Accurate determination of the 3D atomic structure of amorphous materials
Article metadata
Article Date: 28 January 2026
Article URL: https://www.nature.com/articles/s41586-025-09857-4
Article Title: Accurate determination of the 3D atomic structure of amorphous materials
Article Image: (none provided)
Summary
This Nature paper presents a quantitative, end-to-end atomic electron tomography (AET) workflow that enables reliable determination of three-dimensional atomic coordinates and elemental identity in non-crystalline (amorphous) materials. The authors combine robust preprocessing (denoising, projection alignment and normalisation), advanced tomographic reconstruction (RESIRE), atom tracing, elemental classification and atomic position refinement. They validate the workflow using multislice-simulated datasets for amorphous Si, SiGeSn and CoPdPt nanoparticles across realistic noise/dose conditions, and compare performance to an alternative published method. The results give clear practical benchmarks — including atomic identification rates and 3D positional precisions — and the data and code are publicly available on GitHub and archived on Zenodo.
Key Points
- The authors develop a complete AET pipeline (preprocessing, denoising, alignment, RESIRE reconstruction, atom tracing, classification and refinement) tailored for amorphous materials.
- Benchmarks use multislice simulations for amorphous Si, SiGeSn and CoPdPt nanoparticles under realistic dose and noise; the workflow outperforms a recent alternative method.
- For a CoPdPt test particle the workflow identifies ~95.1% of Co, 99.0% of Pd and 100% of Pt atoms with 3D positional precisions of ~29 pm (Co), 12 pm (Pd) and 6 pm (Pt) under realistic dose conditions (abstract figures).
- Ptychographic AET (pAET) and multislice reconstructions achieve high atom-identification rates and sub-10 pm precision for some elements in larger simulated silica particles (examples show multislice pAET giving ~5–8 pm precision for Si/O in a 7,704-atom silica nanoparticle).
- The RESIRE reconstruction with angular refinement is shown to be superior to an updated SIRT implementation under the same noisy-projection conditions.
- All supporting data and source code are publicly available (GitHub: github.com/AET-pAET/Supplementary-Data-Codes; Zenodo DOI: 10.5281/zenodo.17445110).
- The paper provides practical guidelines and quantitative benchmarks for performing high-fidelity AET of amorphous materials; the framework can be applied to other tomographic imaging modalities.
Content summary
The study addresses a long-standing problem: determining full 3D atomic structure in materials that lack crystalline periodicity. The authors build a robust pipeline that corrects common experimental artefacts (noise, misalignment, intensity mis-normalisation and angular errors) and uses RESIRE for tomographic reconstruction. They demonstrate the pipeline on simulated datasets that model realistic imaging conditions, including multislice and ptychographic image formation. Quantitative comparisons show improved atom-identification accuracy and better positional precision versus an alternative reconstruction approach. Detailed extended-data figures document identification rates across doses and particle sizes, while tables list simulation and reconstruction parameters.
Importantly, the paper supplies reproducible resources: all preprocessing, reconstruction and analysis codes are on GitHub and archived on Zenodo, enabling other groups to reproduce the benchmarks and adapt the workflow to experimental datasets.
Context and relevance
Why this matters: amorphous solids underpin many technologies — from thin-film electronics and solar cells to phase-change memory, biomedical metallic glasses and quantum devices. Knowing exact 3D atomic positions and chemistry in these non-crystalline systems unlocks direct links between structure and functional properties, helps validate theoretical models of short- and medium-range order, and informs materials design.
This work advances the experimental toolkit: where traditional diffraction-based methods only provide statistical or radial-average information, AET can now be pushed to give element-resolved, sub-Ångström 3D maps in amorphous nanoparticles under realistic imaging conditions. That will matter to researchers working on disordered alloys, glasses, nanocatalysts and device thin films, and to anyone needing ground-truth data for simulations.
Author style
Punchy: the team lays out a practical, reproducible recipe rather than a theoretical exercise. If you work with non-crystalline materials or use electron tomography, the details and benchmarks here are directly actionable — they show what works, when, and by how much.
Why should I read this?
Short version: if you care about actually seeing where atoms sit in glasses, alloys or amorphous nanoparticles — not just averaged stats — this paper is for you. The authors have done the heavy lifting (simulations, comparisons, error analysis) and put the code and data online, so you can copy the workflow or compare your experimental reconstructions to these benchmarks. In other words: we’ve saved you hours of testing and given clear numbers on accuracy and limits.