Keithley 2000 input bias current (306.2, 306.4 failures) [Repaired]

[pigrew]

Hi,

I'm have a keithley 2000 exhibiting error 306.2 and 306.4. Based on a Keithley post, 306.2 means:

#Start#306#2
The +7V reference is switched to REFBOOT. This +7V,
which is used as the ohms circuit as a voltage
reference, is switched by U133 pins 2 to 3 (/7V control
line low) to op amp U123 unity gain buffer. +7V now appears
as REFBOOT and is routed through R272, Q109 (/HIOHM control
line is low), the 9.9M half of R117, R115, Q101, Q102, K101 pins 3
to 4 (/RESETK2 control line is high), R304, Q104 (LOV control
line is high) to U113 BUFCOM. To actually calculate
the bias current use the formula;

Bias current in pA = (306.2 reading - 306.1 reading) / 9.9E6

This unit passes when first turned on, but fails some minutes later. Therefore, it's likely not be be a cleaning issue?

The unit is only about five years old (though has a slightly dim display).

Doing the test (and holding shift to read the error value), 306.1=6.9500, 306.2=6.9485. This implies a bias current of about 150 pA. I couldn't find the spec, but I expect it's around the spec for the 2700 which is 75 pA.

Should I blindly go about replacing the above mentioned transistors/op-amps? Would replacing them upset the calibration, or is the calibration likely already upset? (I have access to a 3458A, but I don't have a proper calibrator)

Is NDFPD1N150CG the proper replacement for the 2SK1412? My unit was built with the K1412. But, even that FET isn't so available. Other suggestions? Maybe I should order some K1412 from Hong Kong.

TiN's unit on xdevs had the same errors, but I didn't see any mention of them being fixed.

[Kleinstein]

With many of the 2SKxxx parts they only have Kxxx or SKxxx printed. So the K1412 is an 2SK1412.

The NDFPD1N150CG might work, but it's about the uncertainty in the specs. A worst case part won't work well, a typical one likely would.

Even the original 2SK1412 might be screened (like for low leakage) parts - this is part of using the special manufacturers part numbers.

I am not so sure it must the that special FET that is failing, it could be also one of the JFETs failing.

I would suggest to do a test one the bias current in the normal (e.g. 1 V range) mode. So connect an external 10 M (or so resistor) and measure the voltage.

 An alternative method (may be better) is having a low leakage (e.g. polystyrene) capacitor in the 1-10 nF range and look at the drift rate.

As there is a circuit diagram available one might also want to check the voltage of the guard amplifier.

[Macbeth]

My K2000 is from an earlier era, it has the old FPGA and the green bodge wire on the MCU. It passes all tests with the older firmware [06], the newer stuff it fails on these 306.x tests.

You can easily get shut of this K2000 by downgrading the ROM's. It will pass all tests then.   

[pigrew]

I would suggest to do a test one the bias current in the normal (e.g. 1 V range) mode. So connect an external 10 M (or so resistor) and measure the voltage.

 An alternative method (may be better) is having a low leakage (e.g. polystyrene) capacitor in the 1-10 nF range and look at the drift rate.

As there is a circuit diagram available one might also want to check the voltage of the guard amplifier.

I found a 4.8 Mohm resistor (didn't find any others). At startup, the unit reads -60uV in both the 100mV and 1V ranges, so that is about 20 pA.

This is so aggravating. It's passing the test now. I'll have to wait a while longer.....

However, as it warms up (or maybe even intermittently), I've seen voltage of up to -400 uV, approx 80 pA. in 0.1, 1, and 10V ranges.

I wonder if I should shoot warm air at various components to see if that increases the leakage?

EDIT: WOW! Heating up U123 with low-flow 100C air greatly increased the leakage. I'll order one of them, and see what happens.

Should I also order organic-core solder? I've been using resin core, but it's a bit difficult to fully clean.

-Nathan

[Kleinstein]

Extra leakage also in the low DC modes indicates it is not the MOSFET to cause the leakage.
A bias current in the 20 pA range is about normal and good. 80 pA seems to be slightly too much thus just at the border of failing the self test.

