摘要
Modern computational technologies are gradually encounteringsignificant limitations, driving a shift toward alternative paradigms such as optical computing. In this study, novel all-opticalcombinational logic units based on diffractive neural networks(D2NNs) were introduced, which were designed to perform highorder logical operations efficiently and swiftly with the adoptionof only two modulation layers. This innovative design exhibitsincreased processing speed, improved energy efficiency, robustenvironmental stability, and high error tolerance, making itexceptionally well-suited for a broad spectrum of applications inoptical computing and communications. By leveraging thetransfer learning, we successfully developed a fifth-ordercascaded combinational logic circuit for a practical informationtransmission system. Furthermore, we revealed a pioneeringapplication of the device in optical time division multiplexing(OTDM), demonstrating its capability to manage high-speeddata transfer seamlessly without the need for electronic conversion. Extensive simulations and experimental validationsdemonstrate the potential of the model as a foundational technology for future optical computing architectures, which pavesthe way toward more sustainable and efficient optical dataprocessing platforms.
出处
《Chip》
2024年第4期43-52,共10页
芯片(英文)
基金
supported by Research Funds of Hangzhou Hangzhou Institute for Advanced Study,UAS(No.2023HIAS-Y008).
