关键词: 晶体生长;工艺参数优化;计算流体动力学;第二代非支配排序遗传算法;多目标优化

Abstract: In order to obtain higher quality silicon single crystal,this paper presents a hybrid strategy for modeling Czochralski silicon crystal growth and optimizing growth process parameters.This hybrid strategy includes computational fluid dynamics (CFD) method,data neural network based group method of data handing type,and non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ).According to engineering experience and process requirement,the shape variables of solid-liquid interface h and defect evaluation criteria V/G,which containing four design variables that are the heating temperature,the crystal rotating speed,the crucible rotating speed and the pulling speed,are used as the objective functions.Through numerical calculation based on CFD software,the required samples are obtained for GMDH training.Ultimately,the process parameter satisfying the Pareto optimal solution are acquired by using NSGA-Ⅱ.Tested by practical projects,it is proved that the proposed hybrid strategy provides a new numerical method for obtaining crystal growth parameters.

Key words: In order to obtain higher quality silicon single crystal,this paper presents a hybrid strategy for modeling Czochralski silicon crystal growth and optimizing growth process parameters.This hybrid strategy includes computational fluid dynamics (CFD) method,data neural network based group method of data handing type,and non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ).According to engineering experience and process requirement,the shape variables of solid-liquid interface h and defect evaluation criteria V/G,which containing four design variables that are the heating temperature,the crystal rotating speed,the crucible rotating speed and the pulling speed,are used as the objective functions.Through numerical calculation based on CFD software,the required samples are obtained for GMDH training.Ultimately,the process parameter satisfying the Pareto optimal solution are acquired by using NSGA-Ⅱ.Tested by practical projects,it is proved that the proposed hybrid strategy provides a new numerical method for obtaining crystal growth parameters.