Is the temperature coefficient of a component constant over time?

[branadic]

Hi all,

it seems like noone ever investigated if the temperature coefficients of components are constant over time. While it is a material property and thus a constant, on component level where multiplle different materials are being used together and interact with ambinent air I would expect some changes over time due to reaction with ambient air and creating oxids or nitrides.
In some applications, such as SR104, everything is done to prevent any reaction with ambient air. On board level inert surface finishes are used (Ni, Au), while tinned copper tend to create oxids too and create all sources of t.c. effects such as Schottkey diodes and parallel resistors.

Are there any hints, experience or papers available that indicate a t.c. change over time, although the changes might be minor? A web search did lead me to nowhere, but we are looking at ppm's and ppb's, so worth asking anyways.

-branadic-

[tszaboo]

No, it is not, in general.
I remember seeing some LTZ1000 experiments, where they did some oven training of the boards, in a cold-warm-cold-warm fashion, with decreasing the amplitude of the temperature swings, and that resulted better tempco to some components. Also, humidity makes a large difference in components, so being in a humid environment the tempco can be different than in a dry one. FR4 is hygroscopic.
I guess nobody investigated it because it is not fun to make hundreds of temperature dependent measurement over an extended period of time. And if someone would invest the time, they probably wouldn't publish it, but keep it as a trade secret.

[Kleinstein]

The restance and TC of many of the resistor alloys (e.g. NiCr, Manganin) does change with the microstructure. One can use this to fine trim the TC by tempering. However the same process will also happen at normal temperature, just slower.  Here the copper based alloys are a bit more sensitive than the nickle based ones, as diffusion is usually faster in copper. It is a known issue with Manganin shunts, if they run hot with a high current.

FR4 and similar plastics also show aging over time and this can change how much they react to temperature and humidity.

[alm]

There is this case of Nikonoid receiving an SR104 where they had the TC remeasured, and got different values compared to the chart in the lid from 1989. TiN made a plot that makes the curves easier to compare here. There was the theory that the TC was changed by some event that happened to the SR104, but there is no way to verify that.

I'd say an SR104 is the best case where the designers did everything to minimize changes in the temperature coefficient by trying to eliminate stress on the resistance wire as much as possible, hermetically seal things, etc. So I would expect anything that involves more factors, like a PCB with semiconductors, to perform worse.

[Andreas]

Hello branadic,

some hints are given here for resistors from Edwin:
https://www.eevblog.com/forum/metrology/fluke-742a-10k-and-hp3458a/msg1078654/#msg1078654

So heat treatment can change the T.C. depending on alloy used.

For voltage references like AD586LQ (in CerDip package) which I use with temperature compensation
I see much more ageing drift than drift of the temperature compensation.
But these are usually kept in 18-33 deg C temperature range so they are not mis-treated.

with best regards

Andreas

[free_electron]

capacitors are also notorious to drift. BaTi based capacitor ( Barium Titanate ) like multilayer ceramics ( x7r and co ) drift over a time span of  years !
A freshly soldered ceramic cap decays at a rate of 1% the first hour, 1% the next day, 1% the next month and a 1% the next 2 years (it's a 1:25 ratio... 24 hours a day , 24 days a month , 24 months : 2 years).
Heating the dielectric above 148 degree centrigrade resets the decay process !
So you can't make frequency stable oscillators with these suckers ! Worse. If you got the frequency spot on , don't reheat the cap : it'll reset !

http://www.johansondielectrics.com/technical-notes/product-training/basics-of-ceramic-chip-capacitors.html

https://www.eevblog.com/forum/projects/t22291/25/

[e61_phil]

Even very pure things like Pt seems to change their behaviour. Otherwise it would be good enough to calibrate your Pt100 every year against the triple point of water. But maybe it is just due to the impurities in the Pt.

[TiN]

Here's some results of TCR change over long time change. Artifact in question is F742A 10 Ohm. It's the same one repaired last year. Original factory alpha and beta sticker was still present:



Standard was built around 42 week 1997.

Now I've repeat measurements of TCR with a two sweeps, one from 19 to 27 °C and one from 23 to 27 °C.
Measurements were done with DCC against fresh calibrated 1 Ohm standard in airbath. 742A-10 was tested at 10 mA.



Current result for alpha = +0.086 ppm/°C and beta = -0.030 ppm/°C^2. This is pretty close to factory sticker from 25 years ago.

[TiN]

Tested some more resistors, 1 ohm this time. Resistor manufactured circa 1993.



Ran 118 hour log with temperature sweep:



Result



Factory tempco : -0.060 ppm/C and -0.037ppm/C² , measured -0.109 ppm/C and -0.040ppm/C².

I guess not too bad for 29 year old resistor.

[dietert1]

In order to know whether you have seen a change or not, we would need an error estimate of your measurement. For example you could fit the rising temperature and falling temperature curves independently and then calculate a margin.

What are those steps with the 10R standard?

Regards, Dieter

[TiN]

I've understood the words you wrote, but did not understand what you have in mind? Uncertainty of my resistance measurements is present on plots already.
Uncertainty of temperature measurements is relative only, its within 0.5 °C.

What steps you talking about?

[ramon]

Nice plots! I like that very slow temperature slope. What would be the point of lowest TCR for this standard?

[TiN]

"zero" TCR point is already written in my plots, that is Alpha = 0 temperature. +21.6C for 1 Ohm and +24.5 for previous 10 Ohm.

[mendip_discovery]

That amount of drift, could part of that be a difference in measuring accuracy, things have changed in that time period and I know you have some rather stable gear and are testing each one at a time etc.

I also wondered (I have not technical knowhow on this) if the TC has shifted in a relationship to the resistance. For example Calibration resistance has shifted by 10ppm, and the TC shifts by 2ppm.

[dietert1]

I've understood the words you wrote, but did not understand what you have in mind? Uncertainty of my resistance measurements is present on plots already.
Uncertainty of temperature measurements is relative only, its within 0.5 °C.

What steps you talking about?

If you have one result of -0,109 ppm/K and you compare that to factory calibration of -0,060 ppm/K, those values may be in perfect agreement depending on their uncertainty. Without knowing the uncertainty (at least of your own measurement), there is no conclusion.
The steps i was asking for are clearly visible in the 10R curve. Especially at about 22.8 °C, probably drift at stationary temperature.

Regards, Dieter

[e61_phil]

I've understood the words you wrote, but did not understand what you have in mind? Uncertainty of my resistance measurements is present on plots already.
Uncertainty of temperature measurements is relative only, its within 0.5 °C.

What steps you talking about?

If you have one result of -0,109 ppm/K and you compare that to factory calibration of -0,060 ppm/K, those values may be in perfect agreement depending on their uncertainty. Without knowing the uncertainty (at least of your own measurement), there is no conclusion.
The steps i was asking for are clearly visible in the 10R curve. Especially at about 22.8 °C, probably drift at stationary temperature.

Regards, Dieter

Hi Dieter,

what do you think is uncertain? The absolute uncertainty of the resistance measurement doesn't affect the TC measurement.

Regards,
Philipp

[TiN]

Step is due to time delay of resistance element itself reaching for temperature while resistor body already at fixed temperature.
Air is good thermal insulator and resistor does not have any vents or holes, so only thermal conduction is thru resistor wires and hardware inside.
This is obvious looking at time plot. Also 10 Ohm test also much faster than 1 Ohm. Either way that +22.8C shift was just 0.14ppm at worst case, which is negligible over 8 degrees sweep.

Since measurements done with ratiometric bridge against stabilized reference, there is zero tempco or drift of the measurement instruments involved in these plots.
Total expanded uncertainty U at k=2 already listed on plots, based on accredited calibrated reference standard.

[mendip_discovery]

TiN your uncertainty is well documented but the original numbers not so much and that maybe part of the things drifting s little. There are many measurments that were done back then that don't have a full UoM calculated.

[TiN]

From 742A manual Fluke's factory calibration uncertainty is +/-1ppm, so I'd expect they do that or better for their TCR measurement. They also use slightly different process to calculate alpha/beta, by doing only fixed 3 point calibration at +18C, +23C and +28C.

[David Hess]

FR4 is hygroscopic.

Worse than being just hygroscopic, some FR4 is and some FR4 is not.

