Neural Network-Based Permittivity Engineering of Magnetic Absorbers for Customizable Microwave Absorption.

The development of high-performance microwave absorbers featuring ultrathin profiles and customizable bandwidth remains a formidable obstacle for cutting-edge electromagnetic stealth and long-term service applications. Traditional design approaches for absorbers often rely on inefficient trial-and-error methods, a challenge exacerbated in magnetic absorbers by the intricate coupling between permittivity and permeability. This work introduces a neural network-based permittivity engineering strategy, underpinned by a novel "permeability locking-permittivity optimization" paradigm that effectively decouples the interdependent electromagnetic parameters. A high-throughput permittivity feature space was constructed via tensor-based electromagnetic theory calculations, and a dual-task screening strategy was implemented to identify optimal and effective absorption conditions. This data-driven framework facilitated the inverse design of a magnetic composite, culminating in the guided synthesis of flaky carbonyl iron/barium titanate composites. The material experimentally demonstrates an exceptional effective absorption bandwidth of 5.1 GHz at an ultralow thickness of 1.0 mm, with an optimal reflection loss of -45.12 dB at a 1.9 mm. Furthermore, the formation of a protective Si─O─Si surface layer significantly enhances corrosion resistance, confirming practical durability. This study establishes an AI-guided paradigm that successfully bridges electromagnetic theory with materials design, offering a robust and generalizable platform for the accelerated development of advanced microwave absorption materials.