Enhancing the Performance of Diffractive Neural Networks with Second-Harmonic Generation

Diffractive neural networks (DNNs) utilize light diffraction for optical information processing, offering advantages such as speed, energy efficiency, and the ability to process phase-encoded information [1]. However, incorporating a nonlinear activation into a DNN, essential for depth and complexity, remains a challenge. Three-wave mixing parametric processes in the depleted regime of interaction have been shown to be viable candidates for realizing an all-optical nonlinear activation function [2], yet reaching the depleted regime requires strong nonlinearities that are generally more challenging to realize.