Machine Learning-Driven Scanning Probe Microscopy for Ferroelectric Domain Writing

Increasing research interests in ferroelectric memory give rise of efforts in investigating ferroelectric domain writing at the nanoscale level. Given the requirements of ferroelectric memory are generally operating time, bit size, and retention time, studies of domain writing mainly focus on domain size and stability. Scanning probe microscopy (SPM), due to its capability of visualization and control of ferroelectric domains at the nanoscale level, has been extensively used to investigate domain writing in relation to writing bias parameters, such as biasing time, bias voltage, so on. However, until now, most exploration of domain writing parameters depends on human-based decision making, i.e., operators determine the parameters for the next iteration of the domain writing process according to the previous experiment. In this work, we explored the domain size and shape in relation to the writing voltage and writing time using a machine learning (ML) driven SPM workflow, where an ML algorithm is used to control the SPM for ferroelectric domain writing and visualization. In this workflow, a group of initial writing parameters is first used to write domain structure in a ferroelectric BiFeO₃ thin film. The written domain structure is extracted by a threshold filter from the piezoresponse force microscopy image to analyze the domain shape and size. Then, the ML algorithm will optimize the writing parameters according to the previous domain structure.