It is kind of normal that leakage current of JFETs (and many other chips) increases with temperature. So the test with hot air can be misleading, though it's a good idea. To get good results from the heat test, I would limit the temperature rise for the parts to maybe 10-20 K. This would likely increase the leakage of that part by something like a factor of 2 to 5.

Besides a bad JFET, there could also be excessive heating of some parts not directly causing the leakage, thus too much heat to otherwise working parts. So it might be a good idea to also check the supply voltages - just in case.

U123 (AD706)  is an OP used in the Ohms circuit - so it should not contribute to leakage in the volts mode unless other parts are bad too. So it could be another part near by that also got hot.

[z01z]

Nice find with the test description, the description of these newer (>602) tests are missing from the repair manual.

I think one thing is missing from your test, the test case description talks about using a 7V voltage source while measuring the bias.
One way to measure this could be, if you had another meter with a known input resistance (e.g. my Fluke has 10M resistance on voltage setting - btw, this can also be measured with your 4.8M resistor by forming a voltage divider): connect the other meter in series with a battery while reading the voltage drop over the input 10M resistor, then calculate the current (in my case 1pA would read 10mV).
To test the 10V range, simply take a 9V battery (or a 2 cell LiPo) and connect the battery + through the other meter to the meter HI, and the battery - to the meter LO.
Some other methods are also described here. A few post back there's a link from Alex Nikitin, which might also interest you.

If I'd have to choose between cleaning the PCB and start changing parts, I'd definitely start with the cleaning.

[pigrew]

I replaced Q120, one of the switches in the ohms source which was used in the 306 test (the part that was temperature sensitive, it wasn't the amplifier that I previously mentioned) with a similar part (was SNJ132199, but used Fairchild MMBF4393).

After that swap, test 603 consistently passes. Using a 34401A as a 10M load (and cleaning the bananna sockets of the K2000), I read between 100 and 160 uV, which would translate to about 15 pA at V=0.  Using a 4.8 MOhm resistor, I calculate about 18 pA.

I still hear a bunch of switching noise (high-frequency superimposed mixed 60 Hz) when the case is open. I've replaced the electrolytics (probably not worth doing as the DMM is fairly new, but I've read the horror stories on this forum), but the buzz is still there. I guess I'll ignore it.

I also increased the display filament current a bit by putting 68 ohm 0805 resistors in parallel with each of the four 12.1 ohm resistors on the display board. I should have used 1206-size, but I had only 0402 and 0603... It is a slight improvement, but not as bright as new.

One lingering issue is noise. The meter seems a bit ouf of spec (RMS noise is about double that which the datasheet claims). It very occasionally fails test 100.2 (a noise test). I've also seen some popcorn/RTN signals when the input is shorted. I'm wondering if the LM399 is a little bad, or if the aforementioned audible switching noise could be the culprit? I think I'll live with this issue, since I don't want to have to recalibrate the unit.

[Kleinstein]

The audible noise could very well come from the transformer. Has the noise changed when changing the caps in the supply ? A larger capacitance or lower ESR of the new caps could increase the audible noise. To a certain degree some noise is normal - extensive noise could be due to a partially broken rectifier too (making it one way rectification). This would also show as higher (and 60 Hz) ripple.

One can do a measurement to distinguish between noise of the reference, noise of the ADC and noise of the input amplifier. With a shorted input, there should be very little influence of noise from the reference (LM399). In the 10 V range (with a shorted input) most of the noise should be from the ADC. In the 100 mV range most of the noise should be from the input amplifier (especially the buffer).  There is a rather long thread on noise comparison, that should give you approximate numbers on what to expect.

Measuring the reference noise is a more difficult task: it usually needs an external low noise reference voltage. This could be something like a 9 V block measured in the 10 V range. I treated well a 9 V block can be low noise, though with some drift.

[pigrew]

The audible noise could very well come from the transformer. Has the noise changed when changing the caps in the supply ? A larger capacitance or lower ESR of the new caps could increase the audible noise. To a certain degree some noise is normal - extensive noise could be due to a partially broken rectifier too (making it one way rectification). This would also show as higher (and 60 Hz) ripple.

