AI Materials Discovery Platform MatterMind Screens 12 Novel Photovoltaic Candidate Materials in One Month

DeepMind spin-off materials science subsidiary MatterMind announced on May 6 that its AI-driven materials discovery platform completed systematic screening of 2.3 million inorganic compounds in 30 days, identifying 12 candidates with photovoltaic application potential. Three of these materials are predicted to exceed 35% theoretical efficiency — approaching the single-junction solar cell theoretical limit.

MatterMind chief scientist Yolanda Zhang explained that traditional materials discovery typically requires 5 to 15 years of laboratory synthesis and testing cycles. The AI platform combines first-principles calculations with machine learning potential functions, compressing candidate screening time to one-thousandth of the original. "We're not guessing, we're calculating. Every candidate has been validated at the quantum mechanical level."

Three of the 12 candidates have entered laboratory synthesis verification in collaboration with MIT and ETH Zurich. Preliminary results show one tin-based perovskite derivative achieved a measured efficiency of 31.7%, within 2% of AI predictions. MIT materials science professor Robert Harris noted that if remaining candidates pass experimental verification similarly, the photovoltaic industry could see an efficiency leap within three years.