The evolution of materials informatics is crucial for promoting the creation of innovative and sophisticated materials that will drive clean energy technologies. With materials discovery and optimization currently taking the world’s research community decades, this large temporal scale impedes the development of clean energy technology in the time-susceptible fight against climate change. Therefore, we aim to push the opportunities for computer-driven materials discovery.
Materials discovery and design are driven by the exploration of chemical space through synthesis processes. These multifaceted procedures determine materials' structural characteristics which in turn are essential factors influencing key performance properties such as Li+ diffusion for solid-state electrolytes. Contemporary endeavors have led to the establishment of extensive toolkits of first-principles atomistic methods by linking structural properties with first-principles calculations, while advances in natural language processing facilitate large-scale extraction of synthesis processes and results. Concurrently, as laboratory automation progresses, Bayesian optimization-based autonomous experimentation is emerging as a preferred approach for guiding the enhancement of material properties.
Our research focuses on linking computational and experimental materials scientists by making machine learning informed predictions on synthesizability and synthesis parameters. This is accomplished by utilizing extensive first-principles datasets in combination with literature extracted experimental data to determine innovative pretraining and distillation strategies for deep materials representation learning.