Simplified Deep Learning Driven Design of Blue Light Grating Couplers

ABSTRACT In recent years, visible light photonic integrated circuits have emerged as a promising solution for atomic, quantum, and biosensing applications. However, there remains a lack of comprehensive research on fundamental photonic integrated devices like grating couplers. In this paper, a simplified deep neural network (DNN) incorporating a multi‐head self‐attention mechanism is developed to optimize the design of blue light grating couplers. The simplified DNN model takes only 3 min to complete the training process on a home‐grade computer, facilitating efficient access to the predicted response spectra of the device. The trained DNN model has a custom hybrid error of less than 0.03, with a prediction accuracy as high as 0.95 within a precise range of ±0.02, while the maximum global error is below 0.04. Experimental validation of the DNN‐designed structure achieved a coupling efficiency of ‐5.5 dB at 445 nm with a −3 dB bandwidth of 13 nm, which is in good agreement with the simulation. This work presents a reliable design solution for blue light band grating couplers, laying the foundation for incorporating machine learning techniques in designing visible light photonic devices.