T.C. measurements on precision resistors

[juani_c]

A quick question:
I've seen T.C.R expressed as "±5x10-6 /K", is that the same as "ppm/ºC"?

[bdivi]

Yes, ppm is abbreviated Parts Per Million which is exactly 1.10e-6

[Edwin G. Pettis]

Either regular solder or silver solder may be used for hermetic sealing, depends on the application, I sealed many a mil grade hermetic resistor at Ultronix with regular solder, in some cases, particularly the large, oil filled units, sealing could be a bit tricky.  Not all hermetic parts contain oil but they are not nearly as common as the oil filled units.  I have not heard much about the use of no-lead solder for hermetics, most of the time the hermetic parts are exempted anyway.

In real life, it is not that difficult to isolate the LTZ and circuitry from air currents, it is being done many times over in critical applications without any sign of excess drift.  Encapsulating the unit is not difficult, foam works excellently and is not difficult to obtain or use, most anything with very low thermal conductivity.  If the Kovar leads were a real problem, the folks at Fluke, HP, NIST, et al, would have been screaming at Linear Tech about it.  This is not the first time Kovar has had to be used in ICs.  Yes, it does require a bit more effort than if Kovar hadn't been used but obviously Linear Tech deemed Kovar leads necessary.

[texaspyro]

A seller on Ebay is selling currently sheets of aerogel insulation for affordable prices.  Search Ebay for "SPACELOFT".

[Andreas]

Anyways: the results for the Z201 Resistors are ALARMING and i think at this point it we should contact Vishay about this.

I am now thinking about ordering those Audiofoolery naked Zfoils for my Project, since the whole circuit will be hermetically sealed and under transformer-oil anyways.
(Much cheaper then the hermetic zfoils)

Hello,

I do not find this alarming. This is only a feature of the resistor which is not specced within the datasheet like humidity sensitivity. (datasheet = advertisement).  And you still have the possibility to use hermetically sealed resistors.

Usually you should read between the lines: which parameters that are of concern for your application are _not_ covered within the "datasheet".

With best regards

Andreas

[janaf]

Andres, not sure I have seen: the S102 are available in C, K and J where the K alloy should be slightly more temperature stable than C, while J is far behind. You measured the J and it was walking around a lot. But then there are measurements of S102x, x not specified. I think the early S102 later  became S102C?   

So, have you measured C and K alloys, can you see any difference?

[Andreas]

Andres, not sure I have seen: the S102 are available in C, K and J where the K alloy should be slightly more temperature stable than C, while J is far behind. You measured the J and it was walking around a lot. But then there are measurements of S102x, x not specified. I think the early S102 later  became S102C?   

So, have you measured C and K alloys, can you see any difference?

Hello,

From S102 the C and J are the same alloy. Only the lead pitch differs (3.81 mm / 5.08 mm).

I do not know where you looked at but some values differ for the low ohmic and high ohmic types.

Up to now I did not get a K-type alloy (which is not so commonly stocked at the distributors).

With best regards

Andreas

[janaf]

I can send a couple of S102K if you like.
560R, got plenty.
2K, 4K, 10K I could spare 2 each. All NOS.
Sorry, no 1K.
Which values are best for you?
PM your address and I'll put them in the mail.

[barnacle2k]

I do not find this alarming. This is only a feature of the resistor which is not specced within the datasheet like humidity sensitivity. (datasheet = advertisement).  And you still have the possibility to use hermetically sealed resistors.

Usually you should read between the lines: which parameters that are of concern for your application are _not_ covered within the "datasheet".

[rant]

Am i really the only one upset that Vishay creates one of the lowest tempco resistor on the planet and then packages it into a package that completely negates the advances in the resistor technology?

If i pay more then 30 bucks for a resistor i expect superb quality and not horrible hysteresis behavior that makes that resistor completely useless for precision applications.
These results just ruled out anything but their most expensive resistors for my project.
The hermetic's are at least double the price.
(Maybe i'm just mad about blowing the budget again)

[/rant]

[Andreas]

Hello,

to come back to Franks measurements of my Z201#2 resistor with a direct comparison of his and my results.