I used mostly exact replacements (Nichicon VR and VZ). The exception was two 2200uF 35V Matushita NHG-series which I replaced with Nichicon VR. I was thinking about replacing them with low-ESR caps, but I figured that I should maintain the design intent (for fear if disrupting control loops). The amount of noise was not noticeably unchanged. I'll probe the rectifier outputs next time I open up the unit, but I figure it's normal.

One can do a measurement to distinguish between noise of the reference, noise of the ADC and noise of the input amplifier. With a shorted input, there should be very little influence of noise from the reference (LM399). In the 10 V range (with a shorted input) most of the noise should be from the ADC. In the 100 mV range most of the noise should be from the input amplifier (especially the buffer).  There is a rather long thread on noise comparison, that should give you approximate numbers on what to expect.

The noise I saw (and what the datasheet specifies) is with the input shorted. So, I'll assume that there is an issue with the input op-amps. I looked at the other thread which discussed replacing U113 (LTC1050 with a LTC2057).  The input-referred current noise is 100 times higher for the LTC2057 which I don't understand since the input voltage noise is much better. Perhaps the input impedances are different causing the difference? I'm shocked that the LTC1050 only operates at 2.5 kHz, while the K2000 can sample at 2 kSa/s. The LTC2057 is much faster (100 kHz).  Other than the GBW being lower, I don't see any downsides for this substitution?

I'll certainly record the noise before and after the swap, if I choose to do it.

Looking at the schematic, I'm not sure of the purpose of U114. It turns on other FETs when the input is overloaded or something like that? I need to sit down and study the circuit. It uses a lot of interesting concepts (capacitance multiplier or maybe current limiters, clamps, guard rings, filters.

[Kleinstein]

U114 drives the guard traces, bootstraps the over-voltage protection and also drives the supply for the LTC1050. The LTC1050 only operates at low voltage, so it can't work by it's own.

The LTC2057 is a much lower noise version, using a kind of lower in put impedance / larger input switches. With the AZ OPs, there is a trade of between low noise and Bias and current noise. So lower voltage noise types tend to have current noise and also a higher bias current. The LTC1050 is more on the side of low bias, high noise. The LTC2057 is more a low noise, but possibly higher bias type.
One can be lucky with the bias, but chances are the 2057 will have a rather high bias, that causes problems. If one really wants lower noise, I would consider only types with slightly lower noise, maybe the LTC1052 with only slightly higher bias.

It is not clear the noise comes from the input buffer. It is also possible to have excessive noise from a later stage. So one should first compare the noise in the 10 V and 100 mV ranges.

[pigrew]

Kleinstein,

Thanks for your advice.

However, after putting together proper test code to properly measure the instrument's noise, it turned out to meet the datasheet specifications. So, I think that I'm going to leave it alone and be happy with a working K2000. It's quite possible that letting it sit for a few days allowed the PCB to dry out and be happier. It's also consistently passing all of its builtin tests.

Running 1024 measurements (its max buffer size, as fast as it can measure them, 60 Hz line freq), I get the following results:

NPLC	Mean (uV)	SD (uV)	Range (uV)	RMS (uV)	RMS Limit (uV)
0.01	1.21	12.9	76.3	12.9	<150
0.1	-0.8	8.64	36.9	8.68	<22
1	-1.22	0.49	2.83	1.319	<4
10	-1.4	0.181	1.16	1.412	<1.5

(100mV range, shorted input, auto-zero is on, snippet of C# test code is attached.)

(It took me a while to figure out that the instrument doesn't use a standard binary block which starts with its length; these just always start with "#0".)

(This data isn't supposed to be compared to xdev's data because that data is measured over a long time period and would incorporate thermal effects, drift, etc. This is intended to only show amplifier/front-end noise.)

[Kleinstein]

I agree there is no need to fix it, if no broken. The numbers are a little odd: normally RMS and standard deviation should be the same.  Usually the specification limits don't include the DC offset.

The instrument might need an offset adjustment. AFAIK there is a kind of adjustment in software for the offset of the input buffer (the old value might be off due to the old higher bias). The AZ function in the K2000 does not include the input buffer.

