Photonics Research, Volume. 9, Issue 6, B247(2021)
Genetic-algorithm-based deep neural networks for highly efficient photonic device design
Fig. 1. Workflow of the GDNN algorithm developed in this paper.
Fig. 2. Encoding process that uses polar vectors and design rule constrains as a parameter vector to describe the design of a given photonic device.
Fig. 3. Schematic drawing of the DNN models of the forward and inverse design processes.
Fig. 4. Design analyses of a power splitter with splitting ratio of 2:3: (a) the evolution of the qualified population proportion; (b) and (c) the FDTD simulation result of the best devices in the initial population and the final population; (d) the distribution of optical transmission of the initial population.
Fig. 5. GDNN design examples with transmission spectrum and FDTD simulation results: (a) a 1:2 power splitter, (b) a 1:1 power splitter, (c) a TE mode converter, and (d) a broadband power splitter.
Fig. 6. Comparison of GAN and GDNN design results.
|
Get Citation
Copy Citation Text
Yangming Ren, Lingxuan Zhang, Weiqiang Wang, Xinyu Wang, Yufang Lei, Yulong Xue, Xiaochen Sun, Wenfu Zhang. Genetic-algorithm-based deep neural networks for highly efficient photonic device design[J]. Photonics Research, 2021, 9(6): B247
Special Issue: DEEP LEARNING IN PHOTONICS
Received: Dec. 1, 2020
Accepted: Mar. 25, 2021
Published Online: May. 27, 2021
The Author Email: Xiaochen Sun (sunxiaochen@opt.ac.cn), Wenfu Zhang (wfuzhang@opt.ac.cn)