T.C. measurements on precision resistors

[Andreas]

Hello Illya,

from the diagram there is a "warm drift"
during cold phase the resistance is proportional to the temperature.
during warm phase it  does not stabilize within the plateau time.

nearly 50 ppm in 3 days drift is quite a lot.
Cannot rember having that large drift on one of my measurements.

With best regards

Andreas

[CalMachine]

732B cal'd directly against Fluke Everett's JJA in January and they reported a measurement uncertainty of 0.06 µV/V.

That is measurement uncertainty, but your final 732B uncertainty at your site is higher, because it's sum of all uncertainties and 732B stability. Per 732B spec your end uncertainty at the box connector is +/-0.36 ppm/30 days, or +/-0.86 ppm/30 days or 2.06ppm/year assuming that there was no output change due to shipping, and conditions of your lab and Fluke's calibration facility are exactly the same.

Yes, I know this.  He had made the claim that 0.3 PPM was the best uncertainty you were going to find from a lab with access to a JJA.  I provided a case of a lab with a JJA of having much much less uncertainty.  In NO way was I trying to insinuate that was my uncertainty.....

[MisterDiodes]

Yes, I know this.  He had made the claim that 0.3 PPM was the best uncertainty you were going to find from a lab with access to a JJA.  I provided a case of a lab with a JJA of having much much less uncertainty.  In NO way was I trying to insinuate that was my uncertainty.....

About 0.35ppm uncertainty is the typical advertised specified limit -at- Fluke, and that includes the uncertainty of transfer between their lab to and from NIST.  They can also to tighter tolerance measures if you pay for it.  You won't have that uncertainty after the 732b gets shipped to you - every transfer adds uncertainty.  I know they have 732b's shipping to and from NIST and other JJ-array labs almost continuously - and newer methods involve shipping "Compact" Josephson junction array equipment to inter-compare the JJ-Arrays.  That's what it takes to get into fractional ppm world and NIST traceability.

That is not the same as their lab measurement resolution, which is usually finer-grained than uncertainty - they are just telling you what their equipment reading was...and that will have a specified uncertainty of the absolute value, before they shipped the Vref to you.

Even the manufacturers of JJ-Array will give you these measurement uncertainty specs if you order a new JJ-Array, with very good uncertainties in the 10's of nV range on 10V scale.  This gets better with the newer JJ-Array 10V chips.  But in the 0.2~0.3ppm uncertainty on a voltage measure is considered very good for typical alignment to another lab without a government budget.

Normally you will have a 734 unit with at least three other references at your location, and as soon as you get the freshly cal'd 732b back form Fluke you add that to your existing pool and immediately compare with your existing pool members and start running the numbers to see if you see drift in the just-received unit or any of the other three - and then you make a decision on which ref to send in next.

You are always looking at the measured drift vs. the predicted drift to see how your Vref is settling in. Older units are generally much better performers than brand-new units.

Normally on a new 732b you'll run cal cycles every 6 months or sooner until you begin to see it settle down and you can detect a predictable drift rate - and then you can go longer in between cal checks.

If you do all that religiously your lab will eventually have about 2~3ppm real measurement uncertainty capability, traceable back to NIST.  Add more cal'd 732b's to increase confidence and lower uncertainty.

Lots of docs at Fluke site that explain in better detail.

3458a's - as good as they are - aren't really considered a shippable transfer standard.  They can be used as a short term, in-lab voltage measure transfer though, within limits - especially if their drift rate is fairly stable and known.

[CalMachine]

Yes, I know this.  He had made the claim that 0.3 PPM was the best uncertainty you were going to find from a lab with access to a JJA.  I provided a case of a lab with a JJA of having much much less uncertainty.  In NO way was I trying to insinuate that was my uncertainty.....

About 0.35ppm uncertainty is the typical advertised specified limit -at- Fluke, and that includes the uncertainty of transfer between their lab to and from NIST.  They can also to tighter tolerance measures if you pay for it.  You won't have that uncertainty after the 732b gets shipped to you - every transfer adds uncertainty.  I know they have 732b's shipping to and from NIST and other JJ-array labs almost continuously - and newer methods involve shipping "Compact" Josephson junction array equipment to inter-compare the JJ-Arrays.  That's what it takes to get into fractional ppm world and NIST traceability.

That is not the same as their lab measurement resolution, which is usually finer-grained than uncertainty - they are just telling you what their equipment reading was...and that will have a specified uncertainty of the absolute value, before they shipped the Vref to you.

