Intelligent Forward-Wave Amplifier Design With Deep Learning and Genetic Algorithm

In this work, exploitation of artificial intelligence algorithms for the design of forward-wave amplifier is proposed and discussed. The multilayer perceptron (MLP), a quintessential example of deep learning, is employed to get the cost function, i.e., the relationship between the geometrical parameter set and dispersion characterization from only a small proportion of the parameter sets. Due to the structural complexity, there did not exist a closed-form solution and the neural network shows its superiority compared to the time-consuming numerical calculation. Subsequently, the genetic algorithm (GA) is used to find the parameter sets best matching the goals that broadband beam-wave interaction at specific voltages and even specific phases or frequencies, from the huge 4-D parameter-set space. The intuitive goals can be easily accessible, as long as the goals are within the ability of the parameter-set space and converting the goals into the goal functions for GA with some attempts and adjustments. As a practical example, the recently proposed defect photonic crystal waveguide is utilized to demonstrate the design pattern. Through almost full-automatic design with little judgment and analysis, a response of 14-GHz 3-dB bandwidth and 25-dB gain at the 168.5-GHz central frequency has been realized, indicating the effectiveness of the design pattern. This design pattern will ease researchers from numerous simulation work and help to get deeper insight of the structure.