Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AMSnet 2.0: A Large AMS Database with AI Segmentation for Net Detection

Yichen Shi, Zhuofu Tao, Yuhao Gao, Li Huang, Hongyang Wang, Zhiping Yu, Ting-Jung Lin, Lei He

TL;DR

This work tackles the lack of high-quality multimodal data for AMS circuit schematics by introducing AMSnet 2.0, a large dataset containing schematics, netlists, and positional information to train advanced models for schematic understanding. It presents a robust, two-stage net detection and schematic reconstruction pipeline based on instance segmentation that replaces brittle heuristics, enabling accurate netlist generation and OpenAccess schematic reconstruction even in noisy images. A labeling platform is released to expand AMSnet 2.0, and the methodology yields promising netlist F1 scores across difficulty levels, while demonstrating automatic schematic reconstruction with minimal manual intervention. The dataset and tools promise to enhance MLLMs’ capabilities in AMS circuit understanding and pave the way for netlist-to-schematic generation and further LLM-driven design assistance.

Abstract

Current multimodal large language models (MLLMs) struggle to understand circuit schematics due to their limited recognition capabilities. This could be attributed to the lack of high-quality schematic-netlist training data. Existing work such as AMSnet applies schematic parsing to generate netlists. However, these methods rely on hard-coded heuristics and are difficult to apply to complex or noisy schematics in this paper. We therefore propose a novel net detection mechanism based on segmentation with high robustness. The proposed method also recovers positional information, allowing digital reconstruction of schematics. We then expand AMSnet dataset with schematic images from various sources and create AMSnet 2.0. AMSnet 2.0 contains 2,686 circuits with schematic images, Spectre-formatted netlists, OpenAccess digital schematics, and positional information for circuit components and nets, whereas AMSnet only includes 792 circuits with SPICE netlists but no digital schematics.

AMSnet 2.0: A Large AMS Database with AI Segmentation for Net Detection

TL;DR

This work tackles the lack of high-quality multimodal data for AMS circuit schematics by introducing AMSnet 2.0, a large dataset containing schematics, netlists, and positional information to train advanced models for schematic understanding. It presents a robust, two-stage net detection and schematic reconstruction pipeline based on instance segmentation that replaces brittle heuristics, enabling accurate netlist generation and OpenAccess schematic reconstruction even in noisy images. A labeling platform is released to expand AMSnet 2.0, and the methodology yields promising netlist F1 scores across difficulty levels, while demonstrating automatic schematic reconstruction with minimal manual intervention. The dataset and tools promise to enhance MLLMs’ capabilities in AMS circuit understanding and pave the way for netlist-to-schematic generation and further LLM-driven design assistance.

Abstract

Current multimodal large language models (MLLMs) struggle to understand circuit schematics due to their limited recognition capabilities. This could be attributed to the lack of high-quality schematic-netlist training data. Existing work such as AMSnet applies schematic parsing to generate netlists. However, these methods rely on hard-coded heuristics and are difficult to apply to complex or noisy schematics in this paper. We therefore propose a novel net detection mechanism based on segmentation with high robustness. The proposed method also recovers positional information, allowing digital reconstruction of schematics. We then expand AMSnet dataset with schematic images from various sources and create AMSnet 2.0. AMSnet 2.0 contains 2,686 circuits with schematic images, Spectre-formatted netlists, OpenAccess digital schematics, and positional information for circuit components and nets, whereas AMSnet only includes 792 circuits with SPICE netlists but no digital schematics.
Paper Structure (21 sections, 9 figures, 1 table)

This paper contains 21 sections, 9 figures, 1 table.

Table of Contents

  1. Introduction
  2. Schematic Labeling
  3. Element Labeling
  4. Net Labeling
  5. Insight Labeling
  6. Schematic and Netlist Generation
  7. Device Detection
  8. Wire Segmentation
  9. Mocked Marking Data Augmentation
  10. Net Recognition
  11. Netlist Generation & Schematic Reconstruction
  12. Experiments
  13. Experiment Setting
  14. Evaluation Protocol
  15. Netlist Generation Results
  16. ...and 6 more sections

Figures (9)

  • Figure 1: AMSnet 2.0 workflow for multimodal dataset construction.
  • Figure 2: Examples of noisy schematics: (a) overlaid markings and (b) partial highlighting
  • Figure 3: The full schematic processing pipeline: schematic element detection, net segmentation, Spectre format netlist generation, and digital schematic reconstruction in OpenAcess format.
  • Figure 4: The proposed two-stage net detection method. (a) Semantic segmentation on the circuit wires to delineate the wires and background areas. (b) Detecting connected components, then perform split and merge operations at intersection points to identify the nets.
  • Figure 5: Results for element detection (top), net detection (middle), and Spectre format netlist generation (bottom), from the UNet procedure. Columns (a), (b), and (c) present examples from easy, medium, and hard splits respectively.
  • ...and 4 more figures