Low operating power, low activation threshold, and diverse all-optical nonlinear activation based on a PPLN resonator.

Optical neural networks (ONNs) are considered next-generation physical implementations of artificial neural networks, yet their capabilities are constrained by the availability of nonlinear activation functions (NAFs). Current NAF implementations suffer from limited integrability, complex device architectures, and high latency. Here, we theoretically explore diverse all-optical nonlinear activations based on a periodically poled lithium niobate (PPLN) resonator. The device supports passive propagation and features an ultralow activation threshold (0.1 mW) and low operating power (below 6 mW). The underlying physical dynamics are mapped to nonlinear activation, enabling high classification accuracy on the Moons, Iris, and MNIST datasets. These results highlight the potential of this approach for ONNs. We believe this work will facilitate future large-scale photonic intelligent processors with enhanced functionality and simplified architectures.