[Krampmeier]

I just got a K2000 with the same symptoms: 306.2 and 306.4 failures after warm-up. It was actually taken out of order because if had a consistent drift in the 10 MOhm and 100 MOhm ranges over the last 3 years, always failing the calibration one year after adjustment. I suppose these issues are connected, and they seem to be very common for old Keithley 2000s.

Pigrew, I am glad you posted your solution (replacing Q120) here, and indeed, error 306.2 goes away (and the measured voltage during the test rises slightly) when I cool it down. The measured voltage goes down when I heat the Q120 up as well. I'll change it when I get the ordered replacement.

Hopefully 306.4 will go away as well when I also replace Q109 (which I removed out of pure stupidity, along with one of it's solder pads)...

Does anyone have more info about test 306.4? Q109 is switched off during this test (on during 306.2).

[pigrew]

Pigrew, I am glad you posted your solution (replacing Q120) here, and indeed, error 306.2 goes away (and the measured voltage during the test rises slightly) when I cool it down. The measured voltage goes down when I heat the Q120 up as well. I'll change it when I get the ordered replacement.

Hopefully 306.4 will go away as well when I also replace Q109 (which I removed out of pure stupidity, along with one of it's solder pads)...

I usually use two soldering irons simultaneously for SMD repairs. I'll heat up the board with a bit of 125C hot air, and then use two irons (one on each side of the transistor. The manual said to use organic flux to make cleaning easier. I didn't have distilled water on hand, so I used some diluted ethanol to remove the flux. I may still buy some water and clean it up a bit more. I bent the legs of the transistors down so that their body didn't contact the board (fewer leakage paths?). It's probably a good idea except that it removes a thermal sink from the devices.

I also was silly and replaced an unrelated transistor.... I wish I hadn't as I don't trust my board-cleaning skills.

It went away once 306.2 was gone. They do seem connected.

Does anyone have more info about test 306.4? Q109 is switched off during this test (on during 306.2).

I never found any more details on what 306.4 means. It went away once I replaced Q120.

[Krampmeier]

I have replaced Q120 now as well, but the behavior of the meter is still the same as before: it passes the selftest when it is cold, but fails when warmed up.

I continued testing the components in and around the signal path. In order to do that, I connected another DMM (Keithley 2010) to TP104 (buffered ADC input) vs. TP102 (AGND). Then executed the tests 306.x in step mode, rel'd out the reference voltage I got during test 306.1 (~6.95606 V), and watched the voltage drop when the defective meter was stopped in 306.2.

I touched the components which are connected to the signal path with a Q-tip which was soaked with cold spray, and I found that U114 is very temperature sensitive as well. It is a JFET-input Opamp, and any bias current would affect this test. I replaced the part by an OPA2192, which has a very similar bias current spec over temperature.

To my surprise, nothing has changed. The voltage drop during 306.2 is still about 600 µV when the meter is cold or U114 is cooled down, and about 1.2 mV when it is warmed up in the enclosure.

I noticed the internal test passes if there is no more than 1 mV drop between the two steps. 1 mV / 9.9 mOhm are approx. 100 pA.

I have cleaned the area under and around the IC, and there are guard lines, so I don't think there is leakage current via the PCB surface.

Does anyone here have other ideas where to look for the root cause of this problem, or do I just need a better (maybe selected) part for U114?

And just for understanding - Q120 can only reduce the measured voltage during 306.2 if it has excessive gate-source leakage, right? Drain-Source leakage would rather increase the voltage reading, correct?

Thanks for all input on these issues!

[saturnin]

I deal with the same issue in my K2000 unit. During hot days (room temperature >27C) and after warm up, I also got 306.2 error too.

The culprit can be basically any active component connected to the test current path after the 9.9M half of R117, so namely: Q108, Q113, Q105, Q120, U114, U113 (maybe some others, check schematic)... JFETs can be highly suspected, because there is always some leakage from source/drain to gate. After years of service, the leakage can deteriorate so much 306.x tests fail...

