Hydrogen-Bond-Engineered Polymer Dielectrics for 250 °C Operation Enabled by Deep-Learning-Based Virtual Screening

Abstract

Polymer dielectrics face inherent trade-offs among permittivity, insulation, and thermal stability above 140 °C, severely limiting high-temperature capacitive energy storage. To overcome this challenge, we deployed a deep learning framework that screens hydrogen-bonding molecular fingerprints within polyimide (PI) composites. Our model deciphers electrostatic potentials and donor/acceptor distributions across 1,200 candidates, predicting number of hydrogen bonds per unit cell, the hydrogen bond probability, the hydrogen bond energy, per monomer unit, and the electronic bandgap. Guided by these predictions, the top-performing candidate, (4-(3-amino-4methoxyphenoxy)phenyl)(p-tolyl)methanone (O-PU), was synthesized and blended with PI.Critically, the introduced hydrogen bonds direct the formation of low-entropy nanocrystalline networks, which simultaneously enhance mechanical strength (Young's modulus: 6.06 GPa), suppress conduction current at 300 MV/m (1.67 × 10 -7 A/cm 2 ),widen optical bandgaps (from 2.92 eV to 4.62 eV), and improve thermal conductivity. Consequently, the optimal composite achieves record-high performance at 250 °C: a Weibull breakdown strength of 587.1 MV/m and an energy density of 5.99 J/cm3 with > 90% efficiency under 400 MV/m. This work demonstrates a fundamentally new strategy for extreme-condition dielectric design using deep-learning-optimized hydrogen bonds.