Comparison of three different methods to model the semiconductor manufacturing yield

During a semiconductor manufacturing process a large amount of data is stored in databases. These data can be used to model the semiconductor manufacturing yield. A model of the yield according to process measurements is useful to predict the yield before final tests. It is also an help to do sensitivity analysis of the yield to process variations. This paper compares three methods to model the manufacturing yield from test data. Principal components analysis, independent component analysis and partial least squares regression are reviewed. A methodology is then exposed to achieve, efficient manufacturing yield modeling