Keithley 619 current linearity / VC of 100G resistor

[e61_phil]

Hi,

After repairing the Fluke 5440B GPIB interface I could start with some measurements.

I tried to figure out the voltage coefficient of a few high value resistors. The setup was quick and dirty: Fluke 5440B gives the voltages and the 100G resistor was connect via Triax->BNC and BNC->screw_terminal adaptors to the Keithley 619. All measurements were taken in a fixed range (1nA). After setting a new voltage on the 5440B I let everything settle for 30s and after that I take the mean of 10 current measurements.

After the first run I thought the 100G must have a huge VC. To verify this, I connected a 1G resistor to the K619 and run the measurement again (voltage/100). My assumption is 1G should have a smaller VC than a 100G resistor and the voltage is also 100 times smaller.

I scaled the 1G measurement to 100G for better comparision. I looks very similar. Therefore, I think the linearity of my Keithley 619 is quite bad.

Does anyone have experiences with other electrometers?

[e61_phil]

For further investigation I measured the linearity of the 200mV range (I think this range will be used in the 2nA current range). The measurement is done like Dr. Frank described for the 34401A and some other instruments (linear regression to eleminate gain and offset errors).

The linearity is really bad. Maybe it is usefull to measure the analog output of the K619 with a better meter.

Edit: INL corrected (against 100mV)

[guenthert]

I haven't measured the linearity of my 617 yet, but I like to caution that the voltage source in those units is convenient, but not all that accurate.

Overall, I think the 617's claim to fame is its sensitivity, not its accuracy.  At very low currents it's quite difficult to measure with high accuracy.  I believe mine is still within stated specifications (after all those years!).  One could perhaps argue that those 4.5digits are a bit misleading (at low currents), but then the ampere range covers 11 (!) orders of magnitude ...

[e61_phil]

The 617 is much more sensitive than my 619 (2nA lowest range vs 2pA). The 619 can do 5.5digits. Even with 5.5digits the 617 is 100 times more sensitiv.

I didn't dive deep into the 617 manual. I read 2V output for every range only. Is there really a 1TR resistor build in for the lowest current range or is there some amplification behind the first stage?


I think a little box with an AD4530 and a nice 1Gig resistor connected to my 3456A or 34401A will beat the K619 in current measurement easily. Nevertheless, the very high input impedance (>20T) up to 200V is quite interesting and can't be so easy achieved with a few parts in front of a DMM.

[guenthert]

The 617 is much more sensitive than my 619 (2nA lowest range vs 2pA). The 619 can do 5.5digits. Even with 5.5digits the 617 is 100 times more sensitiv.

Mea culpa.  For some reason, I thought the 619 was something like two 617 units in one frame.

I didn't dive deep into the 617 manual. I read 2V output for every range only. Is there really a 1TR resistor build in for the lowest current range or is there some amplification behind the first stage?

The analog output is taken from the input of the ADC, i.e. after the pre-amplifier and gain stages, hence the same output range for all input ranges.  There are large value resistors, but I don't think quite as large as 1TOhm.  The current range is not using a shunt resistor configuration, but rather the input of the instrument is fed into the inverting input of the preamplifier, using a more modest value for the feedback resistor.  The excellent manual has functional overview and circuit diagrams.  They truly don't make them anymore like they used to (manuals, that is).  There is even a parts list, but some resistor values are not stated there (only Keithley part numbers).

I think a little box with an AD4530 and a nice 1Gig resistor connected to my 3456A or 34401A will beat the K619 in current measurement easily. Nevertheless, the very high input impedance (>20T) up to 200V is quite interesting and can't be so easy achieved with a few parts in front of a DMM.

Quite possibly.  I too think that if one just wants to measure very small currents, one is better off using a modern "electrometer" OpAmp as preamplifier very close to, preferably in the same enclosure as, the DUT and a regular DMM as ADC (bypassing the whole expensive triax stuff).  Still, I don't regret having purchased the 617 (for a very reasonable price) as the voltage source and the GPIB make it very convenient to use, not to speak of the automatic switching across so many ranges.

