Research map / Vision-language and multimodal / Multimodal LLMs
Circuit-Think: A Multimodal Reasoning Framework for Automated Circuit-to-Netlist Translation with Trajectory-Guided Reinforcement Learning
Yuqi Jiang, Yupeng Hu, Jinyuan Deng, Xiaotian Qiu, Yucheng Cui, Xuyang He, Ruidong Li, Qi Sun, Cheng Zhuo
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Circuit-Think, a multimodal reasoning framework for translating circuit images into netlists, trained with trajectory-guided reinforcement learning.
Problem
VLMs struggle with circuit images because of diverse component symbols and complex connections, which hinders understanding physical layouts and reasoning about electrical connectivity for circuit-to-netlist translation.
Approach
A multimodal reasoning framework for circuit-to-netlist translation that combines trajectory-guided reinforcement learning to teach VLMs to reason about physical layout and electrical connections.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train with TGRL instead of GRPO | Qwen2.5-VL-7B · Circuit image-netlist test set | SPICE netlist accuracy: 40.23 -> 60.17 | Improved |
| train with TGRL instead of SFT | Qwen2.5-VL-7B · Circuit image-netlist test set | SPICE netlist accuracy: 34.78 -> 60.17 | Improved |
| add strict step-wise format constraint (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 39.21 -> 73.27 | Improved |
| tighten format constraint from soft to strict (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 52.69 -> 73.27 | Improved |
| add device recognition reward after port reward (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 45.93 -> 48.13 | Improved |
| add connection reasoning reward (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 48.13 -> 73.27 | Improved |
| add port/device type classification (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 47.13 -> 73.27 | Improved |
| add port/device bbox localization (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 49.91 -> 73.27 | Improved |
| add step-by-step port-device-connection trajectory (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 35.38 -> 73.27 | Improved |
| add logic consistency reward Rlogic (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 63.25 -> 73.27 | Improved |
| add reflective learning (ablation) | Circuit-Think · Circuit image-netlist test set | SPICE netlist accuracy: 65.76 -> 73.27 | Improved |
| add reflective learning (ablation) | Circuit-Think · Circuit image-netlist test set | logic consistency: 70.10 -> 85.76 | Improved |
Closest papers by meaning
- CircuitSense: A Hierarchical Circuit System Benchmark Bridging Visual Comprehension and Symbolic Reasoning in Engineering Design Process (ICLR 2026 · similarity 0.87)
- Zebra-CoT: A Dataset for Interleaved Vision-Language Reasoning (ICLR 2026 · similarity 0.86)
- CircuitFusion: Multimodal Circuit Representation Learning for Agile Chip Design (ICLR 2025 · similarity 0.86)
- STELAR-VISION: Self-Topology-Aware Efficient Learning for Aligned Reasoning in Vision (AAAI 2026 · similarity 0.86)
- Unveiling Chain of Step Reasoning for Vision-Language Models with Fine-grained Rewards (NeurIPS 2025 · similarity 0.86)
- Reforming the Mechanism: Editing Reasoning Patterns in LLMs with Circuit Reshaping (ICLR 2026 · similarity 0.85)
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.
