Application Of Machine Learning In The Design Of Nano-photonic Structures

Nanophotonic structure is a composite structure that uses nanotechnology to combine two or more natural materials through artificial design.There is a huge design space for such composite structures in terms of material selection and structural parameters.Nowadays,nanophotonic structures play an important role in the fields of photodetectors,perfect absorption,nonlinear optics,thermal emitters,and photovoltaic systems.effect.However,traditional structural design methods are severely limited in design space and design effects due to the designer's experimental conditions and theoretical knowledge.In recent years,the use of machine learning methods to complete the design of nano-photonic structures has become a research hotspot for many domestic and foreign researchers.Compared with traditional design methods,machine learning uses a large amount of data for model training,which greatly improves the design space and design efficiency of the structure.In this paper,aiming at the design of the absorption spectrum of the multilayer film,based on the existing model research,it is innovated based on the idea of machine learning,and two neural network model architectures are designed and built.And in terms of design accuracy,design efficiency,model generalization ability,model migration ability in the face of new problems,etc.,a comprehensive evaluation and simulation verification of the designed model are carried out.A model design architecture based on the idea of generative confrontation is proposed,and the discriminator module is added to improve the convergence efficiency and generalization ability of the model.At the same time,a model design architecture based on the idea of multi-task learning is proposed,which decomposes parallel design problems,and aims to improve the prediction accuracy of the model.For the same verification set,compared with the traditional machine learning model architecture,the model architecture designed in this paper increases the simulation error of the predicted spectrum by 33%and 30%,respectively.At the same time,the model designed in this paper is applied to the design and application of solar absorbers.The purpose is to achieve perfect absorption for a specific wave range.The final model predicts and designs a set of multi-layer thin-film solar absorber parameters.The structure of the solar absorber is verified by simulation.The average absorption rate between 1500nm can reach 96%.