Learning Library Cell Representations in Vector Space
This addresses the problem of efficient cell representation learning for electronic design automation engineers, offering a novel self-supervised method that is incremental over existing representation learning approaches.
The paper tackles the problem of learning meaningful vector representations of library cells for electronic design automation by proposing Lib2Vec, a self-supervised framework that captures functional and electrical similarities without costly labeling. Experimental results show it enables linear algebraic operations on cell vectors to reveal meaningful relationships like vector(BUF) - vector(INV) + vector(NAND) ~ vector(AND) and enhances downstream circuit learning applications with scarce labeled data.
We propose Lib2Vec, a novel self-supervised framework to efficiently learn meaningful vector representations of library cells, enabling ML models to capture essential cell semantics. The framework comprises three key components: (1) an automated method for generating regularity tests to quantitatively evaluate how well cell representations reflect inter-cell relationships; (2) a self-supervised learning scheme that systematically extracts training data from Liberty files, removing the need for costly labeling; and (3) an attention-based model architecture that accommodates various pin counts and enables the creation of property-specific cell and arc embeddings. Experimental results demonstrate that Lib2Vec effectively captures functional and electrical similarities. Moreover, linear algebraic operations on cell vectors reveal meaningful relationships, such as vector(BUF) - vector(INV) + vector(NAND) ~ vector(AND), showcasing the framework's nuanced representation capabilities. Lib2Vec also enhances downstream circuit learning applications, especially when labeled data is scarce.