Advancing Neural Network Applications with a Fully-Optical 4×4 Matrix Operation Architecture: Simulation and Analysis

In recent years, optical computing has emerged as a prominent research focus due to its capabilities of high speed and low power consumption. In this paper, we propose a fully-optical neural network architecture comprising 4×4 optical matrix operation units and optical non-linear unit. Through simulation and analysis across different platforms, we validate the effectiveness of this structure. Compared to other hybrid optical neural networks, our approach minimizes power consumption associated with frequent conversion between optical and electronic signals. Additionally, it utilizes fewer optical components, thereby reducing the impact of thermal crosstalk.