Machine Learning-Guided High-Efficiency and Thermally Stable Capacitive Energy Storage in Dielectric Capacitors With a Simple Chemical Composition.

Abstract

Dielectric capacitors offering ultrafast charge-discharge capability and superior reliability are essential for advanced electronic systems, but achieving both high energy density and efficiency within simple and eco-friendly compositions remains a great challenge. Here, guided by machine learning, we achieve high-efficiency and thermally stable capacitive energy storage in strontium titanate-based ceramics. Through synergistic local structural engineering and optimized fabrication process, the materials exhibit enhanced polarization, reduced hysteresis loss, and improved breakdown strength. The designed materials deliver an ultrahigh energy density of 10.69 J cm^-3 with a near-ideal efficiency of ∼97% and a record high figure of merit of 392 J cm^-3 at room temperature, and retains high performance at 150°C with a figure of merit of 152 J cm^-3 and an efficiency of ∼94%. This remarkable performance arises from the incorporation of Bi³⁺ ions with high polarizability at the A-sites of strontium titanate quantum paraelectrics, which breaks structural symmetry and induces lattice and octahedral distortions, leading to the formation of nanoscale polar clusters with highly dynamic fluctuations. These findings establish a compositionally simple, environmentally benign pathway for developing dielectrics with superior energy storage capability and thermal stability, offering new opportunities for high-performance capacitive energy storage systems.