Image Analysis Methods for Solder-Ball Inspection in Integrated Circuit Manufacturing

Abstract

First-level packaging is an important manufacturing step in the production of integrated circuits. High I/O count devices require more sophisticated techniques than standard packaging methods. A complex packaging technology known as controlled collapse chip connection is revisited. The die-ceramic attachment consists of solder balls lying on the chip surface. The number of solder balls (more than 100 in current products) and high-precision requirements make visual inspection a difficult task for humans. The problem of analyzing images of solder balls automatically is addressed. These images are obtained by using a sensing device that projects shadows of the solder balls onto the integrated circuit surface. Several avenues are explored for segmenting these shadows from the complex and irregular appearance of the surrounding circuitry, with special emphasis given to a feature-based automatic classification method. This segmentation technique, based on decision theoretic classification and feature extraction, has performed with satisfactory results on available data. Experimental results are shown for a variety of chips having complex circuitry. An architecture for fast implementation of the algorithm is also presented. Additionally, some considerations are included regarding subpixel accuracy for computing values of relevant parameters of the solder-ball shapes. © 1988 IEEE