I purchased a RAW-MIPI-IMX462M camera from Aliexpress. This was working correctly on a Radxa Zero 3E for several hours, then began returning zero-byte captures with a large burst of uncorrectable MIPI packet header ECC errors. It finally stopped acknowledging its I2C address entirely.

Hardware and software

• Camera: RAW-MIPI-IMX462M, firmware version 0x1090000.
• Host: Radxa Zero 3E, Debian bullseye, radxa-zero3_debian_bullseye_xfce_b6 image.
• Kernel: 5.10.160-39-rk356x, installed from rk356x_radxa_zero3w.3E_mvcam.tgz, release v1.3_mvcam, from https://github.com/veyeimaging/rk35xx_radxa/releases.
• Driver: veye mv series camera driver version: 01.01.08, compiled into that kernel.
• Device tree node mvcam@3b on i2c@fe5b0000, which is i2c-2.
• Cable: FFC22P-0.5MM-20CM-Y, opposite-face contacts, silver contacts facing outward as the wiki specifies.
• Mounting: Stable mounting to desk. No vibrations or stress points to cable or module.

What worked first

• i2cdetect -y -r 2 showed UU at 0x3b.
• Link came up as 2 lanes at 891 Mbps.
• Default format was Y10_1X10/1920x1080@100/6000.
• Captures succeeded at 2,764,800 bytes each, which is a 2560-byte stride by 1080 lines.
• A live streaming loop ran for several minutes at a time.

The failure, in the order it happened

Stage 1, i2c perfect and MIPI failing. Captures began returning zero bytes. dmesg showed 5449 CSI errors, of which 5390 were ecc2, uncorrectable two-bit ECC failures on MIPI packet headers. Register reads over i2c at 100 kHz stayed flawless throughout, while the 891 Mbps differential pairs corrupted essentially every packet header.

Stage 2, silent bus. i2cdetect -y -r 2 became empty at every address and the driver reported mvcam: failed to read chip id.

Stage 3, no recovery. A new, correctly oriented cable changed nothing. It is now silent on two independent hosts (Pi Zero 2W and Radxa Zero 3E). Tested Radxa and it still works fine with Arducam IMX477 camera.