Support Vector Regression Modeling based Data-Driven Iterative Learning Control for Czochralski Crystal Growth Process

Abstract

In view of the previous Czochralski(Cz) monocrystalline silicon growth control method based on mechanism model, it is difficult to accurately reflect the nonlinear and real-time variation characteristics of crystal growth process. This paper proposes a data-driven iterative learning control method based on support vector regression, which combines crystal diameter prediction model by support vector regression (SVR) off-line training with crystal growth mechanism model, and then control the crystal diameter and output key variable information in the framework of iterative learning algorithm. Simulation results for diameter control of crystal growth process and the analysis of key variable information show that the proposed method is effective and has a fast crystal diameter learning speed, and good performances of setpoint tracking and disturbance rejection, high crystal growth speed.