Measurements of the 7D Precision Resistor Array

[Alex Nikitin]

A little while ago I've offered for sale here on the forum a little gadget, a better looking copy of a unit I use in my lab for many years. The "7D Precision Resistor Array" somewhat unexpectedly is sold now to about ten forum members and requests to build more are still coming. I've decided to make it into a "proper" product, with a better label and added a shorting plug after been asked for it. One thing I didn't manage to do till now is to measure at least one unit in-depth, including temperature coefficients in a reasonable lab range of temperatures. So here are some photos and measurements results, it is much better to have a separate thread about the performance and various application questions (and this small thing has very many uses from my experience).

I hope here is the right place for this topic.

Cheers

Alex

[EC8010]

Fascinating stuff. Curious about the elliptical trace of the 100M, though. Were the temperature sweeps all of the same duration?

[Alex Nikitin]

Fascinating stuff. Curious about the elliptical trace of the 100M, though. Were the temperature sweeps all of the same duration?

Unfortunately, my reptile box "temperature chamber" has different and uncontrolled rates of heating and cooling, so, unless I am watching it carefully, it is not very consistent in this respect. I was trying to alleviate this problem by a natural (not heated) warming up of the box after the coldest point.  Also I measure the temperature of the alloy box on a side with the sensor fixed by a copper tape, but there is a time lag for the temperature of resistors to reach an equilibrium with the box, so the right way would be to measure the resistance only after the temperature completely settled for each point and not in a dynamic fashion, but it would take a very long time and a very stable temperature in the lab as my current sources (Keithley 263 and HP3245A) have a TC on their own, not much but enough to spoil the results for the lowest TC resistors, if the lab temperature changed more than a degree or so during the run. I probably need to note at what temperature the measurements on the label done (26-27C in case of the unit s/n 240800 as pictured).

So please treat these graphs as a rough (but still very useful) estimate of the array performance. Personally I was surprised how well the classic RC55Y 1M0 resistor behaved and on the other hand the 1G Mini-MOX has a somewhat larger TC than the datasheet figure (40ppm/C v 25ppm/C).

Cheers

Alex

[miro123]

Question - are you using liner polynomial for defining TC or squire with Alpha and beta coefficents?
TC measurement require accurate measurement of Tempareture and resistance and air  is bad temparature conductor. Air bath helps but anyway the resistors are placed. in closed box for certain reason.
Higher rate of temperature change creates stress in most resistive materials .  I check when I reach this critical point with extra adding of steady state at Tmin Tmax and later return to Tambient.
Then I can determine T hystarasys and do I go to fast from one to another temepratrure.

To improve the temeprature measurements I used silicone oil bath for my resistor references. See picture below it is 4 Vishay S102 connected in 2s2p configuration. Temperature is monitored with one PT1000 and 10K Omega NTC. Picture is just before sealing the whole assembly. Assembly on the top is more complex - not shown here. And all two reference modules are placed with common Hammon AL case. Aditional  T , RH% and Pressure senosr is added near to the hammon case

[Alex Nikitin]

As you may see from the graphs a linear approximation is good enough for the data quality I am getting and for the price I'm asking for this unit .  I also did a double check on the values after a temperature run and the hysteresis is essentially on par with or below my measurement repeatability error which is probably about 5ppm at 10K and progressively worse with the increase in value. As I've already mentioned, if you want to be as accurate as possible you need to do things slowly point by point reaching a thermal equilibrium every time before taking a measurement. It takes a lot of time and introduces new sources of error due, for example, to the lab air temperature change.

Cheers

Alex

[EC8010]

Yes, thermal time constants can easily be quite long, and for the sort of accuracy looked for here, cause serious problems. My Aldi table top oven had its temperature control replaced by an industrial PID controller plus Type K thermocouple, so it can't do sweeps, but can achieve and hold a stable temperature quite well. And like your reptile box, suffers because it doesn't have active cooling. But it's still way better than nothing and has produced useful data.

[mendip_discovery]

How does Humidity affect the measurements?

If you like to lend me one I can do an interlab for you as I have a 3458A at work.

[TizianoHV]

Hi, another interesting parameter for high value resistors is voltage coefficient.

While in search of good resistors for a multi kV voltage divider I noticed how voltage coeff. can be really important.

Here compared a few high voltage resistors at 0.1 / 4kV against a reference divider.

-Yellow trace: the reference divider (5* 100Mohm SSX-26 resistors in series ) is compared against a 140kV divider, confirming its accuracy.
-The other traces are some SSX-26 4kV precision resistors, with a rather high voltage coefficient.



Note: for every voltage step there are two values: one measured immediately after applying voltage, the other measured after 1 minute settling time.

