STM32F7 Thermal Reset

Bill Finger · ‎2019-09-27

We have had trouble with a STM32F7769BGT6 resetting when heated to around 45°C. The part should be rated for 85°C, as are the rest of the components on our board. We've looked at the power supplies and they don't show any issues. To duplicate the problem, we heat the uC directly with a hot-air gun. Heating other components does not reproduce the issue, but heating the uC immediately causes the reset behavior.

Has anyone seen anything like this before?

(edit) To add some more information, we just performed an experiment, and it appears that bit 30 on one or more of the core CPU registers is switching from a 1 to a 0. Because the register is holding a memory address, the new incorrect address causes a bus fault.

Bill Finger · ‎2020-01-15

In the end, we reduced the clock core frequency to 200 MHz, and the problem disappeared. I recommend using this frequency for any product that must operate over the full temperature range. In fact, since 216 is only 8% larger than 200 MHz, I recommend 200 MHz for all applications.

View solution in original post

Danish1 · ‎2019-09-27

How strongly do you know you’re running within the specifications laid down in the data-sheet and reference manual?

I mean things like having sufficient wait-states and core voltage for the processor speed. Or PLL numbers.

Transistor characteristics change with temperature. Things that work rock-solidly at one temperature might fall over at a different one.

I must admit I don’t know if this is the problem, but it’s something to check.

Piranha · ‎2019-09-28

Have you tested this on more than one board and chip?

waclawek.jan · ‎2019-09-28

Internal temperature of the chip is higher than the temperature of its package.

What's the system clock? What's VDD? What VCAP capacitors are used, what's the internal voltage regulator setting? Check, double check and measure VCAP voltages, cold and also under the heating.

Review power supply paths, grounding arrangement. Check and double check if all VDD/GND (and VCAP of course) pins are soldered rock solid.

Can you perform the test on a "known good" board, such as Nucleo or Disco?

JW

Bill Finger · ‎2019-09-30

Not yet, but we should have another board to test this week.

Bill Finger · ‎2019-10-02

We tested on a second board today. It failed at 58°C in an environmental chamber. The die temperature sensor read about 65°C. Much less than the rated junction temperature of 105°C.

Bill Finger · ‎2019-10-02

> What's the system clock?

216 MHz

> What's VDD?

3.3V

> What VCAP capacitors are used?

Two AVX 12105C225KAT2A, 2.2uF 50V X7R 1210 package.

> What's the internal voltage regulator setting?

Power Scale 1.

> Check, double check and measure VCAP voltages, cold and also under the heating.

This is a good idea. We will do this and report back.

Bill Finger · ‎2019-10-02

I'm pretty meticulous in my designs. I'm using 7 wait states; the datasheet suggests 6 might be possible. The PLL is set: M=25, N=432, P=2, Q=9, R=7. 25 MHz input crystal.

I might experiment with M, though, now that you mention it. It is within specifications, but the N of 432 is right on the hairy edge. Thanks.

Pavel A. · ‎2019-10-02

So you've added that bit 30 in some register flipped and this is actually causing fault in your code, correct?

Which register? How the code relates to its value? Is it a pointer or index?

-- pa

Bill Finger · ‎2019-10-03

Pavel,

The register is R0. When the fault occurs, the register has just been loaded with a memory address 0xCxxx'xxxx, which would be on external memory. However, because of the bit flip the address becomes 0x8xxx'xxxx, which is nowhere and causes the bus fault when the address is loaded.