I did not expect it would take two months before I started posting logs on Hackaday. Since two months have already passed, I will just record what got done in that time.

After the chip selection was frozen, the first real headache was power-on sequencing. If the FPGA controls the sequence, who sequences the FPGA? Handing it to the MPU did not feel stable enough, and I have not seen people do it that way. So I gave it to the MCU — STM32H745XIH6. With proper protection it can still boot safely and handle emergencies.

The plan is: the M4 core plus the FPGA watch hardware status for the whole board and the whole machine, and respond. The M7 core runs on its own: signal processing, computation, and logic. In short, the STM32 owns the earliest sequencing.

Next is power. I plan to feed the whole motherboard from 24 V. First I built a simple protection circuit with a P-MOSFET:

After that, two TI TPS56637 parts do 24 V → 12 V (for the PCIe slot) and 24 V → 5 V (the board bus). Four SY8843 bucks then make 1.0 V, 0.9 V, 1.5 V, and 3.3 V. From 3.3 V, RT9013-12 / RT9013-15 LDO parts make 1.2 V and 1.5 V.

Once every rail exists, four TPS22966 load switches, driven by the MCU boot code, bring the different cores up in order.

Photos of the power tree: