Advanced pattern contour extraction function for see-through BSE images of high voltage SEM

As semiconductor device structures become smaller and more complex, the need to measure not only single-layer patterns but also overlay across multiple layers has become increasingly critical. Traditionally, overlay has been managed by measurement on dedicated patterns. However, due to the growing complexity of manufacturing processes, it is now essential to measure overlay directly on actual device patterns. To address this challenge, TASMIT has developed a high-voltage scanning electron microscope (HV-SEM) system. Achieving accurate contour extraction of lower layer patterns from low contrast and low sharpness see-through BSE images, advanced processing technique is required. To this end, we have developed an advanced contour extraction method that integrates Active Contour Model with deep learning. In this study, we propose a method to generate a synthetic training dataset by leveraging the layout data of the sample, which captures the pattern variations and fluctuations observed in actual sample images. Training the contour extraction model with this synthetic dataset, we successfully obtained accurate overlay measurement results for a sample wafer containing multiple stacked layers in the back-end-of-line (BEOL) pattern, provided by imec. This paper presents the details of our method, evaluation results, and the unique advantages of SEM based die-to-database (D2DB) see-through overlay metrology.