Going beyond High Throughput Screening towards AI in Accelerated Electrolyte Discovery
Discovery of new materials is the key to bringing timely advancements and major breakthrough in the developing technologies. Development of sustainable next generation fast charging batteries for electric vehicles (EV) need a step-forward towards the discovery of non-flammable liquid battery electrolytes. Primary electrolyte systems commonly used for lithium metal batteries are based on ether solvents comprising of 1,2-dimethoxyethane (DME) and a cosolvent 1,3-dioxolane (DOL), both of which are highly flammable. The search and development of sustainable electrolyte formulations for the batteries require collaborative efforts from the disciplines of chemistry, material science and computational simulations. Widely adopted high-throughput screening process for electrolyte discovery involves computational simulations of large set of molecules and screening based on domain intuitive properties. This process manages to accelerate the discovery of new compounds for electrolytes but, the high performance of an electrolyte in a battery goes beyond the sum of the individual compounds. It relies on the complex interactions of those constituent compounds that lead to new, immerging properties of the whole, i.e. the electrolyte formulation. There remain significant questions as to how to relate the properties of the individual electrolyte compounds to that of whole; finding the right formulation based on those screened new compounds; and predicting performance of the formulation that contains those compounds as well as their stability over versus the electrodes in a full cell. Artificial Intelligence (AI) methodologies built upon high throughput screening process can expedite the discovery and optimization of both new electrolytes molecules and electrolyte formulations which contain them. In this talk, we discuss the synergy of high throughput screening and AI methodologies based on graph theory and inverse design in accelerating the process of electrolyte discovery and formulation optimization. To do this, we apply automated simulation workflows capable of simulating multiple molecular properties such as redox potential and stability, and provide in-silico characterization of the interface chemistries at the electrodes on the fly. Post this, surrogate models built on graph theories establish a structure-composition-performance relationship for the new-found electrolyte formulations and enable the search for electrolyte systems with highest capacity retention and long cycle life.