IIRC, I had to replace Q120 (easy to find, it was very sensitive to heat), then it was significantly better (took longer time until the error appeared), but the issue was still there. Then, I removed every single of suspected JFETs (one by one, they are in off state during 306.x tests). When a certain JFET was removed, the issue disappeared. I replaced it with a new component and never saw 306.2 error again.

[Krampmeier]

It is interesting, that Q120 seems to have leakage issues rather frequently!

I have cooled down all the components you mentioned while monitoring the leakage indirectly (as described), and the only ones which were temperature sensitive are Q120 (replaced now) and U114. U114 is replaced now as well (now OPA2192 instead of OP282), but the new part is still just as temperature sensitive as the old one.

I will now try if the issues go away when I replace U114 with a new OP282 or an ada4622-2, which has even better input bias current specs than the OP282 and OPA2192. These seem to be among the best available parts in this respect, with the precondition of 30V supply voltage.

[Kleinstein]

A slight increase in the input current is normal when an OP like OP282 or OPA2192 get warm. There are also a few more parts (e.g. Q105 / Q106) close by, that might also get warm / cold.


It is possible that Q120 is at a position that is not that well protected and thus a little more sensitive to damage (e.g. due to ESD).

[Krampmeier]

I have now checked a couple of K2000s at my company, and found that a brand new unit has a leakage current (calculated from 306.2 result) of about 25 pA, even when warm. Many older units are are 550 pAish, and several others also have about 1200 - 1800 pA leakage, so they fail this test. Some of the meters are 15 ot 20 years old. Is it reasonable to assume that this parameter indicates that these devices are about to reach the end of their intended life cycle?

We calibrate these meters every year, and they usually pass. The 10 MOhm and 100 MOhm ranges are the ones which fail calibration most often, but usually they can still be aligned.
We have now started to look for meters which had to be aligned in 3 consecutive years, and I wonder if

a) there is a connection between 306.2 failing (leakage current increasing) and the high resistance ranges needing alignment - I'd expect that.
b) change of a few JFETs and U114 is likely to solve the issues and give the old meters another couple of years lifetime
c) high leakage currents which do not lead to fails of the calibration present any problem in practise (eg. voltage measurements at high impedance sources)

Unfortunately, the calibration routine does not include any checks of leakage current or input impedance, and also does not include the execution of the self-test. I wonder if a failing self-test should be a "fail" criteria in calibration, even if the meter passes all other steps as descibed in the manual.

I have connected a 10 MOhm resistor across the input terminals in DCV mode, and got readings of about 0.6 - 1 mV on most Keithley 2000 meters. I did not see a clear correlation to the leakage current or age of the meter. That was not what I expected.

Sidenote: I got approximately the same value at several Keithley 2010 and one 2001, but significantly less on a 7510.

[pigrew]

I have now checked a couple of K2000s at my company, and found that a brand new unit has a leakage current (calculated from 306.2 result) of about 25 pA, even when warm. Many older units are are 550 pAish, and several others also have about 1200 - 1800 pA leakage, so they fail this test. Some of the meters are 15 or 20 years old. Is it reasonable to assume that this parameter indicates that these devices are about to reach the end of their intended life cycle?

I assume the planned life cycle is the warranty period, 3 years?

From my reading and experience, there are three major failure causes: capacitor leaking dielectric, display going dim, and input leakage increase.

I really don't know why the Keithley's seem to fail with input leakage, whereas the 34401A seem rock solid...

My feeling is that your assertions are true, but I don't have any data other than a single sample to look at.

I was doing some capacitor DC leakage measurements yesterday, and on the 100 V and 1 kV ranges, the meter's input bias was much larger than the capacitor's leakage (with Vin=10 V). It was definitely screwing up my measurements. The instrument passes its self tests.

I can't find any sort of input bias specification in the Keithley literature for the 2000. It really should be part of the calibration procedure, as it seems to be a common issue they have.

The thermal sensitivity of Q120 could definitely knock the unit out of calibration.

After fixing Q120, my zeros were a little bit offset (barely within calibration limits). I think that my unit was sold because it failed self-tests during calibration (The calibration date was right before it was listed on eBay).

-Nathan

[Krampmeier]

pigrew, did you also check the thermal sensitivity of your unit's U114?

