Performing Intelligent Design of Broadband Pillbox Window for a Terahertz Traveling-Wave Tube by Using Physics-Informed Neural Network and Genetic Algorithm

A physics-informed neural network (PINN) is utilized to advance the research of broadband pillbox window (PW) for a terahertz traveling-wave tube (THz TWT). PINN integrates the physical laws of PW into the neural network training process, thereby enhancing the predictive speed and accuracy between the structure parameters of PW and its S-parameters. Subsequently, the global genetic algorithm (GA) is employed to obtain the structure parameter set best matching the desired S-parameter targets from the huge 5-D parameter-set space. Based on the PINN and GA, a general intelligent PW design method (IPWDM) is developed. With the help of IPWDM, a PW sample operating in 230–260 GHz is quickly obtained within only ~1 h. The predicted S-parameters of this PW sample by PINN clearly show excellent transmission characteristics, which are consistent with those simulated by CST. Experimental results of the PW sample provide strong support for the reliability of IPWDM. These facts demonstrate the superiority of IPWDM over time-consuming simulation softwares. This intelligent design pattern can ease researchers from extensive simulation work and lays a solid foundation for the future development of THz TWTs.