- First attachment is the measurement of Frank.
  Z201_1k.jpg

- to make the comparison easier I have transformed it to my more or less standardized format.
  20150221_TC_Z201_2_LMS_Frank.PNG

- since he uses around 12000 seconds for his 32 degree temperature profile (around 0.36K/minute)
  this corresponds more or less to my "fast" measurement of 0.3K/minute.
  20140729_TC_Z201_2_LMS_temp_corr_fast.PNG

- since I have seen that at least with my setup (with the wires outside of the metal case)
   the results are not so good reproducable than on Franks setup I have reduced the ramp speed to 0.12K/minute.
   Another difference is that on my ramps there is some rest time at the temperature end points so that the
   creeping effect of the mold has more time at the end points.
   And of course the measurements slightly differ in end temperatures
   since with my setup it is easier to reach higher temperatures in summer
   and lower temperatures in winter which has also a slight effect on the hysteresis.
   20140801_TC_Z201_2_LMS_temp_corr.PNG
   20150123_TC_Z201_2_LMS_temp_corr.PNG

All in all there is a good coincidence of the measurements if regarding the boundary conditions.
Of course Franks measurements have less noise with his better setup.
But the resulting LMS curves look very similar under same conditions.
My target is to select from some resistors that I have the best fitting resistors
for my next LTZ1000A reference.
So for the voltage divider (1K/12.5K) this is a similar behaviour with T.C.
And for the other resistors(120R, 70K) I will select for minimum T.C.

With best regards

Andreas

[Andreas]

Hello,

and similar with the S102#1 resistor a comparison with his diagram to my measurements.
I did not standardize and calculate the LMS 3rd order curve since it is simply meaningless.

- Franks measurement with around 0.36 K/minute
  S102JT.jpt

- one of my measurements of summer with 0.3K/minute 3 times
  starting from 25 degrees to negative values and a final setpoint jump from 21 deg to 30 deg.
  20140714_TC_S102JT_1K_1_temp_corr.PNG

- a slow measurement from this year which ressembles more the measurement of Frank.
  20150119_TC_S102JT_1K_1_temp_corr.PNG

so the form of the hysteresis seems to be dependant on the "history"

With best regards

Andreas

[Dr. Frank]

Hello,

here are my measurements on a batch of 6 econistors, 1k, 0,1%.

All have the same date code 1435, and seem to come from the same manufacturing lot.
So I expect similar T.C.s over the whole batch.. and this shows...

As there is practically no hysteresis, and the T.C. behaviour is quite linear over T, I only applied a linear fit for calculation of T.C., in this case.

The T.C. is higher than the typical value, but within the 5ppm/K max. limit, except one sample.

All econistors, also the 12k, 120 Ohm ones, are all very well inside the R25 tolerance, i.e. they are better than 0.01%!

[Dr. Frank]

Here are the measurements for 5EA 12k , 0.1%resistors.

They have much lower T.C., but show pronounced  hysteresis and non linear T.C. over T.

I applied linear fit and box method.
Both values are well below the typical T.C. value of 3ppm/K.

Depending on the shape of the hysteresis and the dependence of T.C. vs. T, the box method may give either lower or higher values than the linear fit, but it will always be on the same order as the linear fit.

This observation is important for the estimation of the T.C. behaviour of the next 120 Ohm resistors.

I would like to use pairs of 1k/12K resistors, matched to a T.C. difference of about 3.0 .. 3.4 ppm/K, for the LTZ1000 reference.
That would give a contribution to the LTZ temperature coefficient of  about -0.04ppm/K (calculatory 1/78 of T.C. of resistor divider)

[Dr. Frank]

And here are the measurement on two (of five) 120 Ohm, 0.1% resistors.
The 4W measurement was done 10mm apart from the resistor tube, therefore the copper T.C. of the leads would not greatly add to the resistor.

The resistance curve is relatively linear over T, and hysteresis is also not very pronounced.
Therefore, these parts are for sure totally outside specification, even if one would measure in the spec limits of -55...85°C, or so.

