The performance of next-generation nanoelectronics is intrinsically linked to our understanding of material functionalities at the nanoscale. While scanning probe microscopy (SPM) platforms are specialized for measuring desired behaviors at these length scales, additional artifacts routinely influence the recovered signal. For instance, in piezoresponse force microscopy (PFM), a voltage-modulated waveform is applied to detect piezoelectric and ferroelectric response (e.g., polarization switching, domain wall effects); however, topographic cross-talk, long- and short-range electrostatic interactions between the microcantilever and the sample surface, as well as numerous tip-induced phenomena (e.g., charge injection, ionic migration, joule heating) can contribute to the signal. Thus, interpretation of a measured PFM signal is not immediate.
Over the past decade, machine learning (ML) techniques have been applied to simplification and interpretation of multidimensional datasets, including those collected via grid-based switching spectroscopy PFM (SS-PFM) experiments. However, the interpretability of ML output can be heavily biased by data curation and/or ML model complexity and optimization. Furthermore, the combinations of input-model configurations can be too exhaustive or time consuming for manual inspection. Thus, the objective of this work is to develop a physics-informed ML approach towards identifying common signal contributors to the PFM response, effectively reducing user bias and effort in data analysis. We aim to classify and/or correlate acquired PFM data to one or more external references where observable behaviors such as ferroelectric domains or surface topography have been captured. In contrast to visual inspection, we quantitatively evaluate the accuracy of several ML models in identifying or classifying the localized effects of different contributors with minimal human intervention. This framework can be further expanded to evaluate model performance. Specifically, feature selection techniques (e.g., variance threshold, Laplacian method) can be leveraged towards identifying which aspects of the measurements had a significant influence on an analysis. We expect the developed framework will not only elucidate the prominent effects of PFM signal contributors on the measurement parameters, but will also aid in uncovering their lesser and more subtle influence on the response. Ultimately, this method is easily applicable to other SPM-based platforms generating multidimensional datasets and can facilitate correlative studies across different characterization methods.

Ferroelectric (FE) materials are widely used in data storage technologies including smart cards, random access memory chips and are being developed as multi-level memory elements in the quest for neuromorphic computing systems. As miniaturization of electronic devices continues, a crucial requirement for materials in data storage applications is the enhancement of their functional properties at thicknesses relevant to Nano-electronics (sub- 10 nm). Typically, FE domains segregate into domains of opposite polarity separated by neutral domain walls. It is widely acknowledged that FE domain structure is the initial governing factor in polarization switching behaviour. The equilibrium domain configuration in a FE thin film arises from the minimization of the elastic and electrostatic energies in the crystal and is influenced by factors such as the film composition, growth mechanism, underlying substrate, and thin film thickness. Therefore, the characterization, understanding and tailoring of FE domain structure in ultrathin films is imperative for controlling electromechanical properties and device applications. In this presentation, we will demonstrate how an atomic force microscopy (AFM)-based Nano-machining method, using a commercially available diamond probe, can remove surface contaminants from ultrathin FE films with nanometer-level precision. Smooth layers are uncovered beneath the as-grown surface, with subsequent piezoresponse force microscopy imaging revealing distinct 45o stripe domain configurations, of stark contrast to the randomly distributed response observed for the pristine film surface. The differences in configuration between the pristine surface and exposed sub-surfaces are attributed to the effectiveness of AFM-based Nano-machining in removing growth surface artefacts that otherwise mask the domain configurations of the underlying planar film. We discuss how sub-surface FE domain configurations are influenced by underlying substrate, growth conditions and film thickness. Furthermore, our investigations indicate that these sub-surface domain structures persist throughout the film depth down to thicknesses of less than half of a unit cell (˂ 2.5 nm), thereby demonstrating the technological potential of the FE materials for future miniaturized memory storage devices.

Dynamics of ferroelectric domain walls associated to domain walls switching are known to depend on bulk structure, being very sensitive to defects, chemical and structural pinning sites, as well as environmental conditions modifying the electronic boundary conditions associated with screening dynamics. However, domain walls also show a sub-coercive field dynamics as a reversible motion, with vibrational states that strongly coupled to local structure and composition of the domain wall. In this presentation, we will show Scanning Oscillator PFM, a novel microscopy mode based on a multifrequency approach which allows us to quantify domain wall oscillations under applied sub-coercive electric field, while simultaneously disentangle electrostatic from piezoelectric signals, which can introduce severe distortions on the net piezoelectric response around neutral domain walls. This technique allows quick visualization of domain wall displacement, their velocities and dependence on pre-existing domain configurations and defects. When applied to lead titanate, this technique shows significant oscillations of the 180° domain walls between the antiparallel c+/c- domains. Further, the displacement and velocities distinctly depend on existing a-c domain structures and relative orientations of domains. We will also discuss about applications of this technique, including a wide range of stimuli such as light, heat, force and bias, to study dynamic phenomena in the second to sub-millisecond range. Finally, we will introduce how domains can be manipulated in new ways to create cumbersome topological structures that can stabilize complex domain walls with singular configurations.