Hello,
I have a strange problem... We have a bunch of equal boards (STM32H743XIH6) from manufacturer and some have temperature dependent problems (3 of 20). The problem seems like a software problem, but many software changes couldn't reproduce this.
I send with help of the QSPI data to an ASIC. The ASIC behaviors a bit like a memory IC. Everything I send could be displayed with help of an industrial bus and an IPC.
In 3 of 20 boards there are the wrong data in some packages (3 - 100 under 2000 packages/seconds are corrupt). We connect a logic analyser to the QSPI and it showed us the same strange thing: If the H7 sends 214 Bytes in upcounting order, at different positions the byte doubles and the last byte is missing. The data is then unaligned:
Decimal example:
... 155 156 156 157 158 ...
The QSPI Clock is perfect (no extra clock pulses), edges are perfect in scope... CS delays correct....
The number of fail packages are rising with chip temperature.
17 of 20 boards are good, but some of this boards have the same problem if the processor is a bit out of spec with 93 °C (85 °C max). This close tolerance makes me think there is a general problem but three boards have a wider spec.
I have memory caching disabled. The source memory for the QSPI is the DTCM memory. We use QSPI bank 2 as single bank.
I have tested software parts:
- don't send via DMA, do it blocking (HAL functions)
- write DMA src addr memory only once at boot time
- do blocking QSPI transfer within an FreeRTOS critical section
- vary QSPI clock (slower)
- reduce overall system load -> don't start some threads
- delete idle hook power save
Has anyone an idea what could be wrong? Because of the output of the QSPI I would say its software problem, but its working on lots of boards and only three failing and is also strong temperature dependent.
We will X-Ray the failing boards and check the BGA connections....
Maybe someone had experience with this kind of strange issues and has good advice to locate the problem closer.
Thx
Sebastian
