Finding a rare pixel corruption when the cause is many frames away from the symptom

Tags: Capture depth, Debug automation

Context A team building digital video processing in FPGA. Video data sets are structurally too large for JTAG tools — video verification requires extended history capture by nature.

The problem The failure mode: a rare visual artifact appearing after hours of operation. The root cause was far upstream from the observation — by the time the corrupted frame was visible, the originating event was long past any capture window. Classic needle-in-a-haystack, except the haystack regrows every time you restart the system.

Why standard tools failed A JTAG capture window of milliseconds cannot bridge real-world gap between cause and symptom. Every restart destroyed the history needed to trace back from the artifact to its origin.

The Exostiv approach Exostiv captured massive data deep inside the FPGA prototype at speed of operation, maintaining multiple frames information of capture history. The flow was fully automated in Python: captures, testing, post-processing — including partial image reconstruction directly from the captured data, turning raw waveforms into visually inspectable frames.

The result The team went from restart-and-hope iterations to root-cause analysis on a single capture.