Even the manufacturers of JJ-Array will give you these measurement uncertainty specs if you order a new JJ-Array, with very good uncertainties in the 10's of nV range on 10V scale.  This gets better with the newer JJ-Array 10V chips.  But in the 0.2~0.3ppm uncertainty on a voltage measure is considered very good for typical alignment to another lab without a government budget.

Normally you will have a 734 unit with at least three other references at your location, and as soon as you get the freshly cal'd 732b back form Fluke you add that to your existing pool and immediately compare with your existing pool members and start running the numbers to see if you see drift in the just-received unit or any of the other three - and then you make a decision on which ref to send in next.

You are always looking at the measured drift vs. the predicted drift to see how your Vref is settling in. Older units are generally much better performers than brand-new units.

Normally on a new 732b you'll run cal cycles every 6 months or sooner until you begin to see it settle down and you can detect a predictable drift rate - and then you can go longer in between cal checks.

If you do all that religiously your lab will eventually have about 2~3ppm real measurement uncertainty capability, traceable back to NIST.  Add more cal'd 732b's to increase confidence and lower uncertainty.

Lots of docs at Fluke site that explain in better detail.

3458a's - as good as they are - aren't really considered a shippable transfer standard.  They can be used as a short term, in-lab voltage measure transfer though, within limits - especially if their drift rate is fairly stable and known.

I apologize to delve more off-topic.  But I feel like as if I am being taking in the wrong light, here.

I think I interpreted your statement differently than everyone else.  As someone who works for an accredited lab, their stated uncertainty on their scope and the stated uncertainty, of the measurement in question, on the calibration certificate are 2 very different things to me. 

Their advertised uncertainty on their scope is for their service lab, in which your reference is compared against their array of 732A/Bs what have you...  This 732B was sent to their standards lab in which it was compared directly against the JJA, which yields the 0.06 µV/V uncertainty.  Not the 0.3 µV/V stated uncertainty on their scope.

When I go to calculate my uncertainty budgets, this matters.


Continue on with the TC measurement talk!  Forget I said anything   

[MisterDiodes]

Hi MisterDiodes,

there were four people in the LTZ thread, who virtually did the same measurement, like you, I assume, i.e. modifying these both resistors by some percentage, and measuring the LTZ1000 output deviation.. this varied from 74 (Andreas) to 105 (somebody on bbs38hot).
So 74 was worst case.

I just mentioned that, because even using these worst case T.C. numbers, the effect and the optimizing of the resistors T.C. are of 2nd order importance only ..
The residual T.C. will always be below 0.05ppm/k, and may be disappointingly too high, if you set your money on expensive ultra - low - T.C. Vishays, or T.C. matched sets.

The LTZ itself, i.e. up to now undisclosed / not yet explained effects, will play an important role, so a T.C. trimming to << 0.05ppm/K might be accomplished otherwise, e.g. by this T.C. compensating resistor

What interests me, in this context, how did you characterize the T.C. of your LTZ1000 devices, as most of the gear accessible to us amateurs has much higher T.C.?

Do you know  a possibility to accomplish that, say trimming to 0.01ppm/K , w/o the aid of a JJA standard?

Frank

Hello Dr. Frank - I want to say first off I appreciate your thoughtful comments.

Not to be shy, but I spoke to my client about what they want revealed about their testing process, and I have to decline an -exact- description of equipment used.  What I can describe is a "Black Box" test jig (capable of measuring dozens of test voltages by various methods) built by the facility, and literally 10's of millions of dollars$$ in calibration gear.  Let's leave it that lots of 732b's that are regularly measured against the facility JJ-array, along with DMM's the likes of 3458a's and such.  The typical uncertainty at this test jig we used is about 0.3ppm or better, and we are > 95% confident we can accurately measure a +-1ppm voltage shift over a several day time span.

The test we performed on LTZs had a basic 13k over 1k resistor ratio, and we had several resistors that could be added in series to the 13k ranging from 0ohm to 1.40ohm via jumper block.  The basic 0-ohm dummy resistor was added to 13k to get a baseline reading, then add in 1.20 Ohm resistor, wait 48~ 72 hrs, look at Vout change.  Everything else being equal.  Then go back to 0 ohm resistor, wait, then add in 1.25 Ohm resistor to 13k and wait 48 hours again.  And so on.

