Design of intelligent data sampling methodology based on data mining

Data mining is the automated discovery of nontrivial, previously unknown, and potentially useful knowledge embedded in a database. With the increase of automated data generation and gathering in semiconductor manufacturing, mining interesting information from huge databases becomes of utmost concern. In this paper, we present a new and better application of data mining techniques by designing an intelligent in-line measurement sampling method for process parameter monitoring in a wafer fab. The sampling method specifies the chip locations within the wafer to be measured, and the number of measured chip locations per wafer in order to represent a good sensitivity of 100% wafer coverage and defect detection. To more effectively detect all the abnormalities of process parameters, we extract the spatial defect features in the historical wafer bin map data and then cluster the chip locations having similar defect features through SOM neural network. To more efficiently design the sampling method, we merge the homogeneous clusters through a statistical homogeneity test and then select the chip location having the best detection power of each of the existing bins through interactive explorative data analysis of SOM weight vectors. We illustrate the effectiveness of the proposed sampling method using actual fab data, and the results indicate that if the sampled chip locations are chosen rationally by optimal data mining techniques, that sampling can provide accurate detection of all defects.