Exploring nonlinear activation function within microring resonators for all-photonic neuromorphic computing

This paper explores a groundbreaking Nonlinear Activation Function (NLAF) module for Optical Neural Networks (ONN), utilizing innovative micro-ring resonators (MRRs). These resonators, capable of generating whispering gallery modes within a single-mode hybrid silicon waveguide, offer tunability via temperature adjustments of phase-changed materials (PCM) on the MRR. This research delves into the nonlinear activation function of these resonators, showcasing their potential as optical neurons in ONNs. It particularly focuses on the temperature-dependent nonlinear response in various phase-changed materials (PCMs), such as MgF₂, CdGeP₂, and LiNbO₃, showcasing how temperature variation alters amorphous and crystalline structures on silicon, thereby impacting refractive index properties. This control leads to a sigmoidal response in transmission, representing the nonlinearity of the structure. We use the numerical analyses based on the Finite Difference Time Domain (FDTD) method to calculate the transmission of MgF₂ at 1336 nm and 1540 nm, CdGeP₂ at 1350 nm and 1560 nm, and LiNbO₃ at 1341.4 and 1547 nm, respectively at various temperatures. Adjusting the temperature from 20 to 300 degrees Celsius results in an average transmission change of 0.000065 for MgF₂, 0.0018 for CdGeP₂, and 0.0188 for LiNbO₃. This indicates a significant variance in the temperature sensitivity and optical transmission properties among these materials, highlighting their potential to satisfy the requisite optical nonlinearity for the NLAF in future all-photonic neural networks.