Demystifying local polarization configuration evolution for high-piezoelectricity chemical design via deep learning

Development of piezoelectric materials through chemical design meets the requirement of the next-generation electronic devices, yet the sensitive piezoelectricity to both chemical components and operational environment call for the trial and error method during material preparation. In order to give an atomic-level understanding about functional unit and assist the chemical design, deep learning was applied to train a novel model based on the most popular BaTiO3 system, as a case study in this work. Through training the atomic force field of calcium and stannum doped solid-solution with Deep Potential method, 3D structure of chemical distribution and corresponding polarization configuration can be constructed for different compositions under different temperatures, which exhibits a high degree of consistency with the local structure quantitatively analyzed from HAADF STEM and reverse Monte Carlo refinement of neutron total scattering data, especially for the critical composition with ultrahigh piezoelectricity of d33 ~ 860 pC/N. Systemic analysis reveals that variations in chemical bond length among various elements with oxygen elements are the primary factors influencing ferroelectric activity and leading to structural evolution. The results and methodology can facilitate the discovery of new ferroelectrics and the design of high-performance piezo/ferroelectrics with atomic-level insights.