Secure information transmission via hybrid optoelectronic encryption with physically embedded diffractive keys
Yuanchao Geng, Wenjian She, Jiacheng Li, Gang Luo, Tong Fu, Qiang Yuan, Deen Wang, Jingqin Su · 2025
The exponential growth of digital communication has underscored significant vulnerabilities inherent in traditional electronic encryption methods, which rely predominantly on computational complexity and thus face increasing security risks in the quantum computing era. Addressing these critical challenges, optical neural networks (ONNs), characterized by inherent parallelism, ultra-high throughput, and physical irreproducibility, have emerged as a compelling alternative for secure data transmission and processing. In this study, we propose a hybrid optoelectronic encryption–decryption regime that leverages a convolutional neural network (CNN)-based electronic encoder to transform input information into secure, phase-encoded optical patterns. These electronically encrypted patterns carry no direct structural information discernible via conventional computational decoding, ensuring robust security against electronic attacks. Subsequently, a physically instantiated diffractive optical neural network performs decoding using dual-wavelength coherent illumination. This optical modulation layer inherently acts as a unique physical key, providing an additional layer of security. Our numerical validation shows that the proposed system attains exceptional classification accuracies of 98.9% on the MNIST dataset and 92.5% on Fashion-MNIST, underscoring its outstanding performance and robustness. This hybrid approach, combining electronic precision encoding and secure optical decoding, provides a promising avenue toward physically secured information transmission, encrypted optical communication, and anti-counterfeiting applications.