Decoupling Mesoscale Functional Response in PLZT across the Ferroelectric-Relaxor Phase Transition with Contact Kelvin Probe Force Microscopy and Machine Learning

Relaxor ferroelectrics exhibit a range of interesting material behavior, including high electromechanical response, polarization rotations, as well as temperature and electric field-driven phase transitions. The origin of this unusual functional behavior remains elusive due to limited knowledge on polarization dynamics at the nanoscale. Piezoresponse force microscopy and associated switching spectroscopy provide access to local electromechanical properties on the micro- and nanoscale, which can help to address some of these gaps in our knowledge. However, these techniques are inherently prone to artefacts caused by signal contributions emanating from electrostatic interactions between tip and sample. Understanding functional behavior of complex, disordered systems like relaxor materials with unknown electromechanical properties therefore requires a technique that allows distinguishing between electromechanical and electrostatic response. Here, contact Kelvin probe force microscopy (cKPFM) is used to gain insight into the evolution of local electromechanical and capacitive properties of a representative relaxor material lead lanthanum zirconate across the phase transition from a ferroelectric to relaxor state. The obtained multidimensional data set was processed using an unsupervised machine learning algorithm to detect variations in functional response across the probed area and temperature range. Further analysis showed the formation of two separate cKPFM response bands below 50 °C, providing evidence for polarization switching. At higher temperatures only one band is observed, indicating an electrostatic origin of the measured response. In addition, the junction potential difference, which was extracted from the cKPFM data, becomes independent of the temperature in the relaxor state. The combination of this multidimensional voltage spectroscopy technique and machine learning allows to identify the origin of the measured functional response and to decouple ferroelectric from electrostatic phenomena necessary to understand the functional behavior of complex, disordered systems like relaxor materials.