Ferroelectric dipole spiral with giant piezoelectric effects predicted with machine learning potential

Perovskite oxides are a versatile material platform for investigating emergent phenomena that arise from the interplay of various degrees of freedom. In this study, we employed a deep neural network with a self-attention mechanism to develop a unified force field, termed UniPero, which facilitates molecular dynamics (MD) simulations of perovskite oxides containing 14 metal elements and their solid solutions with arbitrary compositions [1]. Using large-scale MD simulations enabled by UniPero, we discovered a novel ferroelectric topological structure, referred to as a “dipole spiral.” This structure features canted dipoles that progressively rotate around the out-of-plane direction and was identified in the well-known ferroelectric material PbTiO₃. Our findings demonstrate that complex topological structures can emerge in simple ferroelectric materials without requiring intricate designs or complex compositions. This helical configuration enables collective, small-angle rotations of dipoles in response to external stimuli, resulting in a giant piezoelectric effect, with d₃₃ values exceeding 320 pC/N [2].