Enhancing Adjoint Optimization-Based Photonic Inverse Design with Explainable Machine Learning
نویسندگان
چکیده
A fundamental challenge in the design of photonic devices, and electromagnetic structures more generally, is optimization their overall architecture to achieve a desired response. To this end, topology or shape optimizers based on adjoint variable method have been widely adopted due high computational efficiency ability create complex freeform geometries. However, functional understanding such remains black box. Moreover, unless space high-performance devices known advance, gradient-based can get trapped local minima valleys saddle points, which limits performance achievable through inverse process. elucidate relationships between device nanoscale structuring while mitigating effects trapping, we present an framework that combines optimization, automated machine learning, explainable artificial intelligence. Integrated with numerical simulation method, our reveals structural contributions toward figure-of-merit (FOM) interest. Through explanation-based reoptimization process, information then leveraged minimize FOM further than obtained alone, thus overcoming optimization’s minima. We demonstrate context waveguide splitter 39 74% increases relative state-of-the-art optimization-based across range telecom wavelengths. Our results highlight learning strategies substantially extend enhance capabilities conventional, algorithm revealing deeper insights into algorithm’s designs.
منابع مشابه
Comparative Analysis of Machine Learning Algorithms with Optimization Purposes
The field of optimization and machine learning are increasingly interplayed and optimization in different problems leads to the use of machine learning approaches. Machine learning algorithms work in reasonable computational time for specific classes of problems and have important role in extracting knowledge from large amount of data. In this paper, a methodology has been employed to opt...
متن کاملEnhancing Consistency Based Diagnosis with Machine Learning Techniques
This paper propose a diagnosis architecture that integrates consistency based diagnosis with induced time series classifiers, trying to combine the advantages of both methods. Consistency based diagnosis allows fault detection and localization without prior knowledge of the device fault modes. Machine learning techniques are able to induce time series classifiers that may be used to identify fa...
متن کاملAdjoint-Based Low-Boom Design with Cart3D
We consider analysis and design of low sonic-boom aircraft through the use of an inviscid, embedded-boundary Cartesian mesh method. Adjoint error estimation and adaptive meshing are used in the analysis portion of this study to determine resolution requirements of the computational domain, while adjoint-based gradients in conjunction with a parallel optimization framework are used for design. T...
متن کاملOptimization - Based Machine Learning and Data Mining
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: ACS Photonics
سال: 2022
ISSN: ['2330-4022']
DOI: https://doi.org/10.1021/acsphotonics.1c01636