工作流程
计算机科学
自动化
光子学
光学
硅
电子设计自动化
硅光子学
半导体激光器理论
工程类
光子晶体
信号处理
光电子学
物理
电子工程
集成光学
干涉测量
系统工程
出处
期刊:Optics Express
[Optica Publishing Group]
日期:2026-08-31
卷期号:34 (19): 34999-34999
摘要
Silicon photonic design requires layout generation, simulation, interpretation, and verification across multiple tools, resulting in fragmented workflows that depend on expert operation and limit reproducibility. We present a tool-grounded, rule-guarded large language model (LLM) framework integrating domain knowledge, task parsing, tool invocation, workflow orchestration, and database checking. Starting from a LangChain-style A-flow, we develop an enhanced B-flow through deterministic execution and orchestration. Experiments comparing key system inputs with outputs show that constraining the LLM within a structured workflow improves execution reliability and substantially reduces token consumption. The enhanced B-flow provides a blueprint for model context protocol (MCP) services, enabling cross-client tool interoperability.
科研通智能强力驱动
Strongly Powered by AbleSci AI