Leveraging atomistic simulations and machine learning for the design of ionic liquids as electrolytes for battery application

Abstract

Ionic liquids are classes of salts that are often found in a liquid state composed entirely of ions. They have gained widespread interest in the research community because of several unique and desirable features, such as negligible vapor pressure, environmental friendliness, and high thermal stability. They are currently studied for various industrial applications as a replacement for conventional solvents. Among them, it has caught the interest of the energy community as a potential electrolyte for battery applications. The current electrolytes found in lithium-ion batteries are based on carbonate solvents known for their excellent performance and low material cost. However, they are plagued with numerous safety concerns as the solvent is highly volatile and prone to flammability during thermal runaway or short circuit. Growing demand for lithium-ion batteries for technology such as electric vehicles has mandated the need for safer and more sustainable batteries. This has made ionic liquids a potential electrolyte candidate as they have impeccable thermal and chemical stability with negligible vapor pressure, eliminating any concerns related to safety. However, the performance of ILs is still far behind in matching the performance of current carbonate electrolytes. Finding the appropriate ionic liquid candidate with high stability and performance can be challenging because of the vast ionic liquid chemical space, as synthesizing and testing each ionic liquid would be expensive and unfeasible. Running atomistic simulations to complement the experimental techniques would be tedious as the ionic liquid space is estimated to be in billions, which can be computationally expensive. Instead, machine learning methods can be an excellent tool to search for and design ionic liquids suited for battery applications as they can be easily trained on existing data to generate additional new data at a very low cost in a short time. Thus, the work in this dissertation is focused on developing machine learning models to correlate properties of ionic liquids geared towards battery application. The developed models are then used to generate additional data to search for high-performance ionic liquids that are on par with conventional organic electrolytes, thus expanding the list of potential electrolyte candidates. The latter part of the work utilizes an advanced deep learning method to discover an entirely new family of ionic liquid cations in search of candidates that can operate at high voltage conditions.