Georgia Institute of Technology
Maximizing insights into nanoscale electromechanical behavior: Machine learning approaches for resonant piezoresponse force microscopy
Abstract
dc:description.abstractPiezoresponse force microscopy (PFM) probes the nanoscale electromechanical response of materials via measurement of the sample surface displacement upon application of electric field through a conductive cantilever tip. However, in addition to piezoelectric coupling, many non-piezoelectric phenomena (e.g., electrostatic forces arising from charge injection and electro-chemo-mechanical force arising from ionic and defect motion) can also contribute to the observed response. Hence, analysis of the observed response is hindered in physical interpretation. While such limitations might be marginal in epitaxial ferroelectric ultra-thin films with large piezoelectric coefficients often used in PFM studies, no material is \textit{a priori} impervious to them. Moreover, low signal-to noise ratio (SNR) data is unavoidable in PFM characterization and critical to understanding the electromechanical response of the material, adding to interpretability challenges. In the last decade, machine learning techniques have been increasingly used in the curation and analysis of PFM data to overcome some of these interpretation challenges, but once again most have been used on “ideal” samples with high piezoelectric response. This work explores machine learning approaches to maximize information density within PFM datasets, and aid in the PFM signal interpretation for characterization of “less than perfect'' materials. In the first embodiment, a defect-rich \ce{Pb(Zr_{0.53},Ti_{0.47})O3} thin film was used as a model system for complex electromechanical behavior, and separation of piezoelectric and nonpiezoelectric contributors to the PFM response. Considerations for data cleaning/outlier removal, analysis object preparation (dimensional stacking), and selection of ML methods for investigation of the PFM response were systematically explored with a special emphasis on the physical constraints and implications of user-defined choices on the analysis. An optimized methodology found three major contributors to the electromechanical response: an electrostatic, a piezoelectric (ferroelectric), and a third behavior likely due to a combination of multiple non-piezoelectric phenomena. The localized presence of these behaviors was closely correlated with the film’s topography. After establishing ML-based methods for the separation of piezoelectric and nonpiezoelectric contributors to the PFM response, a second investigation explored in further depth the inherent and expected presence of low SNR data points within PFM datasets. Here, a “cradle to grave” PFM data curation, including reliable methods for the identification and reconstruction of low SNR signals is demonstrated leveraging a sample PFM response of a relaxor-ferroelectric solid solution of \ce{(0.72)Pb(Mg_{1/3}Nb_{2/3})O3-0.28PbTiO3} (PMN-xPT) single crystal collected on a tool with a high amount of instrumental noise. Here, the implications of user defined choices on several required and under reported data curation steps – including initial data cleaning, fitting to the Simple Harmonic Oscillator (SHO) model, and phase calibration as well as the aforementioned identification and replacement of low SNR signals – are thoroughly examined. As a result of improved data curation, probable features due to challenges in extracting PFM parameters from low SNR data points are identified. Specifically, fitting reconstructed low SNR signals to the SHO model minimizes the presence of plateaus in the signal amplitude, horns in the signal phase, notch-like features in the piezoresponse, and large deviations from mean contact resonance frequency. In previous PFM reports, notch-like features in the piezoresponse have been attributed to ferroelastic coupling whereas large deviations in contact resonance frequency have been attributed to nanoscale phase transitions. Furthermore, the methodology developed for low SNR reconstruction provides uncertainty analysis, allowing for a quantified approach to determining the validity of PFM parameters extracted from low SNR data points. In other words, the developed methodology provides a pathway to avoid such artifacts and misinterpretation in PFM characterization of a wide range of materials, including “novel” piezoelectric or ferroelectric materials with small piezoelectric response or multiphase materials with a range of piezoelectric or ferroelectric functionalities.
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Materials Science and Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Williams, Kerisha Nicole
- Advisor dc:contributor.advisor
-
- Bassiri-Gharb, Nazanin
- Committee members dc:contributor.committeemember
-
- Garten, Lauren
- Khan, Asif
- Xie, Yao
- Vogel, Eric
Subjects
dc:subject × 6Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81495