In this dissertation, the process of computational modeling of amorphous oxide materials is studied. Amorphous oxides have random atomic arrangements with no long-range structural order which makes it difficult to study structure-property relationships. Despite the difficulties, amorphous oxides have great uses as semiconductors and dielectrics in thin-film transistors. The ability to develop reliable computational models of amorphous oxides is vital to gain a deeper understanding of structure-property relations and for predictive modeling to identify new amorphous chemical compositions that should be targeted for further study.
To accurately model amorphous oxides, multiple models must be generated to include all possible chemical coordinations into a statistical model. By using InGaZnO₄ has the amorphous material to model, it is shown that it is better to create multiple models with smaller numbers of atoms to achieve the same level of convergence as fewer models that use a larger number of atoms. A large amount of data can be generated from these model collections (i.e. bond lengths, bonding angles, coordination numbers, coordination environments, etc.) and to expedite this process, a portion of this dissertation is dedicated to Python code that was written to extract data from amorphous models.
One of the most common ways of generating amorphous models is simulating the quench of a liquid melt from a high temperature. The most efficient way of doing the liquid quench is by using molecular mechanics for the dynamics process. The downside to molecular mechanics is the requirement of a force field being parameterized for all elements involved in the simulation. Disclosed herein is a methodology for parameterizing Morse pair-potentials for use in amorphous models. This methodology relies heavily on the power of modern computers for searching the parameterization space, but requires little user intervention and can be completed in a short amount of time.
The solution deposition thin-film synthesis method has become popular because it can easily produce uniform films and has a low cost compared to other synthesis methods such as atomic-layer deposition or sputter deposition. Modeling amorphous films from solution deposition requires the consideration of counter-ions and solvent molecules as presented here with the study of amorphous aluminum oxide films and how the inclusion of water and hydroxides is necessary to match the solid state NMR data.
Quantum mechanical computations on amorphous models can be time consuming depending on the size of the system and the number of models that need computing. Herein the benefits of using graphical processing units (GPUs) to accelerate quantum mechanical computations in the Vienna Ab-initio Simulation Package (VASP) are studied. It is shown that if unlimited funding is available, then using GPU servers are the optimal configuration of choice. If funding is limited however, desktop grade components can be used to construct a GPU computing system that will allow for higher throughput and lower cost than server grade hardware.