WSC: Applications of Generative AI and LLMs in Microwave Engineering

Presentation Abstract
Generative AI and Large Language Models (LLMs) are beginning to change how electromagnetic and RF systems are specified, synthesized, and verified. Although these tools are common in software and data science, their use in microwave engineering is nascent and requires careful, physics-aware evaluation. This full-day workshop spotlights state-of-the-art methods that connect AI generation to EM reality, moving beyond proofs-of-concept toward validated models and workflows engineers can use today. Technical content centers on three pillars — (1) Inverse EM / spec-to-layout and end-to-end design: “Generative AI Methods for Wireless Propagation Prediction” (Costas Sarris) shows diffusion and GANs for real-time, generalizable indoor propagation maps and super-resolution; “AI-enabled End-to-End RF and RFIC Design” (Kaushik Sengupta) discusses inverse-design and generative AI approaches for automated synthesis of complex RF passives, multi-port elements, antennas, and spec-to-GDS RFIC flows combining reinforcement learning and inverse design; “Empowering Optimal Design of RF Devices by Generative AI” (Dominique Baillargeat and Francisco Chinesta) introduces rank-reduction autoencoders as generative surrogates for RF circuits and antennas; “An Autonomous Agentic Framework for Deep Inverse Photonic Design” (Willie Padilla) presents an agentic, autonomous inverse-design workflow for metamaterials, illustrating how AI agents can accelerate spectrum-to-structure design paradigms relevant across EM domains — (2) LLM-augmented EDA workflows and ML foundations: “Practical Considerations for Applying AI to RF and Microwave EDA Workflows” (Matthew Ozalas) and “Accelerating Innovation: AI-Driven Advances in Sigrity, Clarity, and Optimality” (Jian Liu) highlight Keysight’s and Cadence’s strategies for GenAI/LLM-aided design; Complementary talks cover attention mechanisms for non-linear circuit modeling (Qi-Jun Zhang) and multiphysics-informed, data-free ML for RFIC design (Dan Jiao) — (3) Multimodal LLMs: “Multimodal LLMs for Electromagnetic Waves” (Zhi Jackie Yao) fuses image-based EM data with text via a BLIP bridge into pretrained LLMs for EM reasoning and design assistance. Rigor and trust will be discussed throughout. Talks and discussion will cover dataset curation, generalization, solver-in-the-loop constraints (passivity/causality/manufacturability), independent EM/measurement validation, and secure integration into EDA flows, along with practical guardrails to avoid hallucinations and constraint violations. For attendees new to this intersection, the workshop includes short primers, reproducible examples, and simple evaluation checklists to separate signal from hype.