Abstract
In this dissertation, a data-driven approach to surface characterization is developed using multifrequency atomic force microscopy (AFM) combined with machine learning to investigate nanoscale tip–surface interactions in III-V photovoltaic (PV) materials. The work is particularly motivated by the goal of enabling wafer reuse in GaAs solar cells through graphene-mediated lift-off techniques, which demand ultra-clean, minimally altered surfaces. A multifrequency AFM framework is implemented to extract rich observables—amplitude, phase, and virial— from multiple vibrational modes, offering higher sensitivity and force-resolution compared to traditional single-mode techniques.To interpret the high-dimensional AFM data, artificial neural networks (ANNs) and clustering algorithms are applied for force classification and model selection. The methodology successfully differentiates interaction regimes—specifically identifying van der Waals-dominated regions critical for evaluating surface cleanliness. Furthermore, the study validates the use of a simplified mass–spring model to approximate cantilever dynamics over the more complex Euler–Bernoulli beam theory, facilitating real-time modeling and machine learning integration. Classification experiments achieve high accuracy, with the neural network model reaching a test accuracy of 98.2% in distinguishing power-law interaction exponents.
The outcomes confirm that multifrequency AFM, guided by machine learning, can reliably assess nanoscale surface integrity, thereby supporting wafer reuse strategies in III-V solar manufacturing. This interdisciplinary framework has broad applicability to nanomaterial systems and sets the stage for real-time, AI-driven surface characterization tools in advanced materials processing.
| Date of Award | 2025 |
|---|---|
| Original language | American English |
| Supervisor | Matteo Chiesa (Supervisor) |
Keywords
- Photovoltaic
- AFM
- Multifrequency
- Machine Learning
- Van der Waals
Cite this
- Standard