Reproduce “CorrectRTL: Boost RTL Generation Quality Through Automated Self-Correction with Instruction Decomposition” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jing Wang, Shang Liu, Hangan Zhou, Wenji Fang, Yuchao Wu, Yugao Zhu, Zhiyao Xie
CorrectRTL generates synthesizable Verilog from hardware specifications by decomposing designs into blocks, retrieving relevant RTL examples, and checking block-level assertions. It rewrites only blocks whose checks fail, using verification feedback. On ScaleBench, it achieves 49.0% design-level Pass@1, versus 20.5% for direct generation with Claude Sonnet 4.5. This improves results on this benchmark, but fewer than half of designs pass.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.