We have an existing custom iMX8 Mini board which is working fine. We call that one the green board.
We've updated the PCB, now red, and done a new build, using literally the same components (same iMX8 mini, exact same Micron LPDDR4), of course still very carefully following the NXP hardware and layout guidelines. However this new red board fails LPDDR4 training. Turning on the extra information within the DDR Tool appears to show the eye diagram being a flatline.
I've attached the logs from both passing and failing boards. Even at much lower speeds it fails and shows all ff's for the eye diagrams. How do I interpret this? Could a bad PCB stackup during PCB fab cause such a thing? What else might cause it? Thanks.