LM311N abuse

[bdunham7]

Motivated mostly by wanting to clean up a bit, I retrieved a couple of old Fluke 8800A DMMs from my shelves and decided to fix or scrap them.  The result from the first one  (they have the same symptoms) is a bad LM311N comparator in the dual-slope integration circuit.  I can eventually get the LM311N, or if I want to spend big bux, the LM111J-8, but there seems to be an issue with the circuit design.

They use +/-18V supplies with the open emitter tied to V_cc, the negative input set to near 0V and the positive input fed a ~+10V signal at the zero crossing point by an op-amp.  The collector output is pulled "up" to -13V, since the 5V logic on these is run from -18 to -13V, which is a common setup to allow the logic to operate N-J-Fets as analog switches. 

Anyway, my poor LM311N is running at the 'absolute max' supply voltage of 36 volts total and the positive input is going to 28 volts above V_cc.  These meters are old and have lasted this long, so if my only option is to just put LM311Ns back in, that's what I'll do.  Does anyone know of a reasonable substitute for this, something perhaps with a 40V supply limit?  The open emitter is not needed since the circuit just ties it to -18V anyway.  The open collector needs to sink about 2.2mA.  And the 200ns response time is probably needed since the meter is doing 20,000 counts with a 100ms integration period.

[moffy]

If they have lasted so long, just go with replacing the LM311, otherwise you'd probably need a small redesign, but do you want to exert the effort for something you say already works?

[Kleinstein]

I don't know a good substitute and newer parts are more like lower voltage, not higher. The LM311 is still a standard part and not expensive. So if they last some 40 years it looks OK.

I would not expect the input voltage at the noninverting input to go all the way to +10 V: there is a transistors (Q62) used as a zener that should limit the voltage to more like 6 - 8 V.

They tied the balance pins to the positive supply and thus seem to want high speed.

[David Hess]

The LT1011 is an improved replacement for the LM311 but it has the same supply voltage limitations.  The inputs can be safely raised to 40 volts above the *negative* supply.

[mawyatt]

Recall there were HV versions of certain analog chips, but can't remember which. Maybe this was originally a HV LM311 which was replaced with a regular LM311N before you acquired the 8800, or maybe Fluke had selected the LM311 for HV use (but would think this would be a special part number and identified as such on the schematic).

Agree with others, just use a regular LM311 and if it fails later replace with another. Maybe get a couple LM311s and tape the replacement inside on the case with a note for later use just in case!!

Best,

[bdunham7]

I don't know a good substitute and newer parts are more like lower voltage, not higher...
...They tied the balance pins to the positive supply and thus seem to want high speed.

Yes, the LM311P doesn't seem to work, for example.  The 200ns rise time and 18V/µs slew rate seem to be important.

I would not expect the input voltage at the noninverting input to go all the way to +10 V: there is a transistors (Q62) used as a zener that should limit the voltage to more like 6 - 8 V.

Now that is interesting, thank you!  I hadn't really analyzed that part of the circuit that thoroughly as I was working backwards from TP17 and found that U6 (the LM311N) had inputs that should have caused it to go low and it didn't.  I didn't even look at Q62 and consider that some other malfunction may be causing the LM311N input to be damaged.

I have a good 8800A, so I put the scope on it to compare.  I've posted an excerpt of the schematic for anyone following along.  CH1 = TP3, CH2 = TP2, CH3 = TP17, CH4 = R88-R89 junction (negative input of U6).  You can see that the good one charges C9 and then discharges it and TP17 goes 'low' right at the zero crossing point as designed.  The faulty one charges C9 and then discharges it past the zero point and beyond until it times out, TP17 never goes low.  TP3 only goes to the 6-7 volts, as  you stated, on the good meter but goes higher on the bad one.  This is all with the meter on the 2V range, with 1V input (actually 10V on the bad one, but it has a ranging issue as well so only 1 volt is getting to the ADC).

[bdunham7]

The LT1011 is an improved replacement for the LM311 but it has the same supply voltage limitations.  The inputs can be safely raised to 40 volts above the *negative* supply.

Thanks, I couldn't find that with a parametric search on Mouser for some reason.  I'd be happer if it was rated for a 40V supply as well, but otherwise it looks good.  Perhaps I'll put a socket in for this chip.

[Kleinstein]

If Q62 is not limiting the output of U5, the op-amp would run into saturation and the recovery time of U5 could cause problems before the LM311 even gets the signal.
I still don't see a damaging high voltage to the comparator

I would still replace Q62. The voltage looks like it is on the high side.  A transistor as kind of zener diode may show some aging effect. The top of the positive phase of U5 does not look flat as it should.
A odd behaving Q62 could also cause noise and transistors in zener mode are known to by noisy.

