An international collaboration involving Cnr Nano has developed an innovative, AI-based approach capable of analyzing electron microscopy images of a material, reconstructing its three-dimensional atomic structure, and calculating its physical and electronic properties—all automatically and within minutes. Tasks that previously required days of specialized work can now be completed rapidly and with high reliability, accelerating research on complex materials. The results are reported in Advanced Materials.
Transmission electron microscopy (TEM) is among the most advanced techniques for characterizing materials, providing atomic-scale information essential for understanding and designing electronic and quantum devices. However, analyzing TEM images typically demands long processing times and specialized expertise.
To overcome these limitations, researchers at the Institut Català de Nanociència i Nanotecnologia (ICN2, Barcelona) and Cnr Nano in Modena have developed a workflow that automates the entire process—from image analysis to 3D reconstruction and numerical simulations of the material. The international collaboration also include researchers from the University of Modena and Reggio Emilia and the CNR-IMM.
“Thanks to a physics-guided artificial intelligence model, the experimental images produced by our electron microscopes are transformed into digital twins of the materials—realistic, three-dimensional ‘virtual’ models faithfully reproducing atomic structures made of millions of atoms,” explains Enzo Rotunno from Cnr Nano. “On these models, we can simulate the electronic, quantum, and mechanical properties of the material as if we were studying the physical sample itself.”
The novelty of the approach is twofold. First, it provides an end-to-end workflow: from the acquired image, to automated analysis, to 3D model construction, and finally to the simulation of physical and electronic properties. Second, it is based on a physics-guided AI model that combines machine learning with principles from materials physics. “It is a unique strategy that unifies microscopy, modeling, and AI in an integrated ecosystem, enabling a direct link between the structure observed under the microscope and the material’s actual behavior,” says Rotunno.
Starting from electron microscopy images, the system automatically identifies crystallographic orientations, elemental composition, and the distribution of strain within the sample. These data are then used to reconstruct a coherent and physically accurate 3D digital model representing the real atomic structure of the material. Numerical simulations of physical and electronic properties can then be performed on this model, allowing researchers to directly correlate the observed structure with the material’s behavior.
Cnr Nano’s contribution—particularly from the TEM microscopy group in Modena led by Vincenzo Grillo—focused on developing and providing the computational and simulation software used to reconstruct the atomic models. “These tools stem from a long tradition of research in electron microscopy and computational modeling, which enabled fully automated data interpretation and the integration of physical simulations within the analysis workflow,” Rotunno explains.
“The key result is not only faster analysis but a fully streamlined workflow that can be applied consistently across many samples,” explains Vincenzo Grillo. “This high-throughput capability is essential for industrial-scale materials screening and for efficiently validating new technological solutions.”
The method has potential applications in both materials research and industry, from next-generation micro- and quantum-device design to the development of advanced materials for energy technologies. The ability to rapidly obtain an accurate digital twin of a material opens new opportunities to accelerate the discovery and optimization of functional materials with tailored properties. “Electron microscopy is like a window into the invisible universe: every image can reveal a secret of matter—and with the help of artificial intelligence, we can uncover these secrets more quickly and in entirely new ways,” Rotunno concludes.
Reference article:
Marc Botifoll, Ivan Pinto-Huguet, Enzo Rotunno, Thomas Galvani, Catalina Coll, Payam Habibzadeh Kavkani, Maria Chiara Spadaro, Yann-Michel Niquet, Martin Børstad Eriksen, Sara Martí-Sánchez, Georgios Katsaros, Giordano Scappucci, Peter Krogstrup, Giovanni Isella, Andreu Cabot, Gonzalo Merino, Pablo Ordejón, Stephan Roche, Vincenzo Grillo, Jordi Arbiol, Artificial Intelligence-Assisted Workflow for Transmission Electron Microscopy: From Data Analysis Automation to Materials Knowledge Unveiling. Advanced Materials. (2025). https://doi.org/10.1002/adma.202506785.
