T.C. measurements on precision resistors

[SilverSolder]

Is the ultimate solution to have precision components in a vacuum...

[alm]

additional NIST article which is lost in their own server, but printed this page using google cache.
NIST pubid 9237 page 11
"wirewounds have no voltage coefficient"

This article from the Encyclopedia of Applied Physics is not lost on the NIST server, just misplaced
https://www.nist.gov/publications/resistors

Is the ultimate solution to have precision components in a vacuum...

I doubt it. Outgassing of the insulators might become a problem. And heat dissipation will be more of a problem. I think putting the resistor in an inert gas like neon or argon would work better. But since there seem to be very few standard resistors employing this except for some high-value resistors, I'm guessing it is not worth the effort.

[e61_phil]

We operate very high ohm resistors (up to 10T) in vacuum for several reasons.

[FriedLogic]

1) a TEC based "resistance standard"

Would the 2PPM over 6 years that Vishay quote for some of their resistors mean that their standard processing for them is effectively a specific one for optimizing the drift for zero load and room temperature?

[Magnificent Bastard]

1) a TEC based "resistance standard"

Would the 2PPM over 6 years that Vishay quote for some of their resistors mean that their standard processing for them is effectively a specific one for optimizing the drift for zero load and room temperature?

That's a good question, but you should be asking a Vishay Precision Group applications engineer.  You can find the contacts you need on their website.  There are two things that will make a resistor change more than "shelf life"-- the first is higher temperature (which can be self generated heat or externally applied heat-- they both will result in the same drift), and temperature cycling.  The reason the SR104 is so stable is that they are in a thermally stabilized enclosure, and the temperature changes are slow and gradual-- as well as the fact that most of them are kept in a calibration lab that has a fairly constant temperature.  If you are exploring making a "standard resistor" with foil resistors, may I suggest you use a ceramic PCB, and then mount multiple FRSM resistors (say nine 10K in series/parallel to make 10K)-- then place this in a hermetic package.  You could even go to the effort to temperature compensate this resistor and trim to very close to 10K.  When you get done, you still will NOT have a primary standard like an SR104, but it will be a very good secondary level standard.

[Andreas]

Would the 2PPM over 6 years that Vishay quote for some of their resistors mean that their standard processing for them is effectively a specific one for optimizing the drift for zero load and room temperature?

I think the specification is a bit unclear.

Shelf life drift is usually specified per year.
The VHP100 spec is:
"Shelf life stability: ± 2 ppm typical after at least 6 years"
So what does 2 ppm after at least 6 years mean?
Does it mean 2 ppm per year after a pre-ageing of 6 years?

What does "typical" mean?
Is it the "average value" of a large lot with min and max values of +50 /-46 ppm?
What is the "stray" or standard deviation?
What is the guaranteed maximum?

So without a measurement specification the 2 ppm value is just advertising.
Similar to the +/-0.05 ppm/K spec within 0..60 deg C which give "typical" 1 ppm/K gradient near room temperature in my Z201 measurements. Up to now I have only found one sample which has near 0.05 ppm/K.

with best regards

Andreas

[Kleinstein]

The testing of the resistors for the KATRIN experiment is to a level better than manufacturer specs. So even the failing ones may still be good. The 200 Pieces batch was likely ordered to have enough spares - not all the not used resistors must have failed.

The scattering in the bad resistor curve looks a lot like resistor excess noise. This normally should not be a problem with metal foil resistors, but there seem to be a few bad ones with poor contacts or other defects (e.g. crack in substrate, contamination that allows parallel paths). At the shown noise level the resistor is no longer good, but a 100% failure. It may still work at lower voltage though.

Resistor excess noise (commonly called current noise - but this a confusing name) can be a problem especially with film resistors. The resistance may change on the sub ppm level just randomly, not only from temperature or humidity effects.