I have now desoldered Q105, Q108, Q113 and Q120 (all at the same time), and the leakage has decreased from ~120 to about 43 pA (temperature not stabilized fully yet).
Compared to a new unit this is still much.

Cooling down U114 (by touching it with a q-tip which was sprayed with cold spray) brings it down to maybe 10 pA.

Looks like all the components contribute to the problem in some way...

The meter has 10 MOhm input impedance in the 100 V and 1000 V ranges - this is not what you are talking about when describing your problems with the capacitor measurements?

Btw, I have not yet seen 2000s (or 2010s) with leaking capacitors. This seems to be a speciality of the 2001 and 2002 models...

[Krampmeier]

I put new MMBF4393 JFETs into the meter, one by one, for Q120, Q113, Q108 and Q105 now. Always let the meter get back to stable temperature after that and checked the voltage drop. It seems like each of these FETs contributes 5 - 15 pA at normal operating temperature, so with all 4 new parts in, I got back from 43 pA to 91 pA.

I compared the SMD marking with the one in a brand new Keithley 2000, and it is the same "6G" marking. U114 is also still the same part as before, but the new meter has almost no voltage drop at test 306.2. I wonder if that means "no leakage", or positive and negative leakage from various sources happens to compensate each other...

I wonder how this is possible if the parts are the same!

Concerning the relevance of the 306.2 and 306.4  errors: I checked the calibration history of a meter at work which had about 250 pA leakage current (as indicated by test 306.2). It has never failed the calibration throughout the last years. Looks like there is no direct connection between this error and the high resistance ranges failing.

[Kleinstein]

Some of the leakage errors can compensate. Some of the JFETs may be just there to compensate on the leakage.

There are several manufacturers of the MMBF4393. For some reason they specify different maximum voltage. Also JFETs tend to have quite some scattering, not just for the gate threshold. There is also a chance that ESD damage to those JFETs might lead to higher leakage.

pA currents can also be due to contamination. So the flux and cleaning after soldering can be important.

[Krampmeier]

I tried not to add contamination by just using hot air (and no new solder) when changing the components. After changing the same part multiple times, I had to add some new solder with fresh flux, but I did not see any indication of additional leakage after that.

I used the MMBF4393 from On Semi / Fairchild, and they don't seem to be too much better than the old ones. Being new parts, they should not be damaged.
I then found the PMBF4393 from NXP, which has a higher voltage rating. Did not find any other similar part though. Kleinstein, which other ones did you have in mind?
The NXP parts may perform better, maybe I'll get some just to try...

I replaced U114 with a ADA4622, and this seems to have made a big improvement. The leakage current went down and does not change when I cool the new opamp.

I'll also write an email to Keithley from my work account. As we have more than 100 of these multimeters, with a significant number of them failing the self-test in step 306.2, I'd really like to know in which cases this could be relevant in real-world applications.

[Kleinstein]

The leakage of JFETs can show quite some scattering. Usually the difference between the typical and maximum values is rather large. So you sometime get lucky and find some good one and sometimes they are all from a not so good waver. This is one reason such FETs might be specially selected / checked and get a special Keithley or HP number.

How much the leakage matters depends on the FET effected. Often it is gate leakage and this only applies when the switch is off.

[Krampmeier]

I had a nice talk to an engineer from Keithley Germany today after their distributor forwarded my questions to the Keithley tech support.

My interlocutor obviously had very deep knowledge of the circuit in (not only) the Keithley 2000 multimeter and took time with me to go through my questions and do some tech nerd chat - a quite pleasurable experience and a very positive surprise, given the negative comments here about Keithley support after they were taken over by Tek! 

In short, as neither leakage current nor bias current of the Keithley 2000 are specified, the failure of the self-test steps 306.2 and 306.4 are not considered a defect. This is true especially if the meter was upgraded from an older firmware which did not contain these tests. That applies to many (if not all) meters which show the error at my employer.

