Optimization and inverse design of nanoscale laser cavities using Deep Reinforcement Learning. Here is link to our paper: https://www.degruyter.com/document/doi/10.1515/nanoph-2022-0692/html
Photonics inverse design relies on human experts to search for a design topology that satisfies certain optical specifications with their experience and intuitions, which is highly labor-intensive, slow, and sub-optimal. Machine learning has emerged as a powerful tool to automate this inverse design process. However, supervised or semi-supervised deep learning is unsuitable for this task due to: 1) a severe shortage of available training data due to the high computational complexity of physics-based simulations and a lack of open-source datasets; 2) the issue of one-to-many mapping or non-unique solutions; 3) the need for a pre-trained neural network model. Here, we propose Learning to Design Optical-Resonators (L2DO) to leverage Reinforcement Learning (RL) that learns to autonomously inverse design nanophotonic laser cavities without any prior knowledge while retrieving unique design solutions. L2DO incorporates two different algorithms– Deep Q-learning and Proximal Policy Optimization. We evaluate L2DO on two laser cavities: a long photonic crystal (PC) nanobeam and a PC nanobeam with an L3 cavity, both popular candidates for semiconductor lasers such as PCSELs. Trained for less than 150 hours on limited hardware resources, L2DO has achieved comparable or even better performance than human experts working the same task for over a month. L2DO first learned to meet the required maxima of Q-factors and then proceeded to optimize some additional good-to-have features (e.g., resonance frequency, modal volume). Compared with iterative human designs and inverse design enabled by supervised learning, L2DO can achieve over two orders of magnitude higher sample-efficiency without suffering from the three issues above. This work marks the first step towards a fully automated AI framework for photonics inverse design.
Deep Q learning (DQN) and Proximal Policy Optimization (PPO).
For PPO, Ray Rllib was used. For DQN, code was written from scratch w/o using Rllib.
For both cases, pytorch was used as the ML library and OpenAI gym was used for building the envs.
Lumerical FDTD was used as the environment for simulating nanobeams.
.
├── ...
├── optim_PhC_dqn.py (or optim_PhC_ppo.py) # main scripts
│ ├── envs # RL environment scripts
│ ├── src # source scripts for nanobeams
├── README.md # Readme file
├── FDTD_PPO-main # Another implementation of PPO
└── ...
Here is the step-by-step instruction for how to reproduce the code in this repo:
detailed implementation procedure of L2DO.pdf
To run either code in the terminal, simply type:
python optim_PhC_dqn.py (or optim_PhC_ppo.py) | tee run.log
Note: you'll need your own .fsp FDTD simulation file in order for this repo to work. To request proprietary simulations files, contact zhangzy@cuhk.edu.cn.
Finally, since RL doesn't require any training data, there's no dataset used or included here.
If you used our code for your research, please consider citing the paper as:
@article{li2023deep2,
options={maxbibnames=99},
title={Deep reinforcement learning empowers automated inverse design and optimization of photonic crystals for nanoscale laser cavities},
author={Li, Renjie and Zhang, Ceyao and Xie, Wentao and Gong, Yuanhao and Ding, Feilong and Dai, Hui and Chen, Zihan and Yin, Feng and Zhang, Zhaoyu},
journal={Nanophotonics},
volume={12},
number={2},
pages={319--334},
year={2023},
publisher={De Gruyter}
}
PS: for a different implementation of PPO written by one of my colleagues, see the FDTD_PPO-main folder, or visit: https://github.com/Arcadianlee/Photonics_RL
This other implementation was found to be less efficient than the one implemented in optim_PhC_ppo.py.