[guenthert]

As chance will have it, just today I received a 100GOhm +/-5% glass-vacuum resistor in the mail from Bulgaria (Fleebay is amazing  ).  The resistor came in what looks original packaging in a sealed plastic bag and I was careful not to touch the glass body, still I measure "only" some 88GOhm (in Ohms mode, some 88GOhm in V/I mode at 100V, some 89GOhm in V/I mode at 10V), so there appears to be some leakage in my set-up.  Not all that surprising really, as I hacked cable and test harness myself with only modest effort.  There's probably room for improvement ...

[e61_phil]

Hmm, 88G is way off. And 100V at 100G is 1nA, that isn't a really small current. I've attached some more measurements. None of these measurements are so far off. The two new ones are done with the HP 3456A at the analog output of the Keithley 619. The scaled 1G shows much better linearity, but below 20V (really 200mV) it is still ugly. The K619 datasheet says the burden voltage is 1mV at max this is already 0.5%. Therefore, I think it is an offset error due to the small voltages on the 1G resistor.
The new 100G measurement looks also something strange. I would expect a higher resistance at lower voltages.

My setup ist really dirty (see attachment).

[acbern]

The plots you are showing are so far off the accuracy of the 619 that they are totally meaningless. The 619 is well above 0.1% accuracy, and assuming measuring in the ppm range with this meter makes just no sense at all.
Use a meter that fits to the accuracy you want to measure. 3458A makes sense, the newer  Keithley 648x e.g. series is mediocre as well, imo, just have a look at the specs.

[guenthert]

Hmm, 88G is way off. And 100V at 100G is 1nA, that isn't a really small current.

Well, in the Ohm range the 617 is specified for an error of 1.5% of reading.  The resistor is 5%, so there is still a gap, but not *that* wide.  Another way of looking at it, assuming the instrument to measure perfectly and the DUT a perfect 100GOhm, there is a parallel resistor of some 800GOhm, which shouldn't be too difficult to introduce accidentally.

I've attached some more measurements. None of these measurements are so far off. The two new ones are done with the HP 3456A at the analog output of the Keithley 619. The scaled 1G shows much better linearity, but below 20V (really 200mV) it is still ugly. The K619 datasheet says the burden voltage is 1mV at max this is already 0.5%. Therefore, I think it is an offset error due to the small voltages on the 1G resistor.
The new 100G measurement looks also something strange. I would expect a higher resistance at lower voltages.

My setup ist really dirty (see attachment).

Haha, that would never work here.  There's way too much RFI.  My DUT is in an aluminium enclosure, keeping the RFI at bay, but (it's internal wiring or the connector to the Triax cable) quite possibly cause some leak currents.

I measured the current through the resistor in a range of voltages and got the attached diagram.  Seems something strange is happening around 20V and I'm fairly sure it's not the resistor causing it (guess this needs to go into the repair section) ...
(looking more closely at the diagram and the source data, the jump occurs around 18V, not 20V and is likely related to the range switching to 2nA)

[guenthert]

I measured the current through the resistor in a range of voltages and got the attached diagram.  Seems something strange is happening around 20V and I'm fairly sure it's not the resistor causing it (guess this needs to go into the repair section) ...
(looking more closely at the diagram and the source data, the jump occurs around 18V, not 20V and is likely related to the range switching to 2nA)

Not sure about you, but I found this quite disturbing.   So I measured once more, this time in more detail and plotted the calculated resistance value around the "discontinuity" as well as the actually measured current.  For the latter the jump doesn't look quite as dramatic (and occurs at 0.2nA, not 2nA as I stated erroneously earlier).  There might be a re calibration needed (but I don't want to do it now, as it is unusually cold around here right now -- only about 18.5degreeC in my 'lab').

