A comparison of data mining methods for yield modeling, chamber matching and virtual metrology applications

Abstract

Statistical modeling methods have become a key tool in yield analysis and chamber matching. As the transition to 45nm and below increases it is becoming difficult to maintain yield and avoid excursions. Generalized and accurate models of process behavior to predict yield can quickly give insight into the cause of yield loss and process excursion. Here we simulate linear and nonlinear models of yield from process data and evaluate the performance of methods like partial least squares, support vector regression, and rules ensemble in predicting these yield models.