Computational methods for small-molecule transparent semiconductors

Abstract

Solar energy has enormous potential to meet global energy demand in a renewable and environmentally sustainable manner. Although silicon-based photovoltaic (PV) devices have become significantly more affordable and accessible in recent decades, there is a need to develop alternative PV technologies which can be deployed more widely and cheaply. Visibly transparent PV devices based on organic semiconductors are well-suited to this role due to their ability to be installed on windows and building facades, their mechanical flexibility, and their high degree of tunability. However, in order for transparent PV to become commercially viable, further research is needed to shed light on the systematic tuning of the optical properties of molecular materials with visible transparency. This work applies computational tools — namely density functional theory (DFT) and graph neural networks — to gain a deeper understanding of how molecular structure impacts macroscopic optical properties and suggest directions for future study.

In this work we employ linear-response time-dependent DFT with optimally tuned and screened range-separated hybrid functionals in order to compute accurate photoabsorption spectra with relatively low computational cost. Additionally, we utilize molecular graph neural networks as a means to leverage quantum mechanical datasets to accelerate the materials discovery process. These methods are combined to make progress on the optical design of organic semiconductors.

This thesis document is organized as follows. Chapter 1 introduces transparent photovoltaics and the associated materials design considerations. Chapter 2 summarizes the computational methods employed in this work. Chapter 3 describes the first-principles modeling of small-molecule transparent absorbers using perylene dimide derivatives as a case study. Chapter 4 studies principles underlying the design of molecular graph neural networks. Chapter 5 applies these modeling techniques to construct a spectral dataset and train a scalable spectral model; screen a large dataset of organic molecules; identify physical trends and structure-property relationships; and suggest promising candidate materials for transparent photovoltaic applications.