Research On CZ Method Silicon Single Crystal Growth Model Identification And Melting Process Detection Method Based On Deep Learning

As the main driving force for the development of the semiconductor industry,silicon single crystal has become an indispensable key material.In order to grow low-defect silicon wafers,it is of great significance to study the existing growth process,and at the same time,the rapid development of deep learning in model parameter identification and image processing has also become a hot research topic at present.In this paper,the equal diameter stage and the melting stage of silicon single crystal growth process are studied emphatically,and the identification model of equal diameter stage and the classification model of melting stage based on deep learning are proposed.In the identification model of equal diameter stage,a diameter model identification method based on LSTM neural network,including network structure identification and training algorithm,which is proposed based on the constant-speed growth strategy and the application of deep learning in parameter identification.Thus,the model has the ability to preserve the key information of the past and present.Firstly,by using the heater power and diameter data of the equal diameter stage and the support vector machine algorithm,the input and output order and lag order of the model are determined.Then,the nonlinear large delay model of heater power-diameter is identified by using the LSTM network structure designed in this paper.Finally,the three layers of BP neural network model and support vector machine model are used as a contrast experiment.The simulation results show that using the support vector machine algorithm to determine the order and lag of the diameter model of the equal diameter stage of silicon single crystal meet the theoretical value,and the LSTM neural network proposed in this paper obtains a more accurate precision diameter model than other methods.In the classification model of the melting stage,the traditional method is mainly based on manual observation,which may cause a series of adverse effects such as misjudgment.In this paper,the convolution neural network is taken as the research object by combining the deep learning method with the image classification of the chemical process.By using the image data collected by the melting process,AlexNet network and its adjustment of the number of convolution layers and the size of the convolution kernel,a CNN-based classification model of the melting stage is finally determined after many experiments.The simulation results show that the classification model of the melting stage proposed in this paper has achieved higher accuracy.