CD-SEM Image Defect Detection and Classification Using Transformers

In the integrated circuits manufacturing process, a critical dimension scanning electron microscope machine provides high-resolution images of geometries on semiconductor wafers and measurements of the printed polygons after the photolithography process. These images and measurements are crucial for calibrating lithography process models such as the etch model, optical proximity correction model, and resist model, which are necessary for simulating the photolithography process and hence reducing the probability of defects occurrence. However, The test chips used for model calibration often contain imperfections or anomalies that require accurate detection and classification to be removed from the model calibration process. Traditionally, engineers manually filter semiconductor wafer images and their measurements, which is a time-consuming and error-prone task. Therefore, there have been attempts to use machine learning to automate image filtering. In this research, we introduce a novel approach using an image preprocessing stage followed by a transformer-based object detection model to automate detecting and classifying defects in semiconductor wafer images containing an array of lines or a matrix of contacts. The proposed approach shows improved accuracy compared to traditional methods.