Machine Learning-accelerated Discovery of Broadband Electromagnetic Absorbers: Uncovering Structural-functional Correlations in Large-scale Heterogeneous Coating Materials

Abstract

The rapid proliferation of high-frequency electronic devices and the increasing demand for stealth capabilities in aerospace and satellite systems necessitate the development of advanced electromagnetic (EM) wave absorbers. Traditional discovery methods, reliant on Edisonian trial-and-error or computationally intensive full-wave simulations (e.g., Finite Element Analysis), are severely bottlenecked by the vast chemical and structural design space. In this work, we introduce a paradigm shift from conventional screening to a data-driven materials informatics framework strictly focused on the thermodynamic screening of candidate materials. Leveraging a massive dataset of 51,182 unique material candidates from the Materials Project, we employ machine learning algorithms to predict thermodynamic stability, specifically Energy Above Hull (eV/atom). By rapidly identifying thermodynamically stable and synthesizable candidates, this framework acts as a critical pre-screening filter to bypass the computational cost of evaluating unstable compounds. This study demonstrates that large-scale informatics not only accelerates the discovery pipeline but also uncovers hidden structure-stability relationships in heterogeneous materials, laying the essential thermodynamic groundwork for the future synthesis of lightweight, wide-bandwidth EM shielding solutions.