Tool/software:
This is a low-incidence issue for us (maybe 1/60 units), but has become a larger issue as our production volumes have grown. The issue follows the serializer board, not the imager (a camera experiencing this issue is always fixed by replacing the serializer board). Removing and re-seating the serializer board of a camera with issues does not resolve the problem. However, this issue does appear like it can come and go - serializers with issues will work sometimes and not others, and these problems usually first appear 1-100 days into the life of a camera, no apparent correlation with environmental conditions.
Hardware defects have been quite exhaustively investigated without finding any clues (PCBAs microscope inspected, probed for shorts and opens, capacitor and resistor values confirmed to be in-spec, voltage rails look clean, connectors replaced to rule out contact issues, boards reflowed and cleaned, etc). The design has also been checked for conformance with the datasheet / app notes and was previously reviewed by TI as well, no issues identified. MIPI traces are proper microstrip diffpairs, length matched and kept short.
In the instances where this is occurring, we see a mix of errors on the CSI_ERR_DLANE01 and CSI_ERR_DLANE23 registers. We see all 3 error types (Control error in HS request, multi-bit error in sync, single-bit error in sync) mixed across all 4 lanes, and it does not appear to be exactly the same on every occurrence.
Our suspicion is that idle CSI lanes are picking up EMI and causing the serializer to get stuck in a bad-state somehow, but we do not have the equipment to properly instrument CSI-2 without potentially introducing new errors.
Is TI familiar with any issues like this? Any guidance on what to test or check, or configuration / process changes to improve robustness / resilience would be greatly appreciated.