Finding better photovoltaic materials faster with AI

Finding better photovoltaic materials faster with AI
by Robert Schreiber
Berlin, Germany (SPX) Jan 27, 2025

Researchers at the Karlsruhe Institute of Technology (KIT) and the Helmholtz Institute Erlangen-Nurnberg (HI ERN) have developed a novel AI-driven workflow that dramatically accelerates the discovery of high-efficiency materials for perovskite solar cells. By synthesizing and testing just 150 targeted molecules, the team achieved results that would typically require hundreds of thousands of experiments. "The workflow we have developed will open up new ways to quickly and economically discover high-performance materials for a wide range of applications," said Professor Christoph Brabec of HI ERN. One of the newly identified materials enhanced the efficiency of a reference solar cell by approximately two percentage points, reaching 26.2 percent.

The research began with a database containing the structural formulas of about one million virtual molecules, each potentially synthesizable from commercially available compounds. From this pool, 13,000 molecules were randomly selected. KIT researchers applied advanced quantum mechanical methods to evaluate key properties such as energy levels, polarity, and molecular geometry.

Training AI with Data from 101 Molecules

Out of the 13,000 molecules, the team chose 101 with the most diverse properties for synthesis and testing at HI ERN's robotic systems. These molecules were used to fabricate identical solar cells, enabling precise comparisons of their efficiency. "The ability to produce comparable samples through our highly automated synthesis platform was crucial to our strategy's success," Brabec explained.

The data obtained from these initial experiments were used to train an AI model. This model then identified 48 additional molecules for synthesis, focusing on those predicted to offer high efficiency or exhibit unique, unforeseen properties. "When the machine learning model is uncertain about a prediction, synthesizing and testing the molecule often leads to surprising results," said Tenure-track Professor Pascal Friederich from KIT's Institute of Nanotechnology.

The AI-guided workflow enabled the discovery of molecules capable of producing solar cells with above-average efficiencies, surpassing some of the most advanced materials currently in use. "We can't be sure we've found the best molecule among a million, but we are certainly close to the optimum," Friederich commented.

AI Versus Chemical Intuition

The researchers also gained valuable insights into the AI's decision-making process. The AI identified chemical groups, such as amines, that are associated with high efficiency but had been overlooked by traditional chemical intuition. This capability underscores the potential of AI to uncover previously unrecognized opportunities in materials science.

The team believes their AI-driven strategy can be adapted for a wide range of applications beyond perovskite solar cells, including the optimization of entire device components. Their findings were achieved in collaboration with scientists from FAU Erlangen-Nurnberg, South Korea's Ulsan National Institute of Science, and China's Xiamen University and University of Electronic Science and Technology. The research was published in the journal Science.

Research Report:Inverse design of molecular hole-transporting semiconductors tailored for perovskite solar cells