Detecting dimple defects of polished semiconductor wafer surfaces

Detecting defects on polished wafers has been one of the interesting problems to be solved in the semiconductor industry for years. We propose building an automated visual inspection system to detect dimples of polished wafers. Dimples are common defects on polished wafers. They appear as circular-shaped bright spots which usually vary in sizes and locations. They may also have irregular boundaries and co-exist with other image features. Identifying dimples has posed a challenging task from both the aspects of practical application and theoretical interest. In this study, we developed an automated visual inspection system to detect dimple defects using the concept of fuzzy mathematics. The algorithm consists of two major processing phases. At the first phase, preprocessing is performed to eliminate noise and to reduce the number of potential candidates of dimple defects. At the second phase, four pattern features are defined based on the consideration of scale-, position-, and orientation-invariance. A fuzzy membership function is utilized to cope with the wide range of shape variations of the dimple defects. A decision-making mechanism is based on the value of the membership function which describes a pattern's closeness to a dimple. The attractive features of the proposed system include that (1) the algorithm is scale-, position-, and orientation-invariant and (2) a multi-dimensional membership function improves identification process by allowing pattern variations.