Genetic-algorithm-based deep neural networks for highly efficient photonic device design
Splitter
DOI:
10.1364/prj.416294
Publication Date:
2021-03-30T21:00:05Z
AUTHORS (8)
ABSTRACT
While deep learning has demonstrated tremendous potential for photonic device design, it often demands a large amount of labeled data to train these neural network models. Preparing requires high-resolution numerical simulations or experimental measurements and cost significant, if not prohibitive, time resources. In this work, we present highly efficient inverse design method that combines networks with genetic algorithm optimize the geometry devices in polar coordinate system. The significantly less training compared previous methods. We implement several ultra-compact silicon photonics challenging properties including power splitters uncommon splitting ratios, TE mode converter, broadband splitter. These are free features beyond capability photolithography generally compliance fabrication rules.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (34)
CITATIONS (64)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....