The supplier is currently investigating on the root cause, and will hopefully supply me resistors with sufficiently low T.C.

This T.C. would give an additional -0.08 .. -0.14ppm/K, (1/100),  maybe much less of that in practise, according to the results of JANAF.
https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/msg615470/#msg615470

That calculatory result is much too high, and also in the wrong direction, to cancel the 1k/12k divider T.C.


PS: the econistors are made in U.S. by General Resistance, which belong to Prime Technology. Here's the G.R. datasheet:
http://www.primetechnology.com/cp/images/PartsDocument/163323.pdf

According to that datasheet, leads of an econistor have 53µOhm/mm, that makes about 5mOhm in total, if I would measure on the ends of the leads (my samples have 50mm long leads) .
Copper has a T.C. of about 4000ppm/K, and that would contribute in worst case 0.17ppm/K for a 120 Ohm resistor.
Therefore, this too high a T.C. can not be caused by the copper leads.

[Edwin G. Pettis]

First, a definition of hysteresis as applied to resistors: A failure of the resistor to return to the initial value of resistance after an applied stimulus.  As I have noted elsewhere, all resistors have hysteresis, only the degree to which they have it is different.  Technically, the initial value of resistance is that obtained after the manufacturing processes, in many cases the ‘final’ value is calibrated after the resistor has been made in the case of precisions.  This is the process of adjusting the ‘final’ value to within the specified tolerance band.

Depending on the resistor specifications, further ‘enhancement’ may be applied to increase its long term stability, the particulars of how this is done tends to vary with the type of resistor but in general thermal conditioning is used.  The success of this depends on how the resistor is designed and built, a loose rule of thumb is the lower the long term drift specification (particularly under power), the better the stability of the resistor.  Stability is a result of the resistor’s design and construction, stability is an inherent characteristic of the design, no amount of external stimulus can compensate for a poor design.

Hysteresis is the accumulation of all previous permanent changes, from the initial to the current.  Drift is a completely separate phenomenon which is generally defined as a change in resistance without stimulus, the ‘sitting on a shelf’ specification.  The long time ‘drift’ due to power stability specification is a misnomer; this is actually hysteresis in action as it is due to an applied stimulation.

When measuring hysteresis, it is important that the initial conditions be accurately specified, ambient temperature and initial value, not to mention the accuracy and repeatability of the measurement system.  If the observer wants to measure the effect of a change in temperature between two points, how fast or slow the external temperature changes is irrelevant, only that the resistor is given enough time to stabilize at the second temperature internally is important.  The same thing applies when returning the resistor to the initial conditions.  Additionally, the greater the stimulus change, the greater the hysteresis, this is not to say that the hysteresis will be a large value but hysteresis tends to be larger with larger stimulus change.

The application of thermal shock cycles will produce a more pronounced hysteresis than just a single change in temperature, even if that temperature change is 125°C, thermal shock can separate the ‘boys from the men’ so to speak, small hysteresis changes are indicative of highly stable resistor designs.

By industry test standards, the relatively gentle temperature shifts of 30°C being used in these measurements will not show much hysteresis except in the probable case of a bad resistor.  If you are going to use these resistors in an application in which the operating conditions are relatively stable, with only a few degrees change, then hysteresis is not going to be of much consequence, the long term drift of the resistor is going to be of much greater importance here.

Primary standard resistors are kept at a constant temperature for a purpose, stability, those Thomas one ohm resistors made of Manganin would otherwise show a significant hysteresis if this was not the case, applied power is kept to a minimum as well, under these conditions any really good resistor will show very little change over time.  In other words, gentlemen, you are basically fussing over a characteristic which in your application will be of little significance compared to other terms.

Long term drift of some of the resistors you have been talking about is much more significant, up to possibly 17 times higher (or more) than any of the hysteresis values you have measured here.  Once the LTZ1000/A or LM399/A have reached their lowest drift coefficients in time, the long term stability of the resistors then become more important as they can and will introduce an additional drift component, albeit not too large but significant, compared to the Vref.  The other 900lb. gorilla in the room is noise, this is the ultimate limitation, and noisy resistors will definitely add to the total noise over and above the Vref’s.

