Abstract Through magneto‐optical materials or spatiotemporal metamaterials, the reciprocity relation between thermal emission and absorption can be broken, achieving the more flexible nonreciprocal thermal radiation (NTR) to even approach the ultimate thermodynamic limit, such as the Landsberg limit. However, most NTR emitters only cover a narrow band, which is unwanted for thermal energy utilization. Here, a material‐informatics framework with a Bayesian optimization (BO) kernel is proposed for designing NTR emitters, which consists of multilayer epsilon‐near‐zero (ENZ) magneto‐optical films on a metal bottom. The optimal structural parameters can be obtained within only 0.5% of all possible structures, demonstrating super‐efficient optimization capability. Additionally, compared to the design method based on the Fresnel formula, the broadband nonreciprocity can be significantly enhanced, with the wavelength‐averaged nonreciprocity improved by 80.4%, which can be attributed to the unequal electromagnetic power dissipation density and mismatched effective impedance at opposite angles. Furthermore, the effects of the dielectric layer, different incident angles, number of layers, and magnetic fields on BO‐based nonreciprocal thermal emitters have been investigated. This study can further promote the development of broadband NTR and can be extended to multilayer structures containing magnetic Weyl semimetals.
Chen et al. (2025) studied this question.