Advanced Edge Detection for Noisy SEM Images: A Hybrid Approach for Enhanced Precision

The precision of lithographic processes in microelectronics production is critical, particularly as transistor dimensions continue to shrink. Ensuring the accuracy of these processes relies on robust quality control mechanisms, including effective edge detection in scanning electron microscopy (SEM) images. Traditional image processing techniques often struggle in noisy or low-resolution conditions, and conventional deep learning models can be computationally intensive. This paper introduces a novel hybrid model that combines Holistically-Nested Edge Detection (HED) [9], Lightweight Dense CNN (LDC) [10] and Pixel Difference Network (PiDiNet) [8] to enhance edge detection in challenging SEM imaging scenarios. The proposed model integrates noise suppression and pixel gradient sensitivity methods to achieve precise edge delineation while maintaining computational efficiency. This approach is particularly beneficial for industrial applications such as microcontroller manufacturing, where rapid and accurate SEM image analysis is essential. The source code is released at httos://github.com/seouhhopar/semimage