Design and machine learning driven optimization of tunable periodic cross-diamond terahertz metamaterial absorber

A tunable absorber is proposed using a periodic cross-diamond (PCD) resonator array integrated with vanadium dioxide (VO 2 ) for terahertz applications. Electromagnetic simulations shows absorption peaks at 2.83 THz and 8.28 THz, with a full-width at half-maximum of 7.43 THz, corresponding to a relative absorption bandwidth of 126.7%. The absorption mechanism is analyzed through magnetic and electric field distributions along with parametric analyses are conducted to assess the effect on the absorber’s performance. To improve design optimization and reduce computational cost, regression-based machine learning (ML) models K-Nearest Neighbors, XGBoost, and Random Forest are employed to predict absorptivity across intermediate frequencies. The KNN model achieves excellent performance with an R 2 of 0.9997 and an MAE of 0.0002, reducing the required CST simulations by nearly 50 % and significantly accelerating the EM design process for terahertz applications.