CLASSIFYING DEFECTS IN TOPOGRAPHY IMAGES OF SILICON WAFERS

In this work, we demonstrate that automatically classifying defects in topography images of silicon wafers is feasible. We process topography images of a set of sample wafers with controlled induced defects in their wafer back surfaces. We group these induced defects into three classes: cavities, cracks, and star cracks. With this sample set, we train and test selected classifiers with suitable feature vectors extracted from their wafer back surface topography images. A comparison reveals, that training and testing linear and quadratic classifiers with two Fisher scores as features, yield the best classification performances. We correctly classify all cavities and can separate them from the critical cracks and star cracks, which show a sufficient signal in the topography images.