Machine learning in design of broadband terahertz absorbers based on composite structures

Abstract

Metasurface absorbers have gained increasing attention in the field of electromagnetic (EM) absorption in recent years. The conventional design method often involves designing various structures and comparing absorption bandwidth and intensity to determine the absorber's structural parameters. The absorption of EM waves by absorbers involves a complicated process of electric field excitation and impedance matching. Although design experience can reduce the number of design iterations, the selection of artificial parameters is often imprecise and can result in design bias. Fortunately, machine learning (ML) algorithms offer the possibility of automating the design of critical parameters for metasurfaces, which can effectively reduce the laborious design process and the high cost of trial and error. In this study, a pyramid-based metasurface absorber (PMA) and an inverted pyramid-based metasurface absorber (IPMA) were designed and optimized using the random forest (RF) algorithm. The results show that the absorption bandwidths of the two absorber models are 2.68 THz and 2.42 THz, respectively, and the mean absolute percentage error (MAPE) is only 0.60%, which is significantly better than other classical ML and deep learning (DL) algorithms. This study offers a new approach for designing complex systems related to EM wave absorption, reflection, and transmission propagation.