Dielectric capacitors, characterized by ultra-high power densities, are considered as fundamental energy storage components in electronic and electrical systems. However, synergistically improving energy densities and efficiencies remains a daunting challenge. Understanding the role of polarity heterogeneity at the nanoscale in determining polarization response is crucial to the domain engineering of high-performance dielectrics. Here, a bidirectional design with phase-field simulation and machine learning is performed to forward reveal the structure-property relationship and reversely optimize polarity heterogeneity to improve energy storage performance. Taking BiFeO3-based dielectrics as typical systems, this work establishes the mapping diagrams of energy density and efficiency dependence on the volume fraction, size and configuration of polar regions. Assisted by CatBoost and Wolf Pack algorithms, this work analyzes the contributions of geometric factors and intrinsic features and find that nanopillar-like polar regions show great potential in achieving both high polarization intensity and fast dipole switching. Finally, a maximal energy density of 188 J cm-3 with efficiency above 95% at 8 MV cm-1 is obtained in BiFeO3-Al2O3 systems. This work provides a general method to study the influence of local polar heterogeneity on polarization behaviors and proposes effective strategies to enhance energy storage performance by tuning polarity heterogeneity.
Keywords: energy storage dielectrics; local polarity heterogeneity; machine learning; phase‐field simulation.
© 2024 Wiley‐VCH GmbH.