基于大语言模型多智能体的端到端领域专用SoC设计

闫沛然; 支沁喆; 刘力峰; 贾天宇

doi:10.20193/j.ices2097-4191.2025.0043

PDF(29686 KB)

集成电路与嵌入式系统 ›› 2025, Vol. 25 ›› Issue (8) : 31-40. DOI: 10.20193/j.ices2097-4191.2025.0043

新兴计算芯片设计研究专刊

基于大语言模型多智能体的端到端领域专用SoC设计

作者信息 +

End-to-end domain-specific SoC design with LLM-based multi-agent

Author information +

文章历史 +

摘要

随着摩尔定律放缓,领域专用片上系统(DSSoC)集成领域专用加速器(DSA)已成为一种极具前景的高能效芯片设计策略。然而,DSSoC的设计流程高度复杂,导致开发周期漫长且人力投入巨大。大语言模型(LLMs)的最新进展为敏捷芯片设计引入了新方法,这一方法在代码生成和EDA脚本编写中展现出巨大的应用潜力。文中提出一种基于LLM的多智能体DSSoC设计框架,覆盖从架构定义到代码生成、再到EDA物理实现的端到端设计流程。最后通过两项案例研究验证了该框架在22 nm和7 nm工艺节点上,仅用2至4周即可完成两个SoC设计。相较于原有流程生成的SoC,文中方案设计的SoC能效分别提升了4.84倍和3.82倍。

Abstract

As Moore’s Law slows down, domain-specific SoC (DSSoC) has emerged as a promising energy-efficient design strategy by integrating domain-specific accelerator (DSA). However, the design process for DSSoC remains highly complex, leading to prolonged development cycles and significant labor effort. Recent advances in large language models (LLMs) have introduced new methodologies for agile chip design, demonstrating substantial potential in code and EDA script generation. In this work, an LLM-based multi-agent framework for DSSoC design is proposed, which consists of end-to-end design stages from architecture definition to code generation and EDA physical implementation. The approach is validated through two case studies involving 2-to 4-week SoC designs at process nodes of 22 nm and 7 nm. The evalautions show the generated SoCs achieve energy efficiency improvements of 4.84× and 3.82×, compared to SoCs generated by the existing framework.

导出引用

闫沛然, 支沁喆, 刘力峰, 等. 基于大语言模型多智能体的端到端领域专用SoC设计[J]. 集成电路与嵌入式系统. 2025, 25(8): 31-40 https://doi.org/10.20193/j.ices2097-4191.2025.0043

YAN Peiran, ZHI Qinzhe, LIU Lifeng, et al. End-to-end domain-specific SoC design with LLM-based multi-agent[J]. Integrated Circuits and Embedded Systems. 2025, 25(8): 31-40 https://doi.org/10.20193/j.ices2097-4191.2025.0043

中图分类号： TN47 (大规模集成电路、超大规模集成电路)

参考文献

列表( 原文顺序 | 文献年度倒序 | 文中引用次数倒序 ) 可视化分析

[1]	ACHIAM J, ADLER S, AGARWAL S, et al. GPT-4 Technical Report[EB/OL].(2023-03-15)[2025-06-17]. https://arxiv.org/abs/2303.08774. https://arxiv.org/abs/2303.08774 本文引用 [1]

[2]	GUO D, YANG D, ZHANG H, et al. DeepSeek-R1:Incentivizing Reasoning Capability in LLMs via Reinforcement Learning[EB/OL].(2025-01-29)[2025-06-17]. https://arxiv.org/abs/2501.12948. https://arxiv.org/abs/2501.12948 本文引用 [2]

[3]	WANG X, WAN G W, WONG S Z, et al. ChatCPU:An Agile CPU Design & Verification Platform with LLM[C]// Proceedings of the 61st ACM/IEEE Design Automation Conference(DAC). New York: ACM, 2024. 本文引用 [2]

[4]	WU H Y, HE Z L, ZHANG X Y, et al. ChatEDA:A Large Language Model Powered Autonomous Agent for EDA[J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2024. 本文引用 [2]

[5]	THAKUR S, AHMAD B, FAN Z, et al. Benchmarking Large Language Models for Automated Verilog RTL Code Generation[C]// Proceedings of the 2023 Design,Automation & Test in Europe Conference & Exhibition (DATE). Dresden: IEEE, 2023. 本文引用 [1]