Dr. Frank, as I recall, the econistors specify 0±3PPM/°C TCR 0°C to +85°C, your batch is out of specification, the 0±5PPM/°C TCR applies above +85°C or below 0°C.  Given that they claim a lengthy thermal conditioning cycle (7 days), something or someone got sloppy on production.

[Dr. Frank]

Hello Mr. Pettis,

thanks for your elaborate analysis, which I mostly agree, apart from two aspects:

If the observer wants to measure the effect of a change in temperature between two points, how fast or slow the external temperature changes is irrelevant, only that the resistor is given enough time to stabilize at the second temperature internally is important.  The same thing applies when returning the resistor to the initial conditions.  Additionally, the greater the stimulus change, the greater the hysteresis, this is not to say that the hysteresis will be a large value but hysteresis tends to be larger with larger stimulus change.

That's a kind of contradiction in itself.. as that's the crucial point , if the  measurement is done in quasi equilibrium, or not..

The hysteresis measurement very well depends on the speed of the temperature change!
Andreas measurements with different temperature change rates illustrate that.

The physical effect, which causes hysteresis, mostly can be described like something like a friction force.
If this force as is "weak", you will get fast recovery or fast creep effects.
If this force is strong, you will get a very stiff hysteresis, and maybe no practical observable creeping (glass state).

Equivalently this may also be described with a relaxation time constant for these hysteresis effects, (strong force <=> long relaxation time constant and vice versa)
I observed something like 1ppm/30min in one of the measurements on Andreas resistors.

Therefore, if the temperature change is much slower compared  to this time constant, the sample can relax to the equilibrium state at the different temperature points, and you may get a very narrow hysteresis loop, compared to the case, when the change is much faster than the time constant.

For the measurement of T.C. curves, that may also give totally different results, if you determine T.C. linearily or by box method.

Dr. Frank, as I recall, the econistors specify 0±3PPM/°C TCR 0°C to +85°C, your batch is out of specification, the 0±5PPM/°C TCR applies above +85°C or below 0°C.  Given that they claim a lengthy thermal conditioning cycle (7 days), something or someone got sloppy on production.

Well, the econistors are not very well specified (similar to Vishay)

That 3ppm/K is a TYPICAL value only, therefore most of the resistors are definitely within spec.

The MAXIMUM 5ppm/K over -55.. + 125°C strongly indicate that the 14 and 7.7ppm/K are outside, but here again, I have the same specification struggle as with Vishay, that is the averaging calculation by a box or a butterfly definition..

For metrological purposes (18..35°C), I need instead the physical definition, i.e. T.C. = dR/dT, or exactly the R(T) curves we are measuring here.

The box and butterfly definition mostly determine the T.C. by measuring at 3 or 5 temperatures only, and then average the resistance change by the temperature range.

This may give much lower T.C. values, if the R(T) curve is not linear over temperature, but has minima and maxima.

Frank

[Andreas]

Hello Frank,

thanks for sharing.
That confirms me that for a good result I will have to
test and pair/select the resistor T.C. s for my references.
And probably the 120R resistor is the most critical.

I always asked me why the LTZ#2 where I paid more
attention on the stability of the cirquit seems to have
a slightly worse 1000hrs standard deviation than #1.

Probably I should measure the resistors some days.
(if I have better references).

With best regards

Andreas

[Andreas]

Hello,

just to see if further reduction of temperature ramp speed could have a significant influence on hysteresis I did a measurement starting with a warm cycle on 24.02. from 20 deg C with 0.06K/minute and a cold cycle on 25.02. with 0.04K/minute for Z201#3.

Other setup is the same as on measurement from 22.02. with 0.12K/minute which is again attached as comparison.

Against the measurement of 22.02. the hysteresis does not change as a factor 2 like the ramp speed for the warm cycle.
For the cold cycle the reduction of hysteresis is larger since there is more time to creep back to the 20 deg point over night and the lower ramp speed.

