This paper addresses the task of automating the visual inspection of contam
ination on the surface of integrated circuits (IC) wafers arising from the
dicing process. Using a set of multi-spectral optical filters and a charged
coupled device (CCD) video camera, several images are acquired from each I
C wafer under different illumination conditions, from which feature space d
ata an then generated. Three conventional classification methods - an artif
icial neural network (ANN) using a back-propagation (BP) technique, a minim
um distance algorithm, and a maximum likelihood classifier are evaluated, a
nd their performances are compared. In addition, important elements of the
feature space, i.e., the optimal illumination condition and appropriate opt
ical spectrum are investigated. The results show that the image-acquisition
technique developed is effective in discriminating feature elements, and t
hat the employed ANN-BP classifier can accurately achieve the required bina
ry (clean/containinnted IC wafers) decisions. (C) 2001 Pattern Recognition
Society. Published by Elsevier Science Ltd. AII rights reserved.