Richard Brutchey, University of Southern California

Host: Julie Fenton (863-1759)

"Harnessing the Full Power of 'Small' Data to Guide Materials Synthesis"

Abstract:

The scale and urgency of developing new materials for next-generation technologies requires that these materials be made and deployed at a more accelerated pace. Traditionally, there has been a significant time gap between discovery and commercialization, spanning multiple decades. However, the Materials Genome Initiative, through efforts like Materials Project, has made significant strides in expediting materials discovery by computationally predicting their properties even before physical production. While this “materials by design” approach has yielded numerous materials with diverse properties, a critical bottleneck now exists in the actual synthesis process. Most materials synthesis lacks a deep mechanistic understanding, making prediction and control challenging. The conventional trial-and-error method, reminiscent of Edisonian experimentation, involves adjusting one variable at a time, leading to prolonged timelines for successful synthesis, execution, and optimization.

An alternative to this slow process lies in data-driven techniques like design of experiments (DoE) and machine learning, which offer an effective and efficient way to predict and control materials synthesis. In this talk, I will share our recent progress in leveraging these techniques to predict and control inorganic materials synthesis. Remarkably, even with “small” data sets related to novel and inherently low throughput chemistries, we can achieve promising results using data-driven methodologies. I will showcase select case studies from my lab that underscore the power of these techniques in optimizing materials synthesis and manufacturability, synthetically navigating complex phase diagrams, and fine-tuning the performance of CO2 reduction catalysts.