Advantest R6581 8.5 digit DMM mini teardown/repair

[szszjdb]

Hi,  Mr. Kleinstein,
Many Thanks!
I know that it is difficult to fully evaluate my 6581 without the higher grade instrument. I just want to make sure the meter now is reliable and trustable with its reading,as it has been repaired. I usally found different reading  between my 6581 and 34401, so want to find out the simple way to detect which is the correct one.

 I will try using the string resistor method later. It 's funny in the way and learn more from you and Mickle T. .Thanks again!

It seems the final solution is to calibrate it with a 3458 and compare it in all range.

Best Regards,
szszjdb

[Mickle T.]

INL of Advantest R6581 is a rather unpredictable thing. I have some measurement results, received by me, by bbs.38hot members and by Advantest Laboratories Ltd. itself.

[szszjdb]

Hi, Mr. Mickle.T and Mr. Kleinstein ,
Thanks for your picture.

I have tested the INL by the string resistor method and get data as below.
RES VOL   0.9982601   0.9983018   0.9982538   0.9982941   0.9982555   0.9982828   0.9982635   0.9983029   0.9982555   0.9982643
           0.9982601   1.9965619   2.9948157   3.9931098   4.9913653   5.9896481   6.9879116   7.9862145   8.98447   9.9827343

6581   0.9982601   1.9965634   2.9948124   3.9931063   4.991357   5.9896427   6.9878998   7.9862089   8.9844558   9.9827393
INL      0              0.07525     -0.11038      -0.08780      -0.16657     -0.09031     -0.16915      -0.07024     -0.15828     0.05017
                              
                              
34401   0.998252   1.99656   2.99481   3.9931   4.99136   5.98963   6.98791   7.98621   8.98447   9.98274
INL    -0.81281   -0.09532   -0.19065   -0.24584   -0.10636   -0.30271   -0.02293   -0.05644   0   0.0571


I am using the LM399-10v board output 9.9827343 as input(LTZ1000 has been sent to my friend for calibration) and 10pcs 10K res. string without switch using but only banana jack.  Every reading of 6581 is the average of 20 sample.

It can be seen that the INL of 6581 is a little better than 34401 in the 10V range.

I also found the big error between each range as I input 1V and get reading from 1V and 10V range of 6581, comparing with 34401. I had just perform the internal calbration 2 hours before.
         6581            34401
1v      0.9998735     0.99987
10v    999.86696   0.999866
err     6.54087E-06   4.00054E-06

Are these data normally with yours?

Best Regards,
szszjdb

[Mickle T.]

I can't understand how you calculated these INL results

[szszjdb]

Hi, Mr. Mickle.T ,
Thanks for your help.

Sorry, I just use 6581 data instead of the best fit to calculate the INL. That 's my mistake.

Could you help to comfirm the issue of the error between each range. The A/D has large tranfer error?
         6581            34401
1v      0.9998735     0.99987
10v    999.86696   0.999866
err     6.54087E-06   4.00054E-06

Best Regards,
szszjdb

[Mickle T.]

1. One of the possible error sources - a different zero offset on the 1 and 10 V ranges. DMM must be nulled before each measurement.
2. As you can see, the voltage artifact source, used in the 10->1 V transfer, have a negative polarity. So INL must be checked in the same negative polarity too.

[szszjdb]

Hi, Mr. Mickle.T ,
Thanks for your soon replay.

1. You means I must short the test line and null the reading every time when I change the range? If does, is there any way to fix the issues  in the hardware ?
2. I will try the negative polarity check.

Further more, I also checked the reading difference in other range and found it exist in most range, data as below.
The error for 100mv-1v is 3ppm and 18ppm for 1ma-10ma range and 37ppm for 10ma-100ma range. I really can't believe it .
It must have some hardware issue that can't be corrected by software, am I right?

