Global electricity usage is rapidly increasing and needs to be addressed sustainably. Developing new materials will provide solar cell materials that are much more efficient than they are today. It is a very thin and flexible material that can wrap anything from a mobile phone or an entire building. Using computer simulation and machine learning, researchers at Sweden's Chalmers Institute of Technology have taken an important step towards understanding and handling halide perovskites, one of the most promising but notoriously enigmatic materials.
According to the International Energy Agency, electricity usage is constantly increasing around the world. The global total energy consumption is expected to exceed 50% in 25 years compared to the current 20%.
“To meet the demand, there is a significant increase in new, environmentally friendly energy conversion methods such as more efficient solar cells, such as more efficient solar cells. Our findings are essential to controlling one of the most promising solar cells for optimal use.
Promising materials for efficient solar cells
Materials located within the group called halide perovskites are considered the most promising for producing cost-effective, flexible and lightweight photoelectronic devices such as solar cells and LED bulbs. However, perovskite materials deteriorate quickly and to know how to utilize them, you need to have a deeper understanding of why this occurs and how it works.
Scientists have long struggled to understand the specific material within the group, a crystalline compound called lead iodide. It has excellent photoelectronic properties. More use of the material is hampered by its instability, which can be solved by mixing two halide perovskites. However, more knowledge of the two types is required so that researchers can have optimal control over the mixture.
Key to material design and control
The Chalmers research group is now able to provide detailed explanations of important stages of materials that were difficult to explain in experiments alone. Understanding this stage is key to being able to design and control both this material and the mixture based on the mixture. This study was recently published in the Journal of the American Chemical Society.
“The cold phase of this material has long been a missing part of the research puzzle, solving fundamental problems with the structure of this stage,” says Sangita Dutta, a researcher at Chalmers.
Machine learning contributed to the breakthrough
Researchers' expertise lies in building accurate models of various materials in computer simulations. This allows the materials to be tested by exposing them to different scenarios, and these are confirmed experimentally.
Nevertheless, modeling materials for the halide perovskite family are challenging. Capture and decode its properties requires a powerful supercomputer and long simulation times.
“By combining standard methods with machine learning, simulations can now be run thousands of times longer than before. Models can contain millions of atoms instead of hundreds, making them closer to the real world,” says Dutta.
Lab observations are consistent with simulations
The researchers have identified the structure of formamidinium lead iodide at low temperatures. They also could see that formaamidinium molecules were clogged in a semi-stable state while the material was cooled. To ensure that their research model reflects reality, they collaborated with experimental researchers at the University of Birmingham. They cooled the material to -200°C so that the experiments matched the simulation.
“We hope that the insights we gain from the simulations will contribute to the future ways to model and analyze complex halide perovskite materials,” says Erik Fransson, Faculty of Physics at Chalmers.
Research details:
Article: “Revealing the cryogenic phase of FAPBI3 Using the possibilities of machine learning” was published in the Journal of the American Chemical Society in 2014th Written in August 2025 by Sanguita Dutta, Eric Franson, Tobias Heiner, Benjamin M. Gallant, Dominique J. Kubikki, Paul Elhart and Julia Wiktor. All of the researchers are members of the Faculty of Physics at the Chalmers Chalmers University of Technology, with the exception of Gallant and Kubicki, who are from the Faculty of Chemistry at the University of Birmingham.
This study was supported by the Swedish Strategic Research Foundation, the Swedish Energy Agency, the Swedish Research Council, the European Research Council, the Knut and Alice Wallenberg Foundation, and the Nano Advance region of the Chalmers University of Secholity. This calculation was facilitated by C3SE's National Academic Infrastructure Resources for Sweden (NAISS) Supercomputing.
