The Potential Surface is a science and technology publication supported by Orbital Materials
Many of the staples of science fiction - invisibility cloaks, infinite zoom lenses, or limitless solar energy - rely on manipulating sound, light or electromagnetic waves. This has historically been incredibly hard - ordinary, natural, materials can’t do this.
However, new scientific breakthroughs are starting to make these materials, called “Metamaterials”, a reality. These materials can bend light backwards, resolve details smaller than the wavelength of light, or even bend light around you to make an “invisibility" cloak”.
Unlike conventional materials, metamaterials’ defining characteristics are derived not from the quantum properties of their chemical composition like most advanced materials , but from their precisely engineered internal architecture. One really exciting implication of this is their suitability for AI design and optimisation. It's possible to generate vast quantities of high-fidelity training data through simulation, and the resulting AI predictions can be readily fabricated and validated in the real world.
This unique approach is being explored for a vast range of potential applications, from future 6G network technologies and compact optics for augmented reality to solutions for managing extreme temperatures in space, developing advanced biosensors, and creating more efficient solar panels.
Because of their potential metamaterials have become a strategic focus for governments around the world. China is investing in developing a national platform for metamaterials. In the US, the Office of Naval Research is pursuing a metamaterials program. And in February 2025, the UK government identified metamaterials as a key research priority, recognising their potential to transform critical technologies in ways conventional materials cannot.
In this article, we’ll unpack what metamaterials actually are, explore how AI could accelerate their development, and consider what it will take to get experimentalists and manufacturers on board with using these tools. To wrap up, we'll explore the science behind everyone's favourite sci-fi concept brought to life: invisibility cloaks.
Structure Over Substance: An Introduction to Metamaterials
Metamaterials are a relatively new class of technology. The term “metamaterial” was first coined by Walser in 2001, who was inspired by the first experimental demonstration of a material with a negative refractive index.
A metamaterial is any material whose structure is intentionally designed and engineered at the meso-scale (nano-to centermeters), built from repeating unit cells like those shown in the figure below.
What makes them unique is that their defining properties arise not from the chemistry of the base material, but from the geometry and periodic arrangement of these unit cells. This deliberate architectural design enables the creation of solids with pre-defined, optimised, and exceptional properties often absent in natural materials. While initially designed to manipulate waves, the approach has been extended to:
Extreme stiffness-to-weight ratios: Creating materials that are exceptionally strong yet lightweight, ideal for aerospace and transport.
Wave cloaking or invisibility: Structures that can steer sound, vibration, or even light waves around an object, effectively hiding it.
Negative Poisson’s ratio: Materials that get thicker when stretched, unlike almost everything else we encounter.
Perfect absorption: Where all incoming sound or light energy at certain wavelengths are 100% absorbed, leaving no reflection.
Enhanced optical chirality: The ability to twist the polarisation of light far more strongly than any natural material, which could lead to new types of optical components.
Antimicrobial and antifouling surfaces: Structures that physically disrupt bacteria or prevent biofilm formation.
Metamaterials aren’t tied to one single fabrication method; the choice depends on the scale (nano, micro, meso) and the type of property you’re targeting (optical, mechanical, acoustic, etc). At the nanoscale, optical metamaterials are often fabricated using lithography and thin-film deposition. At the meso-scale, mechanical and acoustic metamaterials are typically produced through high-resolution 3D printing aka additive manufacturing. In some cases, extraordinary bottom-up self-assembly methods are used, where nanoparticles or polymers organise themselves into repeating structures.
Streamlining the Design Process with AI
So, why aren’t metamaterials everywhere? The main bottleneck is design. Creating structures that live up to their theoretical potential has traditionally relied on slow, costly trial-and-error guided by human intuition. This is because, starting with the desired property and working backwards to discover the geometric pattern that produces is very difficult. This is known as the “inverse design” problem.
AI has real promise in solving inverse design problem for metamaterials. The key thing you’d exploit is that it’s far cheaper to calculate the property of a material than it is to find a material with a certain property. This means you can can generate lots and lots “structure-property” pairs through simulation. To train an AI model to do inverse design you can then just “flip” the data - give the model the property input and ask it to predict the structure. Models like this are called “generative models” (related to “generative AI”) - and are incredibly powerful.