   100mv-1v    100mv-100mv   1ma-10ma   1ma-1ma    10ma-100ma    10ma-10ma
MAX   99.98701   99.98676   0.99932   999.2757   9.99798   9.997452
MIN   99.98672   99.98618   0.999256   999.2635   9.99755   9.99741
AVG   99.98688   99.98658   0.999289   999.2708   9.99781   9.997437
VPP   0.00029   0.00058   0.000064   0.0122   0.00043   0.000042
DEV   8.16E-08   1.27E-07   1.29E-08   3.74E-07   1.05E-07   1.00E-08
err     3ppm                       18ppm                         37ppm

Best Regards,
szszjdb

[TiN]

I wonder if moving this thread to metrology would be good idea?   

[Kleinstein]

Due to the nonlinearity of the ADC and also offset errors the numeric adjustment of the different ranges is not that perfect. This is also reflected in the specs of the DMM. The 10 V range is more accurate for this reason and it needs a really good ADC to make this Autocal mode feasible. Anyway I would have expected better accuracy too.

Especially with the current ranges there can be additional nonlinearity from self heating of the shunts. So uncertainties are expected to be larger with these ranges. Choosing the voltages for the internal ACAL procedure is kind of a compromise between noise and offset problems at low voltage and nonlinarity of the the shunts. Anyway the errors will be higher there.

In theory there could be some improvement by software, but as this it the fixed DMM internal part it's kind of fixed.
For the part of the error that is repeatable (e.g. due to resistor nonlinearity) one might be able to do software correction one the values afterwords. Not very elegant, but an option if really needed. However this requires the errors to be about constant and a reliable measurement of the errors.
Hardware errors like more leakage or internal offsets could also contribute. For this reason the instrument should be well warmed up before doing the ACAL adjustment.

Ideally the INL test should be done for both signs.

[szszjdb]

Hi, Mr. Mickle.T and Mr. Kleinstein ,

1. I had nulled the reading after changing the range and the issue is same and error is about 6.5ppm. I also checked the DCI zero when shorting the test line and the result is normally around 0.
2. The negative polarity test shows improvement and the error is about 1.3ppm, data as below.

-0.9998667 -0.9998673   0.9998842   0.9998874
-999.86541   -999.86619   999.87772   999.88096
1.29017E-06   1.11015E-06   6.48079E-06   6.44077E-06

My 6581 had been  powering up for at least 2 weeks and all calibration are carried out at warm state.
Refer to the spec. attached,the max error for 1v is 2.5ppm and 20ppm for 10ma range and 30ppm for 100ma range. The reading of 1v range must be out the spec. and the one for 10ma and 100ma range is marginal out the spec..

Is there any thing I can further check? How can I improve them?

Further information, I get a message from my friend whose 6581 has the same error with mine , arround 6.5ppm in 1v range. Is it the general issue for 6581?

Best Regards,
szszjdb

[Kleinstein]

The 3 ppm deviation between ranges is larger than specs. For the direct comparison the short time accuracy would apply and thus the even stricter specs.

One first point to check would be to test if the internal cal is repeatable. This would show if there is a noise problem. Normally this should not be the case, but it's good to check - just in case. Another point of the check is to see if the adjustment changes at all - it might need an explicit call of the right internal cal procedure to take effect. For shorter time scale just the divider to make the 1 V my be used.

For the internal cal, the meter should do a similar measurement than the external test. E.g.  measure a test voltage of around 1 V in the 10 V and 1 V range and than calculate the gain ratio. So it is kind of odd to get a different result. I can imagine three possible problems, that could efffect both the DMM internal measurement and the external measurement.

One is a possible offset, that might be caused by leakage currents - so a careful test of the leakage currents might be a good idea. This could be a HW fault. The output resistance of the test source could be important here. For the external test one could measure both a +1 V and -1 V instead of 0 and 1 V. This would reduce the effect of offsets and also noise. It would also be a first test for linearity.

A second point is linearity: the 6581T does not seem to be that good here, and some voltage may be different than others. So one could do the external test with a few more voltages (e.g. +-0.7,+-0.8, +-0.9 , +-1 V). If there is nonlinearity in resistors the test with 0.7 or even less V might give a clue.

