Machine-learning enabled construction of temperature-strain phase diagrams of ferroelectric thin films

Ferroelectric thin films have been explored for many applications such as microelectronics or system-on-a-chip prototypes. It is well established that stability of ferroelectric states of thin films are determined by both temperature and strain between the film and its underlying substrate and the chemical composition for solid solution thin films. A complexity associated with ferroelectric thin films constrained by a substrate is that often the multidomain states of multiple ferroelectric domain variants become stable. Using a combination of high-throughput calculations, classification machine-learning algorithms, and phase-field simulations, we systematically investigate the phase diagrams of (Ba_ x , Ca_1- x )TiO_3 (BCTO) solid solution thin films. We examine several machine-learning techniques to understand the differences in their accuracies and capabilities for the construction of phase diagrams of ferroelectric thin films. We demonstrate that a computational scheme consisting of high-throughput calculations, machine-learning, and phase-field simulations, can be employed to obtain accurate phase-stability diagrams of ferroelectric films. Graphical abstract