Lithographic Hotspot Detection Using CNN

Reducing the feature size of semiconductor technology allows us to incorporate more functionalities in a single chip, making the fabrication more complex. The lithography process fabricates these tiny structures, creating desired structures with the help of masks to transfer the required patterns to the substrate. Due to the optical proximity of the commonly used fabrication method, the chances of occurrence of undesired patterns are high. These undesired patterns are called lithographic hotspots (HSs). Identifying these patterns before fabrication is crucial for fast semiconductor development. In this direction, an accurate HS detection method is proposed using the convolutional neural networks VGG16 and ResNet50. The proposed method shows improvements in performance measures compared to the existing techniques.