What happens inside a micro when it overheats?

[JPortici]

Suppose i work for a company that does chiptuning car modules.
Suppose that we are testing a new product. We tested it on a customer's car for two days and no problems. Customer gets back the car, less than half an hour later he calls us saying that the ECU reports an anomaly.
99,9996% of the time what happens is that the *----* customer disconnected the module while the ECU was still powered (so no more signal from the sensor/s) because he wants to test if there is an effective HP gain and he doesn't care or believe us if we provide an off-on function
BUT
of course he would NEVER do that.

Anyway, we did some preliminary tests and we found out something interesting: with the module installed the 5V power line sits at about 0,5V but if we disconnect the module and connect the prex sensor directly it works. next time you start the engine the ECU clears the error code automatically. And that's unusual.
And if we wait for the ECU to switch off (5V line goes back to 0V) then start the engine the Error code gets cleared automatically. Again, unusual. Convenient, but unusual.

My thought was that the ECU measured too much current coming from the prex sensor so it limited it to see if it was a faulty sensor or something fishy was happening (already happened in other recent cars. the performance chip shares the same supply line. we were actually expecting for something like this to happen but this is the third car, all same model, and the only one to have this problem.)

the increased current consumption that may have caused the problem might have come from excessive heat -> increased current consumption -> greater inrush current from the sensor at engine start -> alarm
we didn't catch this during tests because we were at room temperature in lab and in the car the lid of the box that contains the performance hcip was off so we could program it on board. probably that provided enough airflow.

So the first thing i tried was to monitor the current consumption and the 5V rail while heating the board and MCU with a hot gun. the current increased by about 10%.. until the MCU was really , really hot and it crashed. i don't think it fell into a software trap or the oscillator stopped running because there are two LEDs that are fixed ON during normal operation.. and they were switched off.
The 5V rail was still at 5V but we switched the engine off and on and the rail fell to 0.5V
However, when i disconnected and reconnected the module it started working again.

I am wondering.. what the hell happened? i don't think that there is a thermal shutdown mode in the MCU. Surely it's not mentioned anywhere
it is a dsPIC33EV256GM106 by the way
that's fascinating, what might happen with heat (i am so bloody sure that the customer messed around with the cables, but what happened when overheating the MCU... (150°C rated, rest of the circuit has 125°C parts minimum))
first thing tomorrow i'm adding an ADC scan to the core temperature sensor so we can monitor it but let's hear some toughts

the excessive heat should alter the transfer characteristic of the MOSFETS inside the chip but we are running at a mere 8MHz and again, it wasn't a software crash unless it rebooted and then stalled there or something

[Brutte]

The more generic question would be "what if I violate absolute maximum rating".
What if I zap supply voltage in reverse, what if I bake it at 250degC for an hour, can I clock it 3x what the datasheet says, etc.

There is no definite answer for what happens really. But the core idea of absolute maximum rating chapter is that you are on your own with such experiments. That is why MC bothered with this chapter.

As for checking the temperature - your design has to guarantee the chip won't be operational above some Tmax. So your temperature checking routine would run out of specs and there is no guarantee it would operate correctly. What about some NTC on a RESET pin? Or you could use an LDO with temperature shut-down feature.

[Kjelt]

This is not defined. Bits can toggle, your program counter can slip, even flashbits might just erase, it is undefined.
The same thing as if you do power or clock glitching, it is not defined and unwanted situations might occur.
So that is why if you are operating in a temperature critical environment like an automobile the component it self should monitor the temperature and the integrity of the code.
You never want to get within 10% of the max. temp., you should design your apparatus with enough heat dispersal capabilities never to get close to that zone.
If you still do get in to the red the uC should shut down to a safe situation. On power up check the integrity of the firmware and make sure you are in a safe temperature operating zone.

[JPortici]

spoiler alert: it wasn't an over-current issue. what a surprise

as Brutte said "what if i violate absolute maximum rating" is the subject in question. the backstory was just for the fun of it
in this specific case the only parameter that could be violated was temperature (not really, it shouldn't get hotter than 80°C where we mount the chip, just under the hood, on the side. everything is rated at least at 125°C.)
the power supply is well stabilyzed, no overclocking.
As i said 99,99996% of the times this problems arises is because *somebody* touched the connectors and will swear on his life that he didn't do nothing. We have then to prove to him that he did it so i tested if it was a current consumption problem due to heat and it wasn't. we loaded the line with 5 times the normal current and no problem whatsoever.
all i was able to see is that if i put a ton of really hot air directly to the chip at some points it might hang up, what a surprise!

Funny stories are welcome, such as this

[@rt]

That will be interesting when I’m on the kind of machine that will watch it properly.
I hope the chip at a hardware-isn level doesn’t just stop. That’s not an outcome I’d prefer.

Down to business I’ve heard at radio club that the parts of a transistor that are supposed to be close together get too close together or shorted,
but I suspected a micro would stop being able to read it’s own program storage before that,
but I have encountered the program counter jumping to an unexpected location for a device inside a hot vehicle cab.

[Rasz]

*somebody* touched the connectors and will swear on his life that he didn't do nothing

a small drop of nail polish will suffice to show if connector was opened

[danadak]

There are a number of over stressing situations for micros,
CMOS in general.

1) Temp, can cause local hot spot heating and actually melt Si.
2) Over V and puncture gate oxide.
3) Over V and trigger latchup, many devices fail because this create
large current to flow and again hotspots develop......Can fuse open
chip/die bond wires.
4) Metal migration causing shorts, opens, in chip routes.
5) Radiation causing threshold shifts in transistors, increased leakage....

Regards, Dana.

[hamster_nz]

There are a number of over stressing situations for micros,
CMOS in general.

1) Temp, can cause local hot spot heating and actually melt Si.
2) Over V and puncture gate oxide.
3) Over V and trigger latchup, many devices fail because this create
large current to flow and again hotspots develop......Can fuse open
chip/die bond wires.
4) Metal migration causing shorts, opens, in chip routes.
5) Radiation causing threshold shifts in transistors, increased leakage....

Regards, Dana.

Also temp tends to increase propagation delay within logic circuits. Going from 25C to 125C slows a gate by ballpark 50% (source http://ihome.ust.hk/~eekursun/papers/c25.pdf). This can cause faults that are not repeatable when device is cold.