Keithley 2000 Fails 101.2 test

[Maxis]

Dear Forum Members,

Not so long ago I've acquired 2 well used up Keithleys 2000. Both sold as is for repair. Both had plenty of errors, but finally, I've recapped both, cleaned the board and the terminal switch (no residue, no traces of fluid), etc.
Now, both multimeters are SPOT ON. I've tested both with the calibrator (fluke 5100B) and the readings against Keithley 2002 (borrowed).

The problem is => BOTH Keithleys fail 101.2 TEST (7V ADC IN test).

Both multimeters have 6.90V at the output of the LM399 (quite used up). I've replaced one of the references by the new LM399. Readings are smaller (the reference generates higher voltage), but the autotest error is still there. However, now on the test point TP104 in the test 101.2 I can measure 6.9980V from the new reference. It seems that the reference is not a problem. The problem is somewhere else.

Now I put back the original reference (again the multimeter became spot on), also checked the voltage difference between the reference output U141 and the A/D_IN (TP104) - there is no difference....

Verified the signals on the ADQN[3 : 0] and ADQ[3 : 0] latched by 74HC175(produced by the CPLD Altera), all seem fine, clean, no crosstalk.

The firmware is rev A07 on one unit and rev A02 on the other (very very old).

One more thing - noticed the burn marks on Zener VR109, but the output of the Q118 regulator is 17V (dynamically varies +/- 20mV under the operation)....

Have you came across such problem?

Where else should I look?

How can I request the report in counts of the actual value during the tests 101.2 and 101.1. They say that the expected value is (101.2) - (101.1) = 76275970 counts

When Keithley says that the limit is 1800000, does this mean that the difference between 101.1 and 101.2 should be +/- 1800000, or +/- 1800000/2?

Thank you for your help!!!!

[Kleinstein]

VR109 and Q118 are part of the current source for resistance measurement. This area is a bit prone to damage from over-voltage.
A possible fault (U123 or U133A) could be loading of the 7 V reference level and thus the 7 V to +-14 V stage and this way causeing a wrong ADC gain that the error 101.2 complains about.

Getting 6.9 V for the LM399 is OK. A point to check could be the +-14 V ref. voltages used for the ADC. These should be pretty symmetric and +-2 x the reference voltage with not too much difference (likely 0.1% range or better).

Another easy point to check is the current in the ohms modes.

[Maxis]

Thank you for your suggestions,

The voltages on the resistor array R271 (TF-245) are the following (measured against pin 6 or TP102 AGND):

PIN 1 (+14v) = +13.75550V
PIN 2 (+7v)   =   +6.90317V
PIN 3 (-14V) = -13.69023V
PIN 4 (+14V) = +13.75550V
PIN 5 (-14V) = -13.69023V
PIN 6 (AGND) = 0V
PIN 15 (0V) = 0V (a few microvolts)
PIN 16 (+13.3V) = +13.06810V

VREF = +6.90317V


It seems that you are right! Is there something with the resistor array ? Is that possible? The reverse engineered schematics suggest the exact x2 relationship (and the resistor values).

But look, 6.90317 * 2  = 13.80634V
However, the real value is = 13.75550V => 0.5% ERROR !!!!!

What's the problem? I'll check the Q129/Q130/U139/VR112/VR113, but it seems the resistor array is that it?

How reliable are these arrays?

On the other hand, since the opamps are quite close to the reference, it seems that they run hot (and these two units clocked quite a lot of hours)....
I'll check and try replacing Zeners and the opamps (including the transistors). Another possibility is to unsolder the array and verify the values.

[Kleinstein]

There are a few reports around of the TF245 array failing. However the failure mode is more like a lost connection (e.g. bad bond wire) and not drift. I would definitely not try to unsolder the array - there is a chance it could fail and I don't see a large chance that the resistors are at fault.

The difference from 13.8 to 13.75 V is only some 0.5%, so not that dramatic, though more than I would expect. I don't know the specs on the TF245, but to really make sense of the 1/128 ref current ratio for the fine slope it would make sense to have a more accurate value to start with.

