MAGE: A Multi-Agent Engine for Automated RTL Code Generation * Equal Contribution

Yujie Zhao * , Hejia Zhang * , Hanxian Huang, Zhongming Yu, Jishen Zhao · 2024

Read MAGE: A Multi-Agent Engine for Automated RTL Code Generation * Equal Contribution by Yujie Zhao * , Hejia Zhang * , Hanxian Huang, Zhongming Yu, Jishen Zhao (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-07822.txt.

Description

Yujie Zhao and co-authors present MAGE, a team of AI agents that writes Verilog chip code from plain instructions. Language models can draft register-transfer code, but errors abound. Both syntax and function must be right for real chips.

Single agents juggle too many jobs in this task. MAGE splits drafting, checking, and fixing across agents. Hot sampling creates many code candidates to explore widely. A checkpoint method tests function early in each draft. Early faults return precise feedback for targeted repair. Syntax success rates reached high levels in tests. The system is open source for other teams. It shows teamwork helps models in strict code domains.

Catalog record

CREATORYujie Zhao * , Hejia Zhang * , Hanxian Huang, Zhongming Yu, Jishen Zhao
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-07822

Browse the full library · About the Tower of Babel