[dietert1]

Some weeks ago we had some results on Vishay hermetic resistors, ending with the recommendation of FriedLogic to watch the DUT for extended periods of time, not just some oven cycle. My last result had been a 4 ppm drift after an oven cycle with a time constant of about 45 minutes.
Meanwhile i improved my resistor measurement and observed the HCZ500R00 again.
After heating it to 37 °C it showed that 4 ppm drift again. But then it drifted back with a time constant of about 3 days. After that i did some oven cycles, but it became clear that the results are meaningless, if the resistor shows relaxation at time constants of 3 days. So i started measuring the step response after an oven temperature step 40 °C => 30 °C. Since the oven isn't perfect, the temperatur step was in fact from 38,6 to 29,6 °C, about 9 °C.
The log shows four distinct reactions at different time constants and the effective TC becomes zero twice after the step. On the long term TC is positive and about 0.46 ppm/K.

1) First reaction is a negative resistance change of about 2 ppm after 5 minutes, so that would be a TC of -2 ppm / -9 °C = 0,22 ppm/K
About 11 minutes after the step TC = 0.

2) Next is a positive resistance change of about +1 ppm after 22 minutes. If you stop then, TC = 1 ppm / -9 °C = -0,11 ppm/K
About 40 minutes after the step TC = 0 again.

3) Then we have a resistance change of about -5.2 ppm after 6 hours. This is a TC of -5.2 ppm / -9°C = 0,58 ppm/K

4) About 1.1 ppm of those -5.2 ppm get lost after about 4.5 days. So the long term TC is about -4.1 ppm / -9 °C = 0,46 ppm/K

At the end i left the measurement running some more days logging the linear drift of my bridge, which is about 0,11 ppm/day. Standard deviation of the filtered measurement during that drift was 0,0106 ppm of resistance. In the "slow" diagram residual temperature variations of the oven and the bridge linear drift have been subtracted.

I think the complicated step response mirrors the complicated construction of the Vishay hermetic resistor (foil, glue, carrier...). Andreas already explained that the glue with its creeping feature may constitute a long term memory. For me it is obvious now that all measured TC curves for such resistors are more or less meaningless.

Next i will probably try a similar measurement on a UPW50 thin film resistor. Hope its response will be more straightforward.

Regards, Dieter

Edit: I meant UPF50 (thin film resistor) instead of UPW50 (wirewound).

[Andreas]

Hello,

I would not say "meaningless". At least you get a feeling how repeatable the temperature curve is and how much "hysteresis" is involved in the construction.

UPW = wire wound
do you mean
UPF = thin film?

UPF showed very low hysteresis when I measured them.
https://www.eevblog.com/forum/metrology/t-c-measurements-on-precision-resistors/msg462301/#msg462301

with best regards

Andreas

[dietert1]

I think the Vishay hermetic resistor does not have any memory besides the effects i have shown, except long term aging.
If you manage to get it into the same "state", it will show the same reaction afterwards. Here i can show the two oven cycles i measured. I lost some data on the first one, but it is clear that the agreement is near perfect. You can see that i used a 45 °C temperature setting to "program" the resistor for 40°C instead of waiting another 3 or 4 days after each disturbance.

These cycles are at 0.1°C per minute (lowest setting of the Arroyo 5305). If you use a different speed or steps, the plot will look completely different. In some sense the term TC does not apply to that Vishay foil resistor. Instead of delta R = TC * delta T, you need to calculate the convolution of temperature history with the pulse response = time reversed differential of step response. What i have seen is something like a LTI system.

Regards, Dieter

[dietert1]

In order to convince myself about the LTI character of that hermetic Vishay resistor i repeated the measurement, this time with a +2 °C temperature step. The device shows a very similar response, except the initial peak is somewhat bigger. This should be due to the fact that the oven executes a 2 °C step faster than a 10 °C step.
In order to extract the data for the slow tail i had to change the drift rate and to include a small nonlinear temperature response of the reference resistor inside the SR1010.
I (mainly) attribute the drift to the humidity increase due to leakage of the SR1010 and its box. After inserting the desiccant i had about 4 % HR, the -10 °C step was measured at about 9 % HR and now i have 13 % HR. One may expect that the drift gets slower as it probably is proportional to the humidity difference.

