Keithley 2001 repair attempt -- zener question

[essele]

Hi,

I've just got my hands on a Keithley 2001 reported as failing a load of tests and having a burning smell!

The burning smell appears to have been U316 (LM339D) which had literally blown itself apart. A new LM339, some careful carbon scraping and a couple of small bridge wires and that problem seems to be solved.

From a test standpoint it fails 301.2 and then a large number of the follow-on 300 tests -> 301.2, 302.2 303.2, 304.2-7, 305.2, 307.2, 307.3 and 308.1

On DCV it shows about 340V with no input connected ... connecting a voltage source impacts the voltage by what appears to be the correct amount, i.e a 1V input increases the display to 341V and it seems to be pretty accurate.

DCI appears to work perfectly, so I know the ADC is ok.

I have just taken out the Analog board to replace all the caps (which looked and tested perfectly fine, but I've ordered replacements anyway) and I noticed some strange behaviour on VR302 and VR307, both should be 6.4V temperature compensated zeners (1N4577) ... I wasn't seeing a forward voltage drop, so I've taken both out of circuit and when testing them on my DMM7510 (at 1mA) it shows about 6.2V in the reverse direction, but in excess of 10V in the forward direction ... is there any reason why this might be the case? Normally I'd just suspect a faulty zener, but the reverse voltage is correct, and both zeners are showing the same behaviour.

They are both on a related part of the schematic, both seem to be involved in generating +8VF and -8VF ... which probably not coincidentally are also the rails that feed U316.

But, all of the voltages seemed to be normal when tested, but I don't really understand the bootstrap part of it, so I am very likely missing something.

Any help appreciated,

Lee.

[Kleinstein]

The temperature compensated zeners are a zener (~5.6 V in this case) and a normal (can also be a zener, but it can vary) diode in series. So in the reverse direction one would have the normal diode in reverse. So the behavior with > 10 V in the other direction can be normal.
It is rather strange to have relatively expensive temperature compensated zener diodes for VR302 / 307, as this is just for a BS supply voltage and with a emitter follower - so there is a  +2mV/K anyway.


The in put reading depends on the voltage range. A reading of 340 V suggests that the input divider is used and something is off there, like an input bias current (e.g. some 34 µA) that is too high.
Chances are other ranges, like the 20 V or 2 V range may give better readings with at least a low impedance source.

A blown up chip could be from high current from the input (in case of the U316 this could be the Ohms part portection) or just a latch up causing a short in the supply and this way draw a lot of current.
So it would be woth checking the protection part (e.g. Q321,Q323).

[essele]

Thanks Kleinstein, that's useful -- I think I'm slowly understanding this a bit better.

So U335, the input buffer, drives the mid-point of the +8VF and -8VF rails in a bootstrap fashion with Q330 ensuring the output accurately matches the input.

U341 keeps BSCOM at the same voltage as the input to the input buffer, I'm assuming so you can then use BSCOM and -8VF as the on/off gate drive for the various JFETs which control the input path -- selected by some of the analog switches (U308 for /DIVIDER for example.)

From memory, I was seeing 0V at the bottom end of the 100k part of the 9.9M/100k resistor, but then I was seeing somewhere around 3.5V on the upper part (and at the top of the 9.9M) which makes me think BSCOM is sitting at 3.5V which seems wrong to me for no input. My best guess at this point is that one one of the analog switches or JFETs is on when it shouldn't be.

I wish I hadn't taken the caps out now -- but the new ones are due on Tuesday, so not long to wait!

[essele]

Ok, caps replaced, didn't fix anything.

But Q331 seemed to have a problem, the gate voltage was down at 3.6V rather than closer to 8V as expected, I removed it and it does appear faulty -- my tester showed it as two diodes.

With it removed (I think it's only related to 4W-Ohm measurement) all the 301.x and 302.x tests pass, but there is an -84mV offset on voltage measurement.

