SHADE: Automated Refinement of PCB Component Estimates Using Detected Shadows

Abstract

Automated Visual Inspection (AVI) is an established approach for Printed Circuit Board (PCB) analysis. Several existing AVI methods have been well-developed for quality assurance, but not for hardware security and trust assessment. The main distinction between these two cases is that a golden sample (reference PCB) is available in the former application but is not guaranteed in the latter. Hence, conventional AVI approaches falsely detect many non-components (e.g. traces, vias, board text) as legitimate ones (e.g. resistors, capacitors, transistors). Such false detections cripple the scalability of hardware assurance approaches. In this paper, we propose an algorithm which exploits shadows cast by surface-mount components to accurately distinguish them from their invalid counterparts. It refines an initial component detection estimate by ruling out false positives. Moreover, the algorithm yields a per-component confidence metric which adds flexibility to the process’ sensitivity threshold. Tests on a sample PCB indicate as many as half of the initial false positives from a component estimate can be removed while retaining 96% of the true positives.

Publication
IEEE Physical Assurance and Inspection of Electronics
Olivia Dizon-Paradis
Olivia Dizon-Paradis
Graduate Research Assistant

My research interests include artificial intelligence, machine learning, computer vision, and reinforcement learning