If the "bias current" steps fail, this usually does not mean there a problem with real-world measurements - as long as the meter passes the calibration. The user should be aware that these universal DMMs are not electrometers, and low-voltage measurements at high-impedance sources can be influenced by these currents. 100 pA bias current should be considered a "typical" value, even though it may be higher at some older meters and much lower at some new units.
When we try to measure the bias current, we should be aware that the current depends on the voltage applied to the terminals. When we just connect a 10 MOhm resistor to the inputs, we get the pure bias current. When we add a voltage source in series to that resistor (like the self-test does), we get a leakage current in addition to the bias current. Both of these should not be confused with the input resistance, which is usually somewhere around 100 GOhm.

I was also told that newer revisions of the meters contain some optimizations which lower the leakage current, and the leakage current should be constant over time and should not change much when today's new meters get older. The correlation of multimeter age and bias / leakage current is very likely not only because of component ageing, but also improvements in parts (and maybe PCB layout, but he did not say that).

The internal self-test was rather intended for service at the factory than for the end user, and it sounded a bit like there are regrets that the test was made available on the front panel. As a consequence, that test was "hidden" when they later created the 2010, which is quite similar.

I hope this helps!

I might still do some experiments with the NXP JFETs instead of those from On someday, even though I am aware of the scattering in parameters. However, repair of my Adret 103A calibrator is given priority over that now, as the self-test errors on my Keithley are gone and the remaining leakage and bias current are not critical.

[Kleinstein]

A possible way to test the bias current at a different voltage is to measure the drift rate with a low leakage low DA (e.g. polystyrene) capacitor in the 10 nF range at the input. With 10 nF a 100 pA current would result in a drift rate of 10 mV / second, thus easy to measure.

With those meters that show failure on the bias tests it might be worth checking the bias current in a voltage range, to see how much it actually is. The worst about a high bias is not knowing about it. If you one knows one can be careful with the very few cases where it can matter.

[voltsandjolts]

K2000 failing 306.2 here also.
I'm fairly sure this error can be safely ignored but I also replaced U114 OP282 with a ADA4622 (although that is at the limit of its supply voltage spec).
Definite improvement, leakage starts at 30pA but over a few hours leakage creeps up and over 100pA which causes a self test fail again.
Guess it must be a JFET, hmmm.

Built-in self test
Select test (shift+'filter'), TEST:BUILT-IN, BIT:MANUAL, MANUAL:306, MODE:STEP
When test pauses press shift to view reading.
Bias current in pA = (306.2 reading - 306.1 reading) / 9.9E-6              (as described in attached Tek forum post)
Failure threshold seems to be around 100pA.

[voltsandjolts]

Replacing Q120 has fixed it. Leakage is stable and less than 20pA now.
So, in summary, I replaced U114 with ADA4622 and Q120 with Fairchild MMBF4393.
Thanks to all previous posters for the guidance.

Edit...some years later:

Leakage measurement on another repaired Keithley 2000 which is now passing self test:

Warmed up meter 4+ hours

Built-in self test
Bias current in pA = (306.2 reading - 306.1 reading) / 9.9E-6
On my unit:
(6.950444 - 6.949998) /9.9E-6 = 45pA


Drift with floating inputs
Nothing connected to the inputs
Switch to 100VDC range (10MOhm input impedance gives 0.000V reading)
Switch to 10VDC range (>10GOhm input impedance) and start a stopwatch
Voltage reading shown will likely be moving away from 0.000V
On this unit:
23secs to +5.00 VDC
54secs to +10.00 VDC
(for comparison 34470A took 15mins to get to -3.37 VDC)

[doktor pyta]

Replacing Q120 has fixed it. Leakage is stable and less than 20pA now.
So, in summary, I replaced U114 with ADA4622 and Q120 with Fairchild MMBF4393.
Thanks to all previous posters for the guidance.

My K2000 also encountered this problem.
I replaced Q120 with MMBF4393 but without a success, so I went back to the original JFET.
Next I replaced U114 with ADA4622-2ARZ (at the moment of writing this they are available in Mouser).
After good washing of the soldering area the problem is gone- confirmed after few days of constant operation.

And finally I can get stable readings on high resistance ranges.

Cheers!

[jeffjmr]

Is there any reason you chose the ADA4622-2ARZ vs. the ADA4622-2BRZ?

