AI Framework Detects Hardware Vulnerabilities in Generated Code with High Accuracy

A new embedding-based framework called VeriCWEty identifies common vulnerabilities in hardware description code generated by large language models. The system detects and classifies bugs at both module and line-level granularity, achieving approximately 89% precision in identifying specific Common Weakness Enumerations like CWE-1244 and CWE-1245. While LLMs have significantly improved Register Transfer Level code generation, the resulting code often contains exploitable weaknesses that can evade traditional detection methods. Existing techniques relying on rule-based checks, formal properties, or coarse-grained structural analysis frequently miss semantic vulnerabilities or lack precise localization. This research bridges that gap by providing detailed bug detection that helps prevent attackers from exploiting these hardware vulnerabilities. The framework demonstrates 96% accuracy in detecting line-level bugs, offering a more refined approach to hardware security verification. The work addresses a critical need as AI-generated hardware code becomes more prevalent in computer architecture design.