直拉单晶硅生长过程中放肩阶段常由于断棱的出现导致无法顺利进入等径生长。为研究影响断棱的关键特征参数，提出采用最大互信息系数法(MIC)辨识直拉单晶硅放肩阶段关键特征参数。分别采用最大互信息系数法(MIC)、层次分析法(AHP)计算特征参数与放肩断棱的相关系数，按照降序依次提取前k项直至全部特征参数作为输入参数，在逻辑斯蒂回归模型进行放肩断棱预测评估。结果表明，采用MIC提取特征参数的前13项特征作为输入参数时，模型的准确度最高；且采用MIC法在预测精度上优于AHP法。

The crystal often cannot smoothly enter the body growth process from the shoulder growth process during crowning growth of Cz Silicon monocrystal growth due to the appearance of Broken Edge. To study the key characteristic parameters that affect Broken Edge, a method based on the Maximum Mutual Information(MIC) to identify the key characteristic parameters of the shoulder growth process of Cz-Si monocrystal growth were proposed. The correlation coefficients of the characteristic parameters to a Broken Edge problem each was calculated by the MIC method and the analytic hierarchy process (AHP). Then the first k terms were sequentially extracted in descending order until all feature parameters were used as input parameters for the Logistic Regression model to predict probability of Broken Edge. The results obtained show the accuracy of the model is the highest when the first 13 features of MIC extracted feature parameters were used as input parameters; and the MIC method is superior to the AHP method in prediction accuracy.