Wearable-Compatible All-Optical Physical Unclonable Functions With Hybrid Deep Learning-Based Authentication

Optical physical unclonable functions (PUFs), harnessing the innate entropy of disordered systems alongside the vast expandability of their challenge-response spaces, are emerging as promising alternatives to electronic counterparts in advanced security protocols. Notwithstanding the potential, their widespread practical deployment has been critically hampered by the reliance on cumbersome, precision optical apparatus for excitation, readout, as well as the implementation of capacity-expansion strategies. Here, we introduce a compact all-optical PUF system that integrates centimeter-scale plasmonic nanoarchitectures onto a mechanically compliant chip. Synergized with a lens-free imaging scheme, multi-channel optical encoding is realized upon mitigated hardware complexity, facilitating operation with wearable-grade devices. By proof-of-concept demonstration using oracle bone scripts, a unified framework for information encoding, encryption, and secure readout verification is established. To ensure robustness across multi-round optical acquisition, a ResNet-18 model is employed, achieving authentication accuracy of 91.7%. To further address the challenge of signature encryption and forgery identification, we develop a scale-reconfigurable frequency-domain-enhanced TransUNet model. This hybrid transformer architecture concurrently parses global contextual dependencies and micro-stroke details, elevating the detection accuracy among multiple forgers to 98.75%. This deep learning-enhanced all-optical PUF system offers a wearable-compatible and high-security anti-counterfeiting solution, particularly suited for the Internet of Things ecosystem.