My suspect is more the OP-amp U123, especially in a scenario of some ESD or similar overvoltage causing a latchup that could have damaged the chip and stressed the zener VR109. A stressed chip could have a higher than normal input bias, that than can effect the reference amplification as an added offset there.

Another possible point to check would be if the test actualle gets at least roughtly 7 V to the ADC. So check TP105 with a scope during this test. AFAIK there is a way to run the self tests manual,  one at a time. This could exclude a gross error with U133 or the shift registers.

[Maxis]

Hello! Thank you for the tips!

I also thought that 0.5% error is a way too high..... Did a quick LTSpice simulation, the reference divider/multiplier must be spot on. Only depends of the leakage currents of the opamps powered directly from the TF245 resistor array.
Since I don't have AD706 U123 under the hand to replace, I've played around in the step-by-step test mode to see how the U133 switches will commutate different sources to the U123 inputs and whether I can see any drop when the array output 13.3v or 7.0V is connected or disconnected to the U123.
Actually I don't see any difference in the voltage divider outputs with or without the U123 load. I.e. U123 seem not to affect the output precision.
Hence, the question is - what does...

U139: checked the voltage difference between the differential pins - around 10-20uV.

Then, whatever consumes from 7V will be compensated by rising the +14V (this is the case) and 13.3V as the result for U139B. Since U139A is the inverting amplifier with the gain of x2, it should compensate for any load on the -14V net. However, it seems that -14V, 14V and 13.3V are wrong. I guess this is due to the excessive leakage of the U139 inputs (also it runs very hot).... Another possible explanation is that the gain of the Q130 and Q129 has degraded and therefore U139 tries to pump significantly more current into the transistors to converge the differential at its inputs to zero.

According to some sources, TF245 is the super precise thingy - better than 0.01%, so then unless degraded, it's not the root cause, then. Assuming that the resistor array is to 10ppm exact, such a discrepancy between the +7V and -14V suggests the leakage current of -6uA at the inputs of the U139A!

I'll try replacing U139/VR113/VR112/Q129 and Q130 altogether.

Thank you for your help!

[Jeiiner6]

I have a fluke 5100b calibrator, which shows "error6", how can I fix it?

[Maxis]

We need to discuss it in a separate topic. Quick tip - check the ACK signal from all the installed modules. Absence of it will render ERROR 6

[Kleinstein]

AFAIK the K2000 is some way measures the ratio of the small slope to the main slope. At least there are some extra measurements in between and the slope ratio would be the main part to check. With a measured / corrected small slope part the exact resistor ratio does not matter. One might think that symmetry of the +-14 V ref. is critical, but an error there only leads to an offset and this is corrected anyway, even without the small slope correction.

U139 may effect the reference amplification through it's bias current and there are a few reports of broken parts. The LT1124 is a somewhat strange choice - one may be able to substiture with an LT1013 that has the same, somewhat inusual pin-out. Still I don't think this part should be the problem.

[Maxis]

Hello Kleinstein,

It seems that the asymmetry in the full scale +/-14V causes the ADC to fail..... Also there is some sort of a noise coming from the VR109 Zener (I'm wondering whether it can have any contribution to the ADC algorithm convergence). So, I'll try to replace the LT1124, AD706, transistors and the 17V Zener. Next try also to replace the analog switch DG444. If no result, then unfortunately I'll have to search for the new TF-245.

What's interesting is that both multimeters have wrong readings around TF-245 and both clocked lots of hours. So, it's either the resistor array or the OPAMP aging of some sort (under the harsh operating conditions).

[Kleinstein]

VR109 is only relevant for the ohms mode and normally should not contribute to the DMM noise, as it is only controlling the OP-amp supply. There should be not much noise around the supply - higher than normal noise could be related to things like a bad capacitor / regulator.
I don't see a direct way how VR109 and U123 (AD706) should effect the 101.2 test - this would be a different area. How bad is VR109 looking ? A possible way to stress VR109 would be running the DMM with too high a mains voltage. Such an accident may have damaged other parts, though I don't see many parts (except the voltage regulation) using the unregulated +-20 V.

