Bridging coupled-mode theory and machine learning for rapid design of broadband 3-dB directional coupler

With the growing demand for high-speed and high-capacity optical communications, the development of efficient photonic device design methodologies has become increasingly important. In this paper, a neural-network-assisted coupled mode theory method (NN-CMT) is proposed to accelerate the design process. Neural networks are employed to predict the coupling coefficient and phase mismatch, thereby eliminating repeated numerical mode analyses. The proposed method is combined with the covariance matrix adaptation evolution strategy (CMA-ES) to design a 3-dB directional coupler with a 150 nm bandwidth, where the power splitting ratio is maintained within 50 ± 0.5 percentage points. Devices designed using the proposed approach were fabricated and experimentally characterized. The measured results show good agreement with numerical simulations, and maintain equal power-splitting characteristics over an operating bandwidth of approximately 100 nm, demonstrating that the proposed method can significantly reduce the computational cost while maintaining high accuracy. The proposed approach is suitable for rapid optimization and design of complex photonic devices.