With best regards

Andreas

[Edwin G. Pettis]

Dr. Frank,

Let me restate what I said in a slightly different manner, same conditions.  Using the same 30°C temperature change (roughly) that is generally being used here, are you trying to say that if the resistor is subjected to exactly the same temperature excursions, up to and back down again, that supposedly the rate of change is going to make a difference?  Sorry, I disagree, after seeing thousands of examples of resistor hysteresis over the years, the rate of change is irrelevant (read on below before jumping to conclusions) .  A change of 30°C will produce exactly the same amount of hysteresis no matter how slow or fast the change was obtained IF the resistor obtained equilibrium internally at both temperature extremes and other conditions are met.  The key here is the internal equilibrium of the resistor with the change in temperature, given that the resistors being discussed here has a rather low thermal conductivity, it will take the hysteresis loop of the wire against the mandrel a significant  amount of time to equalise to the ambient temperature.  If insufficient time is not given for the internal stabilization of the resistor then the hysteresis effect will not be consistent from one run to another and that is possibly what may be seen here (again, see below).  How much time, don't know, unknown variables involved.

A further note about the alloy(s) used in PWW resistors, they have an extremely high value of tensile strength, this alloy will not and cannot be stressed in any way that changes it, 150°C is nothing to it, what causes real hysteresis (in shelled resistors) is the hoop stress on the layer of wire against the mandrel, if this stress is not removed during the manufacturing processes and this is important, the bobbin materials must be chosen correctly, the hoop stress is not entirely removed from the resistor during manufacture and further hysteresis can result from temperature changes.  This results from the bobbin material never completely 'relaxing' against the wire turns and this remaining stress can be there for the life of the resistor.  There have been many claims by manufacturers (I know, I used to work for one of them) that they chose their wire and bobbin materials to match each other, this is a steaming pile of hog wash (the British call it rubbish), the materials do not match each other correctly, the problem being that they match in the wrong way leading to different stress and the manufacturers don't seem to understand this because they keep making the same mistakes over and over again.  Molded resistors, which are very popular because they are cheaper than shelled resistors have their own set of problems with internal stress, this partially being caused by the molding process itself which imparts an unpredictable stress onto the entire resistor's windings and this stress changes with time, it can also appear as a significant component of 'drift' in both shelf and powered long term stability.  There is also stress imposed by the bobbin material generally used in this type of resistor as well, more hoop stress and the list goes on.

Another factor that can enter into this apparent hysteresis effect is the resistor's design and construction methods, many precision wire wound resistors will indeed exhibit what appears to be only hysteresis but in fact, other mechanical forces are working internally to cause an additional instability factor which appears as hysteresis during a temperature cycle.  This same effect can also appear as part of a drift term in parts sitting on a shelf long term.  Unless the precise method of design and construction of the resistor is known, it is not possible to identify exactly the cause of this effect but it is definitely inherent in many resistors.  I have seen and identified sources of this effect in many different PWW over the years.  This effect could easily be what is showing up in your tests as variations in hysteresis depending on the ramping speed of temperature as they react differently to the stimulus.

When you are looking at very small term effects in a resistor, it becomes a highly complex combination of electrical and mechanical terms which manifest themselves at the terminals of the resistor, it is like the 'black box' puzzle so many professors like to throw at their students.  There is a component or circuit inside the box and without opening it, you try to figure out what is inside of it.  Unfortunately, that really doesn't work with PWW resistors because so much of it does depend on what is inside of it and how it was made, trying to separate all of the various terms is near impossible from the outside.

At best, a properly designed and constructed PWW will exhibit little hysteresis and will not have any of the effects described above in the second paragraph.  That is not easy to achieve or else all PWW resistors would be pretty much the same in characteristics.  Remember, these parts were intended to be working, not sitting on a shelf or being kid glove treated like a resistance standard, there will be some effect, however small, from the fact that operating conditions are not constant with time.  They may indeed be too small to measure over just one or even several change cycles but tiny changes there will be even in the best, only difference is that those changes are much more likely to be consistent instead of varying response with the same stimulus.