Assymmetry in the +-14 V may be accepteable to quite a degree and the resulting error would be more a DNL problem mainly with short integration time. A DNL test may be a thing to test, before messing with the hardware.  A histogram test with a rather short integration and non AZ mode may be a first try, though it may miss the relevant points. Another DNL test woult be using the 10 V range and low PLC setting to record a large capacitor to slowly discharge from a low votlage (e.g. 50 mV to 5 mV) ideally this would give a smooth curve - a DNL problem form the ref ratios may be visible as repeating breaks.
The effect of ref. Asymmetry on the ADC gain (which the 101.2 error complains about) would be only minor.

If the rest of the DMM works one may even live with that self test error message and consider it a false positive ( maybe messed up internal cal constants ?). Getting 2 DMMs with the same, relatively rare error is a bit odd.

Noise wise I see one weak point that one may improve on: the 7 V to +-14 V amplification adds a bit noise from the not so great LF noise and current noise of the LT1124. Another point is adding filtering for the higher frequency (e.g. 1 -100 kHz range) reference noise that may also add to the ADC noise. A little more filtering at the ref. amplification is not too complicated. This may still be a minor noise part.

[Maxis]

Hello,

I don't really know the expected tolerances for the +/-14V and +7V mismatch. I can only assume that if U123 developed some internal anomaly and also it's connected during the tests (indeed during 101.1 and 101.2 it's connected to 13.3V and 7V), it can overload the reference generator by creating the unwanted leakage current drain/source. That's the theory, but it seems not to be the case (I've tried to go over the various step-by-step tests observing the connection via U133 to the various reference voltages). I can see that U123 doesn't load anything.
Nevertheless VR109 produces some soft of the walk noise, which results at the Q118 collector voltage fluctuation (can it be anyhow related)?

You theory regarding higher than normal unit operating voltage is very probable, since both instruments came from Japan. So, I don't know how much reliability data was gathered during production phase. Likely both multimeters were exposed to the elevated voltage, which results in the higher dissipation in Zeners and also the linear regulators including the transistors. These are very similar (the production date differs) and were used in the same test bench (clocked same number of hours).

Both multimeters are spot on and only have this error.

Therefore, I can only assume:
1) OLD FW, which doesn't tolerate the asymmetry => fix is to get the FW rev 19 and re-calibrate
2) Something is worn up, manifesting the excessive leakage and therefore affecting the references outputs.

My problem is that FW REV 02 or REV 07 doesn't produce more diagnostics (doesn't tell what it actually measures)... If I can read somehow the obtained value during the 101.1 and 101.2 tests, I can judge and correlate the offsets to whether the FW is getting closer to or farther from the value to be accepted. I've connected the RS232 and the unit is mute during the test (is it normal)?

First to be changed is the LT1124 U139 and Q129+Q130...

Second step would be to upgrade to the new FW (keeping the old FW and the calibration data EEPROM backed up). And hope for the best.

I found the repair article where the owner of the failing K 2000 contacted Tek at Oregon and they sent him the new TF-245. But when he measured the new part, the resistor values were quite surprising (some spot on, but some with a large offset).... Nevertheless, after changing the TF-245 everything started working!

Currently I can only say:
1) This is not the reference voltage (tried with the 7V reference) - no change.
2) This is not the corruption of the FW or the calibration data....
3) This is not the problem of the power supplies

Thank you for all your help!

[Kleinstein]

The collector or Q118 is a unregulated 20 V (nominal) supply. So mains variations can produce some variations there and this is totally normal. Similar the stabilization with VR109 is not perfect with a relatively low series resistor and a cheap zener diode that may not be super low resistance. A little variation there should not be an issue. The main cricital point may be no getting enough voltage, so that the 13.3 V case could reach the AD706 common moder limits. 17 V should be high enough though.

A higher than normal supply may explain thermal stress to VR109, but this does not look like it did more than cosmetic damage.
I would not expect any change with exchanging VR109 / Q118.
Similar it is unlikely that the LT1124 is slightly bad in both meters.


One could check the effect of the AD706 on the reference, by looking at the 14 V reference, when switching between voltage and ohms mode (e.g. with a shorted input).

