Inverse design of polymorphic reconfigurable metamaterial absorbers based on a dual-input neural network

Abstract

In recent years, deep learning-based design methods for metamaterial absorbers have attracted much attention; however, the problem of structural homogeneity in inverse design constrains their further development. This paper, proposes a metamaterial absorber composed of the phase change material Ge2Sb2Se4Te1 and titanium. To give the metamaterial absorber a richer structure, we divide its Ge2Sb2Se4Te1 layer and top titanium layer into 36 small squares. In a dual-input neural network-based inverse design, this means that metamaterial absorbers with more types of absorption characteristics can be designed. We utilize this approach to design a reconfigurable metamaterial absorber that exhibits a large absorption bandwidth when the Ge2Sb2Se4Te1 layer is in both the crystalline and amorphous. This absorption bandwidth covers the range of solar wavelengths available to humans. Compared with previous research methods, our method eliminates the step of finding the optimal structure. In addition, we have designed metamaterial absorbers with structural diversity and reconfigurability.