high and lateral PNP mirrors saturate. Cheapy approaches, in certain brands, used split collector arrangements rather than separate-well devices, so one mirror saturating screws all sections of the device. ...Jim Thompson

VCC | - ^ high speed diode + lowest R possible | ___ | +------+|___|+--+---------------+ To Comp input Unknown signal |+ | + - ^ high speed diode | === GND (created by AACircuit v1.28.6 beta 04/19/05

I am just throwing shit again the wall !

Jamie

It's not a speed issue... an LM339 is a micro-second range slug-speed device. To avoid the problem, you need a clamp that holds at a voltage less than a silicon Vbe forward.

AND: Your diode to VCC is useless, you will have already exceeded the allowed common-mode range (about 1.5V BELOW VCC) before it conducts. ...Jim Thompson

How so? the reference isn't passing through 2 diodes, it is only doing one diode, depending on if it's a low or high signal coming in?

The idea is to have the diodes only conduct when the input exceeds either way by a margin of 0.7 volts.?

You being the semi engineer, I most likely am over looking something very obvious to you. I have used that before and it has worked for me. Maybe I got lucky ?

Jamie

After I pressed "Send" I realized that an LM339 will be OK CM-wise if the reference is 1.5V below rail.

Not so for the LM324...presuming you have feedback ;-)

You don't need a diode in the positive direction unless you talking like +30V. The positive direction applies _reverse_ potentials to the lateral junctions... they breakdown at the same voltage as the NPN BVcbo, 30-35V

As I pointed out above, the positive-direction diode isn't needed. In the negative direction, most discrete diode Vbe's are smaller than that of the 339 silicon, so silicon diodes are OK... most of the time. ...Jim Thompson

His "lowest R possible" would be zero.