Application of Principal Components Analysis to improve fault detection and diagnosis on semiconductor manufacturing equipment

With the evolutions in sensing technologies and the increasing use of advanced process control techniques, terabytes of data are recorded today during the manufacturing process of semiconductor devices. These large amount of data are then operated by Fault Detection and Classification (FDC) systems to assess the overall condition of production equipment. However, specific characteristics of semiconductor manufacturing such as highly correlated parameters, time-varying behaviors, or the large number of operating conditions tend to limit the efficiency of current indicators to detect and diagnose a failure occurence. There is therefore a significant requirement for the development and application of new methodologies to improve detection efficiency while reducing the complexity of condition monitoring, without losing detailed insight for efficient failure analysis. In this paper, we use data pretreatment algorithms from signal processing and time series analysis, and Multiway Principal Components Analysis (MPCA) methods to accurately represent equipment behavior and process dynamics and thus overcome issues inherent to semiconductor manufacturing context. A real-case application on a plasma etcher from STMicroelectronics Rousset 8' fab is proposed to highlight benefits of these methods.