I suspect Q525 as the gate is driven by -7.8V (/DIVIDER) through 200k, but is being dragged down to -4.5V at the gate, so I suspect there is some leakage (16uA?) which is causing the -84mV offset. On the 200V range Q525 is not driven and the offset disappears, and the reading seems to be spot on.

Q331 is a "selected TG166", from other Keithley kit I can see it also referenced as a J270 or J175, so I've found J175's at Farnell so have a couple on order.

Q525 is a "selected J210 (TG-167-1)", the J210 does seem to be available from Digikey, but I've gone from a J113 for the time being so I can validate my theory -- the various gate current specs seem a lot worse!

Are there any better options for these two JFET's? Or any other sources?

Thanks,

Lee.

[Kleinstein]

The gate current specs are mainly test limits, so up to what limit they test the parts. The actual leakage is usually way better.
The J113 or PN4393  are classical choices for analog switches. One could try a smaller chip like J201/J202, but I would not consider it that critical. If one has a suitable meter, one could test the parts up front for low leakage.
A j210 is a strange choice, and may be a typo - with only 25 V rating this would be too low.

[essele]

Ok, with a J175 and a J113 just pushed (not soldered) into position everything works perfectly, and all the test pass ... 

So the question is ... are they good enough, or should I be looking for something more? I think I'm going to hang on to this meter, probably replacing my trusty go-to everyday 2000.

For Q331, the J175 ... there was actually a J177 in place, which seems to be a lower Vgs(off) and a higher RDS(on) .. neither of which I suspect will matter in this use case?? I'm assuming the "selected" in this case is likely to be the Igss ... which is better in a J270, but as you suggest probably needs to be measured?

For Q525, the J210 ... there was actually a J2673 which seems to be very obsolete and I can't find a datasheet. I'm assuming the "selected" bit of this is also the Igss, so would I be better off with something like a PN4117 with a much better Igss (1pa or 10pa vs. 1na for the J113) or would that bring other problems?

In both cases I'm assuming Igss (and Id(off) for the N channel) are the potential problems as they will introduce a slight offset into the reading (as demonstrated by the failures) ... presumably this can be calibrated away, but they were "selected" for a reason?

On the voltage front, I'm not sure I understand why 25V wouldn't be enough, all of this section seems to be +/-6V relative to BSCOM, although I may have that wrong.

Many thanks for all the help!

Lee.

EDIT: Just realised there is no continuity or diode test, so this won't replace the 2000, but will supplement!

[Kleinstein]

For Q525 a PN4117 should be OK too.  The downside is the rather high R_on, but with anyway 100 K from the divider this should not be an issue. So I don't see a reason the PN4117 should not work there.
The low capacitance should not hurt.

Nomally there would no be much voltage seen by Q525, but there could be some higher voltage with transients. It is hard to tell what the worst case is for transients. I did not realize that the low side of the divider is switched. So what I though as a worst case is not that bad.

If the J175 is a the high end with the threshold, it should be an issue do to more leakage. Chances are the threshold would not be that high and thing would be fine.

The selection is likely for the leakage current. I don't the the threshold to be so critical to be an issue. With other uses the selection could also be for the threshold.

The FET leakage current is part of the input bias current. There is some compensation for different FETs and clamping diodes. After an exchange / repair one should check the input bias. A rather conventient way is using a capacitor in the 1-10 nF range at the input and than look at the drift rate with different voltages. AFAIK there is not adjustment / trim for the input current.  I don't think the self test can do a full test of the input current, especially not with different voltages.

[essele]

Thanks Kleinstein.

I’ve run through a few tests and it seems to be very similar behaviour to my DMM7510, so I’m reading that as good — a difficult test to get consistent results on, seems to be very sensitive to anything going on in the room!

So, I’ve gone with the two FETs I had, everything seems to be working very nicely.

An enjoyable repair — thanks for all the help.

Lee.