I am not disputing the fact that you gentlemen are measuring differences in hysteresis as such, what I am trying to do is explain why you may be seeing these abberations and are possibly attributing them to the wrong effect.  Feel a bit confused, don't, there are many resistor engineers out there that don't understand it either and/or refuse to understand it.

It would take a book to describe and detail all of the things that go into PWW design and manufacture, one reason you won't likely see one is that to explain some of this stuff, proprietary information would have to be explained and as you know, that isn't likely to happen.  It is most amusing how every PWW manufacturer thinks that their resistor design is proprietary when nearly all of them are a conglomeration of ideas 'borrowed' from competitors over the decades.  General Resistor's design is nothing new nor does it solve many of the problems exhibited by this construction and manufacturing.  I've seen and dissected a lot of resistors over the years and they are all very similar to everybody else's resistors out there.  There are variations of course, but basically they are still all essentially the same result.

[babysitter]

Hi, it seems that you guys are using different definitions of temperature hysteresis leading to misunderstanding, let me summarize:

Edwins definition seems to be considering just the net effect of putting a heat cycle - beginning and ending at the same temperature only. As seen from the resistor, this will earn you a difference in the resistance before the temperature cycle to the resistance after the temp cycle.

The hysteresis error Andreas and Frank are looking for is the maximum difference in resistance at any temperature when approaching the temperature setpoint first with increasing and then with decreasing temperature.
This explanation, supported by the measurements (or vice versa) will give you the idea why the temperature ramping must be slow enough to give the resistor time to get to a equilibrium - caused by right what Edwin describes, the temperature "cycle" must be done before every single measurement. And also giving the data to specify what happens to that certain resistor if you give it the "Edwin style" treatment.

Let me call that "A sensor guys" definition of temperature hysteresis, as I work in sensing and is exactly what I look for at work. The resistor here is treated like a temperature sensor.

[Andreas]

Hello together,

which effect it really is does not really matter for me.

My world is much simpler:

- all what appears as closed curve (so it is not ageing) during
  temperature test which has a "eye" (opening) with different temperature direction I call "hysteresis"
  -> this is something that I cannot correct with a temperature sensor without knowing the history.

- all what leads to a non closed curve during temperature test -> it is ageing or humidity drift

- all temperature dependant what can repeated -> T.C.

Vishay states in the Z201 datasheet that within 1 second the final value of temperature drift is met within 10 ppm.
So 20 seconds should fit to meet final temperature resistance value within sub-ppm.
The hysteresis or creeping effects that last for some hours (or over night) are something that I do not want to have in my references since I cannot calculate it out with a simple temperature sensor.

with best regards

Andreas

[Andreas]

Hello,

after several months now I have found one "typical" (= golden) Z201#6 resistor. (Date code B1305-)
At least when only regarding the 10-45 deg C temperature range.

The LMS approximation (green) in the diagram indicates 1.76 ppm maximum difference between min/max over a 35.5 deg C temperature span. Giving 0.05 ppm/K (average) with the box method.
Hysteresis is around +/- 0.3ppm on the warm cycle when not regarding the noise of the measurement.

With best regards

Andreas

[janaf]

Nice!

Andreas, I have to ask about the terms in you plots. Is this correct:
- With drift you mean the resistance variation over the temperature cycle?
- The hysteresis plot, you use your LMS line as mid-level and plot variations from that mid-level but in a larger scale?
Did I get that right?

[Andreas]

Hello Jan,

Yes. Drift = temperature drift from a virtual "zero" near 25 deg on the left scale.
(since I have no absolute measurement).

Hysteresis (on the right scale) is the difference between LMS value and measured value.
With these low differences of Z201#6 the diagram gets somewhat confusing.

with best regards

Andreas

[babysitter]

... and after all this work, Andreas is in the business of selling golden, selected Resistors. AMG - Andreas' Manufactur für Gute. Vishay is going to buy him, new name will be VISHAY-AMG.