[arXiv]score: 0.09

MCTS-Inspired LLM Summarizer Outperforms Baselines on VHDL and Verilog

June 9, 2026

ROSUM-MCTS applies Monte Carlo Tree Search-style exploration to HDL code summarization, using a composite reward balancing functional correctness, local content adequacy, and fluency. Evaluated on VHDL-eval and Verilog-eval, it consistently outperforms baseline LLM summarization methods through hierarchical candidate expansion.

HOW THIS AFFECTS YOU

●

builderYou can apply this approach to automate HDL documentation in EDA pipelines where LLM baselines underperform on Verilog/VHDL.

●

researcherThe structured bottom-up refinement with composite reward is a transferable technique for other low-resource code summarization domains beyond HDL.

read original ↗arxiv.org

← back to feed