*edit
Maybe for the 100Mohm (and the 1Gohm) resistor could you add one digit to the measured value? I was able to measure accurately down to 100ppm my SSX-100Mohm resistors without special equipment (6.5multimeter + 100kohm foil resistor + low noise voltage source).
After correcting for ambient temperature my measurements were <30ppm off against a Fluke 8508A.

[Alex Nikitin]

Maybe for the 100Mohm (and the 1Gohm) resistor could you add one digit to the measured value? I was able to measure accurately down to 100ppm my SSX-100Mohm resistors without special equipment (6.5multimeter + 100kohm foil resistor + low noise voltage source).
After correcting for ambient temperature my measurements were <30ppm off against a Fluke 8508A.

I am not confident enough to give more digits for high Ohm resistors in the measured values, sorry. I might yet improve my lab abilities in this respect but it takes time more than anything.

Cheers

Alex

P.S. - yes, I know about voltage coefficients, I've worked in the past extensively with high stability and precision HV sources, up to 60kV but fairly low power. 

[Alex Nikitin]

For those who received first few units with Dale RNX 3/8 1% resistors for 100M and 1G sections here are the graphs. I still have some RNX resistors and might yet use these at some point.

Cheers

Alex

[EC8010]

Could you provide some details on how the resistances were measured? Methodology, equipment? Was it simply using the resistance range of an instrument, or was it a potential divider technique from a known voltage? Or Hamon divider? This whole thing has made me think hard about resistance measurements.

[Alex Nikitin]

Could you provide some details on how the resistances were measured? Methodology, equipment? Was it simply using the resistance range of an instrument, or was it a potential divider technique from a known voltage? Or Hamon divider? This whole thing has made me think hard about resistance measurements.

It is complicated. I used several different ways for measuring the resistance and cross-checking the results. For the TC curves above I've used the Keithley 263 as a current source and HIOKI DM7275 as a voltmeter and a thermometer, with the voltage across the measured resistor about 10V (so thermal voltage effects are reasonably benign). This way I have the best resolution as long as the current is stable, so it is a good way to see and measure temperature coefficients, however the absolute accuracy is relatively poor as the accuracy of the K263 current is limited. It is good enough for 100M resistors and above though and I've used it to cross-check absolute measured values. The Keithley 263 has a separate voltage monitoring output from the internal electrometer amplifier, so the load resistance can be measured this way (and that is how HIOKI is connected to measure TC).

The K263 also has a set of reference resistors in nine decades from 1K to 100G, with the values from 100M and above calculated from lower (directly measured) values, step by step using the "ladder" method during the calibration routine (and it needs only an accurate DC voltmeter to do it). So the second method I use to measure 100M and above is a direct measurement by the Keithley 617 electrometer in I/V mode, with 10V across the DUT, after calibrating the 617 from the values of the 263. For examle, I would like to measure 1G as accurately as possible using the value stored in the K263 as the reference. First I measure the reference 1G resistor in K263 using K617. The value displayed on K263 is 0.99945 Gohm and the 617 shows 0.9999 Gohm so roughly 0.5M high. The error is recorded and subtracted from the measurement of the unknown resistor, for example for my reference 7D unit s/n 240800 the K617 shows 0.9975 Gohm and I record the measured value as 0.997 G.

For the lower resistances I have a better reference - the Fluke 742-10k with the measured value 10.00005 K and I use my HP3456A meter to measure the 10K resistance directly (4W, O.C.) after it was calibrated using the Fluke, so the error is smallest this way, despite "only" 6.5 digit resolution of the 3456A . The 100K section is also measured the same way but the HP 3456A is calibrated using a Hamon transfer with SR1010 10K for 100K range and further up to 1M using SR1010 100K. 1M and 10M are measured in 2W mode directly by the HP3456A and I use a self-build Hamon 1M set to calibrate the 10M range. I also double-check the values on the next range up for each measurement, and though it only gives 10ppm resolution it is a good sanity check to see that nothing is amiss.

Cheers

Alex

P.S. - I also measure and select all resistors before I build the array, and use copper clips to reduce the impact of soldering during the assembly. To measure resistors over 10M I use a separate enclosure for shielding.

[EC8010]

Thank you for that. I thought perhaps a current source might be involved for the larger resistors. As you say, although absolute accuracy is not wonderful, it does allow temperature coefficients to be determined. Hamon transfers for lower values understood - they're on my "to do" list.

Glad to hear of copper heatsink clips. I recently bought 100 1M 0.1% resistors, measured and recorded them. I may sacrifice some of the outliers to science by experimenting to see the effect of soldering with and without thermal protection.

I have various screened enclosures that improve measurements. Interestingly, oscilloscope monitoring shows that the screening can for my FET VTO tester mainly benefits by preventing air draughts across the device rather than by reducing hum.

