Generative Image Restoration and Super-Resolution using Physics-Informed Synthetic Data for Scanning Tunneling Microscopy

2025-10-29

AI Summary

Scanning tunneling microscopy (STM) suffers from degraded image quality, low acquisition efficiency, and frequent tip repositioning due to probe tip degradation and serial sampling. Method: This paper proposes a physics-informed generative image restoration and super-resolution framework. Leveraging only 36 real experimental STM images, we construct a physically constrained synthetic data pipeline and jointly train flow-matching and diffusion models, optimized using CLIP-MMD and structural similarity metrics. Contribution/Results: The method enables high-fidelity reconstruction from sparse sampling—preserving atomic-scale details while achieving 2–4× imaging acceleration. It significantly reduces tip regulation frequency and per-frame acquisition time, thereby enhancing STM experimental throughput and high-speed imaging capability. Its core innovation lies in an end-to-end image restoration paradigm that synergistically integrates physical modeling with generative learning under limited-sample conditions.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Scanning tunnelling microscopy (STM) enables atomic-resolution imaging and atom manipulation, but its utility is often limited by tip degradation and slow serial data acquisition. Fabrication adds another layer of complexity since the tip is often subjected to large voltages, which may alter the shape of its apex, requiring it to be conditioned. Here, we propose a machine learning (ML) approach for image repair and super-resolution to alleviate both challenges. Using a dataset of only 36 pristine experimental images of Si(001):H, we demonstrate that a physics-informed synthetic data generation pipeline can be used to train several state-of-the-art flow-matching and diffusion models. Quantitative evaluation with metrics such as the CLIP Maximum Mean Discrepancy (CMMD) score and structural similarity demonstrates that our models are able to effectively restore images and offer a two- to fourfold reduction in image acquisition time by accurately reconstructing images from sparsely sampled data. Our framework has the potential to significantly increase STM experimental throughput by offering a route to reducing the frequency of tip-conditioning procedures and to enhancing frame rates in existing high-speed STM systems.

Problem

• Repair degraded STM images from tip degradation
• Achieve super-resolution with reduced image acquisition time
• Reduce tip-conditioning frequency and enhance STM throughput

Innovation

• Physics-informed synthetic data trains generative models
• Flow-matching and diffusion models restore STM images
• Sparse data reconstruction reduces acquisition time significantly