SCIENTIFIC HIGHLIGHT 9
9 December 2025
How AI Helps Transform (S)TEM Data into Digital Twins
A new workflow led by IMPRESS scientists from ICN2 and CNR automates (S)TEM analysis and builds detailed 3D models that reveal how atomic-scale variations shape the behaviour of advanced materials.
When Every Atom Matters
Inside every smartphone, billions of tiny transistors switch currents on and off. Each one is made from layers only a few nanometres thick. At this scale, even a slight shift in the position of atoms - an interface a bit rougher than expected, a few atoms of strain, a subtle irregularity in composition - can change how quickly a transistor switches or how much energy it wastes. When components are this small, every atom matters. And as devices continue to shrink, the ability to understand these nanoscale variations becomes more important than ever.
From Microscopy Images to Digital Twins
Now imagine creating a digital twin of that transistor layer, an accurate 3D model built from transmission electron microscopy images, showing not only the designed structure but also every tiny deviation introduced during fabrication or operation. A model like this would allow scientists to explore how a slight distortion at an interface or a subtle shift in composition affects the behaviour of the device. But until recently, extracting such detailed information from real (S)TEM data required days of meticulous work by experts, making it difficult to study enough samples to draw reliable conclusions.
A New AI-Assisted Workflow
In a study published in Advanced Materials in October 2025, a team led by IMPRESS scientists from the Catalan Institute of Nanoscience and Nanotechnology (ICN2) and the Italian National Research Council (CNR) presents a workflow that tackles this challenge. Their paper, “Artificial Intelligence-Assisted Workflow for Transmission Electron Microscopy: From Data Analysis Automation to Materials Knowledge Unveiling,” introduces a way to transform transmission electron microscopy (TEM) and scanning transmission electron microscopy (STEM) data into detailed, physically meaningful digital twins of real materials in a fraction of the time previously required.
Automating a Manual, Time-Consuming Process
TEM is one of the few techniques capable of revealing atomic-scale structure in these ultrathin, nanoscale material layers, but interpreting its images has long required slow, meticulous manual analysis. Crystal orientations must be identified, strain fields measured, interfaces mapped, and structural models reconstructed - tasks that can take experts days for a single dataset. The workflow developed by the team automates this entire chain. With a combination of advanced image segmentation, diffraction analysis, crystallography, strain mapping, and physics-guided artificial intelligence, it extracts the key structural information encoded in (S)TEM images and uses it to build three-dimensional digital twins in a matter of minutes.
Linking Structure to Function Through Digital Twins
These digital twins - whether finite-element representations or atomistic models containing millions of atoms - bridge the gap between what microscopes can see and how materials actually behave. They allow scientists to test how subtle variations in structure influence properties such as mechanical response, heat transport, electronic or quantum behaviour. Demonstrated across a range of material systems and geometries, the workflow shows how a single, coherent approach can support many fields where atomic-scale detail shapes performance.
Why This Matters for IMPRESS
The work resonates strongly with the ambitions of the IMPRESS project, which aims to advance interoperable, AI-ready tools for transmission electron microscopy. By automating data analysis and providing a framework for building meaningful digital models directly from microscopy data, the workflow represents the kind of methodological innovation that IMPRESS seeks to make accessible to diverse scientific communities. It shows how combining microscopy, modelling, and artificial intelligence can help transform raw data into deeper understanding, bringing scientists closer to truly comprehensive insight into materials at the atomic scale.
Publication Details
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): e06785. https://doi.org/10.1002/adma.202506785