pith. sign in

arxiv: 2407.10424 · v5 · pith:X7BU7742new · submitted 2024-07-15 · 💻 cs.PL · cs.AI

CodeV: Empowering LLMs with HDL Generation through Multi-Level Summarization

Yang Zhao , Di Huang , Chongxiao Li , Pengwei Jin , Muxin Song , Yinan Xu , Ziyuan Nan , Mingju Gao
show 9 more authors
Tianyun Ma Lei Qi Yansong Pan Zhenxing Zhang Rui Zhang Xishan Zhang Zidong Du Qi Guo Xing Hu
This is my paper
classification 💻 cs.PL cs.AI
keywords llmsgenerationlanguagecodetasksverilogchatdata
0
0 comments X
read the original abstract

The design flow of processors, particularly in hardware description languages (HDL) like Verilog and Chisel, is complex and costly. While recent advances in large language models (LLMs) have significantly improved coding tasks in software languages such as Python, their application in HDL generation remains limited due to the scarcity of high-quality HDL data. Traditional methods of adapting LLMs for hardware design rely on synthetic HDL datasets, which often suffer from low quality because even advanced LLMs like GPT perform poorly in the HDL domain. Moreover, these methods focus solely on chat tasks and the Verilog language, limiting their application scenarios. In this paper, we observe that: (1) HDL code collected from the real world is of higher quality than code generated by LLMs. (2) LLMs like GPT-3.5 excel in summarizing HDL code rather than generating it. (3) An explicit language tag can help LLMs better adapt to the target language when there is insufficient data. Based on these observations, we propose an efficient LLM fine-tuning pipeline for HDL generation that integrates a multi-level summarization data synthesis process with a novel Chat-FIM-Tag supervised fine-tuning method. The pipeline enhances the generation of HDL code from natural language descriptions and enables the handling of various tasks such as chat and infilling incomplete code. Utilizing this pipeline, we introduce CodeV, a series of HDL generation LLMs. Among them, CodeV-All not only possesses a more diverse range of language abilities, i.e. Verilog and Chisel, and a broader scope of tasks, i.e. Chat and fill-in-middle (FIM), but it also achieves performance on VerilogEval that is comparable to or even surpasses that of CodeV-Verilog fine-tuned on Verilog only, making them the first series of open-source LLMs designed for multi-scenario HDL generation.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 16 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. HWE-Bench: Benchmarking LLM Agents on Real-World Hardware Bug Repair Tasks

    cs.AI 2026-04 unverdicted novelty 8.0

    HWE-Bench is the first repository-level benchmark for LLM agents on real hardware bug repair, where the best agent fixes 70.7% of 417 tasks but drops below 65% on complex SoC projects.

  2. CHIA: An open-source framework for principled, agentic AI-driven hardware/software co-design research

    cs.AR 2026-06 unverdicted novelty 7.0

    CHIA introduces a framework for building and deploying agentic AI co-design flows as CHIA loops with tool nodes, reliability mechanisms, and five case-study demonstrations.

  3. CHIA: An open-source framework for principled, agentic AI-driven hardware/software co-design research

    cs.AR 2026-06 unverdicted novelty 7.0

    CHIA is an open-source framework for agentic AI-driven hardware/software co-design using CHIA loops as directed cyclic graphs, a tool library, and features for reliable experimentation, shown via five case studies.

  4. ChipCraftBrain: Validation-First RTL Generation via Multi-Agent Orchestration

    cs.AR 2026-04 unverdicted novelty 7.0

    ChipCraftBrain achieves 97.2% pass rate on VerilogEval and 94.7% on CVDP benchmarks for generating functional RTL code using adaptive multi-agent orchestration and hybrid reasoning.

  5. CASS-RTL: Correctness-Aware Subspace Steering for RTL Generation with LLMs

    cs.PL 2026-06 unverdicted novelty 6.0

    CASS-RTL identifies correctness-linked attention heads, builds a steering subspace from them, and applies a geometry-aware intervention that raises pass@1/5/10 accuracy 10-20% on VerilogEval and 5% on CVDP across mult...

  6. Verilog-Evolve: Feedback-Driven and Skill-Evolving Verilog Generation

    cs.CL 2026-05 unverdicted novelty 6.0

    Verilog-Evolve uses executable feedback from simulation, synthesis, timing, and GEMM metrics to refine LLM-generated Verilog and evolves skills across tasks, improving functional success and downstream hardware qualit...

  7. RTL-BenchMT: Dynamic Maintenance of RTL Generation Benchmark Through Agent-Assisted Analysis and Revision

    cs.AI 2026-05 unverdicted novelty 6.0

    RTL-BenchMT is an agent-assisted framework for dynamically maintaining RTL generation benchmarks by fixing flaws and reducing overfitting in LLM-based EDA applications.

  8. Dr. RTL: Autonomous Agentic RTL Optimization through Tool-Grounded Self-Improvement

    cs.AI 2026-04 unverdicted novelty 6.0

    Dr. RTL's multi-agent framework with group-relative skill learning achieves 21% WNS and 17% TNS timing improvements plus 6% area reduction on 20 real-world RTL designs over commercial synthesis tools.

  9. VeriRAG: A Retrieval-Augmented Framework for Automated RTL Testability Repair

    cs.AR 2025-07 unverdicted novelty 6.0

    VeriRAG is a RAG-based LLM framework that repairs Verilog RTL designs for DFT compliance using a curated dataset VeriDFT and achieves a 7.72-fold higher successful repair rate than zero-shot prompting.

  10. ChipSeek: Optimizing Verilog Generation via EDA-Integrated Reinforcement Learning

    cs.AI 2025-07 unverdicted novelty 6.0

    ChipSeek is a hierarchical-reward reinforcement learning framework with Curriculum-Guided Dynamic Policy Optimization that integrates EDA simulator feedback to improve LLM-generated RTL code on both functional correct...

  11. VeriPilot: An LLM-Powered Verilog Debugging Framework

    cs.AR 2026-06 unverdicted novelty 5.0

    VeriPilot raises GPT-4o Verilog repair success from 54.3% to 85.71% on the CVDP benchmark by using golden-model semantic alignment and CDFG-based signal tracing.

  12. ROSUM-MCTS: Monte Carlo Tree Search-Inspired HDL Code Summarization with Structural Rewards

    cs.CL 2026-06 unverdicted novelty 5.0

    ROSUM-MCTS applies MCTS-inspired hierarchical candidate expansion and a composite reward balancing functional correctness, local content adequacy, and fluency to improve LLM summaries of VHDL and Verilog code, outperf...

  13. EstRTL: Functional Estimation Guided RTL Code Generation

    cs.AR 2026-06 unverdicted novelty 5.0

    EstRTL improves correctness of generic-LLM RTL code generation by 3.2-9% via a Generation-Estimation-Correction loop that uses static functional scoring to decide acceptance, retry, or repair.

  14. Mitigating hallucinations and omissions in LLMs for invertible problems: An application to hardware logic design automation

    cs.LG 2025-11 unverdicted novelty 5.0

    Using LLMs to encode logic condition tables into HDL code and decode back to tables mitigates hallucinations in hardware design automation.

  15. StepPRM-RTL: Stepwise Process-Reward Guided LLM Fine-Tuning for Enhanced RTL Synthesis

    cs.AI 2026-06 unverdicted novelty 4.0

    StepPRM-RTL combines stepwise process-reward modeling, RAFT, and MCTS to fine-tune LLMs for RTL code, claiming over 10% gains in functional correctness on Verilog/VHDL benchmarks.

  16. Fine-Tuning and Serving Gemma 4 31B on Google Cloud TPU: A Technical Comparison with GPU Baselines

    cs.DC 2026-05 unverdicted novelty 4.0

    Empirical port of Gemma 4 31B LoRA training and vLLM inference from PyTorch/FSDP to JAX/Tunix shows TPU 1.61x faster training at 2.12x lower cost and superior long-context inference versus 2xH100 GPU.