Why does sleep current increase significantly after 3 minutes?

Please post a trace of the supply current - as Carole Jacquinot did.

Do you have a timer or Watchdog that could explain the 3 minutes (ie, 180s)?

"in my situation ... it has nothing to do with I/O pins"

how have you verified that?

"If I do a reset, configure the I/O and then go directly to sleep, the high-current behavior does not occur"

So look at what your "real" code does in addition to that test case.

I discovered the source of this issue. It is a problem with the way the ADC peripheral is de-initialized in Cube. Reversing the initialization does not return the ADC to its original reset state. You have to perform an ADC hardware  reset, or it will start to draw current about three minutes into a sleep state. This problem is likely rooted in a silicon error, because a hard reset should not be necessary.

The way to fix it is to go to the HAL_StatusTypeDef HAL_ADC_DeInit(ADC_HandleTypeDef* hadc) function and find the lines:
/* Reset register CALFACT */
hadc->Instance->CALFACT &= ~(ADC_CALFACT_CALFACT);
And directly after, add the lines:
__HAL_RCC_ADC1_FORCE_RESET();
__HAL_RCC_ADC1_RELEASE_RESET();

This will reset the entire ADC system, so if you have other ADC processes going, you will want to consider this.

Hi @Tesla DeLorean ,

Thank you for the flag much appreciated and many thanks for @KRuss.1 for the findings . We escalated this case internally with this tracking number #160332.

Have a great day,

STOne-32.