Design of a Perceptron-Based Optical Neural Network Using Two-Dimensional Materials

Recent advances in machine learning have introduced powerful computational tools that are increasingly transforming materials research, spectroscopy, and optical characterization. Beyond their conventional use in data classification and prediction, artificial neural networks (ANN) serve as universal approximators for nonlinear functions—closely mirroring the intrinsic nonlinearities that govern optical systems. This conceptual overlap has inspired the emergence of optical analogs of computational architectures.

In this work, we propose the design and realization of a perceptron-based optical neural network (ONN) that directly engages with an optical material platform. The proposed architecture harnesses nonlinear effects in bulk optical materials and transition-metal dichalcogenides such as MoSe₂ and MoTe₂ as activation functions paving the way toward a scalable high-speed optical information processing unit.