MicroEvo: Knowledge-Guided LLM Sampling for Efficient Microarchitecture Design Space Exploration

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Jia Xiong, Runkai Li, Chenxu Niu, Guangyuan Gao, Changwen Xing, Yifan Zhang, Xinlai Wan, Jieran Cui, Chen Bai, Yusheng Hua, Ying Wang, Ming Ling, Xi Wang, Tao Xie
• Venue: arXiv: cs.AI
• Published: 2026-08-06
• Collected: 2026-08-07
• URL: http://arxiv.org/abs/2608.06183v1
• PDF: https://arxiv.org/pdf/2608.06183v1
• DOI: UNKNOWN

Plain-Language Brief

A new method called MicroEvo is proposed for efficient exploration of microarchitecture design spaces. This approach utilizes a combination of large language models (LLMs) and Monte Carlo Tree Search (MCTS) to optimize multi-objective microarchitectures. The results show an improvement in Pareto-front quality by up to 36.2% over NSGA-II, with 10.6x higher search efficiency. The method also demonstrates scalability for complex industrial-scale cores.

What the Document Says

• Microarchitecture design space exploration faces issues of expansive search spaces and expensive PPA evaluation.
• Existing methods perform blind search without considering microarchitectural dependencies and do not learn effectively from the iterative search, leading to inefficiencies and weak Pareto convergence.
• The paper proposes MicroEvo, a knowledge-guided framework that combines off-the-shelf LLMs with Monte Carlo Tree Search (MCTS) for multi-objective microarchitecture optimization.
• MicroEvo includes LLM-driven evolutionary operators, a Pareto-aware tree policy, an active knowledge accumulation mechanism, and state-aware directives.
• Experiments show that MicroEvo improves Pareto-front quality by up to 36.2% over NSGA-II and achieves 10.6x higher search efficiency.
• The method also demonstrates strong scalability for complex industrial-scale cores.
• The code repository for MicroEvo is available at GitHub.

Key Concepts Mentioned

• Microarchitecture design space exploration
• Pareto convergence
• Large language models (LLMs)
• Monte Carlo Tree Search (MCTS)
• Multi-objective microarchitecture optimization
• Industrial-scale cores

Claims Made by the Document

• MicroEvo improves Pareto-front quality by up to 36.2% over NSGA-II.
• MicroEvo achieves 10.6x higher search efficiency.

Evidence Used

• Experiments conducted as part of the research.

Methods / Approach

• Use of knowledge-guided framework that combines LLMs and MCTS for multi-objective microarchitecture optimization.
• Inclusion of LLM-driven evolutionary operators, a Pareto-aware tree policy, an active knowledge accumulation mechanism, and state-aware directives.

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The effectiveness of MicroEvo may vary depending on the specific microarchitecture being optimized. Additionally, the quality of the results might be influenced by the performance of the off-the-shelf LLMs.

Open Questions

• How does the performance of MicroEvo compare to other optimization methods for various types of microarchitectures?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The text provides no information about comparisons between MicroEvo and other optimization methods. Further investigation is needed.

Raw Abstract / Excerpt

Microarchitecture design space exploration suffers from expansive search spaces and expensive PPA evaluation, leaving only a small simulation budget for design decision-making. Existing methods perform blind search without considering microarchitectural dependencies and fail to learn from the iterative search effectively, leading to wasted evaluations and weak Pareto convergence. In this paper, we propose MicroEvo, a knowledge-guided framework that couples off-the-shelf LLMs with Monte Carlo Tree Search (MCTS) for multi-objective microarchitecture optimization. MicroEvo combines LLM-driven evolutionary operators, a Pareto-aware tree policy that balances Pareto contribution and diversity, an active knowledge accumulation mechanism that extracts and reuses optimization insights, and state-aware directives that adapt the search behavior online. Experiments show that MicroEvo improves Pareto-front quality by up to 36.2% over NSGA-II and achieves 10.6x higher search efficiency, and also demonstrates strong scalability to a complex industrial-scale core. The code repository is available at: https://github.com/GEAR-SEU/MicroEvo-ICCAD-26.