[Dr. Frank]

Discussing the T.C. change on resistors like the 742A poses several problems:

First, the temperature measurement of the resistors can't be done very precisely, as it lacks a temperature sensor, which would be coupled directly to the resistor case.
There is hysteresis, you have to change the ambient temperature very slowly, and you can't really tell how FLUKE made the determination of alpha and beta. Maybe they used 3 or 5 fixed temperatures only, where they let the whole assembly soak for several hours per point. 
When I made T.C. measurements on 120 Ohm ECONISTOR resistors, from G.R. = General Resistance, there was a very huge difference between the method of G.R. who made stationary measurements at 3 temperatures only, i.e. they got near zero T.C., and my continuous, relatively fast resistance measurement method, which gave about 10ppm/K, or so.

Second, Fluke always uses matched resistor sets with T.C.s of opposing sign, e.g. +0.25ppm/K and -0.25ppm/K, with linear or quadratic temperature characteristic. The sum of both will then give the residual T.C. characteristic, which is printed on the case.
If the individual resistors slightly change their resistance values, that might change the overall T.C., although the T.C. of each individual might stay constant.
This should be calculated, how big such an effect could be. This might also be an explanation for the SR104, as this standard has a proper thermometer, thermally coupled directly to the reference resistors by its oil bath, and therefore problem #1 does not apply. 

Third, what about the OHM stability of the DMM used to measure these resistors over several days? Maybe I missed the explanation, how TiN compensates for T.C. and other drifts of the 3458A.

Therefore, I would assume that you can derive insight on the T.C. drift of resistors from single resistor elements only.

Frank

[TiN]

Dr. Frank I've provided additional details in my article about 742A's about used setup. These measurements were performed with high-resolution 6010B DCC bridge, not DMM.
So there is zero TC from ambient temperature variation and no long-term drift. That is why it was not explained

I used Honeywell HEL-705 Pt100 RTD physically inserted into 742A ground post as temperature data source, and resistance value calculated from ratio to known calibrated reference resistance (1 Ohm for 1 and 10 Ohm tests, 1kOhm for currently ongoing 10kOhm 742A TCR test). Reference resistors are sitting in another air-bath at fixed +23C, while DUT resistor in smaller air-bath with temperature change sweep.

Here's RAW data of 1 ohm run, in case you want to dig into it yourself. Column val4 is ratio to 0.9999985 Ohm standard, val5 is calculated DUT resistance, column nvm_temp is Honeywell RTD temperature readout.

To be fair 742A were never supposed to be primary resistance standards, but a rather working standards, so they have lesser design requirements then say ESI SR104 or L&N 4210. They also cheaper and easier to use

Inside airbath with DUT:



Setup during measurement:

[dietert1]

If you have one result of -0.109 ppm/K and you compare that to factory calibration of -0.060 ppm/K, those values may be in perfect agreement depending on their uncertainty.
The difference is only -0.049 ppm/K. That's a small value yet it might have been the suspected change over time branadic was asking for.
It the uncertainty of that difference is let's say 0.1 ppm/K then no difference has been observed by TIN. I could not find any error estimate in his report (in units of ppm/K for alpha and in units of ppm/K² for beta).
Also we  don't know the error of the factory calibration. For me there is no conclusion yet.

Regards, Dieter

[free_electron]

FR4 is hygroscopic.

Worse than being just hygroscopic, some FR4 is and some FR4 is not.

FR4 simply means Flame Retardant Class 4. It simple means this material is compliant with UL94V-0
It means diddly-squat for anything else. Electrical properties, Glass Transition temperature , weave and strand density and orientation,... all the stuff you really want to know when designing boards is simply not there !

I'm going to design a board and i will use FR4 means as much is i'm going for a drive. You did not specify the important bits : Where, how , why , how far, speed, what direction , are you coming back ?

[mzzj]

Even very pure things like Pt seems to change their behaviour. Otherwise it would be good enough to calibrate your Pt100 every year against the triple point of water. But maybe it is just due to the impurities in the Pt.

Mechanical strain and contamination with impurities are usually given as the usual reasons for standard platinum resistance thermometer (SPRT)drift.
Large temperature swings(hysteresis) require minimally supported pt wire and that tends to make SPRT's sensitive to tiniest shocks.
High temperatures accelerate all sorts of ions to contaminate the high purity pt-wire.
Platinum also oxidizes between 100C and 400C
 
Reference resistors have really easy life compared to SPRT's