Researchers at the Massachusetts Institute of Technology have developed an advanced generative AI model designed to help scientists plan and carry out the synthesis of new materials more efficiently. Traditional materials discovery often gets bottlenecked not at the point of theoretical design but during the actual synthesis process in the lab, where factors like temperature, reaction time, and precursor ratios can dramatically affect outcomes. By training on decades of synthesis recipes, the new AI guides researchers toward promising experimental routes, dramatically reducing the trial-and-error typically involved.
The AI tool, dubbed DiffSyn, uses a type of model known as a diffusion model — similar in concept to models that generate images from noise — to interpret patterns in more than 23,000 historical material synthesis recipes. When a scientist inputs the structure of a desired material, DiffSyn generates a set of plausible synthesis paths, including specific experimental conditions, effectively offering multiple “recipes” for making that material.
To test the approach, the team applied DiffSyn to a class of complex materials called zeolites, which have traditionally been challenging to synthesize due to their high-dimensional parameter space. The model’s suggestions allowed researchers to synthesize a new zeolite with enhanced thermal stability, demonstrating the tool’s potential to not only predict synthesis pathways but also accelerate actual experimental breakthroughs.
Importantly, DiffSyn shifts beyond earlier AI methods that mapped materials to only a single synthesis route. Instead, it maps a material structure to multiple viable synthesis strategies, better reflecting the real experimental landscape and offering researchers a broader set of options to explore. The team believes this approach could extend to other materials and one day interface with autonomous lab systems to further speed materials innovation.