Regards, Dieter

[FriedLogic]

The only resistors (other than an ancient Welwyn foil) that I saw reacting badly to temperature gradients in that tests were the Alpha HCZ ones. The Alpha FLCX100R, FLCX10K and MAY1K ones, and all the Vishay ones were fine. Unlike most of the others, the ebay VHP101's also showed little drift after their temperature was raised to around 60°C, but their TC was borderline.

I tested one of the cheap 20K VHP101 resistors that Farnell had on offer, and other than the first temperature cycle, it also showed little drift after the temperature was raised. It really does have a good TC too, unlike the two ebay ones.

Of course, that's only a test of a few resistors, so I can't draw too many conclusions - but some patterns do seem to be forming...

One interesting thing that I noticed when testing the setup was that the network cables that I was using generated up to around +/- 0.3uV themselves for the 40°C temperature rise.

[WattsThat]

Very interesting thread, especially so as I once upon a time worked for the Vishay Resistive Systems Group (as it was called back in the days before they went into acquisitions mode).

What I find notable is that back in the day (I left in 1981), VC was not a parameter. Never measured, never discussed.

On the calibration side of the business, the primary standards were two ESI 10k’s. Everything in between was derived with transfer standards (Vishay boxes, of course) from those two ESI’s. They never left the building at the same time, alternated them out for cal. They were ten plus years old when I used them and neither had drifted more than 2 ppm absolute from 10k in their entire NBS cal history.

[branadic]

I measured several Alpha Electronics HCZ resistors over temperature with my Prema 5017 SC and its internal multiplexer. Goal was to find some resistors, that can be arranged to make a 10k reference that performs better than a single 10k resistor (one 10k HCZ is included for reference). Also measured one channel shorted for sanity check.
Second temperature run is already in progress to verify the very first results. I found that paralleling resistor on channel 3 and 8 can give a decent reference. Any suggestions on that?

-branadic-

[TiN]

That's crazy amount of noise, but I guess it's normal for this meter? Depends on your end goal, but I'd avoid stressing resistors on such temperature extremes, and rather stick to common +18...+28 range. Perhaps separate test for hysteresis could be helpful, to determine if there are effects introduced to HCZ from going all the way to +60°C.

[branadic]

TiN, you are dirty by 8.5 digit meters and calibrators
I used what I have at hand and the only meter with a multiplexer is this Prema unit which saves my a lot of time as I can measure multiple resistors in parallel.
I'm still in the temperature range that is given in the datasheet, so if a resistor fails here, it would also fail in a smaller temperature range such as 23 ± 5°C, plus it would be harder to see/identify and would need a lower noiser/higher resolution meter, right?
I have expected much more difference as this resistors are from two different manufacturing dates.

-branadic-

[Kleinstein]

The noise level is not great, but it is good enough for the job. There is some hysteresis, but not too bad.

As expected the resistors have a rather similar quadratic term and some individual linear part. So combining same type resistors has a limited effect. One can compensate the linear part, but not the square part. This may be still acceptable for a reference that normally is used in a more limited temperature range.

Going to only 60 C should still be Ok, though it can effect internal humidity.  The nasty parts and changes may start from some 80-120 C when some epoxy/plastics can come close to there glass temperature and thus change there internal structure and this way restart the aging.

[Andreas]

I found that paralleling resistor on channel 3 and 8 can give a decent reference. Any suggestions on that?

Hello,

I would calculate the LMS coefficients for a 3rd order parabolic curve (normalized to 23 or 25 deg C whatever they will have in final application) and then select by the coefficients.

Unfortunately the hysteresis increases (and may lead to permanent shift) with larger temperature excursions.
So I would limit the measurement to the range of final application. (+ a small reserve).

