AMSnet-q: Unsupervised Circuit Identification and Performance Labeling for AMS Circuits
AMSnet-q is a cutting-edge system that operates independently to convert schematic images into a detailed analog and mixed-signal (AMS) circuit database, eliminating the need for human input. Unlike earlier techniques that only focused on extracting netlists, this advanced framework automates the entire verification process. It effectively changes schematics into netlists, creates testbenches that consider circuit topology, and conducts simulations to objectively evaluate performance. Tested with 28 nm technology, AMSnet-q addresses the heavy dependence of AI tools on manually curated datasets, an area where current large language and vision models fall short. This innovation reduces the burden on experts to decipher circuit functions, enabling the efficient development of labeled AMS circuit databases.
Key facts
- AMSnet-q is an unsupervised pipeline for AMS circuit identification and performance labeling.
- It converts schematic images directly into a labeled AMS circuit database.
- The pipeline automates schematic-to-netlist conversion, testbench generation, and sizing validation.
- Validated in 28 nm technology.
- Eliminates human-in-the-loop annotation for AMS circuits.
- Addresses limitations of current LLMs and vision models in automating circuit annotation.
- Prior work stops at netlist extraction; AMSnet-q completes the verification loop.
- Enables AI-driven automation tools to generate candidate topologies without manually curated datasets.
Entities
—