The article presents MABFuzz, a novel framework that uses multi-armed bandit (MAB) algorithms to fuzz processors. The complexities of modern processors make verifying their integrity and security challenging. Current hardware fuzzers often rely heavily on static strategies, which can be inefficient. MABFuzz addresses this problem with a dynamic and adaptive approach that can be applied to any existing hardware fuzzer. It has shown remarkable efficiency in covering a broad spectrum of processors and detecting vulnerabilities. The framework integrates three widely used MAB algorithms into a state-of-the-art hardware fuzzer.
Publication date: 27 Nov 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2311.14594
