01/12/2025
A new artificial intelligence–based system promises to revolutionize the work of scientists studying materials at the atomic scale. In just a few minutes, it can analyze images produced by an electron microscope, reconstruct the three-dimensional structure of the observed material, and even predict its physical behavior. The result comes from an international collaboration between researchers at ICN2 (Barcelona) and Cnr-Nano (Modena), and has been published in Advanced Materials.
Transmission electron microscopy (TEM) is one of the most powerful tools for observing materials at the scale of individual atoms. However, interpreting the images requires time, specialized expertise, and many manual steps. “Typically, days of image analysis are needed to obtain a realistic model of the material. With the new approach, the same work can be completed in just minutes,” explains Enzo Rotunno of Cnr-Nano, co-author of the study.
The system developed by the researchers operates as a fully automated workflow: it starts from the atomic-scale image acquired with the microscope, analyzes it, reconstructs the material’s three-dimensional structure, and finally uses it to simulate its electronic, mechanical, or quantum properties, all with minimal human intervention.
“At the heart of the system is a physics-guided artificial intelligence that combines machine learning with knowledge from materials physics. This ensures that the reconstructions are reliable and consistent with what actually occurs in the sample,” says Rotunno. “The images are transformed into digital twins of the materials—virtual three-dimensional models that faithfully reproduce the atomic structure of the observed material,” he continues. “These models can then be used to simulate the material’s behavior as if it were the real physical sample.”
Cnr-Nano reserachers, working within the TEM group led by Vincenzo Grillo, contributed by developing the computational software needed to reconstruct the atomic models, drawing on years of experience in electron microscopy and numerical simulation.
The potential applications are numerous—from chip and quantum device design to the development of innovative energy materials. “Having an accurate digital twin makes it possible to rapidly test new solutions and accelerate the discovery and optimization of materials with tailored properties. Moreover, this approach can be replicated on many samples in sequence, which is crucial for high-throughput industrial applications in materials development and selection,” Rotunno notes.
“Electron microscopy is a window into the invisible universe, and every image can reveal a secret about matter. Now, thanks to artificial intelligence, we can do this much faster and in an entirely new way,” concludes the researcher.
M. Botifoll, I. Pinto-Huguet, E. Rotunno, et al. “Artificial Intelligence-Assisted Workflow for Transmission Electron Microscopy: From Data Analysis Automation to Materials Knowledge Unveiling.” Adv. Mater. (2025): e06785. https://doi.org/10.1002/adma.202506785
Ufficio stampa:
Maddalena Scandola
CNR - Istituto Nanoscienze
comunicazione@nano.cnr.it
3470778836
Vedi anche: