A microcontroller from TI (TMS570LS0432) has a minimum temperature of -40°C according with its datasheet. I would like to test whether it can work also below (e.g. -55°C) by climatic chamber. What would it be a good test scenario to setup to evaluate at that temperature the MCU still survives ?
What would it be a good test scenario to setup to evaluate at that temperature the MCU still survives ?
Since you have stated in comment that the application is "safety-critical", the answer is simple - there is no good setup, as long as the results will be used to select the parts. You are using the parts outside their rating, and that is a very big no-no.
If you're determined to do it anyways, the answer is to run the part at a temperature significantly below -55 for prolonged periods, preferably for the projected lifetime of the final product. In addition, temperature cycling over the extremes expected for the final product use are also critical. It's one thing to put the box in the deep freeze for a long time, but another to repeatedly freeze and warm it - the latter is much more stressful. And "significantly" varies with test sample size. If you test a large number of parts you can get away with less temperature margin. If you're going to test just a few, you need to include margin for unit variation in the final parts.
Note that this means running the part at temperature, not just freezing it and then testing. You must check performance as you go. This is a performance test, not a storage conditions test.
"But", you whine, "that's going to take a long time and a lot of money!"
Yup. That's why mil-spec parts cost so much. And even then, those are typically qualified by running at temperature for a few hundred hours or so. However, the manufacturer can get away with such short-term production test because the development cycle did the long-term, exhaustive testing and the process and materials have enough of a track record to trust. You don't have that luxury.
As you mention that the part is from TI, they have an extreme environment testing group which works on qualifying existing parts beyond their initial specification. For example extending the temperature range from -40 to 80 °C to -55 to 125 °C.
Go talk to them, the parts won't be cheap but they do have the means to guarantee you that the controller will work under those environments. Or not, but they will tell you.
Of course you have to present a business case for them, if you want to buy 10 controllers a year, they probably won't be interested. But if enough of the parts are demanded (by you and others) they might go for it.
If I remember correctly those parts will be 100 % tested in production, but don't quote me on that.
We sell sensors to customers requiring -50 °C and every sensor gets tested in a climate chamber at that temperature - but even then we don't allow safety critical applications. You can buy the same sensor with SIL certification and try to use it, below -40 °C it will enter the safe state. It is the only thing you can do as a responsible developer if you use parts which are not qualified for -55 °C usage.
Testing for the minimum operating temperature is part of characterization testing. The person doing the testing (usually the manufacturer) needs to have the ability and knowledge to run an exhaustive test on the device, verifying that every function of the device works correctly. There is usually no advance knowledge of which function is most likely to fail, so the exhaustive test is really necessary. Unless you have intimate knowledge of the internal design of the device it is unlikely that you can create the necessary test.
Performing testing at low temperature is usually more challenging that doing it at high temperature because of frost buildup around conductive paths. When I was involved in this we used dry nitrogen gas that was boiled from liquid nitrogen to cool the DUT. You have to keep the gas flowing to prevent frost buildup, and carefully prevent air from getting into the test assembly. Remember that the entire test assembly must operate at clock frequencies that are at least as fast as the final application, which is also difficult because you usually have relatively long connections to the DUT.
As mentioned in the comments, most manufacturers use statistical sampling to verify that a given part type passes the test. For safety-critical applications you might test each and every part.
Since you have mentioned "safety critical" in the comments.
You must provide safety interlocks that prevent the MCU from doing anything that will cause injury or undue property damage no matter what mischief it gets into, including some kind of partial functioning under extremes of temperature, brownout, water splashed on the PCB (if that's possible) etc. Under only very rare circumstances can you depend on the MCU itself.
Just as one ferinstance, a particular MCU was known to malfunction under certain conditions such that the branch instructions would fail first (due to timing or whatever) so the entire code set would get executed, including whatever the GPIO pins were programmed to do.
If it's possible, try fitting a low value tht resistor (no less than 100R if at 5V, you dont want to heat your chamber) over the micro controller and sticking it with silicone (of the conducting kind) to the top (or a scend one under it). Due to resistive heating it should keep your chip warm.
The data sheet for that part lists -40C as both the minimum temperature for "Recommended Operating Conditions" (section 5.4) and "Absolute Maximum Ratings" (section 5.1).
(1) Stresses beyond those listed under Absolute Maximum Ratings may cause permanent damage to the device. These are stress ratings only, and functional operation of the device at these or any other conditions beyond those indicated under Recommended Operating Conditions is not implied. Exposure to absolute-maximum-rated conditions for extended periods may affect device reliability.
There is nothing your test program can do about this. If something goes wrong with your "safety critical" product, the lawyers are going to be all over that paragraph.
So what can you do instead? Build a system that ensure the temperature will never go below -40C, either operating or storage, for the shipped and installed product. Add a temperature sensor on the chip, and use either internal or external circuitry to ensure that it's never below -40 (with some allowance for inaccuracy, etc)