소재 AI 공유함
Menu
← AI in materials

Materials AI · written by AI

Physics-Informed Neural Network for Low-Dose Atomic Electron Tomography Applied to Reconstruction of 3D Atomic Structure and Chemical Species

AI editorPublished 2026.10.23

An npj Computational Materials paper reported that, in low-dose atomic electron tomography (AET), a physics-constrained two-stage neural network (PANN) and GLARE/DAST family models were used to reconstruct 3D atomic positions and chemical species. It suggested simulation-based applicability even for sensitive inorganic crystal systems, but reproducibility under actual experimental conditions depends on simulation and data quality. [1]

1/ AI utilization method: physics-constrained two-stage neural network (PANN) and GLARE/DAST family models
2/ Application task: reconstruction of 3D atomic positions and chemical species in low-dose atomic electron tomography
3/ Reported result: achieved robust atomic structure determination even under low-dose conditions and demonstrated applicability to sensitive inorganic crystal systems
4/ A paper published in npj Computational Materials on 2026-10-23 presented a method for reconstructing 3D atomic positions and chemical species using a physics-constrained two-stage neural network (PANN) and GLARE/DAST family models in low-dose atomic electron tomography.
5/ It reported robust atomic structure determination even under low-dose conditions, and showed applicability to sensitive inorganic crystal systems such as halide perovskites, zeolites, and quantum dots through multi-slice simulations.
6/ This work is an example of materials AI use in structural identification problems that require reduced electron-beam damage, aiming to improve reconstruction accuracy and stability.
7/ In materials terms, its significance is that it can broaden the starting point for structure-property relationship analysis by jointly estimating 3D atomic-level structure and composition in damage-prone samples.
8/ However, the performance presented in the paper is based on low-dose AET and simulation conditions, so the same level of reconstruction quality is not guaranteed when transferred to different instruments, samples, or noise distributions. [1]

How to apply this

Test a neural network pipeline that reconstructs both atomic positions and chemical species from low-dose atomic electron tomography data.

  1. 1. Collect low-dose AET tilt series and matching reference structures, and split them into training, validation, and test sets.
  2. 2. Configure a reconstruction model with physical constraints and train it to predict atomic positions and chemical species simultaneously.
  3. 3. Compare reconstruction results on both simulation and experimental data to check damage suppression and reconstruction accuracy together.
  4. 4. Recheck generalization performance with external data in which noise, dose, and sample type are changed.
Data needed
Low-dose AET tilt series and corresponding reference structures or simulation data, elemental composition information for the sample, and, if available, high-quality reference reconstructions.
Checking the result
On independent hold-out data, check atomic position error, chemical species classification accuracy, reconstruction consistency, and performance variation across different noise levels.
What can go wrong
If the dose, noise, or sample type differs between the training data and the experimental data, the reconstruction results may appear overly optimistic or may produce incorrect atomic placements.

이 가이드는 원문의 방법을 바탕으로 AI가 제안한 적용 방향입니다. 원문에 적힌 내용이 아니며 같은 결과가 재현된다는 보장도 아닙니다.

Related concepts and practice

이 이슈를 읽는 데 필요한 개념입니다. 용어마다 Colab에서 바로 돌려볼 수 있는 실습 노트북이 있습니다.

Sources

  1. Physics-aware neural networks enable robust and full atomic structure determination via low-dose atomic electron tomography | npj Computational Materials

웹 자료를 바탕으로 AI가 자동 요약한 글입니다. 연구의 세부 조건과 결과는 연결된 원문을 확인해주세요.