SRFS: Super Resolution in Fast-Slow Scanning Mode of SEM Based on Denoising Diffusion Probability Model

• Published in: IEEE transactions on instrumentation and measurement 2024, Vol.73, p.1-9
• Main Authors: Wang, Zelin, Feng, Rui, Mao, Jikai, Tao, Xingyu, Zhang, Deao, Shan, Xuliang, Zhou, Jianguang
• Format: Article
• Language: English
• Subjects: Denoising diffusion probabilistic model; Diffusion rate; fast-slow scanning mode; Gaussian distribution; Image quality; Image resolution; Microscopy; Noise reduction; Scanning electron microscopy; scanning electron microscopy (SEM); super resolution (SR); Superresolution; Task analysis; two-stage network
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2024.3351247

Scanning electron microscopy (SEM) is the most commonly used and important tool for material characterization. It provides two scanning mode-fast scanning mode obtain images rapidly, but these images are low resolution (LR) and lack details and slow scanning mode obtain images with high resolution (HR) and rich details, but it is time-consuming. This limits the use of SEM, in order to solve the above trade-off problem, we propose SRFS, a diffusion-based super resolution (SR) network in fast-slow Scanning mode of SEM. To the best of our knowledge, this is the first time of image SR in fast-slow scanning mode of SEM. Comparing to previous SR methods, SRFS can generate SR images with learned perceptual image patch similarity (LPIPS) of 0.1414 and Fréchet inception distance (FID) of 0.59 which provide the best subjective vision quality and the most accurate image details. SRFS allows us to obtain HR SEM images with rich and accurate details fast and continuously. By using SRFS, we can achieve accurate process monitoring under SEM.