Generalized linear models for defectivity related regression modeling

In a plasma etcher, significant defect excursion can be caused by polymer flaking from chamber wall. It is well established that insitu chamber cleaning techniques for maintaining the polymer on the chamber wall at a minimum thickness are critical. The cleaning techniques have proven to improve quality, yield and cycle time. Process chamber dry cleaning using fluorine based chemistry after each processed wafer is the commonly used method. However, if for each wafer a clean and recovery steps are needed, the overall process time will increase significantly. A high efficiency insitu plasma dry cleaning has been developed for a metal etcher. RF power, pressure and gas flows have been optimized to prevent polymer flaking from chamber wall without significantly affecting the process cycle time. The dry clean is only performed after each lot (25 wafers) and produced results that are comparable to cleans performed after each wafer. For defectivity trend modeling, we used Generalized Linear Models (GLM), which extend classical linear regression models and often provide better-fitting with counts data. The GLM is compared to least square (LS) linear regression modeling with and without data transformation. The fitted model plots and 95% confidence intervals are compared to each other to assess the accuracy of each model. The GLM model produced a much better results with no negative estimate and shorter 95% confidence interval.