Three-dimensional modeling of near-field beam profiles from grating couplers using a deep neural network

Abstract

Integrated silicon photonics (SiPh) gratings have been widely studied for the optical addressing of trapped ions. As the form factor of ion traps reduces, the ion-trapping height decreases and may unavoidably fall into the reactive near-field region of SiPh gratings. In this study, a deep neural network (DNN) modeling technique is developed as a rapid alternative to generate reactive near-field beam profiles of light coupled from SiPh gratings, as compared to the conventional finite-difference time-domain (FDTD) technique. The training of the optimized DNN model took 14 minutes, and the generation of beam profiles from the trained model took a few seconds. The time required for model training and beam profile generation is significantly faster than FDTD simulation, which may take up to 2 hours. The generated beam achieved accuracy values of up to 75%. Despite the relatively longer model training duration, it is possible to reuse the trained DNN model to generate beam profiles from gratings with several design variations. In short, this work demonstrates an alternative DNN-assisted technique to rapidly generate beam profiles in the reactive near-field region.