Research map / Multi-agent systems and games / LLM agents
ICLAgent: Integrated Circuit Footprint Geometry Labeling via LMM-empowered Multi-Agent Framework
Yida Wang, Yixin Liu, Taiting Lu, Lanqing Yang, Dian Ding, Juntao Zhou, Yifan Yang, Yi-Chao Chen, Mahanth Gowda
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Stage 1 diagram region detection (ablation) | ICLAgent · ICGeoQA (400 pages) | Overall IoU_IC (%): 72.3 -> 79.0 | Improved |
| add Stage 2 footprint classification and parameter planning (ablation) | ICLAgent · ICGeoQA (400 pages) | Overall IoU_IC (%): 70.2 -> 79.0 | Improved |
| add Stages 1 and 2 on top of Stage 3 only (ablation) | ICLAgent · ICGeoQA (400 pages) | Overall IoU_IC (%): 68.2 -> 79.0 | Improved |
| decompose into key parameter extraction instead of end-to-end pin description generation (ablation) | ICLAgent (Stage 3 only) · ICGeoQA (400 pages) | Overall IoU_IC (%): 65.7 -> 68.2 | Improved |
| use ICLAgent multi-agent framework instead of end-to-end LLM4-IC8K | Qwen2-VL-7B · ICGeoQA (400 pages) | Overall IoU_IC (%): 71.6 -> 79.0 | Improved |
| use ICLAgent instead of LLM4-IC8K for pin count | Qwen2-VL-7B · ICGeoQA (400 pages) Task 1 | pin count MAE: 0.35 -> 0.37 | Hurt |
| use ICLAgent instead of LLM4-IC8K for pin dimensions | Qwen2-VL-7B · ICGeoQA (400 pages) Task 3 | IoU_pin (%): 88.0 -> 95.6 | Improved |
| use ICLAgent instead of LLM4-IC8K for coordinates | Qwen2-VL-7B · ICGeoQA (400 pages) Task 2 | d_pin: 1.11 -> 1.07 | Improved |
| use fine-tuned 7B ICLAgent instead of GPT-5 | Qwen2-VL-7B · ICGeoQA (400 pages) | Overall IoU_IC (%): 40.6 -> 79.0 | Improved |
| use fine-tuned 7B ICLAgent instead of Gemini 3 Pro | Qwen2-VL-7B · ICGeoQA (400 pages) | Overall IoU_IC (%): 68.6 -> 79.0 | Improved |
| use dedicated Diagram Agent instead of general LMM for region detection | ICLAgent · ICGeoQA (400 pages) (per stage) | IoU_diagram (%): 3.9 -> 70.2 | Improved |
| use ICLAgent instead of manual EDA labeling | ICLAgent · ICGeoQA (400 pages) | Overall IoU_IC (%): 44 -> 79.0 | Improved |
Closest papers by meaning
- Functional Matching of Logic Subgraphs: Beyond Structural Isomorphism (NeurIPS 2025 · similarity 0.82)
- DeepLayout: Learning Neural Representations of Circuit Placement Layout (ICML 2025 · similarity 0.81)
- Benchmarking End-To-End Performance of AI-Based Chip Placement Algorithms (NeurIPS 2025 · similarity 0.81)
- CircuitFusion: Multimodal Circuit Representation Learning for Agile Chip Design (ICLR 2025 · similarity 0.81)
- AnalogGenie-Lite: Enhancing Scalability and Precision in Circuit Topology Discovery through Lightweight Graph Modeling (ICML 2025 · similarity 0.81)
- See it to Place it: Evolving Macro Placements with Vision Language Models (NeurIPS 2026 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
