CircuitFormer: LLM for Analog Circuit Topology Design
A team of researchers has introduced CircuitFormer, a language model designed for analog topology design based on natural language inputs. This study tackles two significant challenges in utilizing large language models (LLMs) for analog hardware: the lack of annotated datasets and inefficiencies in tokenization. They assembled the most extensive annotated dataset of analog circuit netlists, featuring 31,341 pairs of netlists and descriptions across key circuit categories. Additionally, they created Circuit Tokenizer (CKT), an innovative graph tokenizer that enhances scalability by encoding netlist connectivity through the identification of frequent subcircuits. The findings have been published on arXiv.
Key facts
- CircuitFormer is a circuit language model for analog topology design from natural language prompts.
- The work addresses two limitations: scarcity of annotated datasets and tokenizer inefficiency.
- Curated dataset of 31,341 netlist-description pairs across major circuit classes.
- Proposed Circuit Tokenizer (CKT) encodes netlist connectivity by mining frequent subcircuits.
- CKT improves scalability for analog circuit design.
- The paper is published on arXiv with ID 2605.05773.
- The work bridges the gap between LLMs and analog hardware design.
- The dataset is the largest annotated analog circuit netlist dataset to date.
Entities
Institutions
- arXiv