Genetic-algorithm-based deep neural networks for highly efficient photonic device design

Ren, Yangming; Zhang, Lingxuan; Wang, Weiqiang; Wang, Xinyu; Lei, Yufang; Xue, Yulong; Sun, Xiaochen; Zhang, Wenfu

doi:10.1364/prj.416294

Cited by 51 publications

(30 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For instance, the nanophotonics polarization beam splitter was demonstrated with a footprint of only 2.4 × 2.4 μm 2 , [35] and synchronously a compact on-chip wavelength demultiplexer was reported with a footprint of only 2.8 × 2.8 μm 2 . [36] Thereafter, various ultracompact photonic devices were demonstrated, such as polarization rotator, [37] polarization splitterrotator, [38] wavelength demultiplexer, [39,40] power splitter, [41][42][43][44][45] waveguide crossing, [46,47] mode (de)multiplexer, [48,49] bending, [50] nonlinear nanophotonic device, [51] twisted light emitter, [52] and even the equation solver. [53][54][55] The inverse-designed method was also applied for densely integrated waveguides and modules.…”

Section: Introductionmentioning

confidence: 99%

Dielectric Metasurfaces Enabled Ultradensely Integrated Multidimensional Optical System

Zhou

Wang

Gao

et al. 2022

Laser & Photonics Reviews

View full text Add to dashboard Cite

Metasurfaces consisted of subwavelength nanostructures can extremely interact with light and manipulate the characteristics of amplitude, phase, and polarization. In particular, on-chip dielectric metasurfaces have attracted significant attention for optical communication and computing, due to its compact footprint, low loss, and broad bandwidth. Herein, an ultradensely integrated multidimensional optical system with a footprint of only 20 × 30 µm 2 based on inverse-designed dielectric metasurface network, incorporating mode-division multiplexing, and coherent optical communication technologies that can multiply the system capacity is demonstrated. It is assembled by the ultracompact multifunction on-chip metasurface devices, including four-mode demultiplexer, optical hybrid, crossing, and bending, which all have a size of only several micrometers. The inverse-designed work can significantly broaden the integrated device applications of on-chip metasurfaces and pave an alternative way for large-scale high-capacity optical communication system.

show abstract

Section: Introductionmentioning

confidence: 99%

Dielectric Metasurfaces Enabled Ultradensely Integrated Multidimensional Optical System

Zhou

Wang

Gao

et al. 2022

Laser & Photonics Reviews

View full text Add to dashboard Cite

show abstract

“…network can process data effectively and learn rich representations by its strong capacity in dealing with high-dimensional massive data, Impressed by the application prospects of DL, some works have tried Multi-Layer Perceptrons (MLPs) [9][10][11][12][13][14][15][16][17] and Convolutional Neural Networks (CNNs) [9][10][11][12][13]18] to design and optimize optoelectronic devices such as nanoscale lasers.…”

Section: Introductionmentioning

confidence: 99%

POViT: Vision Transformer for Multi-objective Design and Characterization of Nanophotonic Devices

Chen¹,

Li²,

Yu³

et al. 2022

Preprint

View full text Add to dashboard Cite

We solve a fundamental challenge in semiconductor IC design: the fast and accurate characterization of nanoscale photonic devices. Much like the fusion between AI and EDA, many efforts have been made to apply DNNs such as convolutional neural networks (CNN) to prototype and characterize next-gen optoelectronic devices commonly found in photonic integrated circuits (PICs). These prior works generally strive to predict the quality factor (Q) and modal volume (V) of for instance, photonic crystals, with ultra-high accuracy and speed. However, state-ofthe-art models are still far from being directly applicable in the real-world: e.g. the correlation coefficient of V (V coef f ) is only about 80%, which is much lower than what it takes to generate reliable and reproducible nanophotonic designs. Recently, attention-based transformer models have attracted extensive interests and been widely used in CV and NLP. In this work, we propose the first-ever Transformer model (POViT) to efficiently design and simulate semiconductor photonic devices with multiple objectives. Unlike the standard Vision Transformer (ViT), we supplied photonic crystals as data input and changed the activation layer from GELU [1] to an absolute-value function (ABS). Our experiments show that POViT exceeds results reported by previous models significantly. The correlation coefficient V coef f increases by over 12% (i.e., to 92.0%) and the prediction errors of Q is reduced by an order of magnitude, among several other key metric improvements. Our work has the potential to drive the expansion of EDA to fully automated photonic design. The complete dataset and code will be released to aid researchers endeavoring in the interdisciplinary field of physics and computer science 1 .Recently, Deep Learning (DL) have been widely utilized in multiple fields such as medical imaging [2,3], NLP [4,5], autonomous driving [6], face recognition [7] and object detection [8]. Neural

show abstract

“…49,50 Optimization algorithms have been proven to be useful tools in tasks such as detection of qudit states 51 and quantum state engineering. 52,53 Machine learning and genetic algorithms have also found many uses in photonics, 54,55 including the use of generative models, 56 quantum state reconstruction, 57,58 automated design of experimental platforms, [59][60][61] quantum state and gate engineering, 52,53,[62][63][64][65] and the study of Bell nonlocality. [66][67][68] Moreover, genetic algorithms have been employed to design adaptive spatial mode sorters using random scattering processes.…”

Section: Introductionmentioning

confidence: 99%

Dynamical learning of a photonics quantum-state engineering process

Suprano,

Zia,

Polino

et al. 2022

Preprint

View full text Add to dashboard Cite

Experimentally engineering high-dimensional quantum states is a crucial task for several quantum information protocols. However, a high degree of precision in the characterization of experimental noisy apparatus is required to apply existing quantum state engineering protocols. This is often lacking in practical scenarios, affecting the quality of the engineered states. Here, we implement experimentally an automated adaptive optimization protocol to engineer photonic Orbital Angular Momentum (OAM) states. The protocol, given a target output state, performs an online estimation of the quality of the currently produced states, relying on output measurement statistics, and determines how to tune the experimental parameters to optimize the state generation. To achieve this, the algorithm needs not be imbued with a description of the generation apparatus itself. Rather, it operates in a fully black-box scenario, making the scheme applicable in a wide variety of circumstances. The handles controlled by the algorithm are the rotation angles of a series of waveplates and can be used to probabilistically generate arbitrary four-dimensional OAM states. We showcase our scheme on different target states both in classical and quantum regimes, and prove its robustness to external perturbations on the control parameters. This approach represents a powerful tool for automated optimizations of noisy experimental tasks for quantum information protocols and technologies.

show abstract

Genetic-algorithm-based deep neural networks for highly efficient photonic device design

Cited by 51 publications

References 34 publications

Dielectric Metasurfaces Enabled Ultradensely Integrated Multidimensional Optical System

Dielectric Metasurfaces Enabled Ultradensely Integrated Multidimensional Optical System

POViT: Vision Transformer for Multi-objective Design and Characterization of Nanophotonic Devices

Dynamical learning of a photonics quantum-state engineering process

Contact Info

Product

Resources

About