MEMS IMU chip thermal stability

[IDEngineer]

Good ol' MEMS chips. Love 'em and hate 'em at the same time.

We received word from a customer that they are experiencing "drift" in one of our IMU devices. We ran an overnight test and sure enough, it appears that the initial output data shifts after 15-30 minutes and then becomes stable. The effect appears as an constant offset in the output measurement. This is most likely a warm-up issue as the assembly starts "cold" and then thermally stabilizes after being powered up for a while.

An online search, and a review of the spec sheets for various MEMS chips we use, reveals nothing authoritative on the topic. There are some hearsay comments but nothing from the manufacturers. Lacking authoritative data, we're loathe to "just gather a bunch of data" and fake-correct it in firmware since the effect may vary by wafer. Our devices are used in mobile applications where we have no absolute orientation reference.

Presuming the effect is real, we're considering how to address it. Our present idea is to thermally stabilize the IC. Our devices are sometimes used in an elevated temperature environment (up to ~100F) so they experience some environmental thermal cycling in addition to the effects of internal waste heat. Most IMU's have a temperature sensor on their die. We're thinking of using a SMD resistor on the opposite side of the PCB, and PWM'ing it to raise and stabilize the die temperature to a target value. Our devices are potted so there would be a relatively large thermal mass surrounding the IMU and an SMD resistor on the other side of the PCB. If the SMD resistor selectively dumped heat, the combination of the thermal mass and the on-die temperature sensor would allow us to run a closed-loop environment. We could ramp up the temperature soon after powerup (to achieve thermal stability ASAP) and then keep it there (to maintain that stability).

Anyone run into this with MEMS devices? How did you handle it? Has anyone done a temperature control like this? Any observations or suggestions?

Thanks!

[daqq]

Has anyone done a temperature control like this? Any observations or suggestions?

Not for MEMS, but for another device yes. It worked well - the item in question was surrounded on the PCB by a bunch of SMD resistors, though they were on the same layer and the temperature transfer was not done via some potting, but rather the copper on the PCB. The temperature stabilised quickly enough. I also did it in analog with hysteresis, so no PWM, the sensing thermistor was pretty much glued to the device being thermostated.

I also added cutouts so the thermal mass of the rest of the PCB had a minimal thermal connection to the device. Not sure if it is applicable in your situation.

For MEMS though I'm not sure how fast it would work - the MEMSy bits are suspended in vacuum, connected to the substrate by the thinnest of threads. The package and most of the die will heat up pretty much instantly, not sure how long the internal bits will take to equalise.

[Kleinstein]

Even in vacuum the thermal insulation is not that good. There is still radiation coupling and the suspended part is vers thin and thus large surface to volume ratio. So the internal time constant would be quites short (e.g. more like 10s of seconds).

The problem with heating is that it may need quite some power and may thus be a porblem for a mobile use.

Another point is that the effect may not be temperature direct. There can be other effects like stress to the case and surface layers on the sensor itself.
Additional heating may speed up the warm up process, but is tricky with a large environmental temperature range.

I would more consider a good temperature measurement and than numeric correction.

[IDEngineer]

I also added cutouts so the thermal mass of the rest of the PCB had a minimal thermal connection to the device.

Yep, I had planned to add routing around three sides of the "rectangle" containing the MEMS chip on one side and the SMD resistor on the other. The idea being to reduce the thermal "wicking" by the PCB laminate. This is often done for MEMS devices anyway, to reduce mechanical stress communicated to the chip by the PCB, since MEMS can also be very sensitive to mechanical strain.

[IDEngineer]

Even in vacuum the thermal insulation is not that good. There is still radiation coupling and the suspended part is vers thin and thus large surface to volume ratio. So the internal time constant would be quites short (e.g. more like 10s of seconds).

Indeed. We've run some tests and even by applying heat to the exterior of our fully potted module and the measurement offset occurred amazingly quickly. This is what made us realize we probably need an actively managed solution to stabilize things because conditions are just too variable to constantly adapt/correct algorithmically. We could get into a footrace with "compensation lag".