[e61_phil]

The plots you are showing are so far off the accuracy of the 619 that they are totally meaningless. The 619 is well above 0.1% accuracy, and assuming measuring in the ppm range with this meter makes just no sense at all.
Use a meter that fits to the accuracy you want to measure. 3458A makes sense, the newer  Keithley 648x e.g. series is mediocre as well, imo, just have a look at the specs.

I'm not hunting single digit ppm's. I've done just some experiments which could be interesting for others. I expected a couple of 100ppm deviation over the swept voltage from the 100G resistor. Therefore, I thought one can  easiliy capture this deviation with a 5.5 digit instrument. In addition to that I haven't expected that the linearity of this instrument is dominated by the ADC, but I learned that know.
I can't understand which improvement a 3458A will bring in this point. The 100nA range 1Year spec for 1nA is 0,34% and the spec for the 619 is ~0,35%. I can't see a huge improvement there. Furthermore, I assummed this 0.35% is caused by long term drift of high value resistors inside the electrometer which doesn't matter within a 30min measurement.

Edit: Keithley 6485 has even worse specs than Keithley 619. Same with 6482. 6487 is the only 648x I found with better specs than my 619, but it is 0.3% instead of 0.35%..

[e61_phil]

(looking more closely at the diagram and the source data, the jump occurs around 18V, not 20V and is likely related to the range switching to 2nA)

I did all the measurements in one current range. Did you measure the output voltage of the 617?

[guenthert]

Heresy!  The 3458A is the answer to everything.


(yeah, yeah, envy speaking here.  well, someday ...)

[Conrad Hoffman]

Not owning any of that fancy stuff, I was thinking about how I'd measure a 100 gigohm resistor. Old school, it turns out the GR 716C capacitance bridge can also be used to measure high resistance values, up to 800 gig. No idea how accurate that would be, and the measurement has to be made at 100 Hz to get that high. Something else I lucked into was an HP 4329A high resistance meter, with voltages up to 1kV. That claims to reach 2 x 10^16 ohms. Again, no idea how accurate. I'd probably just put it in series with some smaller sense resistor and measure it with any 6.5 digit meter. Now I need to search my "odd components" pile and see how big a resistor I've got.

[e61_phil]

Not owning any of that fancy stuff, I was thinking about how I'd measure a 100 gigohm resistor. Old school, it turns out the GR 716C capacitance bridge can also be used to measure high resistance values, up to 800 gig. No idea how accurate that would be, and the measurement has to be made at 100 Hz to get that high. Something else I lucked into was an HP 4329A high resistance meter, with voltages up to 1kV. That claims to reach 2 x 10^16 ohms. Again, no idea how accurate. I'd probably just put it in series with some smaller sense resistor and measure it with any 6.5 digit meter. Now I need to search my "odd components" pile and see how big a resistor I've got.

The easiest way is properly to use the internal 10Meg of your meter (see attachemend, sorry no 3456A. It is >10G below 10V).

9.919mV / 10Meg = 991.9 pA
100V / 991.9pA = 100.8G

The result is very close to the K619 measurements.

[guenthert]

(looking more closely at the diagram and the source data, the jump occurs around 18V, not 20V and is likely related to the range switching to 2nA)

I did all the measurements in one current range.

As far as I understand, that is good practice (in order to avoid issues like the one I just encountered).  Prior, I measured the reverse breakdown voltage of LEDs, for which the auto-range feature was very helpful and didn't think of disabling it.

Did you measure the output voltage of the 617?

I just did and found that for all values (between 1 and 100V) the output voltage was within the stated spec (+/- 0.2% + 50mV).

I repeated the measurement, now measuring the voltage using an HP3478A (not calibrated either, but at least in the 30V range *very* well within its specs), twice -- once with the current measured in auto-range and once in the 2nA range only.

