Predictive Control Of Silicon Single Crystal Quality Based On Soft Sensor Modeling

The information industry and the new green energy industry are currently the two pillar industries of mankind,and silicon single crystal is particularly important as the basic material for the two pillar industries.With the rapid development of the integrated circuit and photovoltaic industries,higher requirements have been put forward on the overall dimensions and crystal quality of silicon single crystals.However,the growth process of silicon single crystal is a dynamic time-varying process with nonlinear,strong coupling,large hysteresis and uncertainty models.Traditional control methods are difficult to ensure that the quality of crystals meets actual industrial requirements.Therefore,the study of silicon single crystal quality predictive control has important value and practical significance.From the perspective of data-driven modeling and control,this paper proposes a soft-sensing model-based silicon single crystal quality predictive control method to achieve precise control of the crystal diameter and ensure that the crystal quality meets the actual process requirements.1?It is difficult to obtain V/G during the actual growth of Czochralski silicon single crystals.Therefore,a soft model based on Hybrid Variable Weighted Stacked Autoencoders Random Forest(HVW-SAE-RF)was established.In this model,the mixed correlation between the input of each layer of the network and the target variable is used to design the objective function,so that the extracted deep features and the target variable form a strong correlation,and the Random Forest model(RF)is used as the output Regression prediction is performed on the layer,and finally the predicted value of V/G is obtained.2?Based on the soft-sensing model established above and considering the uncertainty of the model,this paper proposes a hierarchical control strategy based on the soft-sensing model,aiming at the best output performance of the actual control system,and using the Gray Wolf Optimizer algorithm to solve it.Realize the real-time online monitoring of crystal diameter control and V/G value.Among them,the inner PID control is used to quickly stabilize the system;the outer Model Predictive Control(MPC)is used to deal with system constraints and enhance the control performance of the inner loop;the V/G value monitor is used to ensure The V/G output by the system meets the crystal growth requirements.3?Based on the industrial data of the actual Czochralski silicon single crystal growth process,it is verified that the built model has good predictive performance and generalization ability,and can provide accurate V/G prediction values for the layered control strategy.In addition,the layered control based on the soft-sensing model can not only achieve precise control of the crystal diameter,but also real-time online monitoring of the V/G value of the solid-liquid interface,and control the V/G of the solid-liquid interface within the range required by the crystal growth theory.Inside,it provides effective technical means to ensure the quality of the crystal.