Deep Reinforcement Learning-Based Automatic Test Case Generation for Hardware Verification

Abstract

This paper presents a novel deep reinforcement learning-based framework for automatic test case generation in hardware verification. The proposed approach combines traditional verification methods with advanced deep learning techniques to enhance test coverage and security vulnerability detection. The framework incorporates a modified Deep Q-Network architecture with prioritized experience replay, integrated with static analysis and dynamic mutation strategies. The system utilizes a comprehensive reward mechanism that considers multiple coverage metrics, including line coverage, toggle coverage, FSM coverage, and security asset coverage. Experimental evaluation of diverse benchmark designs, including AES cores, RISC-V processors, and network controllers, demonstrates significant improvements over conventional methods. The results show an average coverage improvement of 17.2% and a 65% reduction in verification time compared to traditional approaches. The framework achieves 95.4% average coverage across benchmark designs and a 94.8% detection rate for security vulnerabilities. Additionally, the system demonstrates good scalability characteristics, maintaining performance efficiency across varying design complexities. The experimental results validate the effectiveness of the proposed approach in automating hardware verification processes while improving test coverage and security vulnerability detection capabilities.