서브메뉴
검색
Novel Optical Metasurfaces: From Machine-Learning Enabled Nanophotonic Design to Colloidal Metamaterials
Novel Optical Metasurfaces: From Machine-Learning Enabled Nanophotonic Design to Colloidal Metamaterials
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211151158
- ISBN
- 9798382757568
- DDC
- 621.3
- 서명/저자
- Novel Optical Metasurfaces: From Machine-Learning Enabled Nanophotonic Design to Colloidal Metamaterials
- 발행사항
- [Sl] : Northwestern University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 163 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-11, Section: B.
- 주기사항
- Advisor: Aydin, Koray.
- 학위논문주기
- Thesis (Ph.D.)--Northwestern University, 2024.
- 초록/해제
- 요약Metasurfaces, 2D counterparts of metamaterial, are planar optical devices that are made of spatially arranged arrays of subwavelength nanostructures. Metasurfaces work by introducing abrupt changes on phase, amplitude, and polarization states of light within sub-wavelength distances and in unique ways. As a result, the metasurface concept, providing a compact and versatile platform for light manipulation, emerged as one of the most promising candidates to replace conventional light manipulation techniques. Metasurfaces have a wide range of applications in imaging, sensing, and communication. To date, numerous optical devices, including flat lenses, polarizers, color filters, and holograms, have been realized by metasurfaces; and various challenges in design and fabrication of metasurfaces have been addressed along the way. However, conventional metasurface approaches are converging to their limitations. Alternative approaches in metasurface design and fabrication are becoming a necessity to achieve advanced applications or properties, such as optical analog computers or active tunability and large device area.Conventional metasurface design process can be divided into two main approaches as forward design and inverse design. Both the forward and the inverse approaches depend on full electromagnetic simulations that are using mesh-based, discretized numeric calculation methods to calculate optical response of nano-photonic structures. Although they provide high accuracy and deterministic results, these methods require high computational costs that significantly increase with structure complexity.Present-day fabrication methods and material platforms are also reaching their limits. Optical metamaterials enable access to effective optical properties that are not found in nature, such as negative or ultrahigh refractive index, by engineering sub-wavelength dielectric and plasmonic structures. Realizing such properties in the near-infrared and visible wavelengths requires complex structures in the nanometer range. Fabrication of such structures is a challenging task despite the tremendous progress in nanofabrication techniques. Lithography-based methods suffer from inherent limitations on structural complexity, spatial resolution, and fabrication throughput. Additive manufacturing techniques, such as 3D printing, can partially address the structural complexity limitations. However, they are limited in terms of available materials and bounded with size constraints. As a result, there is a need for alternative fabrication methods and/or material platforms that can overcome these limitations.As an alternative design approach, deep neural networks (DNNs) assisted design methods are investigated in three progressive steps. In the first step, we modeled all-dielectric cylindrical nanopillars using fully connected DNNs. The wavelength normalization method is introduced as a solution to dimensional mismatch problems between the feature space and output space of the DNNs. As additional advantages of the wavelength normalization method, high prediction accuracy, even with dispersive materials, and spectral generalization capabilities are demonstrated. Following, complex-valued neural networks are used to simultaneously model plasmonic and dielectric metasurface unit cells. Tunable metasurfaces including a resonator and absorber are designed as proof-of-concept demonstrations. Spectral generalization accuracy is also improved. In the last step, convolutional neural networks (CNNs) are used to model arbitrary and freeform metasurface unit cells. A shape generation method is proposed to create a library of freeform and manufacturable cross sections. Generalization of solution for the incident polarization is demonstrated in addition to generalization to dispersive materials and other spectral ranges of interest.Along with these steps, machine assisted and completely machine learning based inverse design and optimization of metasurface unit cells are demonstrated. In the first step, a direct inverse design neural network model is demonstrated, where tandem learning approach is used to overcome non-uniqueness of the solution. In the second step, both DNN-incorporated and completely DNN-based design approaches are demonstrated and compared in terms of final design performance and computational cost. In the last step, as an example of DNN-assisted design, Genetic Algorithm is incorporated with the DNN-based surrogate solver. Objective-first constrained optimization is employed in the latent space of DNN autoencoder in order to design manufacturable metasurfaces. During these projects, several metasurfaces such as metalenses, polarizers, tunable resonators and absorbers, and waveplates are designed and demonstrated through simulations.As alternative material platforms, light-matter interactions for colloidal plasmonic nanoframes and their feasibility as metamaterial/metasurface building blocks are studied. First, optical properties of single-particle and monolayer octahedral nanoframes in comparison to nanoparticles of the same geometry are studied. Single-particle nanoframes exhibited enhanced light-matter interactions and scattering suppression compared to solid nanoparticles. Strong field localization and further suppression of scattering are observed by assembling these structures into close-packed monolayers. As an outcome, polarization