A Novel Inverse Design Method Of Semiconductor Laser Based On Artificial Neural Network And Particle Swarm Algorithm

Semiconductor lasers play a very important role in many fields of modern life,especially in the case of rapid development of information technology today.Semiconductor lasers' economic and social benefits are remarkable.Optical fiber communication networks,including access networks,local area networks,metropolitan area networks and backbone networks,using semiconductor lasers as light sources,have spread all over the world.The rapid development of fiber-optic communication networks enables real-time communication and sharing global resources among individuals,companies,organizations,and countries.Artificial intelligence(AI)has developed rapidly in recent years,and the theoretical knowledge system of computer learning has gradually formed.Output power spectrum is an important reference characteristic of lasers,from which important information such as threshold current and peak power can be obtained.The power spectrum is determined by many parameters in the laser,that is.a set of determined parameters can be obtained by numerical calculation.Inverse design has many applications in scientific research and production.It is a multi-parameter and nonlinear inverse design problem to obtain a set of parameters corresponding to a fixed power spectrum of a laser.The traditional numerical methods of parameter inverse design for semiconductor lasers are easy to fall into local optimum and have a huge amount of computation.Therefore,an inverse design method for semiconductor lasers parameters based on artificial neural network(ANN)and particle swarm optimization(PSO)is proposed in this paper.The main research contents and achievements are as follows:Firstly,the basic principle of numerical simulation algorithm for distributed feedback(DFB)semiconductor lasers is studied,and the numerical simulation algorithm for semiconductor lasers with temperature and thermal effects is further studied.In order to improve the speed of simulation calculation,the Matlab parallel computing function is studied.and the spmd function is selected to speed up the numerical simulation process of semiconductor lasers.Under the condition of only 12 workers,the speed is increased more than 10 times.Secondly,the basic principle of BP neural network algorithm is studied.According to the inverse problem of parameters of semiconductor lasers to be solved in this paper,the basic settings of the neural network are set up.The neural network is trained and learned by using a large amount of data obtained from traditional numerical algorithm.Using this network to predict the power spectrum corresponding to any set of new parameters of the laser.the mean square error can be reduced by about 0.5mW.and the time consumed is only 0.07s.The efficiency is increased by about 1800 times(compared with the numerical algorithm in the same environment,which takes 125.57s)Thirdly,the basic principle of particle swarm optimization is studied,and the basic settings of particle swarm optimization are set up.Combining the trained neural network(instead of the numerical simulation algorithm)with PSO method,the corresponding parameters of target power spectrum can be obtained quickly,that is,inverse design can be realized.The parameters inversely designed is not unique,which proves the characteristics of nonlinear multi-parameters of semiconductor lasers.Compared with the traditional numerical inverse simulation method(whose mean square error is 0.89mW,time consumed is 192 hours),the ANN combined PSO inverse algorithm(whose mean square error is less than 0,04mW,time consumed is 39.45 s)can improve the accuracy by 22.25 times,and the speed by 17521 times,which shows the effectiveness of this method.