“Ground truth" data for metamaterials is generated by calculations such as Finite Element Analysis (FEA) or Finite-Difference Time-Domain (FDTD), which work by sectioning a physical system into a mesh or grid and numerically solving the governing physics equations. They’re very common and used a lot in industry.
Finite Element Analysis of a lattice beam structure with stress distribution (rainbow colormap) and mesh refinement studies from coarse to fine.
In the field of metamaterials, it seems to me that this is one area we could make rapid progress in AI models for advanced materials. The governing physics, while complex, is classical and free from the added complications of quantum effects, making it more tractable for modelling than traditional advanced materials such as semiconductors. Just as importantly, new metamaterial structures can be tested quickly, since many fabrication methods, such as additive manufacturing, are relatively inexpensive and straightforward to implement. The payoff for success is enormous: as highlighted earlier, metamaterials have potential applications across nearly every industry, both current and emerging.
Bridging the Simulation-Reality Gap
From my own experience working in computational science, I know it can be hard to get experimentalists on board with using simulations to guide their research. Many will rightly roll their eyes when they hear that metamaterials designed by AI are primarily evaluated against other simulations, and that these models often operate in an idealised 'perfect vacuum' that doesn't translate to the real world.
These are fair concerns to have, and overcoming this skepticism is the key to adopting AI tools to accelerate research. For an experimentalist or manufacturer to trust an AI model, its designs must be practical and its predictions must be testable. This means ultimately making the models more steerable - being able to incorporate hints and guidance about defects and manufacturing constraints in the design process. There’s a lot of progress in steerable image generation models - so I’m hopeful a lot of the techniques can eventually be transferred here.
Case Study: The Science of an Invisibility Cloak
An object is visible because it perturbs the electromagnetic waves (light) that strike it, creating a shadow and scattering light towards an observer. A true invisibility cloak must therefore guide these waves around a hidden region as if nothing were there, ensuring they emerge on the other side with their original direction and phase fully restored. A useful analogy is a smooth-flowing river parting around a boulder and seamlessly rejoining downstream, leaving no trace of the disturbance.
To achieve this requires unprecedented control over the path of light.
In a vacuum, light travels in a straight line. To bend it, you need to change the properties of the space it's travelling through. The key property that governs how light propagates is the refractive index (n). This index is, in turn, derived from two more fundamental electromagnetic properties of a material:
Electric Permittivity (ϵ) : How a material responds to and is polarised by an electric field.
Magnetic Permeability (μ): How a material responds to and can be magnetised by a magnetic field.
Metamaterials can be engineered so the repeating unit of the structure can have effective values of ϵ and μ that can be positive, negative, near-zero, or anything in between. Crucially, we can make them anisotropic (different values for different directions of wave travel) and spatially variant (the values change from point to point).
This complete control over electromagnetic properties provides the theoretical toolkit to build a cloak. Using a powerful mathematical framework called Transformation Optics, scientists can calculate the precise map of material properties needed to perfectly steer light around a hidden volume.
As cool as the idea is we’re not quite at the stage where we have perfect invisibility cloaks: The resonant nature of these structures means cloaks typically work for only a single frequency (one colour of light), not the entire visible spectrum. Another slight give away for our metamaterial made invisibility cloaks is that the materials used can absorb a small amount of light which would cast a faint shadow (not ideal oops).
However, the very "limitation" of being highly sensitive to a single frequency is precisely what makes this technology incredibly useful for other real-world applications. For example, a metamaterial sensor can be tuned to the exact frequency of a specific molecule's spectral fingerprint, allowing it to detect diseases or explosives while ignoring all background noise. In telecommunications, this same principle allows for highly efficient antennas and filters for 6G networks that are deaf to all interfering signals, saving energy and improving performance.
Thanks so much for reading! If you found this interesting, please consider sharing it and letting us know what you think. The Potential Surface is a publication exploring the frontiers of science, technology and AI. It is support by Orbital Materials. Annabella Wheatley runs the Potential Surface, and is a materials scientist and science writer. You can also find her on Twitter/X
Leave a comment