Atomic-Scale Intermediate Polarization States Enable Superb Energy Storage in NaNbO3 Ceramics via Machine Learning

Abstract

Dielectric energy‑storage ceramics face a fundamental performance limitation stemming from the intrinsic trade‑off between achieving large polarization and minimizing hysteresis loss. This challenge is particularly pronounced in lead‑free NaNbO₃‑based materials, where high electric fields induce irreversible antiferroelectric-ferroelectric phase transitions. To overcome this limitation, we develop a machine‑learning‑assisted design strategy that guides the creation of atomic‑scale intermediate polarization states (IPSs) within dual‑phase NaNbO₃ heterostructures. Aberration‑corrected scanning transmission electron microscopy directly visualizes IPSs between the tetragonal (T-phase) and rhombohedral (R-phase), which reduce polarization anisotropy and flatten the free‑energy landscape, thereby enabling concurrent large polarization and minimal hysteresis. The optimized ceramic delivers a high recoverable energy density of 10.24 J cm⁻³ with an exceptional efficiency of 92% under 920 kV cm⁻¹. This work establishes a materials‑design paradigm that decouples polarization from energy loss through the integration of atomic-scale structural control and machine learning, providing a promising pathway toward advanced dielectrics for high-power energy-storage applications.