Clustered defects in IC fabrication: Impact on process control charts

In IC fabrication, standard process control charts for defects often sound many false alarms, i.e. the chart incorrectly indicates that the process is out of control. It is pointed out that when non-Poisson behavior is encountered in defect data, it is necessary to determine whether this is due to an out-of-control manufacturing process or a manufacturing or data collection procedure that yields clustered defect counts. A procedure that includes the outlier removal method to discriminate between clustered defect data and a process that is out of control is proposed, and its application to two sets of real-world data is shown. One is an example from IC manufacturing where true clustering exists, and the other is a manufacturing example where persistent out-of-control conditions give the appearance of clustering. Simulation results indicate that the procedure works well within reasonable boundaries. A method for constructing defect control charts for processes that yield clustered defects is also presented