[guenthert]

   I'm a happy owner of one of those units and attempted to measure the "voltage coefficient" (voltage dependency of resistance) of the 10GOhm resistor using my Keithley 617 in V/I mode at 3, 10, 30 and 100V.  The reported value increases very slightly, but I now suspect this is due to a (minuscule) leak current of the measurement instrument/set-up which becomes then less relevant with higher driving voltage and resulting total current.

   I don't currently have a 1kV source available for further tests.

[EC8010]

I'm a happy owner of one of those units and attempted to measure the "voltage coefficient" (voltage dependency of resistance) of the 10GOhm resistor using my Keithley 617 in V/I mode at 3, 10, 30 and 100V.  The reported value increases very slightly, but I now suspect this is due to a (minuscule) leak current of the measurement instrument/set-up which becomes then less relevant with higher driving voltage and resulting total current.

I doubt it. If your 617 is similar to my K6517B it's because the voltage coefficient of the thick film resistors in the resistance box is being cancelled by the voltage coefficient of the thick film resistors in the Keithley's electrometer. (I was told by a resistor manufacturer that when you get to really large resistances, there's only one plausible composition, so they're all very similar.) I tried using my K6517B to measure voltage coefficient of some high value thick film resistors and they came out near-perfect. So I then measured a string of 10M metal film resistors and they showed a much larger voltage coefficient because it was no longer being cancelled. I'm now very cautious about the K6517B.

[Alex Nikitin]

If you use the I/V method in the K617, the voltage coefficient of the internal resistors is not important as the meter works as a picoammeter and a) the voltage on the feedback resistor is not more than 20V and b) measuring 10 Gohm at 10V is measuring 1nA and at 100V is 10nA, in the K617 it would mean using 100M feedback resistor at 100mV and 1V.

The main problem when using the I/V method at different voltages is actually the voltage source in the 617 as the error of it directly affects the result, so the only ways to improve the accuracy are either to monitor the voltage source output with a separate accurate voltmeter (and correct for the voltage source error) or use an external precision voltage source.

Cheers

Alex

[EC8010]

Ah. At FSD, the 6517 applies 200V to its feedback resistors...

[Alex Nikitin]

Ah. At FSD, the 6517 applies 200V to its feedback resistors...

Trading linearity for noise performance 

Cheers

Alex

[EC8010]

Not so much noise but it reduces the amount of switching needed to get ranges from 200pA to 20mA FSD. I originally thought it was a clever idea (and was going to copy it). But when I tried the voltage coefficient experiment, realised it wasn't. So your earlier generation electrometer is better.

[Alex Nikitin]

Not so much noise but it reduces the amount of switching needed to get ranges from 200pA to 20mA FSD. I originally thought it was a clever idea (and was going to copy it). But when I tried the voltage coefficient experiment, realised it wasn't. So your earlier generation electrometer is better.

Noise is important too. An increased maximum voltage on the feedback resistor allows to use a larger value feedback resistor for the same current range. Increase from 20V to 200V = x10 feedback resistor value = 3.16 times improvement in the current noise density at the same current in the same range.

Cheers

Alex

[EC8010]

Good point; I had forgotten that. It's not intuitive that you get less current noise from a larger resistor. Of course, if you really look at it, you'd point out that you will get excess noise from the voltage across the (thick film) resistor. Measuring little currents is not trivial.

[Alex Nikitin]

This thread is about measurements of the 7D array on its own, but also about its applications. Some are obvious, as using the array as reference resistors and dividers, some are less obvious. One of such applications is the measurement of the input current for bench multimeters (and some hand held ones) which have a "High Z" input feature.

It is actually very simple, you need to supply a known voltage to a multimeter through a high value resistor of the array and see how much the voltage shown by the meter differs from the voltage supplied. For example, if the input voltage is 10V, the series resistor is 1Gohm and the displayed voltage is 9.95V, the input current of the meter is 50mV/1G = 50pA. for better meters 10G resistor could be used though you might need to wait some minutes for every point measured to get a stable result. Below the results from some of my meters, at 10V range, 100NPLC, from -10V to +10V at 2V steps, using the HP3245A as the voltage source (for the Keithley 2015M I've used 1Gohm resistor, for the other two meters - 10Gohm).

Cheers

Alex

[guenthert]

Funny, I did exactly that, albeit verifying the input impedance of my Keithley 617 electrometer.  While in storage for five years, it developed a bias current far out of specification (~66fA).  I thought that might be humidity related, but the current remained the same after 24h warm-up.  I was able to adjust it to less than ~5fA, but reached the end of the adjustable range.  I suppose a range setting resistor might have drifted too far and needs to be replaced soonish.  Anyhow, now the indicated voltage of a 10V reference drops by 3mv (or less, when patient) over the 10GOhm resistor, which I take to mean a current of 0.3pA or an input impedance of more than 30TOhm (specified are 200TOhm).