Another possible effect could be settling time: the input amplifier may need some time to fully settle - most of the settling will be fast, but there might be a small slow part that can cause trouble. This could cause a slightly (usually < 1 ppm) different gain for different PLC settings. The internal cal is done much faster than the external check. It is a little tricky to test just from the display, unless the DMM offers suitable statistic functions. The Solartron meters seem to have a related kind of problem in it's Az mode - Mickle T has shown this very nice in a different thread.

[szszjdb]

Hi, Mr. Kleinstein ,

Thanks for your advice?

1. I had repeated the internal cal and no error display. How to verify if the internal cal take effects?

2. Offset is the DCV zero point? If so, it is ok by now. I also checked the +-1v  to +-0.7v and found error between +- reading.Data are as below.
0.9998879   0.8999006   0.7999162   0.6999385
-0.999881   -0.8998955   -0.7999122   -0.6999343
-6.90082E-06   -5.66732E-06   -5.00055E-06   -6.00056E-06

3. The linearity result are as above and the error is nearly the same number.

4. How to check the  effect of settling time in detail?

Best Regards,
szszjdb

[Kleinstein]

An error in the 5 ppm range just from changing sight is really bad. This would be really poor INL. One has to expect not so good INL values in the 1 V range, but these numbers look odd. So maybe there is still a HW fault causing poor linearity, or maybe to much input bias in some ranges.
With a JFET based MUX the JFETs that are turned on may not contribute to leakage current - so there is a slight chance that the input bias could be higher than normal when not reading the DCV input.

[szszjdb]

Hi, Mr. Kleinstein ,
Thanks a lot!

How can I test the bias? By connect the 10M res ? I had checked it yestoday and got 0.2mv when AZON and 0.02mv when AZOFF on 10V range. I forgot to test them on 1V and 100mV range and will comfirm it tonight.

Further information,my firend had comfirmed that his 2pcs 6581 also have the same issue in 1V/10V INL ,which is about 5-7ppm, refer to 1V range. And he told me that the error is about 1.5-1.6ppm when in cold state and become larger when in warm state. He use the 732b output 1v for testing.

Best Regards,
szszjdb

[Mickle T.]

R6581T: preheated during 6 hours, Int CAL, AZERO ON, PROT ON, 100 PLC, LO-G, NULL ON. Voltage source - LTZ1000 10-1-0.1 V

1 V Range polarity reversing errors:
-0.9999918; +0.9999929; 0.9 ppm error

10 V Range polarity reversing errors:
-9.999904; +9.999910; 0.6 ppm error

1V -> 0.1V transfer error:
+0.10000225; +0.1000016; 6.5 ppm error (in 24 h spec)
-0.10000180; -0.1000013; 5 ppm error (in 24 h spec)

10V -> 1V transfer error:
+0.9999949; +0.9999929; 2 ppm error (not in 24 h spec)
-0.9999912; 0.99998995; 1.25 ppm error (in 24 h spec)

100V -> 10V transfer error:
+9.999910; +9.9999082; 0.18 ppm error (in 24 h spec)

[szszjdb]

Hi, Mr. Mickle.T,
Thanks for so detail test!

Could you give me some advice on the further check? There has 3pcs 6581 have the same issue with the 10V to 1V transfer error, plus my unit.

Best Regards,
szszjdb

[Kleinstein]

Hi, Mr. Kleinstein ,
Thanks a lot!

How can I test the bias? By connect the 10M res ? I had checked it yestoday and got 0.2mv when AZON and 0.02mv when AZOFF on 10V range. I forgot to test them on 1V and 100mV range and will comfirm it tonight.

Further information,my firend had comfirmed that his 2pcs 6581 also have the same issue in 1V/10V INL ,which is about 5-7ppm, refer to 1V range. And he told me that the error is about 1.5-1.6ppm when in cold state and become larger when in warm state. He use the 732b output 1v for testing.

Best Regards,
szszjdb

0.2 mV over 10 M would be around 20 pA. This is OK. However there could be still an issue at higher voltage (e.g. to test with capacitor charging).

