DgcityDesigning a new solar material used to mean years of lab trial and error. A team in China has compressed that search by handing much of the work to artificial intelligence, then proving the computer’s top choices in real devices.
Researchers at the University of Science and Technology of China, working with partners at the Southern University of Science and Technology and City University of Hong Kong, built a screening system that combines AI, high-volume theoretical calculations and chemists’ judgment. The target was a self-assembled molecular layer, an ultra-thin coating that sits between a perovskite solar cell and its electrode and controls how efficiently charge leaves the device.
The search space was vast. From functional fragments of carbazole compounds catalogued in the PubChem database, the team identified more than 90,000 possible structures. They first assembled 300 candidate molecules, narrowed those to 124 based on energy levels, and then scored each on eight measures: how easily it could be synthesized, its energy alignment, the energy cost of moving charges through it, aromaticity, heteroatom content and more.
Crucially, whether a molecule could actually be made in a lab was part of the ranking from the start rather than an afterthought. That matters because a compound that looks ideal on screen is useless if no practical synthesis route exists.
The team synthesized the three highest-ranked molecules. When built into cells, their measured performance tracked the ranking almost exactly, a strong sign the scoring system captures what matters. The best molecule powered a cell with a 27.2 percent conversion efficiency. After 2,000 hours of operation at 65 degrees Celsius under continuous illumination and maximum-power tracking, the device retained roughly 95 percent of its initial output.
Perovskite cells have climbed in efficiency faster than almost any solar technology, but long-term stability has been the main barrier to widespread use. Interfaces, where different materials meet, are often where cells first degrade. A method that can predict durable interface molecules before anyone steps into a lab could cut the development cycle considerably. The researchers say the modular framework is not limited to perovskites and can be extended to other material systems where multiple competing requirements must be balanced at once.
