Empowering Hardware Security with LLM: The Development of a Vulnerable Hardware Database

Authors

Dipayan Saha, K. Yahyaei, Sujan Kumar Saha, Mark Tehranipoor, Farimah Farahmandi

Abstract

This paper addresses the significant challenge of a lack of comprehensive hardware design databases and benchmarks specifically tailored for security tasks. Such databases are essential for developing and evaluating machine learning-based hardware security solutions. We propose utilizing Large Language Models (LLMs) as AI agents to efficiently generate extensive datasets and introduce Vul-FSM, a database comprising 10,000 vulnerable finite state machine (FSM) designs. These designs incorporate 16 distinct security weaknesses and vulnerabilities, generated using a framework called SecRT-LLM which leverages in-context learning guided by specific prompting strategies.

@inproceedings{saha2024empowering,
  title={Empowering hardware security with llm: The development of a vulnerable hardware database},
  author={Saha, Dipayan and Yahyaei, K and Saha, Sujan Kumar and Tehranipoor, Mark and Farahmandi, Farimah},
  booktitle={2024 IEEE International Symposium on Hardware Oriented Security and Trust (HOST)},
  pages={233--243},
  year={2024},
  organization={IEEE}
}