Date of Award

6-2026

Degree Name

Doctor of Philosophy

Department

Engineering and Applied Sciences

First Advisor

Robert A. Makin, Ph.D.

Second Advisor

Steven M. Durbin, Ph.D.

Third Advisor

Nancy S. Muyanja, Ph.D.

Keywords

Applied machine learning, data science, materials informatics, nonequilibrium synthesis, plasma informatics, quantum machine learning

Abstract

Machine learning is a valuable approach for the processing and analysis of complex information. By estimating relationships from recorded data, machine learning methodologies can be effective strategies for pattern recognition, enabling investigations and technological applications based thereon. The potential for improved understanding of high-dimensional data has drawn interest towards machine learning from across the sciences, including the research and development of new and improved material systems. In the context of experimental materials research, much of the reported efforts to incorporate machine learning into conventional practice have been primarily focused on either the enhanced analysis of characterization experiment data or the accelerated identification of improved experiment designs.

This body of work focuses on the application of conventional data science techniques as well as both classical and quantum machine learning technologies to the modeling and analysis of synthesis pathways in epitaxially grown nitride semiconductors as well as the coinciding laboratory nitrogen plasmas. The investigations exclusively focus on experimentally acquired data sets that were recorded under a common hardware configuration. Relationships between the designs of GaN and InN plasma-assisted molecular beam epitaxy synthesis experiments and the presence of grain boundary defects in the grown samples are analyzed from in-situ reflection high-energy electron diffraction patterns. The formation of an atomically-flat GaN growth surface is qualitatively modeled and assessed in the same fashion. Antisite defect concentrations within GaN thin film crystals as quantified by a Bragg-Williams measurement of lattice ordering (S2) are also investigated as a function of epitaxy experiment design. Calculations of p-values, Pearson correlation coefficients, and tree algorithm nodes are implemented to assess the statistical significance of the experiment design variables for influencing the select figures of merit in each of the assembled data sets. Quantum and traditional learning models are trained upon the experiment records and implemented to generalize beyond each data set, forecasting synthesis-structure relationships for every investigated material composition and figure of merit. SHAP values are calculated from trained algorithm predictions in order to further analyze the relationships that are estimated from the data. The predicted processing spaces corroborate and show good agreement with published experimental literature while mapping previously unexplored domains of the epitaxy synthesis space for these materials. Furthermore, a hybrid quantum-classical machine learning algorithm, a quantum support vector machine, displays superior generalization performance over hyperparameter optimized classical models when tasked with predicting the occurrence of grain boundary defects in GaN.

These methodologies are subsequently implemented to study the relationships between radio-frequency, inductively-coupled nitrogen plasma source operation conditions and the attributes of the generated plasma. Optical emission spectra are recorded and processed to characterize the composition and electron temperature of radio-frequency, inductively-coupled nitrogen plasma under high vacuum. The relative intensities of emission features within the recorded spectra are used to develop population estimations of the various emitting species within the formed plasma. The Boltzmann plot method is implemented to estimate nitrogen plasma electron temperature. For each spectrum, the figures of merit are correlated with the set of plasma source operating conditions, and the resulting data are subjected to the same analyses as the nitride epitaxy data. The simultaneous maximization of plasma power and minimization of pressure is forecasted to maximize electron temperature within the plasma, coinciding with conditions resulting in maximized GaN lattice ordering as measured by S2.

Access Setting

Dissertation-Open Access

Download

Share

COinS