DiffSyn: A Generative Diffusion Model for Materials Synthesis Planning

This article introduces DiffSyn, a generative diffusion model designed to predict synthesis routes for crystalline materials, specifically zeolites. Trained on a large dataset of synthesis recipes from scientific literature, DiffSyn addresses the challenges of high-dimensional synthesis spaces and complex structure-synthesis relationships. It generates probable synthesis pathways conditioned on desired material structures and organic templates, achieving state-of-the-art performance. The model was applied to differentiate competing phases and optimize synthesis routes, leading to the successful experimental synthesis of a UFI material with improved thermal stability.