[6]	THAKUR S, AHMAD B, PEARCE H, et al. Verigen:A Large Language Model for Verilog Code Generation[J]. ACM Transactions on Design Automation of Electronic Systems, 2023. 本文引用 [2]

[7]	ASANOVIĆK. The Rocket Chip Generator: UCB/EECS-2016-17[R]. Berkeley: University of California,Berkeley, 2016. 本文引用 [1]

[8]	NVIDIA Corporation. NVDLA:NVIDIA Deep Learning Accelerator[EB/OL].[2025-06-17]. http://nvdla.org/. http://nvdla.org/ 本文引用 [1]

[9]	ZARUBA F, BENINI L. The Cost of Application-Class Processing:Energy and Performance Analysis of a Linux-Ready 1.7 GHz 64-Bit RISC-V Core in 22-nm FDSOI Technology[J]. IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2019.DOI:10.1109/TVLSI.2019.2926114. 本文引用 [1]

[10]	AMID A. Chipyard:Integrated Design, Simulation, and Implementation Framework forCustom SoCs[J]. IEEE Micro, 2020, 40(4):10-21. 本文引用 [2]

[11]	BALKIND J. OpenPiton:An Open Source Manycore Research Framework[J]. ACM SIGPLAN Notices, 2016, 51(4):217-232. 本文引用 [1]

[12]	SANTOS M. A Scalable Methodology for Agile Chip Development with Open-Source Hardware Components[C]// Proceedings of IEEE International Conference on Computer-Aided Design (ICCAD). San Diego: IEEE, 2022. 本文引用 [2]

[13]	PULLINI A.Mr. Wolf:An Energy-Precision Scalable Parallel Ultra Low Power SoC for IoT Edge Processing[J]. IEEE Journal of Solid-State Circuits, 2019, 54(7):1970-1981. 本文引用 [2]

[14]	FAUST A. Automatic Domain-Specific SoC Design for Autonomous Unmanned Aerial Vehicles[C]// Proceedings of the 55th IEEE/ACM International Symposium on Microarchitecture (MICRO). Chicago: ACM/IEEE, 2022. 本文引用 [1]

[15]	SANTOS M C. A 12nm Linux-SMP-capable RISC-V SoC with 14 accelerator types, distributed hardware power management and flexible NoC-based data orchestration[C]// Proceedings of IEEE International Solid-State Circuits Conference (ISSCC). San Francisco: IEEE, 2024. 本文引用 [1]

[16]	XI Z H, CHEN W X, GUO X, et al. The Rise and Potential of Large Language Model Based Agents:A Survey[EB/OL].(2023-09-14)[2025-06-17]. https://arxiv.org/abs/2309.07864. https://arxiv.org/abs/2309.07864 本文引用 [1]

[17]	BACHRACH J, VO H, RICHARDS B, et al. Chisel:Constructing Hardware in a Scala Embedded Language[C]// Proceedings of the 49th Annual Design Automation Conference. New York: ACM, 2012:1216-1225. 本文引用 [1]

[18]	LOCKHART D, ZIBRAT G, BATTEN C. PyMTL:A Unified Framework for Vertically Integrated Computer Architecture Research[C]// Proceedings of the 47th Annual IEEE/ACM International Symposium on Microarchitecture. Cambridge: IEEE/ACM, 2014: 280-292. 本文引用 [1]

[19]	AJAYI T, BLAAUW D. OpenROAD:Toward a Self-Driving,Open-Source Digital Layout Implementation Tool Chain[C]// Proceedings of Government Microcircuit Applications and Critical Technology Conference. Orlando: GOMACTech, 2019. 本文引用 [1]

[20]	LI X Q, TAO S M, HUANG Z R, et al. ieda:An Open-Source Intelligent Physical Implementation Toolkit and Library[EB/OL].(2023-08-03)[2025-06-17]. https://arxiv.org/abs/2308.01857. https://arxiv.org/abs/2308.01857 本文引用 [1]

[21]	CHANG K Y, WANG Y, REN H M, et al. ChipGPT: How Far Are We from Natural Language Hardware Design[EB/OL].(2023-05-23)[2025-06-17]. https://arxiv.org/abs/2305.14019. https://arxiv.org/abs/2305.14019 本文引用 [1]

