Decoding disorder: exploring invisible materials with AI

Amorphous materials – substances whose structure appears disordered at the microscopic level – are ubiquitous across a range of technological sectors, including organic solar cells, photonic materials and certain pharmaceutical formulations. However, their behaviour remains difficult to predict. Unlike crystals, which have a regular atomic arrangement, amorphous materials have extremely complex structures that are difficult to reproduce using traditional simulation methods. Lena Simine from the Department of Chemistry at McGill University is seeking to better understand this “organized disorder” with the help of artificial intelligence. Her aim is to link the internal structure of materials to their physical properties, to eventually design new materials with targeted performance characteristics.

Rather than using AI simply as a tool for analysing big data, Simine’s team is using it to improve existing physical simulations. The researchers are thus developing algorithms capable of speeding up calculations and reproducing amorphous structures at scales previously unattainable. In particular, their method makes it possible to simulate materials at mesoscopic scales of several tens of nanometres, bringing numerical models considerably closer to real materials. The team has also designed an algorithm inspired by data compression techniques to predict interactions between small molecules and DNA sequences with computational efficiency far greater than that of conventional approaches. Finally, the researchers have adapted generative models to more effectively explore the vast chemical space of possible molecules.

This work paves the way for a much more predictable design of amorphous materials. In the long term, the tools developed could enable the design of materials with precisely defined optical, electronic or biological properties before they are ever synthesized in the laboratory. The applications are vast: improving photovoltaic devices, developing DNA-based biosensors, discovering new therapeutic molecules, or designing advanced photonic materials. For Lena Simine, it is not a question of simply optimizing existing materials: it is about exploring a vast chemical landscape that is still largely inaccessible and opening the door to a new generation of materials designed using AI.

Références

Liu, Y., Madanchi, A., Anker, A. S., Simine, L., et Deringer, V. L. (2025). The amorphous state as a frontier in computational materials design. Nature Reviews Materials, 10(3), 228-241. https://doi.org/10.1038/s41578-024-00754-2

Madanchi, A., Azek, E., Zongo, K., Béland, L. K., Mousseau, N., et Simine, L. (2025). Is the future of materials amorphous? Challenges and opportunities in simulations of amorphous materials. ACS Physical Chemistry Au, 5(1), 3-16. https://doi.org/10.1021/acsphyschemau.4c00063

Gastellu, N., et Simine, L. (2025). Disentangling morphology and conductance in amorphous graphene. The Journal of Physical Chemistry Letters, 16(18), 4522-4528. https://doi.org/10.1021/acs.jpclett.5c00458