"IRIS Functional Cell Level Classification Dataset for Hardware Assurance"

"The absence of practical techniques that enable end users to independently verify fabricated silicon motivated the development of the Infra-Red In-Situ (IRIS) Inspection of Silicon technique. IRIS is a non-destructive, open-source optical imaging method that allows end users to inspect meso- and macro-scale structural features of integrated circuits (ICs), including embedded functional blocks. However, current IRIS-based inspection relies heavily on expert manual interpretation, limiting scalability and accessibility. In this work, we develop a pipeline for automated IRIS-based hardware inspection that integrates IRIS imaging, labeled functional cell data, and computer vision (CV)\u2013based classification, and evaluate its performance through cell-level classification. We create a functional cell-level IRIS dataset containing 246,410 labeled cell images for cell-image classification using chip-level IRIS images of nine functional blocks. Our experiments demonstrate that cell-level classification yields strong performance, with ResNet achieving an accuracy of 88.99%. These results underscore the potential of automated IRIS analysis and represent a step toward scalable, accessible, and expert-independent chip inspection. The curated IRIS functional cell dataset and associated source code are available at AICAS_IRIS to facilitate reproducibility and encourage further research by the broader community. To the best of our knowledge, this is the first work to explore functional cell-level classification on IRIS imagery."