Automatic Detection of Via Arrays in AFM Images for CMP Dishing Evaluation

We present an efficient algorithm for the detection of array-like patterns of circular vias in AFM images. It combines a new variant of the Hough transformation with an efficient implementation of brute-force search to ensure good results and high performance, compared to a simple intuitive approach. A set of manually labeled benchmark data is used to systematically evaluate the accuracy and improve the algorithm. We apply the new via detection method to measure dishing in AFM images of copper vias after CMP.