Unfortunately STM uses a lot of external silicon IP (all MCU designers have to) from different firms that don't really work well together or are straight up inf

Unfortunately STM uses a lot of external silicon IP (all MCU designers have to) from different firms that don't really work well together or are straight up inferior as you mentioned. Their eratta sheets are shorter but I've run into a number of problems that I've worked around with help from their engineers and a year later I still couodnt find any notes on them. I'm guessing they underreport edge case silicon bugs like almost all silicon manufacturers. My last project used a STM32F437 with 100 mbit ethernet, just about saturating the connection at 80mbits while maxing out the processor to interleave the bits of all incoming data and bit banging every GPIO unused by peripherals to control 30k individually adressable LEDs at 45 fps which is just incredible for a $7-10 microcontroller. Then after a few weeks of field testing we identified half a dozen problems all resulting in nondeterministic failures of the DMA or MAC peripheral requiring a hard reset. I think the most problematic one, DMA1 or DMA2 stalling when they tick within 10-20 cycles of each other, is the only one that was added to the errata and still plagues the STM32F4x7 parts. That said if you care about part swapability, developer productivity, and need a very powerful ARM that you're not pushing to the edge, the STM32 line of chips is a great choice.