DONUT: Physics-aware Machine Learning for Real-time X-ray Nanodiffraction Analysis

2025-07-18

AI Summary

In scanning X-ray nanodiffraction, severe artifacts arising from the convolution of divergent incident beams with local structural features, coupled with computational bottlenecks in real-time analysis, impede high-throughput nanoscale characterization. To address this, we propose an unsupervised, physics-informed deep learning framework that embeds a differentiable geometric diffraction model into a neural network, enabling end-to-end joint inversion of lattice strain and orientation—without labeled data or pretraining, and inherently incorporating optical diffraction priors. This work presents the first unsupervised, interpretable, and real-time end-to-end inversion method for X-ray nanodiffraction. Experimental results demonstrate a >200× speedup over conventional fitting-based approaches while fully preserving structural feature fidelity, thereby enabling high-throughput, nanoscale, in situ imaging.

Technology Category

• Search and Optimization: Distributed Search
• Machine Learning: Structured Learning
• Computer Vision: Diffusion Models for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Coherent X-ray scattering techniques are critical for investigating the fundamental structural properties of materials at the nanoscale. While advancements have made these experiments more accessible, real-time analysis remains a significant bottleneck, often hindered by artifacts and computational demands. In scanning X-ray nanodiffraction microscopy, which is widely used to spatially resolve structural heterogeneities, this challenge is compounded by the convolution of the divergent beam with the sample's local structure. To address this, we introduce DONUT (Diffraction with Optics for Nanobeam by Unsupervised Training), a physics-aware neural network designed for the rapid and automated analysis of nanobeam diffraction data. By incorporating a differentiable geometric diffraction model directly into its architecture, DONUT learns to predict crystal lattice strain and orientation in real-time. Crucially, this is achieved without reliance on labeled datasets or pre-training, overcoming a fundamental limitation for supervised machine learning in X-ray science. We demonstrate experimentally that DONUT accurately extracts all features within the data over 200 times more efficiently than conventional fitting methods.

Problem

• Real-time analysis bottleneck in X-ray nanodiffraction microscopy
• Artifacts and computational demands hinder coherent X-ray scattering
• Convolution of divergent beam with sample structure complicates analysis

Innovation

• Physics-aware neural network for nanobeam diffraction
• Differentiable geometric diffraction model integration
• Unsupervised learning without labeled datasets