Machine-Learning-Designed BCZT-SBT Heterointerface Unlocks Fatigue-Resistant Energy Storage

Dielectric capacitors are attractive for advanced energy storage owing to their ultrafast charge-discharge capability, yet their practical use is hindered by severe fatigue under repeated operation at ultrahigh electric fields. Achieving fatigue-free performance therefore represents a key challenge in dielectric design. Here, guided by machine learning (ML), SrBi₂Ta₂O₉ (SBT) is introduced into Ba_0.85Ca_0.15Zr_0.1Ti_0.9O₃ (BCZT) to construct a (1-x)BCZT-xSBT solid solution. At x = 0.10, the coexistence of perovskite and tungsten bronze phases gives rise to an epitaxial interfacial layer only a few unit cells thick, formed by lattice mismatch. Atomic-scale analyses reveal that this hetero-barrier effectively suppresses carrier migration, while the tungsten bronze phase promotes polarization homogenization, together enhancing both voltage endurance and reliability. As a result, 0.90BCZT-0.10SBT achieves a recoverable energy density (W_rec) of 9.94 J cm^-3 with 92.1% efficiency, and more strikingly, maintains stable performance after 10⁹ charge-discharge cycles without degradation, enabled by an elevated Schottky barrier. This work not only uncovers the atomic origin of fatigue resistance in lead-free dielectrics but also establishes a ML-guided strategy for designing next-generation high-performance, fatigue-free capacitors for reliable energy storage.