In our tests the 10ea. LTZ performed within all datasheet spec for heater resistor sensitivity of an attenuation of slightly over of a factor of 100 to see a 1ppm change in output voltage by this method.  IT SHOULD BE NOTED though that the heater ratio stability is greatly influenced by the heat flow into and out of the LTZ device.  You can get different results based on A) A particular LTZ and its age / stress level and B) How thermally insulated the LTZ is (or not) to ambient.  HINT:  You can over-insulate an LTZ to the point where the heater circuit does not function well, and that's just as bad as drafty or no insulation.  This effect can show up as what looks like an unstable heater resistor ratio problem.

Remember, the LTZ is a power out vs power in device.  You will get a stable Vout only when the circuit has good, solid thermal stability and the heater circuit can servo to setpoint temp correctly.

RE: Add TC compensation resistor.  We Looked at this once, but here's the caveat to chasing 0ppm TC, especially on LTZ1000A - You can do that, but we found in long run that was never profitable to try to characterize the LTZ circuit over time (and over several units) and adjust that TC resistor individually for each LTZ.  The real trouble comes 5 years later when you realize the die has stress-relieved itself, and that TC resistor that worked originally now really isn't doing the job. 

If you're playing with just a one-off LTZ circuit on a well-aged LTZ die, this might be fun to try for entertainment - but at least in my limited experience the correct TC compensation resistor value doesn't last forever.

We found other ways to compensate for slight TC adjustment if our circuit is running warmer or cooler than anticipated - and only if required by customer specs.  Normally we are after a very -quiet- voltage source for the equipment we are building.  Yearly drift rate or extreme low TC over say a few 10's degrees C is not the primary requirement.  Normally the datasheet-spec TC of an LTZ on a well-built circuit board and enclosure will do just fine as-is.  We use only LTZ1000a units.

The advice to hobbyist: The more tests the merrier.  3458a's (and similar DMM's) are great instruments for getting an initial approximation, but if you collect 732b's and 7-decade KVD's and even null meters you can get even better approximations - especially when you run critical tests on battery power.  Or at least get rid of all local switcher noise.   If you are chasing PPM-level absolute values traceable to NIST or whatever government lab you like, keeping your equipment calibrated often is the best approach - you generally want a Z540- type calibration so you can see what your actual equipment measures at - A typical regular low-cost calibration will just tell you if your equipment is within spec or not.  Again: Low-noise Battery powered tests are your best friend, and yes on the production line tests I see these methods are used as well on very critical test points.

My other suggestion: Keep track of your 732b calibrations, and keep track of how well the predicted drift matches the actual Z540 cal information when you get a freshly cal'd 732b back from Fluke.  That measurement data over time will really help that 732b become more valuable as it gets more stable over time.

[MisterDiodes]

Back to resistors, and Edwin's comment on the resistance wire TC:

We use Edwin's resistors as well as General Resistance - and yes you have to look at multiple factors if you're measuring resistors... Dr. Frank touched on this also:  There is the resistance of the wire, and then then is the real-world effects of everything else:  Bobbin & winding stress, how well it is sealed (or not), welds, etc.  It is not only the basic TCR of the wire changing resistance with temperature, it's also the stresses added to that that are changing as well.

In other words you really have to look at the total "Apparent Resistance" of the resistor, which might be different from the initial wire rated TCR.  And if you're looking at resistors used as a divider ratio, you really want to look at the entire Divider Ratio Change vs Temperature.

As Dr. Frank pointed out:  You can have the resistance divider set wound with the same wire, so that the wire TCR is the same for both resistors in the divider.  The little caveat to that is in the case of LTZ heater resistors, the power dissipated in each resistor is not the same, and that means the heat-induced wire stress levels are not the same - so you'll almost never get a theoretical 0ppm TC resistor ratio change.  You'll come close with careful construction, but it's very hard to chase after that 0ppm TCR.

Fortunately the LTZ circuit is forgiving in that you don't need 0 TCR ratios, just low TCR will be OK over your anticipated operating temperature range.

What my point is:  We used to try and characterize heater ratio resistors like was TiN is doing and at least for us - that method wasn't profitable.  That doesn't mean to take away from TiN tests, not at all.  It's just hard to pin down what you see for apparent resistor ratio stability and how it affects the LTZ - ON YOUR APPLICATION LTZ CIRCUIT which will have its own thermal characteristics.   You'll see one thing on TCR charts and measures, and then you'll find out something else when you run the same resistors with your LTZ.

Just take the TCR measures with a little grain of salt on how the lTZ is really affected.

