Accelerating the discovery and design of advanced functional materials using AI/ML and exascale computing
Through this project, we will develop a series of exascale-capable computational codes and workflows that integrate materials theories and methods with Artificial Intelligence/Machine Learning (AI/ML) tools and materials databases, as well as the software stack developed through the Exascale Computing Program (ECP). The goal is to dramatically increase the speed of predicting new materials in new composition-structure-property spaces and identifying synthetic pathways for new functional materials. We will produce AI/ML integrated and validated open-source, public-access community codes, and associated databases on exascale computing facilities.
We bring together experts with complementary skills from two national laboratories in AI/ML for material research, ab initio theory and computation, atomistic modeling/simulation of complex materials, and exascale computing. This team is unique and competent for addressing basic energy sciences and exascale computing challenges. We organize our research work into two synergized thrusts: thrust 1—AI/ML-assisted materials discovery and design framework; thrust 2—computational prediction of synthesis pathways. This project will also leverage close collaborations with experimental groups at Ames Lab to form a timely feedback loop for validation of the computational codes and predictions.
The project aligns well with and supports the U. S. Department of Energy’s missions on exascale computing and artificial intelligence for science initiatives. We demonstrate the power of AI/ML and exascale computing in significantly reducing the time-to-solution for energy and advanced technology innovations. The open-source community codes, and associated databases developed from this project will enable science-based predictive design and discovery of a wide-range of functional materials that would otherwise be impractical or impossible to investigate in a timely manner due to their complexity.
