Automated feature detection and characterization in sort wafer maps

This paper describes the technique used by a wafer map analysis system that the author has developed to characterize spatial features on gray scale cumulative wafer maps. The technique combines image processing with neural networks: the former enhances features and extracts certain attributes from the map, and the latter determines the categories, extents, and locations of features based on these extracted attributes. The paper also presents the results of running this wafer map analysis system on data derived from real Intel Fabs. Prototypes of this system have been operational in several advanced Intel Fabs for more than two years and have enjoyed considerable success.