MoS2 as Nonlinear Optical Material for Optical Neural Networks

Abstract

Despite the significant advancements in photonic computation in recent years, the inadequacy of optical nonlinearities limits the scalability of optical deep networks (ONNs). Molybdenum disulfide (MoS <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> ), with excellent nonlinear properties, is emerging as a promising candidate for nonlinear processing. Here, we investigate the saturable absorption of MoS <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> by continuous wave lasers and illustrate the capability of MoS <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> as an activation unit for nonlinear mapping in ONNs. Moreover, a simulation-based fully connected neural network is fabricated for mimicking the operation of ONNs and demonstrating image classification. The results show that the recognition accurateness ranged from 89% to 94%, depending on the morphology of MoS <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> . This article provides a guideline for the selection of nonlinear units and opens up the possibility of implementing all-optical neural networks.