Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learning
Finding an optimal device structure in the vast combinatorial design space of freeform nanophotonic design has been an enormous challenge. In this study, we propose physics-informed reinforcement learning (PIRL) that combines the adjoint-based method with reinforcement learning to improve the sample...
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
De Gruyter
2024-02-01
|
| Series: | Nanophotonics |
| Subjects: | |
| Online Access: | https://doi.org/10.1515/nanoph-2023-0852 |
| _version_ | 1797219093270495232 |
|---|---|
| author | Park Chaejin Kim Sanmun Jung Anthony W. Park Juho Seo Dongjin Kim Yongha Park Chanhyung Park Chan Y. Jang Min Seok |
| author_facet | Park Chaejin Kim Sanmun Jung Anthony W. Park Juho Seo Dongjin Kim Yongha Park Chanhyung Park Chan Y. Jang Min Seok |
| author_sort | Park Chaejin |
| collection | DOAJ |
| description | Finding an optimal device structure in the vast combinatorial design space of freeform nanophotonic design has been an enormous challenge. In this study, we propose physics-informed reinforcement learning (PIRL) that combines the adjoint-based method with reinforcement learning to improve the sample efficiency by an order of magnitude compared to conventional reinforcement learning and overcome the issue of local minima. To illustrate these advantages of PIRL over other conventional optimization algorithms, we design a family of one-dimensional metasurface beam deflectors using PIRL, exceeding most reported records. We also explore the transfer learning capability of PIRL that further improves sample efficiency and demonstrate how the minimum feature size of the design can be enforced in PIRL through reward engineering. With its high sample efficiency, robustness, and ability to seamlessly incorporate practical device design constraints, our method offers a promising approach to highly combinatorial freeform device optimization in various physical domains. |
| first_indexed | 2024-04-24T12:28:10Z |
| format | Article |
| id | doaj.art-09906914e8804a9ab7a3e63e333e1b24 |
| institution | Directory Open Access Journal |
| issn | 2192-8614 |
| language | English |
| last_indexed | 2024-04-24T12:28:10Z |
| publishDate | 2024-02-01 |
| publisher | De Gruyter |
| record_format | Article |
| series | Nanophotonics |
| spelling | doaj.art-09906914e8804a9ab7a3e63e333e1b242024-04-08T07:36:18ZengDe GruyterNanophotonics2192-86142024-02-011381483149210.1515/nanoph-2023-0852Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learningPark Chaejin0Kim Sanmun1Jung Anthony W.2Park Juho3Seo Dongjin4Kim Yongha5Park Chanhyung6Park Chan Y.7Jang Min Seok8School of Electrical Engineering, 34968Korea Advanced Institute of Science and Technology, Daejeon34141, Republic of KoreaSchool of Electrical Engineering, 34968Korea Advanced Institute of Science and Technology, Daejeon34141, Republic of KoreaKC Machine Learning Lab, Seoul06181, Republic of KoreaSchool of Electrical Engineering, 34968Korea Advanced Institute of Science and Technology, Daejeon34141, Republic of KoreaSchool of Electrical Engineering, 34968Korea Advanced Institute of Science and Technology, Daejeon34141, Republic of KoreaKC Machine Learning Lab, Seoul06181, Republic of KoreaSchool of Electrical Engineering, 34968Korea Advanced Institute of Science and Technology, Daejeon34141, Republic of KoreaKC Machine Learning Lab, Seoul06181, Republic of KoreaSchool of Electrical Engineering, 34968Korea Advanced Institute of Science and Technology, Daejeon34141, Republic of KoreaFinding an optimal device structure in the vast combinatorial design space of freeform nanophotonic design has been an enormous challenge. In this study, we propose physics-informed reinforcement learning (PIRL) that combines the adjoint-based method with reinforcement learning to improve the sample efficiency by an order of magnitude compared to conventional reinforcement learning and overcome the issue of local minima. To illustrate these advantages of PIRL over other conventional optimization algorithms, we design a family of one-dimensional metasurface beam deflectors using PIRL, exceeding most reported records. We also explore the transfer learning capability of PIRL that further improves sample efficiency and demonstrate how the minimum feature size of the design can be enforced in PIRL through reward engineering. With its high sample efficiency, robustness, and ability to seamlessly incorporate practical device design constraints, our method offers a promising approach to highly combinatorial freeform device optimization in various physical domains.https://doi.org/10.1515/nanoph-2023-0852metasurfaceadjoint-based methodreinforcement learningphysic-informed neural networkfreeform designinverse design |
| spellingShingle | Park Chaejin Kim Sanmun Jung Anthony W. Park Juho Seo Dongjin Kim Yongha Park Chanhyung Park Chan Y. Jang Min Seok Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learning Nanophotonics metasurface adjoint-based method reinforcement learning physic-informed neural network freeform design inverse design |
| title | Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learning |
| title_full | Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learning |
| title_fullStr | Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learning |
| title_full_unstemmed | Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learning |
| title_short | Sample-efficient inverse design of freeform nanophotonic devices with physics-informed reinforcement learning |
| title_sort | sample efficient inverse design of freeform nanophotonic devices with physics informed reinforcement learning |
| topic | metasurface adjoint-based method reinforcement learning physic-informed neural network freeform design inverse design |
| url | https://doi.org/10.1515/nanoph-2023-0852 |
| work_keys_str_mv | AT parkchaejin sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT kimsanmun sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT junganthonyw sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT parkjuho sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT seodongjin sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT kimyongha sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT parkchanhyung sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT parkchany sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning AT jangminseok sampleefficientinversedesignoffreeformnanophotonicdeviceswithphysicsinformedreinforcementlearning |