1. Introduction
Metamaterials—artificially structured media with tailored subwavelength features—are redefining the limits of electromagnetic and thermal control.
Their ability to precisely tailor how light and heat move, interact, and are directed enables unprecedented performance in solar energy harvesting, thermophotovoltaics, passive radiative cooling, and thermal management in smart buildings. Yet, despite their transformative potential, the design of high-performance metamaterials remains an intractable challenge: the vast, high-dimensional parameter space and strong multi-physics coupling make traditional optimisation prohibitively slow and limited in scope.
In reality, modern lithography can produce essentially arbitrary 2D or 3D shapes, aperiodic arrangements, and multiscale combinations without added fabrication cost — yet this flexibility is rarely exploited because human-guided design cannot feasibly search such an immense, non-convex parameter space. The result is a performance ceiling defined not by physics or fabrication limits, but by the constraints of our design tools and workflows. This project is not just an acceleration of current methods: by combining high-fidelity experimental data with AI-enabled inverse design, it will unlock solutions beyond human intuition, expanding exploration from the narrow confines of conventional resonator arrays to the full diversity of patterns that lithography can offer.
2. Research Aims & Challenges
The research will develop next-generation AI-driven inverse design frameworks for functional metamaterials in sustainable energy systems.
The PhD project will aim to address three critical technical challenges:
Multi-physics, multi-objective optimisation — simultaneously maximising optical/thermal performance, structural robustness, and manufacturability, incorporating fabrication constraints directly into the optimisation loop.
Generalisation and robustness — developing AI models capable of extrapolating to unseen parameter regimes, incorporating uncertainty quantification to mitigate performance degradation under fabrication imperfections.
Design–fabrication–characterisation integration — establishing a closed-loop pipeline in which AI-generated designs are fabricated via scalable nanomanufacturing techniques (e.g., nanoimprint lithography, roll-to-roll processing), optically and thermally characterised, and iteratively refined via active learning.
3. Methodology
Datasets will be generated using finite-difference time-domain (FDTD) full-wave electromagnetic simulations, calibrated against experimental measurements to ensure high fidelity and physical realism. These datasets will then be used to train advanced AI models, including deep generative architectures (e.g. conditional GANs), reinforcement learning agents, and physics-informed neural networks. The resulting framework will deliver orders-of-magnitude acceleration in the exploration and optimisation of complex metamaterial design spaces, uncovering device architectures and performance regimes that lie beyond the reach of traditional human-guided design.
The project sits at the convergence of computational photonics, multi-physics simulation, and machine learning, leveraging recent advances in AI for inverse design (Liu et al., Nature Photonics, 2021; Malkiel et al., Light: Science & Applications, 2019; So et al., Nature Computational Science, 2022) and physics-informed learning (Raissi et al., Journal of Computational Physics, 2019). It will generate both fundamental advances in AI-enhanced materials design methodology and high-impact prototypes for sustainable energy applications. To accelerate translation and maximise societal benefit, we will engage with industrial partners across different sectors, including Jaguar Land Rover , Malvern Optical and Pilkington, to co-create application contexts such as smart windows and energy-efficient building envelopes.