insensitive, broadband absorption above %90 is demonstrated from self-assembled monolayers of Pt-Au octahedral nanoframes. Then, optical properties of their 3D superlattices are discussed. We proposed that these structures are RLC circuit analogous supporting both capacitive and inductive resonances. Following, negative effective refractive index, an unnatural and long-desired property, with engineered 3D crystals in the near-infrared wavelengths is verified through electromagnetic simulations and experiments.In the last project of this thesis, we investigated metasurface based all-optical image processing. Due to increasing demand for real-time, continuous data processing, metamaterial and metasurface based all-optical computation techniques emerged as a promising alternative to digital computation. Metasurfaces have been shown to enable real time edge detection with low to no power consumption. However, the previous demonstrations were subjected to the several limitations such as need for oblique-incidence, polarization dependence, need for additional polarizers, narrow operation bandwidth, being limited with processing in 1D, operation with coherent light only, and requiring digital post-processing.All-optical edge detection is chosen as the target application as it is a fundamental step of most image processing tasks. A metasurfaces for 2D isotropic and polarization independent edge detection based on Fourier optics principles that overcome aforementioned limitations and challenges is proposed and demonstrated. The proposed metasurface enables a carefully chosen spatial transmission profile matching the Fourier transformation of the second-order differentiation. Polarization-independent, broadband edge detection with high transmission efficiency under both coherent and partially coherent illumination along the visible frequency range is experimentally confirmed. Edge detection by the same metasurface in near-IR wavelengths is also confirmed through additional simulations and measurements.This thesis comprehensively explores the advancements in metasurfaces in various aspects. Limitations and challenges of the conventional design and fabrication approaches are identified. Alternative techniques are proposed to overcome these challenges. Emerging application areas for metasurfaces are also investigated. Our findings reveal the potential of deep learning methods and colloidal plasmonic nanoframes in metasurface engineering. We envisage that our findings pave the path towards advanced applications and properties for metasurfaces such as, naturally unavailable effective optical properties and real-life image processing.
- 일반주제명
- Electrical engineering
- 일반주제명
- Nanotechnology
- 일반주제명
- Optics
- 일반주제명
- Computer engineering
- 키워드
- Metasurfaces
- 키워드
- Fabrication
- 키워드
- Machine learning
- 기타저자
- Northwestern University Electrical and Computer Engineering
- 기본자료저록
- Dissertations Abstracts International. 85-11B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017161072
■00520250211151158
■006m o d
■007cr#unu||||||||
■020 ▼a9798382757568
■035 ▼a(MiAaPQ)AAI31236752
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621.3
■1001 ▼aTanriover, Ibrahim.▼0(orcid)0000-0003-1207-7016
■24510▼aNovel Optical Metasurfaces: From Machine-Learning Enabled Nanophotonic Design to Colloidal Metamaterials
■260 ▼a[Sl]▼bNorthwestern University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a163 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 85-11, Section: B.
■500 ▼aAdvisor: Aydin, Koray.
■5021 ▼aThesis (Ph.D.)--Northwestern University, 2024.
■520 ▼aMetasurfaces, 2D counterparts of metamaterial, are planar optical devices that are made of spatially arranged arrays of subwavelength nanostructures. Metasurfaces work by introducing abrupt changes on phase, amplitude, and polarization states of light within sub-wavelength distances and in unique ways. As a result, the metasurface concept, providing a compact and versatile platform for light manipulation, emerged as one of the most promising candidates to replace conventional light manipulation techniques. Metasurfaces have a wide range of applications in imaging, sensing, and communication. To date, numerous optical devices, including flat lenses, polarizers, color filters, and holograms, have been realized by metasurfaces; and various challenges in design and fabrication of metasurfaces have been addressed along the way. However, conventional metasurface approaches are converging to their limitations. Alternative approaches in metasurface design and fabrication are becoming a necessity to achieve advanced applications or properties, such as optical analog computers or active tunability and large device area.Conventional metasurface design process can be divided into two main approaches as forward design and inverse design. Both the forward and the inverse approaches depend on full electromagnetic simulations that are using mesh-based, discretized numeric calculation methods to calculate optical response of nano-photonic structures. Although they provide high accuracy and deterministic results, these methods require high computational costs that significantly increase with structure complexity.Present-day fabrication methods and material platforms are also reaching their limits. Optical metamaterials enable access to effective optical properties that are not found in nature, such as negative or ultrahigh refractive index, by engineering sub-wavelength dielectric and plasmonic structures. Realizing such properties in the near-infrared and visible wavelengths requires complex structures in the nanometer range. Fabrication of such structures is a challenging task despite the tremendous progress in nanofabrication techniques. Lithography-based methods suffer from inherent limitations on structural complexity, spatial resolution, and fabrication throughput. Additive manufacturing techniques, such as 3D printing, can partially address the structural complexity limitations. However, they are limited in terms of available materials and bounded with size