Secure Hardware Assurance Using Visual AI on AOI Imaging of Electronic Assemblies

High-reliability electronics demand full-coverage verification of component integrity, yet conventional inspection methods remain limited in scope and unable to detect subtle or undocumented modifications. This work presents a secure hardware assurance framework that leverages visual AI, specifically deep learning applied to high-resolution images from Automated Optical Inspection (AOI) systems, to detect component-level anomalies in printed circuit board assemblies (PCBAs). The modular system integrates object detection, semantic segmentation, and a second-stage anomaly classifier, enabling detection of unauthorized modifications, substitutions, and structural deviations with high precision. Trained using a bootstrapped procedure on multi-domain datasets, including clean production boards and degraded scrap units, the model generalizes across real-world conditions with over 99% detection accuracy and sub-second board-level analysis times. In light of recent hardware-level cyber threats, such as the discovery of rogue communication devices embedded in commercial infrastructure, the need for scalable, image-based verification has never been more critical. This method transforms existing AOI data into an intelligent layer of visual forensics, enabling manufacturers to detect covert hardware changes and enforce trust in deployed electronic systems.

Significant vulnerabilities have been found in chips. Computer programs and methods have been developed to prevent, find, and mitigate them. We proposed Secure Hardware Assurance Reference Dataset (SHARD) as a repository of reference examples (test cases) of both vulnerable and “clean” hardware IC designs. SHARD test cases will enable tool makers to test their techniques and allow chip designers to evaluate those tools. SHARD received preliminary funding through NIST Building The Future (BTF). This reports what we have achieved with that funding.