Terahertz (THz) radiation, situated in the spectral gap between microwave and infrared frequencies (0.1–10 THz), offers distinct advantages including broadband coverage, low photon energy, and unique penetrability. Given that this regime encompasses the fingerprint vibrational modes of numerous biological macromolecules, it presents transformative potential for high-precision sensing. Metamaterials, comprising periodically arranged subwavelength units, can induce robust electromagnetic coupling under THz excitation, facilitating the development of high-sensitivity, non-destructive THz sensors. Precise alignment between the sensor's resonance frequency and the characteristic absorption peak of the analyte maximizes the light-matter interaction at the interface. However, achieving such precise alignment remains a key challenge. Conventional design paradigms predominantly rely on empirical parameter tuning, which lacks precision in frequency targeting, whereas deep-learning-based inverse designs are often constrained by the requirement for massive training datasets and prohibitive computational costs.
This study proposes an efficient optimization framework for THz metamaterial sensors utilizing the Non-dominated Sorting Genetic Algorithm II (NSGA-II) with an elitist strategy. By employing fast non-dominated sorting to stratify the population and integrating crowding distance metrics for individual selection, the algorithm achieves robust frequency selection and performance optimization. To enhance design flexibility, two surface geometries with different spectral responses were integrated, allowing for multi-dimensional modulation of the electromagnetic properties through geometric parameter tuning. An automated co-simulation platform was established by interfacing MATLAB with CST Microwave Studio. Taking the resonant frequency, peak absorptivity, and quality factor (Q-factor) as multi-objective functions, the algorithm successfully optimized three distinct sensor configurations tailored to specific spectral requirements. Numerical results demonstrate that the resonant peaks of all optimized designs align with the target frequencies within a marginal error of ±0.05 THz, achieving a peak sensitivity of 351.43 GHz/RIU. Based on impedance matching theory and the analysis of surface electric fields and induced current distributions, the underlying physical mechanism is identified as the synergistic effect of electric dipole resonance and magnetic resonance.
The application of NSGA-II to the frequency-selective design of THz metamaterials enables the realization of tunable multi-band responses by merging structures with heterogeneous electromagnetic properties. This optimization framework significantly enhances design efficiency and provides a systematic methodology for developing multi-band sensors tailored to specific molecular vibrational characteristics, holding substantial promise for applications in biosensing and material characterization.