The problem with heating is that it may need quite some power and may thus be a porblem for a mobile use.

This is a mobile application, by which I mean on an automotive platform. We run on 12VDC nominal, and we estimate we will need <100mA additional current to run the heating system. A watt of heat deep in a potted environment can get surprisingly warm.

Another point is that the effect may not be temperature direct. There can be other effects like stress to the case and surface layers on the sensor itself.

See my previous post about routing around the PCB footprint. We don't have stress issues now thanks to thick potting (done partially for that exact reason) but routing around the footprint will further decouple the MEMS part physically from its environment, both mechanically and thermally.

[Kleinstein]

For automobile use active heating is probably OK. The Lm399 reference uses some 200-300 mW to keep it nice hot at some 90 C.  The sensor is a little larger, but no need to get so hot.

For the PCB often the copper traces carry more heat than the actual board.
Heating would also help against some of the thermal stress and some humidity problems, but there can be more aging - at least different processes that get relevant. Some part that are now nasty slow will settle in reasonable time, while others  are so slow we don't care at RT will become the new nast ones with a time constant of some years.

I have just build a small oven for a voltage reference, some 20x20x10 mm. From that I found a transistor a more suitable heater, as it is linear and more efficient, as no heat is lost elsewhere in the regulating element. PWM with a resistor may work simmilar, but causes extra EMI. The key to keeping the power needed low is to keep it small and reduce the normal power consumption low. As a rule of thumb the power needed is something  like at least 2 times the power consumption of what is inside the oven already. One can't get much below as otherwise the self heating would be to much and thus staritng quite high in temperature. With a larger part one may be limited by the insulation. Isolation essentially scales similar to the log of inner to outer radius - so it gets rather inefficient to just add much thicker insulation than the inner diameter.

[IDEngineer]

I agree copper is a far better thermal conductor than laminate. However, the target footprint is a 14-LGA which has 14 outer pads with no center pad, so the metal pathways into the package are extremely limited. OTOH, the assembly is potted so the potting compound does serve as a modestly better thermal conductor than free air and helps to couple a greater surface area between the laminate and the 14-LGA and the SMD resistor. It's definitely not optimal but could be worse.

Yes, I'm planning to use PWM but at a very low frequency. Probably 100Hz since we have an existing convenient 10mS timebase on which to base it. Given the thermal masses involved there's little chance that a 10mS period will be slow enough to allow temperature variations due to the off time of the PWM. The last step would be to slow the rise/fall time of the current through the resistor. I'll worry about that if EMI proves troublesome.

[IDEngineer]

It worked well - the item in question was surrounded on the PCB by a bunch of SMD resistors, though they were on the same layer and the temperature transfer was not done via some potting, but rather the copper on the PCB.

Followup: I ran some experiments yesterday and confirmed that PCB laminate is a reasonable (and thus undesireable) thermal insulator. That lends weight to your approach of keeping the heat source on the same side as the target.

You're right, ideally we'd use a copper pour to thermally couple the components. But a 14-LGA package doesn't have a big center pad, only the 14 tiny little pads around its periphery. So not a lot of thermal transfer. Thus we believe the potting compound will be the primary thermal conductor in this situation. It's not a bad approach, since the potting completely surrounds the components and thus will be coupling heat via at least five sides of the chip and each resistor. And we hope to use a large copper pour on the resistor(s) to improve thermal coupling from the heat sources to the potting.

Thanks for the feedback!

[Siwastaja]

Is it a 4-layer board? FR4 isn't a great thermal conductor, but on 4 layers, the prepreg between the outer and inner layer tends to be very thin (100 to 300 um typical), this short distance helps compensate poor thermal conductivity, so if you can have your copper pour on the inner layer next to the component layer, that does the trick.

[IDEngineer]

Presently two layers and hoping to keep it that way.

The ideas are evolving, combining suggestions from this thread with some internal ones. Based on experiments yesterday and what's being discussed now, I think we can make it work well on two layers.