Deep learning techniques for transmissive metasurface design
| dc.contributor.author | Niu, Chen | |
| dc.contributor.examiningcommittee | Jeffrey, Ian (Electrical and Computer Engineering) | |
| dc.contributor.examiningcommittee | Henry, Christopher (Computer Science) | |
| dc.contributor.examiningcommittee | Li, Maokun (Tsinghua University) | |
| dc.contributor.supervisor | Mojabi, Puyan | |
| dc.date.accessioned | 2024-11-22T21:07:13Z | |
| dc.date.available | 2024-11-22T21:07:13Z | |
| dc.date.issued | 2024-11-22 | |
| dc.date.submitted | 2024-11-22T20:17:25Z | en_US |
| dc.degree.discipline | Electrical and Computer Engineering | |
| dc.degree.level | Doctor of Philosophy (Ph.D.) | |
| dc.description.abstract | Electromagnetic (EM) metasurfaces are considered as two-dimensional artificial structures since they are thin with respect to the wavelength of operation. These thin devices are designed to tailor the incoming EM waves into the desired outgoing ones. This transformation is achieved through lattices of engineered subwavelength scattering particles, known as meta-atoms or unit cells. The properties of these unit cells provide the mechanism to control the amplitude, phase, and polarization of the incoming EM waves. It is through this control that the EM radiation pattern of the metasurface can be shaped in terms of different far-field (FF) criteria such as main beam directions and sidelobe levels. However, the current metasurface design process has been complex and iterative, often requiring extensive computational resources. In light of this challenge, this thesis investigates and presents an end-to-end systematic approach for transmissive metasurface design, aims at shorten the design cycle. To achieve this, it develops deep learning (DL) frameworks and integrates them into the two key stages of metasurface design: macroscopic and microscopic. At the macroscopic design stage, the primary goal is to infer the necessary metasurface properties from the desired FF specifications while ensuring that the metasurface remains lossless, passive, and reciprocal. This thesis explores how these requirements can be met through DL techniques, either by addressing them in separate steps or simultaneously in a single step, with the benefits and drawbacks of each approach examined. Subsequently, the microscopic design stage aims to yield physical designs that realize the properties obtained in the macroscopic stage. To achieve this, this thesis develops a DL framework to synthesize metallic traces on three-layered unit cells, which can be fabricated using the printed circuit board (PCB) technology. Finally, this thesis integrates the macroscopic and microscopic DL frameworks to create an end-to-end design pipeline. The performance of this pipeline is verified through full-wave EM simulations and experimental evaluation. | |
| dc.description.note | February 2025 | |
| dc.description.sponsorship | Mitacs Accelerate Entrepreneur | |
| dc.identifier.uri | http://hdl.handle.net/1993/38682 | |
| dc.language.iso | eng | |
| dc.subject | Metasurfaces | |
| dc.subject | Deep learning | |
| dc.subject | Electromagnetic inverse design | |
| dc.subject | Beam shaping | |
| dc.subject | Pattern synthesis | |
| dc.title | Deep learning techniques for transmissive metasurface design | |
| local.subject.manitoba | no | |
| oaire.awardNumber | CGSD3 - 547807 - 2020 | |
| oaire.awardTitle | Alexander Graham Bell Canada Graduate Scholarships - Doctoral | |
| oaire.awardURI | https://www.nserc-crsng.gc.ca/Students-Etudiants/PG-CS/CGSD-BESCD_eng.asp | |
| project.funder.identifier | NSERC: https://doi.org/10.13039/501100000038 | |
| project.funder.name | Natural Sciences and Engineering Research Council of Canada |