There are two points of confusions.
1. VREFINT_CAL is a 12-bit precision value
Unlike predecessor of H7 family (H743, H7A3, etc.), the VREFINT_CAL value stored in VREFINT_CAL_ADDR is a 12-bit precision value stored in a 16-bit (8-bit x 2) location.
For example, VREFINT_CAL values from devices in different H7 sub-families are as follow.
STM32H743ZI: 0x5f8a (24458) at 0x1ff1e860-0x1ff1e861
STM32H7A3ZI: 0x5ea8 (24232) at 0x08fff810-0x08fff811
STM32H725ZG: 0x5ea (1514) at 0x1ff1e860-0x1ff1e861
I can understand this change, as H72x, H73x and H730 only supports 12-bit precision on ADC3 where VREFINT is connected to in most applications. However, as RM0468 Rev 2 suggests that VREFINT is also supported on ADC2 where precision can be 16-bit, there is another level of confusion here.
The same applies for other calibration values.
In my opinion, nature of the VREFINT_CAL and other calibration values should be noted where necessary, in the reference manual and/or data sheet.
2. Different values of VREF+ is suggested for ADC1,2 and ADC3.
In RM0468 Rev 2, value used for target VREF+ calculation (corresponds to VREFINT_CAL_VREF) is given differently for ADC1,2 and ADC3.
p.1057 (ADC1,2) uses 3.3V
p.1164 (ADC3) uses 3.0V
Given a singular VREFINT_CAL value, there is clearly something wrong with these descriptions (one of the values, most likely 3.0V from actual measurement and experimental code shows, is wrong, or different VREFINT_CAL must be provided for each of these target VREF+ values).
Thoughts?
