stm32h750-weact

WeAct STM32H750VBT6 core board

Board-specific firmware knowledge. The two reference files hold the detail — read the one you need rather than guessing, because most of the failure modes below are silent (the board runs, just wrongly).

• reference/board-hardware.md — complete hardware reference: schematic-level pin map, every connector, power tree, memory map, SDK/examples inventory, plus a full development guide (Part II: §12 PlatformIO, §13 clocks, §14 peripheral cookbook, §15 M7 gotchas, §16 flashing, §17 pitfall table).
• reference/recipes.md — copy-paste-ready code: platformio.ini, clock config, LCD driver with the fast blit path, backlight PWM, ADC temperature sensor, USB CDC, DWT frame pacing.
• template/ — a complete working project that builds and flashes, plus a scaffolding script. It is where every recipe was extracted from; see template/README.md.

Orientation

Rules that prevent the expensive mistakes

Check these before writing code — each one produces a failure that looks like something else.

1. -DHSE_VALUE=25000000 in build_flags. The stock framework-stm32cubeh7 CMSIS template already defaults HSE_VALUE to 25 MHz — matching this board's crystal by coincidence, the same way the F4 template does for the Black Pill — so this currently works without the flag. It stops working the moment a CubeMX-generated stm32h7xx_hal_conf.h from a different H7 board (most Nucleo/Eval H7 boards run their HSE from an 8 MHz ST-Link MCO) gets dropped in: the PLL still locks, but every derived timing is wrong by ~3×. Set it explicitly so the project doesn't depend on which hal_conf.h happens to be in play.
2. PWR_LDO_SUPPLY — this board has no SMPS. Configuring one leaves the core under-supplied and unresponsive to SWD until a BOOT0 power cycle. The setting latches on first write after reset.
3. Fix the board definition's swapped sizes in platformio.ini: board_upload.maximum_size = 131072 (flash), board_upload.maximum_ram_size = 524288 (RAM). Otherwise nothing warns you when an image exceeds the real 128 KB.
4. board_build.stm32cube.disable_embedded_libs = yes if you vendor an ST7735 driver into lib/ — Cube's embedded BSP ships its own and they collide at link time.
5. FLASH_LATENCY_1 at HCLK 120 MHz. Too few wait states → hard fault on the first flash read.
6. Cache enable before HAL_Init(), clock config before any peripheral init.
7. D-cache vs DMA: buffers __attribute__((aligned(32))), size a multiple of 32, clean before TX / invalidate after RX. And DMA1/DMA2 cannot reach DTCM or ITCM — a buffer there silently never updates. Default linker puts everything in RAM_D1 (AXI SRAM), which is fine.
8. Backlight is TIM1_CH2N, a complementary output: HAL_TIMEx_PWMN_Start(), not HAL_TIM_PWM_Start(), and HAL_TIMEx_ConfigBreakDeadTime() + HAL_TIM_MspPostInit() are mandatory or the pin never drives.
9. USB needs HAL_PWREx_EnableUSBVoltageDetector() plus HSI48State = RCC_HSI48_ON. Missing it means the host sees nothing at all. 9a. The AF macro is GPIO_AF10_OTG2_FS, not GPIO_AF10_OTG1_FS. This part has two USB_OTG controllers, so stm32h7xx_hal_gpio_ex.h defines USB2_OTG_FS and compiles out the OTG1_FS macro name entirely — code copied from an H7A3/H72x example (which has only one controller) fails to build with an undefined-identifier error that doesn't say why.
10. LCD and camera reset lines are tied to the board reset net — no GPIO control, ever. A wedged panel needs a full MCU reset.