Review of nonlinear activation functions in optical neural networks

Recently, the rapid development of electronic neural networks (ENNs) has enabled the widespread application of artificial intelligence in fields such as computer vision, natural language processing, and autonomous systems. As an emerging computing paradigm, optical neural networks (ONNs) have become a promising alternative to their electronic counterparts, offering advantages such as ultrahigh speed, low latency, and inherent parallelism. Nonlinear activation functions in ENNs are known to accelerate network convergence and improve accuracy across various tasks. Similarly, incorporating optical nonlinear activation functions into ONNs is crucial for achieving fully optical-domain neural network computing, which is an essential step toward leveraging the high-speed and high-capacity computing potential of ONNs. In this work, we first introduced several methods for implementing optical nonlinear activation functions. We then propose approaches for measuring their activation curves and exploring their interactions within network architectures. Finally, we demonstrated their roles in ONNs and discussed future development prospects and remaining challenges.