Designing Microwave Circuits Using Genetic Algorithms Accelerated by Convolutional Neural Networks

Reduction of solution search time of automatic design of an RF-DC conversion circuit by Genetic Algorithms (GA) is argued in this paper. Accelerating GA by Convolutional Neural Networks (CNN) is proposed. Already simulated circuits' patterns are learned by CNN and electromagnetic simulation in GA is partially replaced by it. This approach reduces circuit generating time by 90% compared to the general GA design method to generate an RF-DC conversion circuit with the same conversion efficiency.