Fast and accurate design based binning based on hierarchical clustering with invariant feature vectors for BEOL

As design rules continue to shrink, systematic defects have become a serious problem. It becomes very important to review systematic defects effectively by a defect review SEM (scanning electron microscope) and to modify the design and the process to keep or improve an yield. In this paper, we propose a fast and accurate design based binning (DBB) method that is based on the hierarchical clustering with invariant feature vectors for back end of line (BEOL). In order to improve classification accuracy, we employ the hierarchical clustering method. Shift-, rotation-, and flip-invariant feature vectors are extracted from layout data. We propose two variations of DBB methods: one-step method and two-step method. The one-step method employs solely the hierarchical clustering. It can improve classification accuracy. However, computational time of the hierarchical clustering is high so that it is not practical to classify many defects by this method. In order to achieve both high accuracy and fast computation, we also propose two-step method that employs the hierarchical clustering after classifying defects by the DBB software used in the production line. We apply the proposed two methods to volume production data. The results show that the proposed two-step method can significantly improve accuracy against the production line DBB software, despite of slight decrease in purity and slight increase in computation time.