Atom identification in bilayer moire materials with Gomb-Net

2025-02-13

AI Summary

In bilayer moiré materials, the superimposed moiré pattern severely impedes layer-resolved atomic-scale imaging in scanning transmission electron microscopy (STEM), hindering accurate atomic positioning, elemental identification, and mapping of strain/doping distributions. To address this, we propose a novel deep learning framework—Gomb-Net—a custom-designed convolutional neural network integrating multi-scale feature extraction with layer-aware attention mechanisms. This is the first method to achieve decoupled atomic localization and element-specific identification for individual layers beneath strong moiré interference. Applied to WS₂–WS₂(1−x)Se₂ₓ heterostructures, it enables quantitative, layer-resolved identification of Se substitutional sites, yielding high-fidelity, layer-specific strain and doping maps. Remarkably, it reveals that dopant atom positions are insensitive to local moiré potential modulation—a previously unobserved physical phenomenon. Our approach overcomes the fundamental limitation of conventional image analysis in strongly coupled moiré systems and establishes a generalizable tool for atomic-scale characterization of such materials.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Computer Vision: Multi-modal Vision
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Web Mining and Content Analysis: Machine learning and data science for the Web

Abstract

Moire patterns in van der Waals bilayer materials complicate the analysis of atomic-resolution images, hindering the atomic-scale insight typically attainable with scanning transmission electron microscopy. Here, we report a method to detect the positions and identity of atoms in each of the individual layers that compose bilayer heterostructures. We developed a deep learning model, Gomb-Net, which can distinguish atomic species in each individual layer, effectively deconvoluting the moire pattern to enable layer-specific mapping of strain and dopant distributions, unlike other methods which struggle with moire-induced complexity. Using this approach, we explored Se atom substitutional sites in a twisted fractional Janus WS2-WS2(1-x)Se2x heterostructure and found that layer specific implantation sites are unaffected by the moire pattern's local energetic or electronic modulation. This advancement enables atom-identification within material regimes where it was not possible before, opening new insights into previously inaccessible material physics.

Problem

• Identify atoms in bilayer moire materials
• Deconvolute moire patterns for layer-specific mapping
• Enable atom-identification in complex material regimes

Innovation

• Deep learning for atom identification
• Layer-specific strain mapping
• Moire pattern deconvolution