Temperature-stable high-κ oxides for miniaturized capacitors

The miniaturization of capacitive components in advanced electronics requires dielectric materials that simultaneously have a high permittivity, low loss tangent and near-zero temperature coefficient of capacitance. However, improving one property typically negatively affects the others, and the design of new dielectrics is challenging due to the vast compositional space that could be searched. Here we report a hierarchical machine learning-guided strategy that first screens oxide candidates to identify promising structural families and then optimizes local atomic features to enhance the dielectric properties. The approach identifies tungsten bronze structures, followed by targeted bismuth and germanium co-substitution. The resulting Ba4(Nd0.5Bi0.5)9.33(Ti0.99Ge0.01)18O54 ceramic had a dielectric constant of around 185, a loss tangent below 0.0005 and a temperature coefficient of ≤±30 ppm per degree Celsius. We also study the mechanism of the temperature-insensitive behaviour using in situ atomic-resolution scanning transmission electron microscopy and density functional theory analysis, and we find that strong Ge–O covalent bonds sustain off-centre polar displacements during thermal expansion. The material is used to create prototype single-layer chip capacitors with volumetric efficiency higher than leading commercial benchmarks. A machine learning-assisted materials design method can navigate large compositional search spaces and guide structure design, allowing a temperature-stable, low-loss, high-κ ceramic for miniaturized capacitors to be identified.