![[2602.16953] LLM4Cov: Execution-Aware Agentic Learning for High-coverage Testbench Generation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16953] LLM4Cov: Execution-Aware Agentic Learning for High-coverage Testbench Generation
Summary
The paper presents LLM4Cov, an offline learning framework for high-coverage testbench generation, addressing challenges in hardware verification with execution-aware learning techniques.
Why It Matters
This research is significant as it proposes a novel approach to improve the efficiency of hardware verification processes, which are critical in ensuring the reliability of complex systems. By leveraging execution-aware learning, it aims to reduce the costs and time associated with traditional reinforcement learning methods.
Key Takeaways
- LLM4Cov utilizes offline agent-learning to enhance hardware verification.
- The framework introduces execution-validated data curation and policy-aware data synthesis.
- A compact model achieves a 69.2% coverage pass rate, outperforming larger models.
- The study addresses the limitations of online reinforcement learning in high-coverage scenarios.
- A reality-aligned benchmark is developed to assess the proposed framework's effectiveness.
Computer Science > Artificial Intelligence arXiv:2602.16953 (cs) [Submitted on 18 Feb 2026] Title:LLM4Cov: Execution-Aware Agentic Learning for High-coverage Testbench Generation Authors:Hejia Zhang, Zhongming Yu, Chia-Tung Ho, Haoxing Ren, Brucek Khailany, Jishen Zhao View a PDF of the paper titled LLM4Cov: Execution-Aware Agentic Learning for High-coverage Testbench Generation, by Hejia Zhang and 5 other authors View PDF HTML (experimental) Abstract:Execution-aware LLM agents offer a promising paradigm for learning from tool feedback, but such feedback is often expensive and slow to obtain, making online reinforcement learning (RL) impractical. High-coverage hardware verification exemplifies this challenge due to its reliance on industrial simulators and non-differentiable execution signals. We propose LLM4Cov, an offline agent-learning framework that models verification as memoryless state transitions guided by deterministic evaluators. Building on this formulation, we introduce execution-validated data curation, policy-aware agentic data synthesis, and worst-state-prioritized sampling to enable scalable learning under execution constraints. We further curate a reality-aligned benchmark adapted from an existing verification suite through a revised evaluation protocol. Using the proposed pipeline, a compact 4B-parameter model achieves 69.2% coverage pass rate under agentic evaluation, outperforming its teacher by 5.3% and demonstrating competitive performance against mod...
