Multivariate fault detection (MVFD) EP/FDC implementation

The increase in the complexity of interrelations between multiple process input variables has led to the rise of multivariate fault detection (MVFD) as the next generation statistical process control methodology. Two or more variables (e.g. flow, power, temperature, pressure...) may appear to be within standard population while their combination indicates a non standard condition within the multi-dimensional space, which leads to discrepant material. Eight dimensional MVFD applied to the "excursion prevention" (EP/FDC) tool parameters of a medium current implanter has been demonstrated at an Intel fab. This paper describes how multivariate statistics such as principal component analysis (PCA). Hotelling T² SPC charts and contribution plots are being used to provide an on-line multivariate statistical control and excursion alert system. The system has detected problems that traditional univariate SPC systems could not detect.