Thanks,
Jeff

[Kleinstein]

Is there any reason you chose the ADA4666-2ARZ vs. the ADA4622-2BRZ?

Thanks,
Jeff

The A version can be cheaper and better available than the lower offset B version.
There is no need for a really low offset in this application. The original AD822 is higher offset than the A version and a 10 x high offset would likely still be no problem.

[GnomeZA]

I'm replying here because this thread is a wealth of information.

I bought a K2000 on eBay.  I swapped out the capacitors as always advised.  I also noticed a SMD capacitor was broken off under the board, but I suspect it was that way from factory, so I replaced a pair.

Anyway, after upgrading the firmware to 19 I also got the classic 306.2 and 306.4 after the meter had been on for roughly 15 minutes.
You can tell it will fail because in DCV mode with nothing connected it will start cycling from ±~0.4-1.1vDC very quickly, it gets faster as the meter gets hotter.

I bought some MMBF4393LT1G and replaced most of the JFETs.  I couldn't replace about 4 of them because a precision resistor board featuring caddock resistors was sitting over those JFETs.

Anyway after powering up 306.2 and 306.4 failed immediately and the cycling I talked about earlier was very fast and actually erratic (jumping around)
It also started to fail on 4 wire short.
When I looked into the code, the JFETs related to that error code was situation under the caddock board.

I didn't want to touch that precision resistor board with an iron but I did, I removed it and replaced the rest of the JFETs.
Went crazy with a toothbrush and some 99% iso to clean everything as well as a I could and I was wearing gloves (basically just all over)

Anyway after this, the meter only failed 06.2 and 306.4 consistently.
Next day same thing.

It is now a week later and I've had the meter on all day, and noticed the cycling I'm talking about above is now very slow.
Did the self test and everything passes now.
Even on cold start it now passes.

I suspect me cleaning the board introduced moisture that had to dry out.
I also had a voltage reference that I'd measured multiple times since I got this K2000 and it is still giving the same reading after all this crazy work.

So at least as far as the K2000 goes, it seems like you can get away with making quite a number of "repairs" and your calibration will remain relatively unchanged (whether it is in spec or not)

[Kleinstein]

Changing the JFETs should not effect the 10 V range very much. If at all the expected change is more like a small offset from thermal EMF. This is easy to check in the 200 mV range with a short. The other part that could be effected is the high value resistance (due to the input impedance in parallel) and the high voltage ranges that use the divider Q136 and Q114 are at the low end of the divider and there resistance does matter, especially if too high.

For the calibration there are 2 parts that get confused sometimes. One is getting the right scale factors and the other is to have the meter checked in a more legal sense that it performs as specified to have confidence in it to meat the specs for the next time. A repair always has a chance to cause some accidential changes also to part that should normally not be effected (e.g. via ESD or dirt or mechanical stress to the PCB). So the calibration in the more legal sense is void after a repair or just opening the meter.  For private use one could get away with a few checks like the ref. reading before and after.

[GnomeZA]

The other part that could be effected is the high value resistance (due to the input impedance in parallel) and the high voltage ranges that use the divider Q136 and Q114 are at the low end of the divider and there resistance does matter, especially if too high.

Good to know, I'll do some more tests around there.  I don't have precision resistors so I compare the K2000 against my Tonghui TH2830 in resistance mode.  It is new and calibrated in December.  Unfortunately, at the moment it is the closest thing I have to a comparable "standard".  Thankfully in 4 wire mode I get very much the same readings from both meters.  After the decimal things start diverging pretty quickly, but if that did not happen I'd be mightily impressed.

[voltsandjolts]

Just adding a little more info...

I have seen three K2000s that needed U114 replaced, seems quite common failure. As mentioned it's a OP282 but I replace it using ADA4622 with good results.

I have also replaced one U113. On older units this is an LTC1050, but on newer units it is a Nat Semi LMP2021 which has a reduced power supply voltage range of 2.2-5.5V. Zeners for the bootstrapped power supply VR107 VR108 have changed accordingly and are marked W50 = NXP BZX84-A2V4  2.4V Zener.