A machine learning-based inverse design method for frequency-selective surface microwave absorbers

Electromagnetic compatibility (EMC) and electromagnetic interference (EMI) shielding in the electric vehicle have attracted more and more attention. As an effective solution for EMC and EMI in the electric vehicle, microwave absorbers have attracted more and more attention. In the design of microwave absorbers, more degrees of freedom (DoFs) can enhance performance, but the associated design cost increases. To address this issue, this work proposes a machine learning-based inverse topological design method. Different from the traditional methods, i.e. equivalent circuit methods, the proposed method designs the absorber using topological and decimal variables, offering more DoFs and higher efficiency. Furthermore, mixing on multi-layer perceptrons (MLP-Mixer), is employed to map the desired electromagnetic response to the corresponding absorber structure. To validate the proposed method’s design capability, a predicted design sample is fabricated and measured. Measurements indicate that under normal incidence, the absorber operates within a frequency range of 1.33–7.31 GHz, demonstrating a significant improvement in absolute bandwidth. The relative bandwidth is 138.2%, with a thickness of 0.114${\lambda }_L$, where ${\lambda }_L$ is the wavelength at the lowest operating frequency. At the incidence angle of 45°, both transverse electric (TE) and transverse magnetic (TM) polarizations maintain a broad relative bandwidth.