LEVERAGING AI FOR AUTOMATED FUNCTIONAL COVERAGE AND TEST SEQUENCE GENERATION IN GPU AND AI ACCELERATOR VERIFICATION

Abstract

Achieving comprehensive functional coverage is crucial in design verification, particularly for complex architectures like GPUs and AI accelerators. Traditionally, writing functional coverpoints and generating test sequences is a manual and skill-intensive process, often requiring directed test cases to hit complex cross-functional coverpoints. This paper presents an innovative approach utilizing AI to automate the creation of functional coverpoints and the generation of test sequences. By providing AI with transactional and protocol information in a simplified format, the system can generate base and complex sequences to cover intricate scenarios. This methodology is extended to create versatile coverpoints for different protocols, which can be reused across interfaces following similar protocols and packet types. Additionally, the AI script facilitates the creation of a high-level verification plan (HVP) that measures effective coverage across various levels of verification abstraction. The automated process significantly reduces manpower requirements, minimizes the risk of missed datapath coverage, and adapts sequences dynamically based on feedback, ensuring thorough functional coverage and efficient verification.