The reports of failing TF245 I have seen so far were with a more definitive damage ( open pins), not a more gradual drift. AFAIR getting replacemanent was tricky / expensive.
In theory one could first try adding parallel resistance (solder away from the TF245 chip) to shift the references.

A linearity test may be good idea. If the linearity is OK the meter either compensates for the reference ratio (which is likely) or the measured ratio is normal (e.g. included in a factory calibration).

In theory one could sniff the signal send between the analog(ADC) and digital (output side) part of the meter. Not sure if the protcol is already analyzed / known though.

[Maxis]

The collector or Q118 is a unregulated 20 V (nominal) supply.

Sorry my typo - wanted to say the emitter voltage fluctuation. It's like a walk noise. Low frequency stepping..... (not related to the driving of the AD706 and its output current). Everything is static during the test, but VR109 and therefore the positive voltage supply of AD706 fluctuates.

OK, first I'll try replacing the LT1124 when it will arrive. If it will not help, will try to balance out the voltages by adding the external resistances (not directly to the exterior of the TF-245 package).
If after the two above I'll have ZIP, then try to make the FW upgrade and recalibrate (complete the calibration even roughly just to see whether there is any difference to the failure contribution). Then roll back the FW changes and calibration....

Also I'll simply remove AD706, add the load resistance and see if the walk noise is still there. If YES, then I'll try replacing VR109 and the transistor in series regulator.


Under all circumstances I'm not going to unsolder TF-245, even though I'm more than equipped for this kind of work (low temp unsoldering).

[Kleinstein]

Replacing VR109 would be a simple task - it should be a normal cheap zener, seems to be some 18 V. If one does not have one at hand 2x9.1 V in series should also do and may even work better (lower TC and maybe less series resistance).  Still I see no easy way how VR109 would effect the voltage readings that much.

I would not fully exclude corruption of calibration data (the expected value for the 7 V reading). This could be due to a faulty calibration attempt or such and possibly the same procedure used with both meters.
So there is a chance that just the expected value is wrong and the error message would be a false positive.

As the readings are still corret chances are there was no major change with the references after the last adjustment.

[Maxis]

I would not fully exclude corruption of calibration data (the expected value for the 7 V reading). This could be due to a faulty calibration attempt or such and possibly the same procedure used with both meters.
So there is a chance that just the expected value is wrong and the error message would be a false positive.

This is a very good point! I also thought that this might be a plausible theory. Both multimeters are from the same organization. After examining the stickers I can see that the calibration wasn't done by Keithley Japan. It was some sort of a generic lease/repair/calibration service company: Toyo CO LTD. The last calibration was done in 2015 same month, same company for both instruments (so, then something developed from 2015 to 2017 when the units were screened again and then ditched. Maybe during the last calibration attempt something became misaligned (like a technician) and they ditched both units. When I got these a few month ago, the electrolytic capacitors started leaking, but haven't destroyed anything yet. I removed all the traces of the spilled over electrolyte. Replaced electrolytic capacitors by the very same type and manufacturer, cleaned up the terminal commutator switch (many errors attributed to the bad switch). After this welcome treatment both units came up with a single fault 101.2 (all other faults were gone).

Therefore if I'll have zero progress with LT1124 replacement, then I'll try upgrading the FW and attempting a rough calibration (just to converge) with my Fluke 5101B calibrator (not precise enough for 6.5 digit). If after this step the error will disappear, then this is the calibration problem and I can safely send both units for re-calibration attempt to calibration service.

Otherwise, if after the rough calibration attempt and FW 19 I get NOP, I'll roll back to the original FW and calibration data, add the resistances to achieve +/-2xREF and see if the error will disappear.

Actually both multimeters are quite spot on and don't drift..... But out of curiosity I'd like to get at least the understanding of where the FAULT 101.2 comes from.


Thank you for your helpful suggestions again!

[Maxis]

Short update.

Before modifying anything, run the DC calibration. It successfully completed. The error reduced vs 8 1/2 multimeter. But, the FAULT101.2 persists unfortunately, no other errors are detected...

Next step - change the LT1224.