If I interpret the specifications correctly (1.6% of rdg. in 200pA range) - for the moment ignoring that this is a 30y/o instrument with unknown calibration history - this "discontinuity" is still in the range of acceptable values (the current jumps less than 1%.

[guenthert]

[..]
The easiest way is properly to use the internal 10Meg of your meter (see attachemend, sorry no 3456A. It is >10G below 10V).

9.919mV / 10Meg = 991.9 pA
100V / 991.9pA = 100.8G

I did that too once in the case (essentially just replacing the 617/Triax with my 34401/BNC) and once with one lead 'through the air' avoiding leaks (using 'slow' integration, average of 100 rdgs to combat RFI).

The result is very close to the K619 measurements.

Here too: 11.4mV at 100.17V across resistor and 34401, i.e. about 88GOhm.  I believe now, that my 100GOhm resistor is busted: one end is a bit milky, which I didn't notice before -- I'm afraid I might have broken the glass there and it isn't so much a glass-vacuum resistor anymore, but a cheap hygrometer 

[Conrad Hoffman]

Perhaps gently bake it out and see what the resistance does?

[VintageNut]

If you plug a 2-lug BNC into the 2-lug triax jack of the 617, there is a chance that the 2-lug triax jack is damaged. The center pin of a BNC is larger than the center hole of the 2-lug triax jack.

This is a common user error with the KE7065 hall effect card. Novice users see the 2-lug adapters on the rear of the card and just use coax not realizing that the jacks are triax. That breaks the jacks of the $10k card. I have witnessed this damage first-hand during a visit to a university using the 7065 card.

[..]
The easiest way is properly to use the internal 10Meg of your meter (see attachemend, sorry no 3456A. It is >10G below 10V).

9.919mV / 10Meg = 991.9 pA
100V / 991.9pA = 100.8G

I did that too once in the case (essentially just replacing the 617/Triax with my 34401/BNC) and once with one lead 'through the air' avoiding leaks (using 'slow' integration, average of 100 rdgs to combat RFI).

The result is very close to the K619 measurements.

Here too: 11.4mV at 100.17V across resistor and 34401, i.e. about 88GOhm.  I believe now, that my 100GOhm resistor is busted: one end is a bit milky, which I didn't notice before -- I'm afraid I might have broken the glass there and it isn't so much a glass-vacuum resistor anymore, but a cheap hygrometer 

[guenthert]

Perhaps gently bake it out and see what the resistance does?

I suppose, you didn't have self-heating in mind.   
So, in an oven?  I don't think I have such, warm enough to have an effect and not so hot to cause more damage than good (although, this unit can be considered fubar already).  Further, it's unusually cold here and the sun came out the last few days after a period of rain, so the air should be fairly dry.  It feels dry at least, nose-bleeding dry in fact.

Then, I paid less than $10 for the unit, I might just order another one (but haven't decided yet, whether I really need a 100GOhm resistor).

[guenthert]

Well, since the R100G appears to be fubar anyhow, I had no reservations to experiment on it.  Lacking a proper oven, I used my gas driven heat gun (actually more like a mini-torch) in an attempt to not only heat (and drive out humidity), but then also seal the apparently broken glass.  I managed to get the glass to yellow glow (the wire was then red glowing too) and it deformed a bit.  The surface looks now more smooth, so it did at least partially melt.  It is however also now silvery, instead of milky, which I take as some metal deposits from the inside.  Uh, oh, what did I breathe there?  I'm old, so it won't matter, but kids: don't do that at home!

Perhaps surprisingly, it still works and the resistance went up a bit, see attached.

I don't have a nice high voltage source, just a (dangerously) cheap DC converter from eBay.  I don't want to run that when I'm not present (although I don't want to be too close, when it is running ;-} hence only a few data points and no error bars this time.  I used auto-range again, hence the bent graph (there's a jump in the curent reading when the instruments switches from 200pA to 2nA and less pronounced so when switching to 20nA).  Despite that, there is a clear voltage dependency visible (which was the topic of this thread after all .