基于偏最小二乘法回归的工序质量建模

针对制造工序质量控制问题,应用多元统计分析中的偏最小二乘回归法建立了质量模型.利用该模型可以定量分析加工工序与最终成品率之间的关系,进而通过将大量的工序影响因子约简得到主要影响因子子集.根据在线生产的相关质量数据,采用非线性迭代偏最小二乘法获得影响因子的权重.得到偏最小二乘因子权重可以在线预测成品质量变化,避免离线测试.在半导体制造实例研究中,以工序质量水平为自变量,成品质量水平为因变量,建立了质量水平传递模型,应用该方法可实现多工序质量异常的在线诊断和预测,为质量控制提供了定量依据.

Manufacture process quality modeling based on partial least square regression

Partial least square(PLS) regression,a multivariate statistical approach,is applied to the manufacture process quality control,so that quality modeling can be established.The quality modeling can be used to analyze the relationship between finished product ratio and individual manufacture process,and reduce the large set of process quality variables to a much smaller set of PLS components.The PLS component scores can be frequently computed by the nonlinear iterative partial least square method from the on-line process data.With PLS component scores the quality variables can be predicted on-line before conducting off-line tests.In terms of the semiconductor manufacturing research,quality level transfer model was established on the basis of taking process quality levels as independent variables and finished quality levels as dependent variables.The application of this approach can detect and forecast the multi-process abnormity and provide a quantitative analysis foundation for quality control.