# Can an NTC sensor drift over time?

I was reading stuff on NTC self heating and I was thinking if the thermal runaway is something that could really affect a measurement. So, not an offset, but actually a drift on the resistance.

So I tried to think what happens in a circuit like this:

The maximum power absorbed by the thermistor is when has the same resistance as Rref due to the maximum power transfer theorem. Plotting this power absorbed versus temperature (i.e. resistance) and facing the graphs, I obtain something like this:

where of course at 25°C both resistance are at the same value. Now, for temperatures higher than 25°C, I can see that there is some negative feedback that keeps the sensor from drifting: as temperature increases, the power dissipated decrease and temperature decrease again, and I expect that it will stay in one balance point lower than the offset without this negative feedback.

On the other hand, for temperatures lower than 25°C, as the sensor increase its temperature, it will increase the dissipated power, increasing further, up to arrive at stationary point of 25°C.

Now, if those assumptions are correct, how this is taken into account in the design? Is this effect dumped out somehow?

This question I think applies also on some "fixed" resistors, but with significant parasitic thermal dependence.

• I removed my answer for 2 reasons. It was only a link and you explicitly commented that you wanted something about the drift, which was nearly nonexistent in the linked document. If you are interested in the drift due the aging and how the self heating accelerates it, you should write it into the question. – user287001 Nov 16 '17 at 10:45
• @user287001 The question is related to how (or how not) positive feedback due to self-heat could affect the sensor. If it affect it, it will drift to a certain point, if don't it will not drift. Either case, would be nice to know why does not happens or if it is happening, how to mitigate it. (but usually does no happens) So it is not related to ageing. +1 for the effort to try to understand better the question. – thexeno Nov 16 '17 at 10:55
• Ok. You want something about the dynamics of the generation of the offset, which is the resulted steady state. And especially the possible feedback effect. Right? – user287001 Nov 16 '17 at 11:01
• Correct, more specifically, whether there is a feedback and why. How to calculate the offset and what is it is clear to me. What is not, is how the NTC reaches the steady state if, in theory, there is a feedback (positive below 25°C) – thexeno Nov 16 '17 at 12:30

The self heating can slightly affect the measurement but in measuring circuits the change is, by design, small so the effect of the additional or reduced dissipation due to the change is not important and "runaway" is not a concern. For example if a 1 degree C error occurs, the resistance will change by perhaps 5% and the dissipation will change by less than 1% near the maximum, so still essentially 1 degree C rise,

Self heating is greatly affected by the thermal resistance between the sensor and the medium being measured. A sensor in intimate contact with water flowing at 1m/s will experience much less self heating than the same sensor in still air, all other things being equal.

• So, basically this effect dumps out the drift. If so, then makes sense to have little to no info about that. – thexeno Nov 16 '17 at 8:01
• But I don't see well why this little change does not accumulate over time, if it is supposed to be a positive feedback. – thexeno Nov 16 '17 at 11:00
• To get runaway the incremental increase in self heating would have to approach 100%, so 100x worse than my example. Think of an op-amp inverting amplifier with gain -1 -- you can add some positive feedback and the gain will change a bit but it is not until you get to net positive feedback that it will rail. The analogy to negative feedback is the heat loss to the environment, which increases with increasing temperature delta, and can be expressed in a (perhaps locally linearized) number of degrees C/W. To get runaway dissipation has to increase faster, which is far from normal sensor mode. – Spehro Pefhany Nov 16 '17 at 13:25

When using an NTC you should make sure self-heating is a non-issue by:

• making the measurement current small so power dissipation can be neglected. It might help to choose an NTC with a high resistance value.

• measuring at intervals and only providing power to the NTC when measuring so that it can cool off when not measuring

You could use an alternative component:

A PTC does not have this thermal runaway issue as it counteracts it's own self-heating

You could use a more complex sensor based on a chip, example: DS18B20

• Sure, the self-heating can be easily mitigated, but I was not so confident on the run-away effect and why it seems to not happens that often. – thexeno Nov 14 '17 at 14:06
• why it seems to not happens that often Because if a design suffers from run-away, I consider that a bad design, so much that it becomes unusable. And unusable designs have a tendency to disappear over time. – Bimpelrekkie Nov 14 '17 at 14:09
• +1 for the small current. A 10k NTC is pretty typical and most of our designs use a 3.3V reference voltage, so max power dissipated is <1mW and can be ignored. – Redja Nov 14 '17 at 14:52
• @Bimpelrekkie Nice for the design that tends to disappear. In any case the point was exactly to see if, with a proper design, there is a contribution of drift and if not, essentially why. I know the basic techniques to design with NTC, and I also noticed that drift seems not to appear, and I wonder why – thexeno Nov 14 '17 at 15:34
• When I talk to drift, I refer only to self heating. Because the maths shows a positive effect before the maximum power point (Rth/Rs > 1), so I though that even with very little heat in a proper design, it will degenerate to stall in Rth/Rs = 1. But this seems not to happens from tests and is a bit difficult to me to know why. – thexeno Nov 16 '17 at 10:59