What I am missing is one of the resistors kept at constant temperature during measurement to do a sanity check for the T.C. of the instrument.

with best regards

Andreas

[branadic]

Thanks for the feedback. I thought that measuring a short would be good enough as sanity check (gray curve lower left). However, looking at the values the 20k resistor on channel 9 looks like a good candidate for reference, as it shows only small t.c. and I only have one temperature chamber so no possibility to keep it at constant temperature, but I can measure it together with ambient temperature sensor. Otherwise I would need to machine some aluminium and add a modified crystal heater to it, to make a constant temperature reference resistor. Unfortunately my P331 10k is still waiting for arrival at my location, which probably could have served as a 10k reference

In the next run I will reduce the temperature range to 10 - 40°C, which I think is needed to see some curvature and which also speeds up the measurement by about 6h. Final verification of single resistors can be done using R6581 with lower noise.
This is all preparation to measure FLC resistors for a Hammond divider based on the design by Frank, so each run is a leaning step for me.

-branadic-

[TiN]

branadic
I really don't have any experience with Prema, but I expected less noise from 7.5 digit meter. It seems like similar to Keithley 2001. Does the Prema mux have external outputs? If so, you could use R6581 to read resistors and Prema as simple mux switch. Like Kleinstein said, for TC only tests noise level is not very important, as far that you are able to get filtered curves.

Hysteresis and resistance shifts from thermal cycling is easier to see if you run multiple sweeps, like 3 or 5 times for the same resistor set. Wirewound resistors of all kinds are often prone to show more sensitivity to thermal cycling, as they are much larger physically and more sensitive than tiny 5x5 BMF element. So P331 might not be optimal as reference. I have learned from my tests that saving time for resistance measurements often backfire and cause unreliable data  . Hence I ended up with multiple 8.5d meters and calibrators, to verify that and build up confidence in measurement setups when obtain same results by different methods/instruments.

Zero short would not help as sanity check, because it does not show you gain errors from DMM, but purely TEMF offset and ADC noise. 

[guenthert]

Oh, just average the values, so that TiN's sensitive eyes won't get hurt looking at those jagged graphs.

[branadic]

TiN
Unfortunately the MUX doesn't have external outputs and it would need modification on the meter that I'm not willing to do. I'm still angry about myself, that I sold my Prema 2080 (20/40/80 channel MUX), when I received this Prema 5017SC, as I thought it wouldn't be of any use. But now that I have R6581D it would be a great expension for it. I couldn't find a decent replacement for it by now.

Two sweeps are already done. Made some modification to the setup, put AE20k from channel 9 to channel 11, which is outside the chamber, placed 10k Ohmite resistor to channel 9 and installed 10k Z201 to channel 10, which is the last port available inside the chamber. Also modified the snake script to run 3 profiles in the range of 10 ... 40°C. So I will know more in about two days.

However, the final run on a resistor arrangement will be done with R6581.

-branadic-

[branadic]

I currently measure on NOMCA1603 networks. I ran into some issue before, as I use a zero force socket. Obviously the socket was not able to reliable break through the oxide on the leads, which resulted in strange t.c. curves. So with a bit of flux and an almost clean solder tip I had to break the oxide, now I get more reasonable results. Here is a first view on a first samle. The t.c. seems to be almost linear.

Edit: Calculated some first values

R1: 4.333ppm/K
R2: 5.452ppm/K
R3: 4.983ppm/K
R4: 4.667ppm/K
R5: 6.072ppm/K
R6: 5.409ppm/K
R7: 5.986ppm/K
R8: 4.823ppm/K

-branadic-

[branadic]

The resistor values measured can now be used to calculate a low t.c. 13:1 divider for e.g. LTZ temperature setpoint made out of 7 resistors as shown previously. Did that already and solved 8 possible equations as an example, though not all. And indeed I could find one combination resulting in a divider with very low temperature influence (blue curve), though power coefficient is not faced here and heat distribution within the network still has to be thought of, but I found it interesting to show anyway. The combination here is a series connection of R3+R4+R5 and parallel connection of R1||R6||R7||R8.

-branadic-

[dietert1]

Last summer i used this little software to find the best combination. Hope it helps.

Regards, Dieter