CA-PPO-AXI: Coverage-Aware Deep Reinforcement Learning for Adaptive AXI Bus Protocol Verification
Fundamental Scientific Reports in Multidisciplinary Areas · 25.06.2026 · 2 (02) · 159-169
DOI: 10.66238/fsrma102 ↗Kurz gefasst
Studienprotokoll in der Zeitschrift Fundamental Scientific Reports in Multidisciplinary Areas, online erschienen am 25.06.2026. Verfasst von Jingsi Zhang und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
The verification of Advanced eXtensible Interface (AXI) bus protocols, such as AXI4 and AXI5, plays a crucial role in ensuring the functional correctness and reliability of modern System-on-Chip (SoC) designs, including CPUs, NPUs, and high-performance server processors. However, the increasing protocol complexity, characterized by burst transactions, multiple outstanding requests, out-of-order responses, and quality-of-service (QoS) mechanisms, makes traditional constraint-random verification inefficient and heavily dependent on manual testbench development. In particular, achieving high functional and assertion coverage for rare corner-case scenarios remains a major challenge. To address these issues, this paper proposes CA-PPO-AXI, a Coverage-Aware Proximal Policy Optimization framework for adaptive AXI bus protocol verification stimulus generation based on deep reinforcement learning. The AXI verification process is formulated as a Markov Decision Process, where a reinforcement learning agent dynamically generates transaction-level stimuli by jointly observing protocol state information and real-time coverage feedback. A coverage-aware state representation and a reward function explicitly designed to encourage coverage improvement and corner-case exploration are introduced to guide policy learning. By leveraging the stability and efficiency of the PPO algorithm, CA-PPO-AXI enables closed-loop, self-adaptive stimulus generation without manual constraint tuning. Experimental results on representative AXI verification scenarios show that CA-PPO-AXI reaches 95.6% functional coverage in 4,700 simulation cycles, reducing coverage convergence time by approximately 52% compared with constraint-random verification. The method also triggers 53 corner-case events, nearly doubling the detection rate of CRV. These results demonstrate that CA-PPO-AXI offers an efficient and scalable approach for intelligent, coverage-aware AXI protocol verification in advanced SoC design flows.
Zitieren
APA
Zhang, J., & Ding, J. (2026). CA-PPO-AXI: Coverage-Aware Deep Reinforcement Learning for Adaptive AXI Bus Protocol Verification. Fundamental Scientific Reports in Multidisciplinary Areas, 2(02), 159-169. https://doi.org/10.66238/fsrma102
Vancouver
Zhang J, Ding J. CA-PPO-AXI: Coverage-Aware Deep Reinforcement Learning for Adaptive AXI Bus Protocol Verification. Fundamental Scientific Reports in Multidisciplinary Areas. 2026;2(02):159-169. doi:10.66238/fsrma102
BibTeX
@article{Zhang2026,
author = {Zhang, Jingsi and Ding, Jixiang},
title = {CA-PPO-AXI: Coverage-Aware Deep Reinforcement Learning for Adaptive AXI Bus Protocol Verification},
journal = {Fundamental Scientific Reports in Multidisciplinary Areas},
year = {2026},
volume = {2},
number = {02},
pages = {159--169},
doi = {10.66238/fsrma102}
}
EUVIMED