We use cookies to ensure that we give you the best experience on our website. By continuing to browse this repository, you give consent for essential cookies to be used. You can read more about our Privacy and Cookie Policy.

Durham e-Theses
You are in:

Design and Optimisation of Optical Metasurfaces Using Deep Learning

GHASEMI, AMIR (2024) Design and Optimisation of Optical Metasurfaces Using Deep Learning. Doctoral thesis, Durham University.



This thesis centres on the design, processing, and fabrication of tunable optical metamaterials. It incorporates physics-based simulation, deep learning (DL), and thin film fabrication techniques to offer a comprehensive exploration of the field of optical metamaterials. Placing stiff resonators on a flexible substrate is a common type of mechanically tunable metasurface, whose optical responses are tuned by dynamically adjusting the spacing between resonators by applying mechanical force. However, the significant modulus mismatch between materials causes stress concentration at the interface, leading to crack propagation and delamination at lower strain levels (20-50%), and limiting the optical tunability of the structure. To address this challenge, we propose two designs to manipulate stress distribution. Under mechanical force, the structure enables localised deformation, redirecting stress from critical areas. This mechanism minimises the accumulation of stress in the interface, thereby diminishing the risk of material failure and improving stretchability up to 120% compared to traditional designs. This extreme stretchability leads to a 143 nm resonance shift, which is almost twice as large as that of conventional geometry. A universal machine learning (ML)-based approach was developed to optimise the metasurface design across three key aspects: geometric parameters, material development, and free-form shape configuration. In design parameters optimisation, a fully connected neural network (FCNN) was developed with a mean absolute error (MAE) of 0.0051, recommending a single geometry with a 104 order of magnitude decrease in computational time when compared to finite element method (FEM) simulations used for data generation. The suggested structure provides extensive coverage of the colour space, encompassing 27.65% of the standard RGB (sRGB) space. For the materials development part, an inverse design (ID) network was combined with effective medium approximation (EMA), navigating infinite materials composition space to identify new compositions for custom applications. The last network was tasked to explore boundless free-form shape space to propose the one for the on-demand optical properties with MAE of 0.21. The accuracy of all networks was experimentally validated.

Item Type:Thesis (Doctoral)
Award:Doctor of Philosophy
Keywords:Active Metamaterials, Inverse Design, Machine Learning, Mechanically Tunable Metasurface, Materials Discovery
Faculty and Department:Faculty of Science > Engineering, Department of
Thesis Date:2024
Copyright:Copyright of this thesis is held by the author
Deposited On:16 Apr 2024 12:56

Social bookmarking: del.icio.usConnoteaBibSonomyCiteULikeFacebookTwitter