The classification methodology of chip quality using canonical correlation analysis-based variable selection on chip level data

The semiconductor manufacturing industry produce lots of information about performance of chips. Among them, process control monitoring (PCM) data that are measured at test element group before probe test are multiple-dimensional information. PCM data are including the device characteristics such as a resistance, capacitance, current, and so on. Fail bit count (FBC) that is the number of defective cells failed by function items of probe test is also multi-dimensional information and gives a direct impact on the yield loss at the probe test step. In this study, we proposed classification methodology using a canonical correlation analysis as variable selection method on chip level data. Through this proposed method, we were able to extract important 22 variables from 77 PCM variables by using the correlation between the multiple FBC variables and PCM variables. As a result, the accuracy of quality classification for a chip is dramatically improved on the probe test.