Computational approaches to understanding structure-property relationships in single plasmonic nanoparticles

Shiratori, Katsuya

Permalink

https://hdl.handle.net/2142/129566 Copy

Description

Title

Computational approaches to understanding structure-property relationships in single plasmonic nanoparticles

Author(s)

Shiratori, Katsuya

Issue Date

2025-04-23

Director of Research (if dissertation) or Advisor (if thesis)

Link, Stephan

Doctoral Committee Chair(s)

Landes, Christy Fae

Committee Member(s)

• Makri, Nancy
• Wagner, Lucas

Department of Study

Physics

Discipline

Physics

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• plasmonics
• machine learning
• single nanoparticles
• structure-property relationship
• electromagnetic simulaiton

Abstract

Plasmonic nanoparticles are promising building blocks for applications ranging from solar cells to cancer therapy, owing to their highly tunable optical properties that depend on morphology and environmental conditions. For optimal applications, it is crucial to understand the structure-property relationship of single plasmonic nanoparticles. However, conventional characterization methods rely heavily on electron microscopy, limiting their applicability for in situ and topographical measurements. In this thesis, I leveraged computational approaches, including machine learning (ML) and electromagnetic simulations, to overcome these limitations and gain deeper insight into the structure-property relationship of single plasmonic nanoparticles. First, I demonstrated that a decision tree regressor can predict gold nanorod (AuNR) dimensions directly from single-particle dark-field scattering spectra, eliminating the need for correlated electron microscopy. Trained on∼450 AuNR geometries and their corresponding spectra obtained via finite-difference time-domain (FDTD) simulations, the model predicted AuNR dimensions within∼10% of their true values when tested on experimental data. Analysis of the model structure reveals that resonance energy (Eres) and line width are sufficient to determine particle size, outperforming more complex models. Building on this foundation, I developed a domain adaptation approach that allows ML models trained on AuNRs in one environment to predict sizes in another. By leveraging the linear relationship between aspect ratio and Eres, I adapted training data for AuNRs on glass substrates to predict sizes on indium tin oxide (ITO) and alumina (Al2O3). This method eliminated the need for new training data and achieved comparable accuracy to models trained directly on each substrate, demonstrating a scalable approach for nanoparticle characterization across different environments. Finally, I investigated a novel plasmonic meta-atom, the Sphere-in-Frame (SpiF), designed to mitigate interparticle coupling effects while maintaining a stable Eres. Single-particle dark-field spectroscopy and correlated electron microscopy confirmed that SpiF exhibited an Eres comparable to isolated gold nanospheres, despite structural variations. FDTD simulations further revealed that SpiF maintains stable Eres even in dimer configurations, unlike conventional nanoparticle assemblies. This design offers new possibilities for plasmonic applications requiring precise field control and enhanced stability. By integrating ML and computational simulations, this work advances the capability to predict nanoparticle structure from optical spectra, adapt models across different experimental conditions, and design robust plasmonic architectures. These approaches pave the way for new methodologies in nanoparticle characterization and engineering, particularly in environments where traditional electron microscopy is impractical.

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129566

Copyright and License Information

Copyright 2025 Katsuya Shiratori

Owning Collections