Abstract

We propose an all-optical nonlinear activator based on the fiber-optic ring resonator (FORR) to implement physically nonlinear activation at low optical signal intensity. The FORR is fabricated from a simple passive optical element, suggesting that optical nonlinearity can be generated by the inherent properties of the device without the need for additional energy consumption. Our device exhibits advantages of excellent nonlinear traits, high stability, versatility, and compatibility in combination with existing optical chips. These features have potential advantages for the scalability of optical neural (ONNs). As a proof of concept, we show the capability of FORR as a nonlinear unit for nonlinear mapping in ONNs. For classification and super-resolution reconstruction tasks, the performance of our model, obtained from experimental measurements, is comparable to that of activation functions commonly used in computers. Our work may provide a feasible avenue for novel layout and wider deployment of all-optical neural networks.