2020-05-21 07:11 AM
According to the STM32H7 reference manual (RM0433), there is only a single 128kB sector of user flash space available for STM32H750xB devices (0x08000000 - 0x0801FFFF).
Other devices in the STM32H7 family (H742xI, H743xI, H753xI) have 2 banks of 8 128kB sectors, totaling 2MB of user flash available (0x08000000 - 0x081FFFFFF).
During initial testing with an STM32H750xB device, it was discovered that flash beyond bank 1 sector 0 (0x08000000 - 0x0801FFFF) could be written to and read from, contrary to the indication in the reference manual for this device. Flash sectors starting at 0x08020000 and 0x08060000 could be used.
Seeking an explanation for why flash sectors beyond the stated range seem usable for STM32H750xB devices and if there are any precautions to take when using these sectors.
Has anyone encountered something like this with any of the STM32 MCU families?
Solved! Go to Solution.
2020-05-21 08:24 AM
At least 1MB on the die looks usable.
Typically testing the erase, write, read functionality fully takes a lot of time on the IC test equipment, by testing only 128KB they can probably increase machine throughput by 10x or more. Usually not a case of it being tested and failed.
It is a "Use at Your Own Risk" / "No Lifeguard On Duty" situation. You should test for gross failure, and CRC/Sign the firmware image you plan on running. QSPI is also a very viable route.
2020-05-21 07:44 AM
Actually H750, H753, H753, H745, ... seem to share the very same die. But depending on the device, some features are (certainly irreversibly, due
to legal concerns) disabled, others (like flash) not fully tested. That's nothing new (see STM32F1xx or STM32G07x) and not specific to ST at all. ST's website currently lists 1059 types ... For personal use you could risk using the full flash, but for production purposes you should not rely on the undocumented features.
2020-05-21 08:24 AM
At least 1MB on the die looks usable.
Typically testing the erase, write, read functionality fully takes a lot of time on the IC test equipment, by testing only 128KB they can probably increase machine throughput by 10x or more. Usually not a case of it being tested and failed.
It is a "Use at Your Own Risk" / "No Lifeguard On Duty" situation. You should test for gross failure, and CRC/Sign the firmware image you plan on running. QSPI is also a very viable route.
2020-05-22 06:59 AM
Thanks!
Can you share how you know 1MB on the die looks usable but perhaps not the full 2MB? Is this a limitation of some other hardware in the H750xB for accessing memory?