[22]	FU Y G, ZHANG Y A, YU Z Z, et al. GPT4AIGChip: Towards Next-Generation AI Accelerator Design Automation via Large Language Models[C]// Proceedings of IEEE/ACM International Conference on Computer Aided Design(ICCAD). San Francisco: IEEE, 2023:1-9. 本文引用 [1]

[23]	BLOCKLOVE J. Chip-Chat:Challenges and Opportunities in Conversational Hardware Design[C]// Proceedings of 2023 ACM/IEEE 5th Workshop onMachine Learning for CAD (MLCAD). New York: ACM/IEEE, 2023:1-6. 本文引用 [1]

[24]	WU B Y, SHARMA U, KANKIPATI S R D, et al. EDA Corpus:A Large Language Model Dataset for Enhanced Interaction with OpenROAD[J]. IEEE, 2024.DOI:10.1109/LAD62341.2024.10691774. 本文引用 [1]

[25]	LIU M J. ChipNeMo:Domain-Adapted LLMs for Chip Design[EB/OL].(2023-11-01)[2025-06-17]. https://arxiv.org/abs/2311.00176. https://arxiv.org/abs/2311.00176 本文引用 [1]

[26]	HUANG H X, LIN Z H, WANG Z X, et al. Towards LLM-Powered Verilog RTL Assistant: Self-Verification and Self-Correction[EB/OL].(2024-06-01)[2025-06-17]. https://arxiv.org/abs/2406.00115. https://arxiv.org/abs/2406.00115 本文引用 [1]

[27]	HO C T, REN H X, KHAILANY B. VerilogCoder: Autonomous Verilog Coding Agents with Graph-based Planning and Abstract Syntax Tree (AST)-based Waveform Tracing Tool[EB/OL].(2024-08-12)[2025-06-17]. https://arxiv.org/abs/2408.08927. https://arxiv.org/abs/2408.08927 本文引用 [1]

[28]	Anthropic. Claude 3.7 Sonnet and Claude Code[EB/OL].(2024-06)[2024-06-17]. https://www.anthropic.com/news/claude-3-7-sonnet-and-claude-code. https://www.anthropic.com/news/claude-3-7-sonnet-and-claude-code 本文引用 [1]

[29]	WEI J, WANG X Z, SCHUURMANS D, et al. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models[J]. Advances in Neural Information Processing Systems, 2022, 35:24824-24837. 本文引用 [1]

[30]	YAO S Y, ZHAO J F, YU D, et al. ReAct:Synergizing Reasoning and Acting in Language Models[EB/OL].(2022-10-07)[2025-06-17]. https://arxiv.org/abs/2210.03629. https://arxiv.org/abs/2210.03629 本文引用 [1]

[31]	ROGERS J, EECKHOUT L, JAHRE M. HILP:Accounting for Workload-Level Parallelism in System-on-Chip Design Space Exploration[C]// Proceedings of the International Symposium on High Performance Computer Architecture (HPCA). Denver: IEEE, 2025. 本文引用 [1]

[32]	CHE S, BOYER M, MENG J, et al. Rodinia:A Benchmark Suite for Heterogeneous Computing[C]// Proceedings of 2009 IEEE International Symposium on Workload Characterization(IISWC). Austin: IEEE, 2009:44-54. 本文引用 [1]

[33]	HONG S R, ZHENG X W, CHEN J, et al. MetaGPT: Meta Programming for Multi-Agent Collaborative Framework[EB/OL].(2023-08-01)[2025-06-17]. https://arxiv.org/abs/2308.00352. https://arxiv.org/abs/2308.00352 本文引用 [1]

[34]	CLARK L T, VASHISHTHA V, SHIFREN L, et al. ASAP7:A 7-nm FinFET Predictive Process Design Kit[J]. Microelectronics Journal, 2016, 53:105-115. 本文引用 [1]

[35]	XU Y N. Towards Developing High Performance RISC-V Processors Using Agile Methodology[C]// Proceedings of the 55th IEEE/ACM International Symposium on Microarchitecture (MICRO). Chicago: IEEE/ACM, 2022:1178-1199. 本文引用 [1]