Parameter Extraction And Inverse Design Of Semiconductor Laser Based On Deep Learning And Particle Swarm Optimization

As a research field,deep learning is a very important technical means.In recent years,deep learning has been widely used in the field of Photonics due to its accuracy and low time cost.As a typical optoelectronic device,semiconductor laser has the characteristics of small volume,light weight,wide wavelength range,and high reliability.It is one of the key devices widely used in the field of optical communication,imaging,detection,and industrial processing.Output power,small signal response,and wavelength characteristics are important output characteristics of semiconductor laser,which are determined by many parameters of semiconductor laser.In the manufacturing process of semiconductor laser,if the output spectral line does not meet the production requirements,it is necessary to extract the internal parameters of the problem(i.e.inverse design).This process can potentially guide actual industrial production.The traditional inverse design process of semiconductor laser adopts numerical calculation method,which is based on a variety of physical processes inside the semiconductor laser and solved by relevant formulas.This method needs to iterate the equation for many times,which will occupy a lot of computing resources and eventually lead to high time cost.In this paper,a method of semiconductor laser parameter extraction and reverse design based on DL and particle swarm optimization algorithm is proposed.By combining DL and particle swarm optimization algorithm,the fast and accurate reverse design of the seven design parameters selected in this paper is realized,and the relevant data are counted,which verifies the effectiveness of this method.The main research contents and achievements of this paper are as follows:The first is to study the theoretical method used in this paper,namely,TWM,which takes into account temperature.The method is to combine the carrier and photon in the semiconductor laser effectively in the form of matrix,so as to simulate the output characteristics of the electric and optical of the semiconductor laser.The input and output calculated by TWM simulation method are used as the data set of DL for the later training process.In this paper,the L-I curve and small signal response curve of semiconductor laser are selected as the output indexes of neural network to map the laser system.The second is to establish a fully connected deep learning neural network,extract the data set generated by TWM method,and improve the network structure according to the actual situation.The test data set is imported into the trained network,the output curve generated by the network is compared with the original data,and the relationship between the test error and the amount of data is analyzed.The results show that the network has good fitting ability.By comparing the test results of the network at different temperatures,the fitting ability and prediction ability of the DL method are proved.The third is to study the basic principle of particle swarm optimization(PSO)algorithm,and the algorithm is set according to the problem of reverse design in this paper.The combination of particle swarm optimization algorithm and deep learning method can quickly and accurately reverse design the parameters of semiconductor laser.Due to the multi solution characteristics of the problem studied in this paper,the parameters to be reversed are designed many times,and the statistics and error analysis of the parameters are carried out.The results show that the distribution of the parameters obtained by the DL combined with the PSO algorithm is not completely random,but close to the distribution of the original data within a certain error range.Thus,the effectiveness of the inverse design method is verified.