Machine-Learning-Guided Polarization-Lattice Decoupling Enables Ultrahigh Energy Storage in Lead-Free Dielectric Ceramics.

Abstract

Achieving ultrahigh energy storage in lead-free dielectric ceramics is fundamentally constrained by the intrinsic trade-off between large polarization and high dielectric breakdown strength. Here, we establish an interpretable machine-learning-guided design framework that quantitatively links ionic descriptors with polarization behavior in ABO3-based dielectric matrices, enabling the rational identification of compositions with intrinsically high polarization potential. Guided by this strategy, a (Bi0.275Na0.2255K0.0495Ba0.3)(Ti0.985Hf0.015)O3-0.15(La0.5Sm0.5)2Ti2O7 (BNBT-3) composition is discovered that exhibits an exceptional maximum polarization of 50.19 µC cm-2. When processed via a viscous polymer process, the resulting BNBT-3-VPP capacitors achieve an ultrahigh breakdown strength of 1400 kV cm-1 and a recoverable energy density of 25.1 J cm-3 with high efficiency, placing them among the best-performing lead-free dielectric ceramics reported to date. Structural characterization combined with phase-field simulations reveals that the outstanding performance originates from polarization-lattice decoupling, where nanoscale polarization clusters and multiphase coexistence suppress long-range ferroelectric order while enabling reversible polarization rotation. This work establishes a generalizable strategy that integrates interpretable machine learning with physically grounded materials design, providing a powerful route for discovering high-performance dielectric energy storage materials.