It was just for us, it was much easier and faster us to build the LTZ circuits and let them run - and then we could see how different resistors worked or not.  Generally they all work well, and that's when we found out the expensive Vishay's really offered nothing for the cost / benefit bottom line.  Even $1 PTF56's will work if your required temperature operating range is not too big.  It just depends on what final spec you are aiming for.

[Kleinstein]

The TC measurement does not need a well calibrate meter. It is only the short time (e.g. a few hours to days) drift of the ranges used that is really important. It would be only if you care about an accurate values for a high TC (e.g. measuring PTCs) that this might be an issue - but even than temperature accuracy is way more important. So the calibration question is off topic here.

How sensitive the LTC1000 circuit is to changes in the temperature setting resistors, can depend on the individual reference. The TC without temperature regulation in supposed to be at 50 ppm/K, but it can vary with samples (e.g. zener voltage), set temperature, ref current and a little on the lead length. Units with a lower TC should also respond less to a temperature change. There is also an influence from the "400 K" resistor used in the non A circuit to compensate for residual TC - this resistor (and if used the thermal layout) will also modify the way the circuit reacts to changes in the divider. The use of the compensation resistor also suggest that heater power could have an indirect influence. So it's possible that it sometimes takes 100 ppm and sometimes 70 ppm divider change for 1 ppm output change - this number is expected to scatter. I also think that trying to adjust the unregulated TC to a lower value is probably adding more trouble (another sensitive resistor) than it helps.

In theory there is even the possibility to allow for an external measurement of the set-point voltage and if needed maybe correct for the effect of a possible drift of the divider. Detecting a 100 ppm drift in a divider ratio is easy compared to a 1 ppm drift in an reference voltage.

With the 1 K resistor is might be important on how much of the lead wires is included in the measurement. The curves shown by TIN don't look like very good resistors - the TC might be acceptable, but the drift looks really bad. Since this is much worse than expected, it might be worth using a different setup / meter or check a different known good resistor of similar value.

[Andreas]

Since this is much worse than expected, it might be worth using a different setup / meter or check a different known good resistor of similar value.

Hello,

why do you expect better values?
Illya has relative high humidity in his measurement plots.
compared to my sensor values of below 40% rH at my place (in winter).
So its natural that a epoxy packaged device will give different results in different climate conditions.
By the way: I think that the dew point temperature has more relevance than the rH-value.

In theory there is even the possibility to allow for an external measurement of the set-point voltage and if needed maybe correct for the effect of a possible drift of the divider. Detecting a 100 ppm drift in a divider ratio is easy compared to a 1 ppm drift in an reference voltage.

With the original datasheet cirquit its really only in theory.
As soon as you connect a antenna (measurement wire) to this sensitive point the temperature of the LTZ + output voltage will change making the measurement obsolete.

With best regards

Andreas

[Andreas]

Hello,

to go further on topic my measurement of the last VHP202 resistor (a 10K resistor).

VHP202 10K #1

This candidate has a very low T.C. (below 0.1 ppm/K) but at the price of a relative large ageing drift.
And a around 2 ppm hysteresis for a 10-40 deg C temperature ramp.

First I thought the drift could be due to the fact that I had to exchange (solder) the 12K5 Z201 reference resistor.
But then I made some measurements without temperature ramping and with only 18-32 deg C ramp.
During this time (from day 9-14) the value of the resistor was rock stable.
After that continuing with 10-40 deg C ramps I got again a drift.
So at maximum the first 2-3 days could be influenced by the exchange of the reference resistor.

Attached: chart of drift of the 25 deg C LMS value.
the 2nd day  of the measurements (Day 1 in the drift chart).
the last day of the measurements.

One day of the "drift gap" with 18-32 deg C ramp

with best regards

Andreas

[TiN]

Some first data from 8hour step response. Rather interesting, yet still apparent TCR is reverse for 12.5K one.

[mimmus78]

TiN, as predicted by using a fast rump up you get what look like my overshoots ... check my charts too.

Inviato dal mio Nexus 6P utilizzando Tapatalk

[Edwin G. Pettis]

If I am reading your chart correctly, once your TECBOX reaches temperature 40°C and stays at that temperature, the resistors should also level out and stay constant once they reach thermal equilibrium.  Why are there slants in the resistor tracks when the temperature is apparently constant?  Is it just too late in the evening for me or am I missing something?

[MisterDiodes]

If I am reading your chart correctly, once your TECBOX reaches temperature 40°C and stays at that temperature, the resistors should also level out and stay constant once they reach thermal equilibrium.  Why are there slants in the resistor tracks when the temperature is apparently constant?  Is it just too late in the evening for me or am I missing something?

