Quantum and Classical Supervised Learning Study of Epitaxially–Grown ZnO Surface Morphology

Andrew Steven Messecar; Robert Makin; STEVEN DURBIN

Download Paper | Permalink

Conference: 2024 ASEE North Central Section Conference
Location: Kalamazoo, Michigan
Publication Date: March 22, 2024
Start Date: March 22, 2024
End Date: March 23, 2024
Page Count: 10
DOI: 10.18260/1-2--45633
Permanent URL: https://peer.asee.org/45633
Download Count: 23

Request a correction

Paper Authors

biography

Andrew Steven Messecar Western Michigan University orcid.org/0000-0002-1515-6206

visit author page

Andrew Messecar is a Ph.D. candidate at Western Michigan University's Department of Computer Science. He works with Dr. Robert Makin in the College of Engineering and Applied Sciences' Molecular Beam Epitaxy Laboratory. His research interests include materials and process informatics, the epitaxial synthesis of novel electronic materials and devices, and the simulation of physical systems using quantum and classical computation.

visit author page

author page

STEVEN DURBIN

author page

Robert Makin Western Michigan University

Download Paper | Permalink

Abstract

Material synthesis parameter spaces typically have a very high dimensionality and are often intractable in size. Exploring these vast, multi–dimensional parameter spaces by trial–and–error experimentation – even for well–studied materials – is not feasible on reasonable time scales. Thus, considerable interest exists in the development of machine learning–based approaches for the rapid and accurate identification of optimal materials designs and synthesis conditions. In this work, data describing over 125 plasma–assisted molecular beam epitaxy (PAMBE) synthesis experiments of ZnO thin film crystals have been organized into a single data set. For each growth record, the complete set of PAMBE operating parameters for ZnO synthesis are associated with a measure of crystallinity as determined by in-situ reflection high–energy electron diffraction (RHEED) patterns. Additionally, a Brag–Williams measure of lattice disorder (S) is included as a second figure of merit for investigation. Quantum and classical supervised learning algorithms – including linear models, tree–based algorithms, and quantum variational circuits – are trained on the data and used to study which growth parameters are most statistically important for influencing crystallinity and S in epitaxially–grown ZnO thin films. Comparisons are drawn between the generalization performances of the various algorithms that are trained on the data. The supervised learning algorithm exhibiting superior generalization performance is used to predict the class conditional probability of obtaining a monocrystalline ZnO thin film crystal across a processing space defined by the most statistically important synthesis parameters. These machine learning–predicted experiment results are compared with PAMBE operating parameters that have been previously reported in published literature to result in single crystalline ZnO samples. S is also predicted across the same processing spaces in order to draw comparison to the crystallinity predictions. The machine learning predictions identify the various growth conditions that are of interest depending on whether a single crystalline ZnO sample or a well–ordered lattice (as measured by S) is desired. These supervised learning–based predictions yield experiment design rules which can be used to inform future ZnO PAMBE growth experiments. This analysis offers a valuable perspective on the mechanisms that are active during the PAMBE synthesis of ZnO and other related oxide compounds. Furthermore, the methodology exemplified in this work can be extended to study synthesis–structure relationships in additional PAMBE–grown oxide materials.

Citation
Format

Messecar, A. S., & DURBIN, S., & Makin, R. (2024, March), Quantum and Classical Supervised Learning Study of Epitaxially–Grown ZnO Surface Morphology Paper presented at 2024 ASEE North Central Section Conference, Kalamazoo, Michigan. 10.18260/1-2--45633

TY  - CPAPER
AB  - Material synthesis parameter spaces typically have a very high dimensionality and are often intractable in size. Exploring these vast, multi–dimensional parameter spaces by trial–and–error experimentation – even for well–studied materials – is not feasible on reasonable time scales. Thus, considerable interest exists in the development of machine learning–based approaches for the rapid and accurate identification of optimal materials designs and synthesis conditions. In this work, data describing over 125 plasma–assisted molecular beam epitaxy (PAMBE) synthesis experiments of ZnO thin film crystals have been organized into a single data set. For each growth record, the complete set of PAMBE operating parameters for ZnO synthesis are associated with a measure of crystallinity as determined by in-situ reflection high–energy electron diffraction (RHEED) patterns. Additionally, a Brag–Williams measure of lattice disorder (S) is included as a second figure of merit for investigation. Quantum and classical supervised learning algorithms – including linear models, tree–based algorithms, and quantum variational circuits – are trained on the data and used to study which growth parameters are most statistically important for influencing crystallinity and S in epitaxially–grown ZnO thin films. Comparisons are drawn between the generalization performances of the various algorithms that are trained on the data. The supervised learning algorithm exhibiting superior generalization performance is used to predict the class conditional probability of obtaining a monocrystalline ZnO thin film crystal across a processing space defined by the most statistically important synthesis parameters. These machine learning–predicted experiment results are compared with PAMBE operating parameters that have been previously reported in published literature to result in single crystalline ZnO samples. S is also predicted across the same processing spaces in order to draw comparison to the crystallinity predictions. The machine learning predictions identify the various growth conditions that are of interest depending on whether a single crystalline ZnO sample or a well–ordered lattice (as measured by S) is desired. These supervised learning–based predictions yield experiment design rules which can be used to inform future ZnO PAMBE growth experiments. This analysis offers a valuable perspective on the mechanisms that are active during the PAMBE synthesis of ZnO and other related oxide compounds. Furthermore, the methodology exemplified in this work can be extended to study synthesis–structure relationships in additional PAMBE–grown oxide materials.
AU  - Andrew Steven Messecar
AU  - STEVEN DURBIN
AU  - Robert Makin
CY  - Kalamazoo, Michigan
DA  - 2024/03/22
PB  - ASEE Conferences
TI  - Quantum and Classical Supervised Learning Study of Epitaxially–Grown ZnO Surface Morphology
UR  - https://peer.asee.org/45633
DO  - 10.18260/1-2--45633
ER  -