Abstract

Diffractive photonic structures exhibit high-dimensional, device-specific variability arising from coherent interference, wavelength-dependent phase accumulation, and fabrication-induced disorder. These properties have motivated increasing interest in optical physical unclonable functions (O-PUFs); however, converting analog optical responses into compact and repeatable binary representations suitable for practical authentication remains challenging.

In this work, we address this limitation using graphene diffractive zone plates (GDZPs) combined with a vision transformer-based model. Rather than compensating for diffractive chromatic aberration, we exploit the wavelength-dependent focusing behavior of GDZPs as an additional optical challenge dimension. Under multi-wavelength excitation with predefined RGB power ratios, the devices generate spatial-spectral diffraction patterns shaped by device geometry, material dispersion, and inherent fabrication variations, enabling spectrally programmable optical readout. A self-attention-based encoder captures global diffraction features and maps the optical responses into a compact 64-dimensional representation, from which stable 16-bit binary response units are extracted through statistical thresholding and reliability-based bit selection.

Experimental evaluation demonstrates highly repeatable binary response units, near-random cross-configuration Hamming-distance behavior, low configuration-level error estimates, and clear separability across spectral challenge conditions. These results establish a proof-of-concept spectral-challenge GDZP-transformer pipeline that converts structured diffractive optical responses into repeatable and composable binary response units, highlighting the potential of programmable spectral-spatial optical authentication.