Abstract: At present, most of studies of wafer manufacturing focused on discrete data defect pattern recognition, while chip lithography manufacturing was a continuous stacking process, and wafer overlap error monitoring based on continuous data was challenging and necessary. The data interpretability was fully considered in the processes of data monitoring, and new penalty terms were added in combination with wafer data characteristics and physical significance. A robust and sparse principal component analysis technique with high flexibility was proposed on basis of the improved LTS-SPCA dimensionality reduction model. Considering the geometric characteristics of wafers, optimal multivariate coupled statistic based on Copula properties of reflection and permutation symmetry was established to monitor the stacking process anomalies of wafer manufacturing. The accuracy of proposed method may reach 91.75%, which has high engineering application values.

Key words: abnormal monitoring, improved LTS-SPCA model, Copula property, wafer manufacturing

摘要： 目前，大多数晶圆制造研究集中在基于离散数据的缺陷模式识别上，而芯片的光刻制造是连续叠加过程，因此基于连续数据的晶片重叠误差监测具有挑战性和必要性。在数据监测过程中充分考虑数据的可解释性，同时结合晶圆数据特性及其物理意义加入新的惩罚项，改进LTS-SPCA降维模型，提出了灵活度较高的稳健稀疏主成分分析技术；然后基于Copula的置换对称、反射对称两种性质，考虑晶圆的几何特征，建立了最佳多元耦合统计量，用于监测晶圆制造的叠加过程异常。所提方法监测异常数据的准确率可达91.75%，具有较高的工程应用价值。

关键词: 异常监测, 改进LTS-SPCA模型, Copula性质, 晶圆制造