This paper describes the application of model-based principal component analysis (MBPCA) to the identification and isolation of faults in NMOS manufacture. In MBPCA, multivariate statistics are applied to the analysis of the portion of the data variance that is unexplained by models based on material and energy balances carried out on the unit operations used in manufacture. It is demonstrated that the failure detection and isolation performance achievable using the model-based procedure exceeds that of commonly used univariate SPC or conventional PCA approaches. Index Terms-Model-based fault diagnosis, model-based PCA, NMOS device fabrication.

This paper describes the application of model-based principal component analysis (MBPCA) to the identification and isolation of faults in CMOS manufacture. Some of the CMOS fabrication processing steps are well understood, with first principles mathematical models available, which can describe the physical and chemical phenomena that take place. The fabrication of the device using a known industrial process is therefore first modeled "ideally," using ATHENA and MATLAB. Detailed furnace models are used to investigate the effect of errors in furnace control on the device fabrication and the subsequent effect on the device electrical properties. This models the distribution of device properties resulting from processing a stack of wafers in a furnace, and allows faults and production errors to be simulated for analysis. The analysis is performed using MBPCA, which has been shown to improve fault-detection resolution for batch processes. The diagnosis method is demonstrated on an industrial NMOS transistor fabrication process with faults introduced in places where they might realistically occur.