Artificial intelligence is emerging as a powerful design engine for terahertz (THz) metamaterial devices (meta-devices), according to a review published in PhotoniX Synergy . The review examines how machine learning, evolutionary algorithms, deep learning, generative models, and large language models are being used to design THz meta-devices are artificial structures that manipulate THz waves for applications ranging from 6G wireless communication and high-resolution imaging to biosensing, spectroscopy, optical computing, and security inspection.
THz waves, typically defined between 0.1 and 10 THz, offer high penetration, low photon energy, and strong sensitivity to material properties. However, designing high-performance THz meta-devices is challenging because practical targets such as low loss, broad bandwidth, polarization selectivity, and fabrication constraints correspond to a high-dimensional design space. Multiple geometries can produce similar spectral responses, a one-to-many mapping problem, and small geometric deviations can degrade performance. Conventional parameter sweeps and local optimization are therefore computationally expensive and may miss feasible solutions.
The review organizes recent progress around device families, AI methods, and applications. It covers meta-absorbers, meta-modulators, meta-filters, meta-sensors, meta-detectors, and multifunctional devices, and discusses how AI can accelerate both forward optimization from structural parameters and inverse design from target electromagnetic responses. Representative AI approaches include random forests, artificial neural networks, convolutional neural networks, evolutionary and swarm optimization, reinforcement learning, physics-informed neural networks, generative diffusion models, transformers, and large language models.
The authors note that generative models are particularly promising for inverse topology design, where a single target spectrum can be matched by many distinct geometries. Diffusion models and latent-space generative frameworks can produce multiple candidate designs and can incorporate fabrication constraints, helping to avoid physically invalid or non-manufacturable structures. Large language models and multi-agent frameworks are also being explored for “chat-to-chip” workflows, in which natural-language prompts are translated into design constraints and candidate metasurface topologies.
Despite rapid progress, the review identifies several challenges. AI-based meta-device design still faces significant computational pressure as datasets and network sizes grow. Many deep-learning models lack interpretability, which can limit trust in sensitive applications. Current models often generalize poorly to new structures or multifunctional devices, and high-quality THz experimental datasets remain limited because of costly measurements and complex fabrication. The authors highlight physics-preserving data augmentation, active learning, transfer learning, and physics-informed neural networks as strategies to reduce data requirements and improve physical validity.
Looking forward, the review outlines directions including physics-informed and fabrication-aware models, multi-scale analysis, adaptive algorithms, and AI-assisted optical computing hardware. Platforms such as diffractive neural networks, optical neural networks, and MAFT-ONN processors could accelerate full-wave simulation, design optimization, and real-time signal processing for THz systems, with potential benefits for 6G cognitive radio and edge inference. The authors also point to emerging intersections between metasurfaces and quantum technologies, such as metasurface-based optical tweezer arrays for neutral-atom quantum computing.
“AI is changing how we explore the high-dimensional design space of THz meta-devices,” said Yu-Sheng Lin, corresponding author and professor at Aerospace Information Technology University. “The next challenge is to embed physical laws, fabrication constraints, and interpretability into these models so that intelligent design can translate into practical THz systems.”
See the article:
A review: the development of terahertz meta-devices using artificial intelligence technology in advanced optoelectronic applications, PhotoniX Synergy (2026) 1:11. https://doi.org/10.1007/s44519-026-00004-8
PhotoniX Synergy
Literature review
Not applicable
A review: the development of terahertz meta-devices using artificial intelligence technology in advanced optoelectronic applications
10-Aug-2026
Yu-Sheng Lin is an Editorial Board Member of PhotoniX Synergy and was not involved in the editorial review or the decision to publish this article. All authors declare no competing interests.