Work

Optically reconfigurable physical unclonable functions based on 2D MoS2 ring-oscillator arrays for attack-resistant hardware authentication

Year: 2026

Type: article

Abstract: Hardware-level security is crucial for establishing trust in the rapidly expanding Internet-of-Things (IoT) and edge computing systems. A promising approach employs physical unclonable functions (PUFs) which leverage intrinsic process variations to generate unique and irreproducible identifiers. However, conventional silicon-based PUFs often suffer from limited entropy and lack of reconfigurability, becoming vulnerable to machine learning attacks. Here, we present an optically reconfigurable PUF based on a 64-cell array of five-stage ring oscillators fabricated from wafer-scale monolayer MoS2. The system exhibits spectrally selective frequency shifts under red, green, and blue (RGB) illumination, establishing a dynamic optical entropy dimension that enables on-demand, reversible rekeying without hardware modification. We develop a robust key-generation pipeline combining within-chip normalization, random-projection q-ary quantization, and the Secure Hash Algorithm 256 (SHA-256) privacy amplification, followed by hash-based message authentication code (HMAC)-based key derivation. The resulting keys demonstrate near-ideal uniformity (~50%) and inter-device Hamming distance (~0.677 at q = 3, 39 ~ 0.761 at q = 4), while remaining resilient to advanced machine learning attacks (≤ 52% accuracy). We further demonstrate image encryption and authentication with noise-like ciphertexts and reliable tamper detection. This work introduces a promising class of material-intrinsic, optically addressable security primitives for trusted edge computing applications. Conventional Si-based physical unclonable functions (PUFs) are normally limited by limited entropy and lack of reconfigurability, exposing them to machine learning attacks. Here, the authors report optically reconfigurable PUFs based on 2D MoS2 ring-oscillator arrays, demonstrating machine-learning-attack-resistant image encryption.

Source: Nature Communications

DOI: 10.1038/s41467-026-72557-8