Optimization of Grating Coupler over Single-Mode Silicon-on-Insulator Waveguide to Reach < 1 dB Loss through Deep-Learning-Based Inverse Design

Abstract

Grating couplers are essential components in silicon photonics that facilitate the coupling of light between waveguides and fibers. Optimization of the grating couplers to reach <1 dB loss when coupling to single-mode fibers (SMFs) has been reported in the literature, but this was based on silicon-on-insulator (SOI) waveguides supporting multi-modes. In this paper, using a deep-learning model combined with an inverse-design process, we achieve <1 dB losses for grating couplers implemented over single-mode SOI waveguides, i.e., a maximum efficiency of 80.5% (−0.94 dB) for gratings constrained with e-beam (EB) lithography critical dimension (CD), and a maximum efficiency of 77.9% (−1.09 dB) for gratings constrained with deep ultraviolet (DUV) lithography CD. To verify these results, we apply covariance matrix adaptation evolution strategy (CMA-ES) and find that while CMA-ES yields slightly better results, i.e., 82.7% (−0.83 dB) and 78.9% (−1.03 dB) considering e-beam and DUV, respectively, the spatial structures generated by CMA-ES are nearly identical to the spatial structures generated by the deep-learning model combined with the inverse-design process. This suggests that our approach can achieve a representative low-loss structure, and may be used to improve the performance of other types of nanophotonic devices in the future.

Keywords

• SOI
• gating coupler
• single-mode and multi-mode waveguide
• deep learning
• inverse design