Establishing Control over VO2 Phase Transformations and Perovskite Quantum Confinement Through Targeted Exploration of Materials and Processing Design Spaces

Abstract

Establishing mechanistic and causal understanding of a material’s function requires the ability to solve the inverse problem of relating function to atomistic properties and to subsequently examine how synthetic processes can be designed to arrive at desired atomistic configurations. Though detailed mechanistic study and data-driven methodology the processing-structure-property relationships of a synthetic design space can be leveraged to converge on the desired functional properties. Through the examples of VO2 and metal halide perovskites, materials of great current interest due to their desirable properties of a metal-insulator transition (MIT) and high photovoltaic performance respectively, we have established mechanistic understanding and predictivity of relationships between synthesis and materials properties. This work explores chemical doping in the VO2 system as a vector to modify the MIT properties and establish mechanistic understanding of the modulation observed when doping with tungsten, boron, germanium, and iridium. The interplay of defect dynamics, phase transition kinetics, and crystallographic modification in these doped systems result in mechanisms that lower and raise MIT temperatures, afford control over the MIT hysteresis, create a dynamical response, and stabilize entirely new metastable crystal structures. The synthesis of CsPbBr3 perovskite nanoparticles proves to be a successful case study in the utility and flexibility of machine learning to predict outcomes, such as nanoparticle thickness, from sparse and incomplete data as well as to provide quantitative insight into the mechanisms driving the shape of the reaction landscape. Sparse expensive synthetic data benefit from machine-learning-directed navigation as it is more efficient and tolerant to uncertainties than simpler experimental exploration methods. The utility of this type of the machine learning and feature analysis when combined with Bayesian learning for chemical and material syntheses has potential for high-quality predictions with small and/or high-dimensional datasets with implications for automated experimentation and engineering where solving the inverse problems may provide human-like insight to automation processes.