We have seen that before also.  Part of that is the thermal shock that resistor and bobbin goes thru during very fast temp changes, and that's why you want to dwell the oven temperature long enough for the resistor internal stress to relieve itself - you want that resistance line to get flatter.  This is the part where you find out to some extent a resistor stability / apperent TCR also is affected by the -rate- of temp. change of surrounding ambient.

We see similar effects at the crystal substrate level with diffused resistors.  Generally, the faster you over-run the device's capability to stay in thermal equalibrium, the more energy gets stored as stress in the device - and the longer it takes to recover back to a settled state.

If you change the temp at a slower rate and then wait, usually you'll see the resistor equalize faster, and with less spikes at the leading edge.

In other words:  Change temp, and don't change temp again until your resistance measure line gets flat again.  It can take a while.  This is another area where bobbin size, wire size and resistor mounting style / thermal heat flow into (and out of) the PC board comes into play.

[Kleinstein]

The resistors TiN measured show quite a lot of dirft / relaxation effects. This seems to be more important than the TC and in precision applications it would also be a big problem. I am surprised to see such a poor performance in what is supposed to be a precision resistor. There are several possible causes for mechanical stress on the wire and this way a change in resistance. Besides thermal expansion, there is intake of humidity and if the resistors are still relatively new is might even be residual reaction in the epoxy or at least structural relaxation.

My guess is that something is wrong with these resistors: maybe too much wire tension on the bobbin or a faulty potting. There is the general trend for the resistance to go down, even in the cold temperature part. So this does no look like relaxation of something like thermal stress (this would go up and down, depending on the temperature). The tendency down in resistance is more like relaxation of a stressed wire that has not got to an end. Also shrinking of the potting might be a cause. At least is looks like things are getting better - the drift in the first curves looked even worse.

[TiN]

I've currently attached Fluke hermetic thin-film resistors which specified for max 1ppm/K with zero TC at +22C to test the setup operation. Same meters measuring 9K and 90K resistors, using same cables and configurations. We'll get data in 16 hours or so.

[branadic]

Plotting resistance over temperature, ambient temperatur, humidity and dew point could help to identify the different influences of your measurements. Plotting the data in time domain makes it hard to seperate each influence.

[Andreas]

Hello,

full ack but:
The humidity related parts are difficult to catch since the time constants in epoxy are typical between 3-14 days.
Its more a project to differ between summer and winter.
(ok the graph has only to be long enough).

With best regards

Andreas

[branadic]

Plotting resistance over temperature, ambient temperatur, humidity and dew point could help to identify the different influences of your measurements.

Did that for the given data: https://xdevs.com/datashort/vpg21_tcr1_nplc100_tin.csv

Dew point was calculated using the Magnus formula: https://www.wetterochs.de/wetter/feuchte.html

[Andreas]

Interesting: the 12K5 vs pressure. accident or linear dependancy?

with best regards

Andreas

[Andreas]

Dew point was calculated by the Magnus formula: https://www.wetterochs.de/wetter/feuchte.html

I hope this was done with temperature and humidity from the same sensor (BME280)

with best regards

Andreas

[branadic]

I used the ext_temp data not his rtd_temp data, so presumably it's Illyas BME280 data I used. 
Double checked all plots for confidence, no mistakes found.

[branadic]

I have updated to plots... Illyas website is somewhat confusing, but I managed to find the data of his 3 temperature cycles, a file with exactly the same name as before but full data of the cycles.
Excluded are the data of the temperature jump. It's now a somewhat different picture.

[branadic]

And here are the plots for the temperature jumps... interpretation is up to you

[TiN]

Test complete on Fluke thin-film, no surprises detected, resistance value came back to pre-test value within 1ppm, as would be expected.
Here's a chart. Speed on this test is same as first two runs on Edwin PWW's, around 0.05K/minute, or 2.5K/hour.



As result, worst TCR of 9K resistor is +0.4ppm/K. 90K is similar with +0.6ppm/K.
This should give little bit more confidence of measurements, which surely might have errors, but not quite 20ppm+ numbers like we saw last few days.

DSV-file

Reference resistor photographs below. Two most rightside resistors are tested on chart above (90 Kohm, and 9 Kohm).



Note that this is hermetic resistor, resistive element is covered with glass seal.

[branadic]

Of course these data look somewhat different... I would have expected same temperature behavoir for both resistors as they are on the same substrate and manufactored during same process, obviously I was wrong. I used 3. order fit to calculate residuals.