Would you expect poor reliability from repeated 0 to +65C thermal cycling?

[peter-h]

There is a particular avionics product which has a near 100% failure rate within say 5 years.

No obvious reason other than it has a heated module (housing three accelerometers, three ADI ADXRS613 yaw gyros, and a multichannel delta sigma ADC) which is heated to +65C when the thing powers up - to stabilise the gyro drift. An old technique for these components...

The ADC - a £20 chip - AS1258IRT, fails as often as the gyros, and there is no reason for it to be underneath the heated block.

The accelerometers, ADXL321J,  don't seem to fail.

How often does it get turned on? Maybe once a week on average.

I won't name the product because I don't want to piss off the company, whose help I still need here and there

[UnijunctionTransistor]

Have you sent the failed components for a failure analysis?
That would be step one, to validate your assumptions.

It has happened to me, more than once, that I assume a root cause for a particular problem. But when the FA comes back, one can see that the actual failure mechanism wasn’t related, or at least not closely related, to the original assumption.

[peter-h]

You mean send the gyros to ADI?

I would not even know where to. Also that particular P/N is discontinued.

[incf]

Would you expect poor reliability from repeated 0 to +65C thermal cycling?
...
How often does it get turned on? Maybe once a week on average.

I did design work on much less sophisticated bits of avionics for a number of years.

No, I would not expect that. If that was true it would invalidate a lot of assumptions underlying the sorts of testing and approval processes that I was exposed to. (We tested against similar types of thermal cycling, and it took forever to do)

If I were to speculate given the limited information you have provided: what may "really matter" is the rate of change. If large thermal gradients develop across the die or leadframe, maybe that could cause issues.

Probably wouldn't be too hard to get some XRAYs done to look at the internal structure of the failed ICs to see if it could be a mechanical failure.

edit: to clarify, mostly talking about mechanical failure of ICs like the ADC.

[UnijunctionTransistor]

Isn’t ADI the acronym for Analog Devices, Inc?

They do have failure analysis capabilities which they will do for important customers. Certainly an avionics companies qualify as an important customer, as critical performance applications are ADI’s core business segments.

Even if the device in question is obsolete.

[Niklas]

Based on the ambient temperature alone I would not expect that amount of failures. However, the packages, and their solder joints, of both the gyro and the ADC might be susceptible to mechanical stress.

Mechanical assembly
- uneven seating plane for the PCBA.
- fixation screws close to the components

Thermal cycling
- how fast is the heating element ramping up the temperature?
- Does it only heat a small section of the board?
- possible mismatch in heating between component and PCB? Mismatch in thermal expansion can stress the solder joints.
- the solder joints might look good, but how about the solder pads and their connections to traces? Lead free solder is stiffer than leaded solder, so there are cases when cracks are forming in the laminate underneath the pad.

Via hole breakage
- long shot but this could lead to open circuit when the board heats up. Might be possible to diagnose by putting the board in the freezer to shrink and reconnect the via.

[incf]

One last thing comes to mind, vibration. Combining localized thermal stress (as Niklas suggests, due to screws prohibiting thermal expansion of the PCB) with cyclic vibration just might be able to do something like that over the course of half a decade. PCB failures do seem far more likely than IC failures.

[peter-h]

The big aluminium heated block goes to +65C in about 30 seconds, which seems fast.



I've just opened up another failed one of these boxes - a later model - and note that they no longer use that arrangement. They have two of these chips on a PCB, and that area of the PCB is heated with a MOSFET on the back of it, again at a similar speed.



The two chips are marked

MXG3A
13281
763AA

and

8519
DSH
091LD

which don't easily google. Well, the 2nd one is probably this: https://www.ictransistors.com/ic/lis3dshtr-marking-dsh-lga16-accelerometer.html i.e. this: https://www.st.com/resource/en/datasheet/lis3dsh.pdf

The 1st one is this: https://www.slideshare.net/slideshow/maxim-integrated-max21000-3axis-mems-gyroscope/37422796 i.e. MAX21000.

The way the PCB is getting heated from the underside seems primitive from the thermal expansion POV, especially with that fixing screw right next to the heated area.

I see two chips away from that area, marked 41XR A60 which are probably accelerometers and which never needed temperature stabilisation.

Curiously, I no longer see any chip which looks like an A-D converter. But the two chips are SPI/I2C anyway. I know from serial data debug that the accelerometer chip is OK (the 3 axis G readout is fine) so the problem is likely the MAX21000 chip which lists as obsolete This is hilarious
https://ez.analog.com/other-products/w/documents/20421/is-there-a-replacement-for-the-max21000

I found some on Aliexpress, not cheap...

[moffy]

There is probably a history behind the MAXIM MEMS being withdrawn, and reliability would be a top suspicion.

[David Hess]

Regular temperature swings will exasperate problems with mismatches of temperature coefficient of expansion between the printed circuit board and parts.  Presumably they used a special printed circuit board material to avoid this.

Aircraft are also subject to vibration which will make the above even worse.

Or maybe the MEMS were withdrawn because of poor reliability, which is made worse by regular temperature swings and vibration.

The design should have been run through a shake and bake test to destruction before being put into production.

[Siwastaja]

How is that PCB mounted to the aluminum block? If the coupling is hard, I can imagine there will be huge forces because the coefficient of thermal expansion is significantly different for aluminum and FR-4. And 65degC in just 30 seconds is pretty fast!

[peter-h]

The PCB is attached to the ally block with several screws.

I wonder if the MAX21000 even needs any temp stabilisation at all.

A google on MAXIM MEMS reliability digs out enough to suggest they had loads of problems with these chips. But this
https://www.analog.com/media/en/technical-documentation/reliability-data/max21000.pdf
denies any issues.

[peter-h]

Looking at this again after a bit of digging around:

Cockup #1: heating up to +65C very fast (there are operational reasons for wanting a fast warm-up, because the aircraft heading system is useless before it gets a "valid" signal, but...)

Cockup #2: in the 2nd design, it seems an extreme bodge to be heating just that little bit of the PCB, not realising thermal expansion issues

Cockup #3: in the 2nd design, the spec sheet of the MAX21000 suggests that this level of temperature stabilisation is not actually needed. The device claims to have a very low drift. But anyway, the main purpose of this box is to stabilise the heading data from a fluxgate magnetometer (which jumps around by 1-2 degrees P-P otherwise) and the fluxgate output is the primary variable.

Regards the choice of +65C, this was probably picked to be above the highest ambient the devices will see. The box itself warms up by about 20C due to various electronics inside, and with max likely ambient being +40C or so...

Astonishing, given that this is a TSOd certified product, sold to the military as well.

[peter-h]

It's been suggested that the least bad way to do this bodge is by slotting the PCB, to prevent the inner rectangle buckling (when heated to +65C) and breaking either the solder joints or the actual chips