[Maxis]

Short update.... got long
In a course of Keithley 2000 recalibration, the Fluke 5101b broke! Had to identify the fault and fix it. That was extremely difficult. Now 5101b is back (but still I have to re-calibrate it later).

However, unfortunately after the replacement of the LT1124, the FAULT 101.2 is still there.....

Next step - verify the hFE of Q129 and Q130, since the replaced LT1124 runs extremely hot (around 65C). Only one explanation - degraded transistor Beta (from running hot for a long long time)? Is that possible? Try to verify VR113 and VR112 too.
If the transistor and zeners are fine then unfortunately, IMHO, the problem is in the TF-245 degradation.

[Kleinstein]

The LT1124 naturally run rather hot due to the relatively high speed and normal power consumption. So 65 C may be normal. The references don't provide much current and thus even quite some degradation of Q129 / Q130 would not cause much more current. The load to the OP is mainly set by the resistors and the Zener and transistor combination is more there to get a larger output range, not so much isolating the load.
A possible cause for excessive power use would be if the circuit oscillates (e.g. due to a lost capacitor).

[Maxis]

Checked both MMBT3904 and MMBT3906 + Zeners - all fine.

Finally, I've ordered two TF-245 for the both multimeters from Tektronix. This part is still active and available.

Once received, I'll measure the each individual one and make a replacement.

We will see whether the resistor assembly aging is the direct contributor to the FAILURE 101.2. Hopefully it is (otherwise my search will continue).

[vsmith]

Hi Maxis
Have you made any further progress with this error (101.2)?
I have a 2000 with the same issue as what you have described. Did you change the R271 resistor network, and did it help the error?

Thanks!

Vic

[David Hess]

Next step - verify the hFE of Q129 and Q130, since the replaced LT1124 runs extremely hot (around 65C). Only one explanation - degraded transistor Beta (from running hot for a long long time)? Is that possible?

65C is not that hot, although I do not remember the LT1124 running that hot normally.  Transistor hfe increases with temperature, but the LT1124 has input bias current cancellation.

It is not difficult to measure the input bias current of an operational amplifier, which would confirm subtle ESD damage.

[vsmith]

Merry Christmas everyone! (and Happy Holidays for those who don't observe Christmas!)

I had some success with my problem 2000, and wanted to share my findings with the board.

After trying to eliminate any obvious causes for errors 101.2 and also 304.1, I was left looking at the precision resistor network R217 which is used (among other things) to set the gain in the U166 preamp to the A/D circuits. Since 304.1 uses the x100 mode of the A/D Mux and preamp, and R217 supplies the gain setting resistances for that mode, I compared some in circuit resistance measurements for R217 between my problem unit and another working 2000 I (fortunately) had. I found the following:

pins    nominal    "bad unit"    "good unit"
6-9    1K             1.002K        .954K
9-7    9K             8.570K        8.263K
7-8    90K           16.71K        16.40K

The discrepancies between these in circuit readings emboldened me enough to remove the R217 from the "bad" unit and remeasure those pins. I discovered 9-7 reading 9.364K which seemed way out for a "precision" network. A "parts" unit I have donated it's R271, which read 8.997K between 9-7. Swapping the new R271 into the "bad" unit gave a passed all self tests result.

So, yes, R217 can (with no visible signs of damage either to it or to the inputs or power supplies in the unit) change and cause errors in tests which check the a/d preamp stage gain.

Did I read that someone in the forum was able to order this part (R217) from Tektronix/Keithley?

Maxis : would love to hear from you regarding your unit!

Vic

[vsmith]

Folks:

I need to correct my previous post. The precision resistor network (which I changed to resolve the self test errors) is designated R271, NOT R217.

Thanks

[vsmith]

And a final note, as of this date (12-28-23) The resistor network R271 (Part No. TF-245) is available from Tektronix (Keithley) for $31.00 ea. Pricey, but unobtanium otherwise. This part is also used in the model 2010, and for all I know, in other DMMs designed at Keithley in the same era.

[coromonadalix]

and eventually this part will get obsolete   i dont know  how much they have in stock ...