Deep learning assisted inverse design of metamaterial microwave absorber

Abstract

To accelerate the design of metamaterial microwave absorbers (MMAs), in this work, we developed a deep neural network model to predict the spectrum based on the known structural parameters at the beginning. Then, a tandem network was constructed, which can predict the geometries of an unknown MMA based on a desired absorption characteristics with a small mean square errors of validation set (8.3 × 10−4). With the help of the tandem network, a dual band absorber that achieves an absorption rate greater than 85% in the range of 5.1-14 GHz was obtained. By comparing with traditional methods, the demonstrated methodology can greatly accelerate the whole process and realize an inverse design.