Harnessing Automation and Machine Learning in Scanning Probe Microscopy to Accelerate Physics Discovery

In this study, we developed automated experimental approaches in scanning probe microscopy (SPM) and investigated ferroelectric polarization switching in response to varied pulse biases (i.e., bias magnitude and duration). Our approach consists of (1) a high-throughput experimentation that offers detailed insights into the correlation between pulse biases and the growth of ferroelectric domains; (2) an autonomous, machine learning-driven experimentation to optimize experimental conditions in real-time based on immediate results.

SPM has long been recognized as a powerful tool for the nanoscale manipulation and visualization of ferroelectric domains. Deepening our understanding of ferroelectric domains and stability can enhance the application of ferroelectrics in memory devices. However, traditionally SPM measurements are labor-intensive, often necessitating expert intervention for repetitive tasks and on-the-fly adjustments of measurement parameters. Here, we first implemented an automated high-throughput experiment, applying a spectrum of bias pulse conditions to write ferroelectric domains. This was followed by piezoresponse force microscopy (PFM) to visualize the resultant domain structures. This method allowed us to systematically modulate bias pulse parameters, revealing polarization states under a spectrum of bias conditions. Second, we integrated a hypothesis active learning (HypoAL) algorithm to the SPM. This algorithm is based on structured Gaussian process (sGP), it evaluates the correlation between bias pulse parameters and the size of the ferroelectric domain in real-time. It then selects the bias parameters for subsequent experiments. The overarching goal of HypoAL is to deduce the most accurate physical model from a set of hypotheses regarding the material's behavior, using the fewest experimental steps. We anticipate the methodologies can be adapted for other microscopy techniques, paving the way for swifter breakthroughs in materials science and physics.