Study On Identification Method Of Crystal Diameter Model In Czochralski Silicon Single Crystal Growth

The rapid development of the solar photovoltaic industry and large-scale integrated circuits have put forward requirements for the semiconductor industry to produce larger diameter and higher quality single crystal silicon.The most common method for preparing high quality silicon single crystals is the Czochralski method(CZ method).In the actual growth of silicon single crystals using the Czochralski method,the crystal diameter is usually controlled by changing two key variables-the pulling speed and the thermal field temperature.Moreover,there are three characteristics in the crystal growth process:one is that the crystal diameter has a time lag in response to the change of the pulling speed and the thermal field temperature,the other is that the crystal growth process exhibits nonlinear characteristics,and the third is that the parameters of crystal growth process vary slowly over time.In order to obtain a silicon single crystal of equal diameter and high quality,it is more effective to control the crystal diameter by establishing a model between two variables and crystal diameter.Therefore,the focus of this paper is to obtain a model under the influence of the diameter on the tensile velocity and the thermal field temperature through data analysis.Since the crystal growth process involves multiple change processes and multiple variables,the model variables established by the mechanism modeling method are more difficult to solve accurately.Therefore,this paper adopts the idea of building a system model based on the real data of the system.In terms of model identification,it is divided into three parts,namely,the identification of the speed-crystal diameter model,the identification of the thermal field temperature-crystal diameter model,and the model identification of the influence of the thermal field temperature and the pulling speed on the crystal diameter.The identification step is:first,determining the model order and the time lag of output response to the input quantity change,and the second step performs the model parameter identification.First,during the crystal growth process,the data of the phase of significant change in crystal diameter caused by the change of two variables is collected,and the data is preprocessed;Secondly,to identify the structure of the model,the first step is to determine the model order of the pulling speed and the thermal field temperature by using the three-layer feedforward network and the fuzzy approximation,and the second step is to determine the lag order by using the determinant ratio method;Finally,after determining the model structure,the model parameters of the pulling speed-crystal diameter model and the thermal field temperature-crystal diameter model are identified based on the improved deep belief network algorithm.Taking the pulling speed and the thermal field temperature as inputs,the crystal diameter is the output of this nonlinear model,the model parameters are identified based on the improved convolutional neural network algorithm.Simulation experiments show that the lag order and model order obtained by this paper are in line with the empirical value,and the deep belief network identification method and convolutional neural network identification method improved in this paper have higher model identification than the traditional identification method.