constraints. As a result, there is a need for alternative fabrication methods and/or material platforms that can overcome these limitations.As an alternative design approach, deep neural networks (DNNs) assisted design methods are investigated in three progressive steps. In the first step, we modeled all-dielectric cylindrical nanopillars using fully connected DNNs. The wavelength normalization method is introduced as a solution to dimensional mismatch problems between the feature space and output space of the DNNs. As additional advantages of the wavelength normalization method, high prediction accuracy, even with dispersive materials, and spectral generalization capabilities are demonstrated. Following, complex-valued neural networks are used to simultaneously model plasmonic and dielectric metasurface unit cells. Tunable metasurfaces including a resonator and absorber are designed as proof-of-concept demonstrations. Spectral generalization accuracy is also improved. In the last step, convolutional neural networks (CNNs) are used to model arbitrary and freeform metasurface unit cells. A shape generation method is proposed to create a library of freeform and manufacturable cross sections. Generalization of solution for the incident polarization is demonstrated in addition to generalization to dispersive materials and other spectral ranges of interest.Along with these steps, machine assisted and completely machine learning based inverse design and optimization of metasurface unit cells are demonstrated. In the first step, a direct inverse design neural network model is demonstrated, where tandem learning approach is used to overcome non-uniqueness of the solution. In the second step, both DNN-incorporated and completely DNN-based design approaches are demonstrated and compared in terms of final design performance and computational cost. In the last step, as an example of DNN-assisted design, Genetic Algorithm is incorporated with the DNN-based surrogate solver. Objective-first constrained optimization is employed in the latent space of DNN autoencoder in order to design manufacturable metasurfaces. During these projects, several metasurfaces such as metalenses, polarizers, tunable resonators and absorbers, and waveplates are designed and demonstrated through simulations.As alternative material platforms, light-matter interactions for colloidal plasmonic nanoframes and their feasibility as metamaterial/metasurface building blocks are studied. First, optical properties of single-particle and monolayer octahedral nanoframes in comparison to nanoparticles of the same geometry are studied. Single-particle nanoframes exhibited enhanced light-matter interactions and scattering suppression compared to solid nanoparticles. Strong field localization and further suppression of scattering are observed by assembling these structures into close-packed monolayers. As an outcome, polarization insensitive, broadband absorption above %90 is demonstrated from self-assembled monolayers of Pt-Au octahedral nanoframes. Then, optical properties of their 3D superlattices are discussed. We proposed that these structures are RLC circuit analogous supporting both capacitive and inductive resonances. Following, negative effective refractive index, an unnatural and long-desired property, with engineered 3D crystals in the near-infrared wavelengths is verified through electromagnetic simulations and experiments.In the last project of this thesis, we investigated metasurface based all-optical image processing. Due to increasing demand for real-time, continuous data processing, metamaterial and metasurface based all-optical computation techniques emerged as a promising alternative to digital computation. Metasurfaces have been shown to enable real time edge detection with low to no power consumption. However, the previous demonstrations were subjected to the several limitations such as need for oblique-incidence, polarization dependence, need for additional polarizers, narrow operation bandwidth, being limited with processing in 1D, operation with coherent light only, and requiring digital post-processing.All-optical edge detection is chosen as the target application as it is a fundamental step of most image processing tasks. A metasurfaces for 2D isotropic and polarization independent edge detection based on Fourier optics principles that overcome aforementioned limitations and challenges is proposed and demonstrated. The proposed metasurface enables a carefully chosen spatial transmission profile matching the Fourier transformation of the second-order differentiation. Polarization-independent, broadband edge detection with high transmission efficiency under both coherent and partially coherent illumination along the visible frequency range is experimentally confirmed. Edge detection by the same metasurface in near-IR wavelengths is also confirmed through additional simulations and measurements.This thesis comprehensively explores the advancements in metasurfaces in various aspects. Limitations and challenges of the conventional design and fabrication approaches are identified. Alternative techniques are proposed to overcome these challenges. Emerging application areas for metasurfaces are also investigated. Our findings reveal the potential of deep learning methods and colloidal plasmonic nanoframes in metasurface engineering. We envisage that our findings pave the path towards advanced applications and properties for metasurfaces such as, naturally unavailable effective optical properties and real-life image processing.
■590 ▼aSchool code: 0163.
■650 4▼aElectrical engineering
■650 4▼aNanotechnology
■650 4▼aOptics
■650 4▼aComputer engineering
■653 ▼aMetasurfaces
■653 ▼aFabrication
■653 ▼aMachine learning
■653 ▼aConvolutional neural networks
■690 ▼a0544
■690 ▼a0652
■690 ▼a0752
■690 ▼a0464
■71020▼aNorthwestern University▼bElectrical and Computer Engineering.
■7730 ▼tDissertations Abstracts International▼g85-11B.
■790 ▼a0163
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17161072▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


