Abstract
Using machine learning, we optimized an ultrasmall photonic crystal nanocavity to attain a high $Q$. Training data were collected via finite-difference time-domain simulation for models with randomly shifted holes, and a fully connected neural network (NN) was trained, resulting in a coefficient of determination between predicted and calculated values of 0.977. By repeating NN training and optimization of the $Q$ value on the trained NN, the $Q$ was roughly improved by a factor of 10–20 for various situations. Assuming a 180-nm-thick semiconductor slab at a wavelength approximately 1550 nm, we obtained $Q={1},\!{011},\!{400}$ in air; 283,200 in a solution, which was suitable for biosensing; and 44,600 with a nanoslot for high sensitivity. Important hole positions were also identified using the linear Lasso regression algorithm.
© 2020 Optical Society of America
Full Article | PDF ArticleMore Like This
Takashi Asano and Susumu Noda
Opt. Express 26(25) 32704-32717 (2018)
Renjie Li, Xiaozhe Gu, Ke Li, Yaoran Huang, Zhen Li, and Zhaoyu Zhang
Opt. Mater. Express 11(7) 2122-2133 (2021)
André-Pierre Blanchard-Dionne and Olivier J. F. Martin
Opt. Lett. 45(10) 2922-2925 (2020)