Deep learning assisted heat-resistant metamaterial absorber design

Metamaterial absorbers (MMA) have great potential in civilian and military applications because of their large degrees of design freedom to achieve broadband absorption with thin thickness. In modern applications, great-performance absorbers require both broad-band absorption with low thickness and highly heat-resistant characteristics. In this paper, a deep-learning-based composite-network (DLCN) is proposed to efficiently design the MMA working at S to X bands with 10-mm-thick thermal barrier layer. The trained DLCN can accurately predict the reflection coefficients of the meta-atoms that constitute the MMA by their configuration parameters, including geometric and material parameters. Using the evolutionary algorithm (EA) to search the parameter space with the mapping function given by DLCN, we obtain the optimal configuration parameters of the meta-atoms and their filling ratio in the MMA that satisfies the demands best. The design method using EA and the DLCN is highly efficient and can be extended to design other multi-functional metamaterials.