LLM-guided automatic design of passive, paper-printed metasurfaces

This project aims to develop a Large Language Model (LLM)-based system that automatically designs passive Metasurfaces based on high-level user specifications. The goal is to make advanced electromagnetic (EM) surface design accessible to a broader community, including researchers and engineers without deep expertise in full-wave simulation tools such as Ansys HFSS, Feko, or CST studio suite. Users will provide natural-language requirements such as operating frequency (e.g., X GHz), desired polarization (e.g., linear or circular), and constraints such as minimizing signal gain or attenuation (e.g., 10dB) in a specific direction or scattering at a target angle for a given channel condition. The system will translate these descriptions into parameterized electromagnetic models using analytical equations and heuristic initializations that define unit cell geometry, periodicity, and substrate properties for spectrum spanning FR1, FR3, and newer FR2 up to 120GHz. These initial designs will be passed to a collaborative ML–HFSS Python language pipeline, where automated Monte Carlo simulations iteratively optimize performance metrics such as reflection phase, amplitude, polarization conversion, or radiation pattern shaping. The LLM will guide parameter exploration and interpret simulation feedback, enabling intelligent refinement of the Metasurface geometry. Ultimately, the framework will automatically generate fabrication-ready PCB layouts for the optimized design. This project integrates electromagnetics, machine learning, optimization, and automation, providing undergraduate
researchers with hands-on experience in RF design, computational modeling, and AI-driven system design. The long-term vision is to democratize Metasurface design by transforming it from a highly specialized manual process into an intelligent design system.