Machine Learning‐Driven Ultra‐High Energy Storage Performance in Ferro‐Superparaelectric Capacitors

Abstract The development of high‐performance lead‐free dielectric capacitors based on BCZT ceramics has traditionally relied on compositional doping to enhance energy storage density and efficiency, though this strategy introduces considerable uncertainties. In this study, a machine learning‐assisted screening strategy is implemented, utilizing a AdaBoost model to predict the recoverable energy density of (1– x )BCZT– x BNZN ceramics. It is found that compositions with doping levels above 0.10 display superparaelectric (SPE) behavior at room temperature, and an optimized relaxor‐ferroelectric composition with 0.15 BNZN is identified by the model as having outstanding energy storage potential. The viscous polymer process (VPP) is subsequently employed, leading to further improvement in dielectric properties and the unprecedented formation of a core‐shell structure within the grains. Consequently, a record energy density of 11.6 J cm ⁻³ with 92% efficiency and an ultrahigh breakdown electric field of 700 kV cm ⁻¹ is achieved in a superparaelectric relaxor ferroelectric (SPE‐RFE). Moreover, a discharge energy density of ≈3.2 J cm ⁻³ at an ultrafast rate of 34 ns at 150 °C is demonstrated. Superior temperature, frequency, and fatigue stability highlight the potential of these ceramics for extreme environment applications, opening new avenues for advanced lead‐free dielectric energy storage materials.