I thought this problem was gone
You said it seems to be related to midi activity.
Both DIN5 and USB ?
Don't you think it can be realated to Display activity that follow midi note ?
Have you tried to use the sequencer with all level metter on ?
Let it play (without midi data) with the display showing all level metters and see if it's stable.
When the TFT does not change, no data is sent to it. So it's stable.
The CPU only send data when the display has something new to show.
From all the tests i've made (and i've made a lot), the problem was some random corrupted data (due to external noise), that puts the TFT in a bad state.
The TFT displays is today very stable on the preenfm3s i have. (I had to tweak many parameters in the firmware).
Is your TFT soldered ? or do you use a connectors between the main board and the TFT ?
Let me know if you found something,
I can make a firmware with slower Data transmission to the TFT, to see if that makes change.
Xavier