The other point is that the test in the DCV range does to include gate leakage for the FETs turned on - thus the one for the DCV range, which is the one that is exposed the most and thus the most likely one to get damaged (e.g. from ESD or overvoltage events). It would be really hard to test these FETs in the MUX.  It should show up when comparing the lowest current range with the inputs open compared to short. However I am not sure the resolution is sufficient and there might be additional contributions, e.g. from the input protection.
If the problem is due to an offset / bias current the 1 V - 0.1 V transfer should be similar bad, with an even larger error.

Besides offsets, it could also be an INL problem that could cause the 10 V - 1 V transfer to not work that well. From looking at the ADC circuit known so far, the ADC does not looks like it is made for good INL, more like for good stability and in parts for low noise with the old parts. My guess is that DA in the integrating cap could be a major contribution to INL, especially the more complicated irregular shaped part that is difficult to correct for.  It depends on the way the ADC works, but usually the rather large integration cap is not a good sign. There is the possibility the cap when getting old will slowly take up humidity and this way the DA goes up. If really bad excess DA should also show up in histogram tests - not with all, but at a few critical points.

[szszjdb]

Hi, Mr. Kleinstein ,
Thanks a lot!

1.  Yes the FET is hard to test. I had test the DCI in 100na range, data as below. It is not sufficient  to find out something strange.
   OPEN DCI   short DCI   short DCV
MAX   0.0271   0.029   0.0319
MIN   0.0191   0.0281   0.0245
AVG   0.0246   0.0286   0.0289
VPP   0.008   0.0009   0.0074
DEV   2.10E-12   2.60E-13   2.00E-12

The 1V -100mv transfer have 3ppm error, just in spec.
N=100, 10plc   100mv-1v RG   100mv-100mv
MAX   99.98701   99.98676
MIN   99.98672   99.98618
AVG   99.98688   99.98658
VPP   0.00029   0.00058
DEV   8.16E-08   1.27E-07
ERR      -3.00039E-06

2. The stange is the error is larger in the positive polarity than the negative  one, which is almost in the spec. If the cap has problem , why it just contribute to the positive polarity?
-0.9998667   -0.99988   -0.9998673   0.9998842   0.9998874
-999.86541   -0.999867   -999.86619   999.87772   999.88096
1.29017E-06   1.30017E-05   1.11015E-06   6.48079E-06   6.44077E-06

I also finished the 1-10v positive /negative  check, data are as below. The most critical area is the 1-3v ,which is beyond 2ppm.

-0.9982572   0.9982602   3.00523E-06
-1.9965605   1.9965682   3.85662E-06
-2.9948251   2.9948311   2.00345E-06
-3.9931183   3.9931236   1.32728E-06
-4.9913767   4.9913802   7.01209E-07
-5.9896593   5.9896629   6.01035E-07
-6.987925   6.9879291   5.86726E-07
-7.986225   7.986221   -5.00863E-07
-8.9844896   8.9844866   -3.33909E-07
-9.9827622   9.9827581   -4.10708E-07

Any further check  should  I do ?

Best Regards,
szszjdb

[Kleinstein]

The current zero test (29 pA) seems to be consistent with the bias measured in DCV mode (around 20 pA). The effect of shoring also is small, so likely no problem here.

The turn over test kind of indicates a problem, either with an offset or with INL errors. A few more points towards even smaller voltages (in the 10 V range) could help to distinguish offset and true INL a little better.

One source of nonlinearity are resistors, e.g. due to self heating. This effect should be smooth, mainly an U³ contribution. The main candidates are the resistor at the ADC input and the gain setting resistors in the 1 V range. Self heating takes some time and thus the 1 PLC mode might act different from a true 100 PLC mode.  A comparison of 100 PLC mode an averaged 1 or 10 PLC mode might give some clues. However slow settling might have a similar effect - thermal effect can be one part of a slow settling contribution.
From Mickles description the internal cal for the 10 V 1 V step is done in 10 PLC mode. So it would be worth a test at 10 PLC.

A second source for INL is DA in the integration cap. In a multi-slope converter this is a rather odd jagged function with possible hundreds of points going up and down over the range. This is kind of difficult to measure and test, especially without PC connection. One could repeat the turn over test with a few (e.g. 5 points close together, like 1 V, 1.02 V, ...1.1 V). DA caused INL would give a more random up and down, while resistor caused INL would give a smooth curve. Depending on how the converter is made, there may be an additional smooth function INL caused by DA. Here positive and negative direction can be quite different as there is a constant current added at the input of the ADC.
The capacitor can effect different voltage different.

In theory there might be an error if the resistors for the current source change in ratio. I am not sure if and when the DMM checks those and measures them - it should be possible to use measured values, but I don't know if such a measurement is done, and if so when. It could be part of the ACAL procedure or a separate point.

[Mickle T.]

As in case of HP 3458A, R6581 don't have a true 100 PLC integration time. All of the 20...100 PLC modes are simply averaging of 10 PLC one.
DMM don't measures the ADC current weighting resistors ratio in the ACAL procedure. I have check an every ms of the DCV and DCI/OHMs ACAL timeline

[szszjdb]

Hi, Mr. Kleinstein and Mr. Mickle.T ,
Thanks a lot!

The transfer error is more like the offset one which is almost 6uv around. Why it is corrected in the negative  polarity and left undo in the positive one? I had try both in the 10PLC and 100PLC, the same result got, allway 5-6uv, 6ppm error, very strange!

For the check of the DA in the integration cap, you means to test 5 points close to1V and record the 1v-10v error and check if it is in random number?  I will try it.

Is there any posible the U105 work abnormal, as it is been replaced by the former?

Any further advice?

Best Regards,
szszjdb

[Mickle T.]

A fastest way to check the DA - is a replacement of C206 (SOSHIN NQP 20 nF) to the new MKP one

[Kleinstein]

It is still possible to have a problem with the DCV amplifier. If you have a suitable meter (the 34401 might work) one could check the input an output of the DCV amplifier (e.g. output might be output of U105) and input could be the input terminals in non AZ mode. The difference should be pretty constant with input voltage.

A changed chip (U105) might at least point to a problem in that area. So it might be worth checking the input amplifier, that seems to be using discrete JFETs for the input and an OP (U105 ?) for the output.  A offset adjustment might also help, though with JFETs there is no such simple relation from offset to drift as with BJT based amplifiers.

As there is no real 100 PLC mode (does not make much sense due to 1/f noise of the integrator) - the 10 PLC and 100 PLC mode should give the same readings - it is only the 1 PLC or similar mode that might be off a little, but hard to tell without the computer.

Not checking the exact ratio of the reference currents could be a problem, or at least a kind of odd solution. For me it would be kind of the obvious way to use the ADC in a special mode to measure those ratios to get the exact numbers. The first ones might need to be relatively accurate - the small one likely don't matter that much. It might still be something only done at a special factory cal or similar.

[szszjdb]

Hi, Mr. Kleinstein and Mr. Mickle.T ,
Thanks a lot!

I had try another 3 type cap inhand as attached , and the original is the best ,which has the minimal error.
22NF CBB    22NF mica    20NF mica    SOSHIN 19.6n   CL 20.1N
999.84399   999.85551   999.8478           999.8682            999.85
0.9998821       0.9998825   0.9998799   0.9998752   0.99988
-3.81159E-05   -2.69932E-05   -3.21039E-05   -7.00087E-06   -3.00036E-05

What is the substitution type for the SOSHIN ,I will find it later.
From the data above , is there still any issue with the cap?

I had test the U105 with 34401, but the output is square wave, so can got the pricision result.

Any further advice?

Many Thanks and Best Regards,
szszjdb

[Mickle T.]

In my measurements R6581T Input Amplifier shows a very little INL (about 0.1 ppm, InpAmp.png), so the main problem is in ADC.

I did not even expect such a strong DA influence SOSHIN NQP have the same polypropylene dielectric as CBB.
By the way, at slow integration times (>200 uS and up to 100 PLC) both of the integrator capacitors C206 and C210 are connected in parallel. If C210 have a large DA, it is possible to temporarily cut it and looks to INL changes.