An Evaluation of an Automated Detection Algorithm to Count Defects Present in X-Ray Topographical Images of SiC Wafers

Full semiconductor wafer defect/dislocation characterization is difficult to implement manually. We present an analysis of an automated algorithm used to extract Threading Screw Dislocation defect data from Synchrotron White Beam X-Ray Topographical images of SiC wafers. This extraction involves a two-fold process; firstly the algorithm highlights the appropriate defect and secondly updates the counter to provide a final result of defect count. The result of the automated algorithm is compared to hand counts in all cases, thus allowing a critical analysis of the technique. Improvements to this algorithm have been made since last reported by the same authors [1], which are discussed. The analysis herein was also performed on a much larger sample of SiC wafer images than previously used by the same authors [1] allowing a better judgment of performance and critical evaluation. The algorithm is also compared with the original previous algorithm that was used [1]. The success of this methodology paves the way for a complete analysis of whole SiC wafers, which previously was extremely difficult due to image analysis inaccuracy or the bottleneck presented by manual counting.