研究目的
Investigating the phenomenon of structure loss in Czochralski monocrystalline silicon growth through statistical analysis of production data to identify causes and control parameters.
研究成果
Manual temperature adjustments by operators reduce the risk of structure loss. Ingots with structure loss have higher average pull speeds and lower temperature fluctuations. A binary logistic regression model using the standard deviation of temperature fluctuations can predict structure loss with high accuracy.
研究不足
The analysis is based on industrial data which may have uncontrolled variables. The study focuses on statistical correlations rather than direct causal relationships.
1:Experimental Design and Method Selection:
Statistical analysis of industrial data including hypothesis testing, feature selection, and binary logistic regression.
2:Sample Selection and Data Sources:
Data from approximately 14000 n-type ingots produced over a year at NorSun factory.
3:List of Experimental Equipment and Materials:
Industrial Czochralski silicon growth furnaces.
4:Experimental Procedures and Operational Workflow:
Calculation of standard deviations of heater power and temperature fluctuations, pull speed averages, and fluctuations. Comparison of these parameters between ingots with and without structure loss.
5:Data Analysis Methods:
Two-sample t-tests, test of proportions, Sequential Feature Selection with Support Vector Machine and Gradient Boost Tree classifiers, binary logistic regression.
独家科研数据包,助您复现前沿成果,加速创新突破
获取完整内容