HALO: A Physics-Aware LLM Agent Framework for Nanophotonic Design
Multiagent Systems
Summary
The authors created HALO, a system that uses language models to help design nanophotonic devices by combining planning, simulations, and feedback in a loop. They tested HALO on a set of 52 design challenges called HALO-Bench, comparing different ways the system plans and runs simulations. One method was very efficient and reliable, while another could achieve better success but sometimes made errors. They also found that learning from past failures helped the system improve faster in repeated tasks. Overall, the authors showed how different approaches affect the balance between efficiency, success, and learning in AI-driven scientific design.
Authors
Yubo Zhang, Jinlin Xiang, Zijun Zhao, Yang Zhao, Eli Shlizerman, Arka Majumdar
Abstract
Language models have recently been applied to nanophotonic design, but it remains unclear whether they can reliably translate optical objectives into simulation-ready designs, execute electromagnetic analysis, and revise decisions from numerical feedback. We introduce HALO, a physics-aware framework that couples language-model planners with typed design specifications, electromagnetic simulation, diagnostic evaluation, and optional reuse of prior failure trajectories in an iterative design loop. We further introduce HALO-Bench, a 52-task benchmark spanning lab-derived, paper-derived, and open-ended nanophotonic design tasks under a shared evaluation protocol. We compare three planner configurations: a Fixed Structured Workflow, an Autonomous Structured Agent using the same simulation interface, and an Autonomous Coding Agent that directly writes and executes simulation code. The Fixed Structured Workflow is the most token-efficient and exhibits no observed code- or path-level failures, while autonomous coding can achieve higher task success with stronger models at the cost of additional operational failures. We also study reuse of prior failed trajectories. On targeted multi-round tasks, retrieved failure feedback reduces both iterations to first success and total token use. These results clarify the tradeoffs between explicit interfaces, autonomous execution, and reusable design experience in scientific agents.