Ultra-efficient machine learning design of nonreciprocal thermal absorber for arbitrary directional and spectral radiation

Abstract

Development of nanophotonics has made it possible to control the wavelength and direction of thermal radiation emission, but it is still limited by Kirchhoff's law. Magneto-optical materials or Weyl semimetals have been used in recent studies to break the time-reversal symmetry, resulting in a violation of Kirchhoff's law. Currently, most of the work relies on the traditional optical design basis and can only realize the nonreciprocal thermal radiation at a specific angle or wavelength. In this work, on the basis of material informatics, a design framework of a multilayer nonreciprocal thermal absorber with high absorptivity and low emissivity at any arbitrary wavelength and angle is proposed. Through a comprehensive investigation of the underlying mechanism, it has been discovered that the nonreciprocal thermal radiation effect is primarily attributed to excitation of the cavity mode at the interface between the metal and the multilayer structure. Moreover, the impact of factors, such as layer count, incidence angle, extinction coefficient, and applied magnetic field on nonreciprocal thermal radiation, is thoroughly explored, offering valuable insights to instruct the design process. Additionally, by expanding the optimization objective, it becomes feasible to design fixed dual-band or even multi-band nonreciprocal thermal absorbers. Consequently, this study offers essential guidelines for advancing the control of nonreciprocal thermal radiation.