AI Decodes Metasurface Genome, Achieving 3% Precision with Meta-GPT Technology

The design of nanoscale materials that control light, known as photonic metasurfaces, presents a significant challenge for scientists, demanding both precision and creativity. Now, David Dang, Stuart Love, and Meena Salib, alongside colleagues, are pioneering a new approach that combines artificial intelligence with the fundamental laws of physics. Their work introduces METASTRINGS, a symbolic language that describes these complex nanostructures as simple text, and builds upon this with Meta-GPT, a powerful AI model trained to understand and generate metasurface designs.

The team demonstrates that Meta-GPT can accurately predict how a metasurface will interact with light, achieving exceptional precision and generating novel prototypes with responses closely matching desired characteristics, representing a crucial step towards automating the discovery of advanced photonic materials and ultimately unlocking a ‘metasurface genome project’.

Machine Learning Accelerates Metamaterial Design Exploration Researchers have pioneered a new approach to designing metamaterials, artificial materials engineered to have properties not found in nature, by combining machine learning with reinforcement learning. To address the time-consuming and often trial-and-error nature of traditional metamaterial design, the team developed METASTRINGS, a string-based language that allows the model to generate designs in a compact and manageable way. The core of this new method is Meta-GPT, a generative model trained to create METASTRINGS representing metamaterial designs. This model learns to generate designs and then uses reinforcement learning to refine them, optimizing for specific performance criteria such as light absorption. A technique called chain-of-thought prompting improves the model’s reasoning ability, allowing it to explain its design choices.

The team created a dataset of metamaterial designs and their electromagnetic properties using computer simulations, then trained and tested Meta-GPT on this data.

Results demonstrate that this framework successfully automates the design of metamaterials with desired properties, and in some cases, the generated designs outperform traditional designs. The model also demonstrates the ability to generalize to new design spaces and optimize for different performance criteria. Experimental validation confirms the accuracy of the framework, as fabricated structures show good agreement with simulation results.

This research highlights the potential of artificial intelligence to revolutionize metamaterial design, offering a powerful new tool for materials scientists and engineers.

Photonic Metasurface Design with Generative AI Scientists have developed a new method for designing photonic metasurfaces, nanoscale structures that control light, by leveraging the power of generative artificial intelligence.

The team introduced METASTRINGS, a symbolic language that represents these complex structures as textual sequences, mirroring the way molecules are represented in chemistry. This language encodes crucial information about materials, geometries, and lattice configurations, establishing a framework that connects human understanding with automated design principles. Building on this foundation, the researchers developed Meta-GPT, a generative model trained on a large dataset of METASTRINGS. The model underwent further refinement using supervised learning, reinforcement learning, and a chain-of-thought strategy, incorporating intermediate reasoning steps to enhance design accuracy and performance. Unlike previous approaches, Meta-GPT leverages the inherent structure of METASTRINGS, enabling more efficient reasoning about photonic structures while maintaining human intelligibility. Experiments involved generating diverse metasurface prototypes using Meta-GPT and then fabricating these structures for experimental verification. The resulting designs achieved high accuracy, with a mean-squared spectral error of less than 3% and greater than 98% syntactic validity. Crucially, experimentally measured responses closely matched the target spectra, validating the effectiveness of the language-driven design paradigm and establishing a new approach to scalable and interpretable metasurface design. This work represents a breakthrough in AI-driven photonics, paving the way for similar linguistic representations in other physical sciences. Meta-GPT Designs High-Fidelity Photonic Nanostructures Researchers have developed a new approach to designing photonic nanostructures, nanoscale structures that control light, by combining a novel symbolic language with a powerful artificial intelligence model.

The team introduced METASTRINGS, a language that represents these structures as textual sequences, mirroring the way molecules are represented in chemistry. This language encodes materials, geometries, and lattice configurations, establishing a framework for capturing the structural hierarchy of metasurfaces and enabling a new approach to AI-driven photonics. The work introduces Meta-GPT, a foundation model trained on METASTRINGS and refined using supervised learning, reinforcement learning, and chain-of-thought reasoning. Experiments demonstrate that Meta-GPT achieves high accuracy, with a mean-squared spectral error of less than 3% across various design tasks, while maintaining greater than 98% syntactic validity. This means the model generates diverse metasurface prototypes whose experimentally measured optical responses closely match their target spectra, confirming the model’s ability to accurately predict and design optical behavior. The generated designs are not merely accurate, but also interpretable, as the textual representation allows for human understanding of the underlying structure and function. This approach moves beyond previous methods that relied on natural language processing or simple parameter lists, instead utilizing a language specifically designed for photonic structures. This allows Meta-GPT to reason more efficiently about photonic designs while remaining readily understandable to human researchers. The breakthrough delivers a rigorous foundation for a “metasurface genome project”, enabling scalable and interpretable design of photonic structures and representing a significant step toward automated discovery in the field. Metasurface Design via Language Model Training This work introduces METASTRINGS, a new textual language for representing photonic metasurfaces, and demonstrates its effectiveness when integrated with a large language model called Meta-GPT. By encoding materials, geometries and lattice configurations as symbolic sequences, METASTRINGS provides a framework connecting human interpretability with automated design, mirroring approaches used in molecular chemistry. The researchers then trained Meta-GPT on this language, further refining its capabilities with physics-informed learning techniques. The resulting model achieves high accuracy in generating metasurface designs, consistently matching target spectral responses with less than 3% mean-squared error while maintaining a high degree of syntactic validity. Different training approaches, including reinforcement learning and chain-of-thought reasoning, yielded complementary strengths, with reinforcement learning prioritizing optimization and consistency, and chain-of-thought enhancing design diversity and interpretability. These results demonstrate that Meta-GPT can learn the fundamental rules governing light-matter interactions through the METASTRINGS framework, moving beyond simple pattern recognition towards physics-aware design. Future work aims to expand the framework to encompass a wider range of photonic systems, including beam-steering and holographic devices, by incorporating additional layers and polarization descriptors into the language. This scalable approach offers a promising route towards fully automated, language-based design in photonics, potentially paving the way for a comprehensive “metasurface genome project”. 👉 More information 🗞 Meta-GPT: Decoding the Metasurface Genome with Generative Artificial Intelligence 🧠 ArXiv: https://arxiv.org/abs/2512.12888 Tags: