You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, I’m not a hardware-security engineer, but your compare and cross-board analysis features made me wonder whether the same approach could be extended from pairwise board comparison to population-level anomaly detection.
The idea is to collect many high-resolution images of the same PCB model/revision, learn the range of normal manufacturing variation, and then flag outlier components, rework, added ICs, wires, shields, or other post-manufacturing physical modifications.
The goal would not be to claim that a board is “malicious”, but only to identify physical anomalies that differ from known-normal samples and may deserve manual or vendor review.
Do you think this is technically sensible, and does it fit the direction of this project?
I’m mainly contributing the threat model and concept; I don’t have the hardware or computer-vision expertise to implement it myself.
Thank you very much for taking the time to read this. I also appreciate your work on this project and for making it publicly available. Any feedback, correction, or suggestion would be greatly appreciated.
Use Case
A possible use case would be supply-chain or post-manufacturing integrity checking.
For example, a user could upload high-resolution images of a PCB and compare them against a collection of images from other boards of the same model and revision. If most known samples share the same physical layout, while one board contains an extra component, unusual rework, different wiring, or another unexplained physical difference, the system could flag that region for manual review.
This would not prove that a board is compromised, but it could provide a low-cost first-pass screening method for detecting unusual board-level modifications.
Alternatives Considered
Possible alternatives would include:
Comparing a board against a single known-good reference board.
Using manufacturer-provided PCB/BOM reference data.
Using X-ray or CT inspection for deeper physical verification.
Performing manual visual comparison by experienced hardware researchers.
However, these approaches can be difficult for ordinary users because a trusted reference board may not be available, manufacturer data is usually not public, and X-ray/CT inspection is expensive.
A crowdsourced image baseline could potentially provide a lower-cost and more accessible first layer of integrity checking, while still allowing suspicious cases to be escalated to more advanced inspection methods.
Activity