Simulation of EUV mask defect classification based on optical neural network

Optical neural networks perfectly combine the high parallelism, high-speed transmission, and low energy consumption characteristics with the powerful computing capabilities of neural networks, demonstrating unprecedented potential and advantages in the field of edge computing. However, when it comes to actual industrial application scenarios, the accuracy of optical neural networks still needs to be verified. In order to expand the application of optical neural networks, this work conducts a simulation study on EUV mask defect classification based on optical diffraction neural networks. By simulating the EUV lithography mask defect dataset and training an 8-layer optical diffraction neural network, high-precision classification of amplitude-type defects and phase-type defects was achieved. After 200 epochs of iterations, the accuracy of the model reached 100% on the validation set. The simulation results show that the optical diffraction neural network can achieve high-precision classification of amplitude and phase defects in EUV masks, opening up a new technical route for the problem of EUV mask defect detection.