The signal to the LM311 sees some filtering from R86 and C16 with some 4 µs time constant. So from this side the delay / speed should not be that critical.
It makes absolute sense to limit the BW, as this also effects the noise of the ADC.

A socket for the LM311 makes definitely sense, just in case.

[David Hess]

Thanks, I couldn't find that with a parametric search on Mouser for some reason.  I'd be happer if it was rated for a 40V supply as well, but otherwise it looks good.  Perhaps I'll put a socket in for this chip.

Parts built on that old junction isolated NPN process could support operation to 44 volts, and companies like Tektronix did sometimes grade incoming parts like the 741 or 301A for operation at 40 or 44 volts themselves to save money.

If you have a random assortment of 311s, then you could do the same thing and grade for the ones with the highest breakdown voltage, but I suspect any 311 you buy will work fine and I would not spend that much effort without better reason.

[bdunham7]

I would still replace Q62. The voltage looks like it is on the high side.  A transistor as kind of zener diode may show some aging effect. The top of the positive phase of U5 does not look flat as it should.
A odd behaving Q62 could also cause noise and transistors in zener mode are known to by noisy.

I had attributed--without detailed analysis--the issues with the output of U5 to the fact that the integrator was going way above zero when in normal operation it would have stopped at zero.  Looking at it again that may be wrong, but in any case Q62 is a MPS6520 and the LM311N is still in production, so I may as well just order up a few of each.

[bdunham7]

After a bit of a hiatus, I've finally restarted some clean-up projects.  I got my parts and installed them in the meters with mixed results. 

In both meters I replaced Q62 and U6 with a new Central Semi MPS6520 and an NOS 311N from All Electronics.  Meter #1 now integrates properly but has other issues so was set aside.  Meter #2 was unchanged and I don't remember how carefully I diagnosed this--it is likely I assumed for some reason that it had the same problem.  It doesn't, so that's a new rabbit hole.

I tested the old Q62 transistors to see if they were actually defective somehow.  Meter #2 did not have an MPS6520 installed and I have been unable to identify what was in there.  In the photo, it is the one with the "7503" marking.  I'm assuming that this one is actually not defective and probably the 311N from that meter wasn't defective either.  In a transistor tester, the MPS6520 from Meter #1 showed as a normal NPN with an hFE of 138 and the BE junction tested as a zener was 8.8V @ 100µA.  The "7503" from Meter #2 showed an hFE of 65 and a BE zener of 7.1V @ 100µA. 

I did a curve trace of both, the presumed-good "7503" is the blue trace and the "bad" MPS6520 is the orange.  I have not tested the new transistors at all yet. 

[moffy]

Lovely curve trace plots, the orange set do look degraded with respect to the blue traces, perhaps a degraded BE junction.

[David Hess]

The orange one is what I would call "soft".  I wonder what a leakage test would show.

The blue one does not look very good to me either, but that might be an artifact of the curve tracer which is using a 10k source resistance for the base drive instead of a current source.

What curve tracer hardware is that?  I guess with such a limited collector voltage it is one of those little semiconductor identifier things.

[bdunham7]

What curve tracer hardware is that?  I guess with such a limited collector voltage it is one of those little semiconductor identifier things.

It is the curve tracer add-on board for the Analog Discovery, which has a two-channel AWG and two-channel scope.  The board uses these, some relays to switch in resistors and some software.  It's new to me and I haven't really set it up or scaled it correctly.  The maximum V_CE is 10V and it does just use a resistor instead of a current source, but I think it still may be useful.  I'm trying to improve my troubleshooting skills and transistor malfunction understanding. I was curious how the reverse V_BE zener would correlate to other things, like this set of curves. 

As far as leakage goes, do you mean the BE junction reverse voltage leakage?  The zener voltage on the "bad" MPS6520 is actually higher than the other.  When I get some time again I'll measure a new MPS6520 and anything else you suggest.

[Kleinstein]

In the circuit Q62 is used as a diode and maybe including the zener breakdown, not as a normal transistor function. Too high a voltage in zener mode may be the problem.

[David Hess]

As far as leakage goes, do you mean the BE junction reverse voltage leakage?

I mean the leakage of both junctions.

The easy way to measure it takes just a power supply and a common multimeter.  If the multimeter is in voltage mode, then it operates like a current meter with a 10 megohm shunt resistance, so the measured current is 100 picoamps per millivolt.