Single-narrowband graphene-based metamaterial absorber for bio-sensing applications with absorption prediction using machine learning analysis

• In this study, a new metamaterial perfect absorber structure for hemoglobin concentration sensing is designed and analyzed. • The structure proposed here contains three different layers. The bottom layer is a metal gold layer; the middle one is a SiO 2 layer, and the top layer is composed of four sets of circular disk graphene resonators connected to each other and a ring graphene resonator. • The FDTD simulation results show that the structure produces a narrowband absorption peak at 1.755 THz in an almost wide frequency range of 1–3 THz with the absorptivity and FWHM values of 99.9 % and 125 GHz, respectively. • The values of two important parameters of sensitivity and figure of merit for the structure are 601 nm/RIU and 4.9 RIU -1 , respectively. • To predict the behavior of the metamaterial perfect absorber, the Extreme Randomized Tree regression model as a machine learning method is used. This method can decrease the resources by 90 %. Bio-sensors are among the most suitable candidates for detecting hemoglobin (Hb) molecules in human blood due to their high accuracy, fast performance, high reliability, low cost, and high sensitivity. Accordingly, a new metamaterial (MTM) perfect absorber (PA) for Hb concentration sensing is presented. The structure contains three different layers. The bottom layer is a metal gold layer; the second (middle) one is a SiO 2 layer, and the top layer is composed of four sets of circular disk graphene resonators connected to each other and a ring graphene resonator. Due to the symmetric topology of the MTM PA, it is a polarization-insensitive structure. The finite difference time domain (FDTD) method is utilized for numerical simulations of the MTM PA. A single narrowband absorption peak at 1.755 THz in an almost wide frequency range of 1–3 THz is created in the absorption curve. The absorptivity and full width at half maximum (FWHM) values of the created single band are 99.9 % and 125 GHz, respectively. Due to the single-mode and narrowness of the generated mode, the performance of the MTM PA as a sensor for Hb concentration sensing is also investigated. The sensitivity of 601 nm/RIU and FoM of 4.9 RIU -1 are obtained. Furthermore, to predict the behavior of the MTM PA, the Extreme Randomized Tree regression model (ERTRM) as a machine learning method is used. This method can decrease the resources by 90 %. The simple structure of the presented topology, and its high sensitivity make it a suitable option for biosensing applications.