Search papers, labs, and topics across Lattice.
This paper introduces HiFuzz, a hierarchical reinforcement learning framework that enhances CPU fuzzing by employing a two-layer generation process involving a Program Agent for global layout and a Basic Block Agent for detailed instruction filling. The framework addresses the challenge of reward sparsity by incorporating an adaptive coverage reward mechanism alongside a semantic-aware basic block encoder, which provides intrinsic feedback during the fuzzing process. Evaluations on real-world RISC-V cores show that HiFuzz significantly surpasses existing fuzzers in both coverage and bug detection capabilities.
HiFuzz outperforms traditional fuzzing techniques by leveraging hierarchical reinforcement learning to achieve deeper architectural state exploration and improved bug detection.
Modern processor verification struggles to reach deep architectural states due to the inefficiencies of traditional mutation-based fuzzing. We propose HiFuzz, a novel hierarchical reinforcement learning framework that replaces mutation with a structured, two-layer generation process: a Program Agent for global layout and a Basic Block Agent for precise instruction filling. To overcome reward sparsity, HiFuzz integrates an adaptive coverage reward mechanism and a semantic-aware basic block encoder providing intrinsic feedback. Extensive evaluations on three real-world RISC-V cores demonstrate that HiFuzz significantly outperforms state-of-the-art fuzzers in coverage and bug detection.
