Research On Batch Process Modeling And Diameter Control Of Czochralski Silicon Single Crystal

With the rapid development of integrated circuits,communications,aerospace and other fields,the size of semiconductor silicon wafers has gradually increased,and the manufacturing process has become more and more advanced,which puts forward higher requirements for the manufacture of silicon single crystals.The Czochralski method is the main method for growing silicon single crystals,which mainly includes the stages of materialization,seeding,shoulder setting,equal diameter and finishing.Among them,the crystal diameter control in the equal diameter stage is the most important control link,which directly determines the crystal energy.No high-quality growth.However,silicon single crystal growth is a typical batch process,which has the characteristics of.batch production,multiple varieties,intensive technology,and complex growth process,which requires precise control during the batch production of silicon single crystals.Therefore,in order to realize the batch growth control of large-size,high-quality semiconductor silicon single crystals and meet the crystal growth process requirements,it is of great significance and practical value to study the modeling and diameter control of the Cz silicon single crystal batch process.Based on the idea of data-driven and real-time learning modeling,this paper establishes a crystal growth process model that takes pulling speed and heater power as input,and melt temperature and crystal diameter as output.First of all,in order to reduce the real-time learning modeling time,this paper proposes a data search strategy based on fuzzy c-means clustering to ensure the rapidity of modeling.At the same time,the model retention and update mechanism is introduced to improve the utilization rate of the model.Secondly,considering that the data collected by the sensor may have anomalies,an outlier processing mechanism is introduced to enhance the robustness of the model.Then,in order to accurately describe the nonlinear characteristics in the industrial data,an extreme learning machine is used to realize local online modeling,which is used to improve the identification accuracy of the model.Finally,the simulation results show that the established model has higher accuracy than other models.Based on the above-mentioned crystal growth process model,in order to achieve the crystal diameter control requirements during the batch growth of silicon single crystals,this paper first adopts an iterative learning control strategy.Aiming at the strict repeatability of iterative learning control and the difficulty of dealing with the interference problem on the time axis,a model predictive control method on the time axis is introduced,and an iterative learning predictive control method for crystal diameter of silicon single crystal batch process is proposed.Among them,the MPC control method based on improved instant learning modeling is used to deal with the interference and constraint problems on the time axis,and the model-free adaptive iterative learning control is used to correct the repeatability error of the crystal diameter between batches to ensure that the batch axis is The crystal diameter can be precisely controlled.The simulation results show that the proposed batch process control method can achieve precise control of crystal diameter.