Preliminary Exam Seminar: Junyuan Lu
"Deep-learning-assisted Optical Characterization of Disordered Nanostructures for Noise-resilient Authentication"
Junyuan Lu, MSE PhD Candidate
Advisor: Jinglang Li
ABSTRACT
Physical unclonable functions (PUFs) are devices utilizing the inherent randomness of materials for hardware security. They can be widely applied in industrial production and daily life as anti-counterfeiting authenticators. In recent years, optical PUFs, a major branch of PUFs based on optically active materials, have attracted numerous research interests because of their high non- linearity and outstanding machine learning attack resistance. An optical PUF exploits the distinctive optical response of disordered nanostructures to incident light. The wavelength, polarization, and direction of illumination can be encoded as challenges, and various optical characterization methods, such as bright-field/dark-field imaging, photoluminescence, and Raman spectroscopy, can be used to capture the responses. However, the imaging noise arising from mechanical drifting, ambient light, and misalignment during characterization affects the authentication accuracy since the algorithm-based readout lacks flexibility. This Critical Review provides an overview of recent advances that integrate deep learning models in optical PUFs for robust readout. We first discuss existing optical characterization methods in PUFs and the selections of deep learning architecture for specific response dimensions. Then, we systematically analyze the implementation of discriminative learning techniques in optical PUFs, their inherent limitations in closed-set classification, and the recent shift toward representation learning for open- set and noise-resilient readout. We further summarize current research gaps and potential future developments such as standardized benchmarks, small models for mobile operation, and deep learning for reconfigurable PUFs.
2026-11-19 09:00:00 2026-11-19 10:00:00 America/Indiana/Indianapolis Preliminary Exam Seminar: Junyuan Lu HAMP 2101