A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a machine learning framework that can significantly accelerate the process of material discovery. The system, called DiffCSP, focuses on predicting the crystal structure of materials from their chemical composition alone, a historically complex and computationally expensive challenge. By employing a diffusion model that learns to generate stable atomic arrangements, the researchers were able to predict structures for novel materials orders of magnitude faster than traditional physics-based simulations. This advancement holds promise for accelerating the development of next-generation batteries, semiconductors, and catalysts by rapidly identifying materials with optimal properties for specific applications. The team validated their approach by successfully predicting the structures of materials that were later confirmed by experimental data. For the full details, read the article at https://technologyreview.com/2024/07/18/1094985/ai-accelerates-material-discovery-crystal-structures/