aigverse: Toward Machine Learning-Driven Logic Synthesis
Date:
The integration of machine learning into logic synthesis has long lagged behind, hampered by a fundamental conceptual divide. Logic synthesis demands mathematical precision and guarantees of functional correctness, while machine learning embraces probabilistic reasoning and statistical generalization. This contrast is further exacerbated by a practical ecosystem mismatch: high-performance logic synthesis tools are typically implemented in C/C++, whereas the machine learning community predominantly operates in Python.
This talk introduces aigverse, an open-source library designed to address this divide. aigverse wraps leading C/C++ logic synthesis frameworks and exposes their capabilities through a clean, Pythonic interface. More importantly, it provides native infrastructure for machine learning integration via modular adapters, enabling ML models to interact directly with logic synthesis workflows. We demonstrate seamless compatibility with major ML frameworks such as PyTorch. As a side benefit, aigverse also brings logic synthesis capabilities into the Python ecosystem, where such tooling has historically been scarce, thereby opening the field to the broader Python community.
The presentation will outline the architecture of aigverse and illustrate its use through practical examples. Attendees will learn how to use the library to generate datasets, extract features from circuits, and train models. aigverse paves the way for a new era of ML-guided logic synthesis in the free and open silicon ecosystem, offering both accessibility and extensibility for researchers and practitioners alike.
Learn more here.