NoCSNet: Network-on-Chip Security Assessment Under Thermal Attacks Using Deep Neural Network
Meisam Abdollahi, Mohammad Hossein Chegini, Mahdi Hasanzadeh Hesar, Samaneh Javadinia, Ahmad Patooghy, Amirali Baniasadi · 2024
As the demand for high-performance computing continues to rise, Network-on-Chip (NoC) architectures play a crucial role in enabling efficient data transmission within complex systems. However, the sensitivity of NoCs to intentional thermal fluctuations opens doors to conducting Denial of Service (DoS) attacks that can alter the system’s reliability and security. In this paper, for the first time, we introduce NoCSNet as a novel database of NoC traffic collected under various network configurations and thermal attack scenarios. We also use Deep Neural Networks (DNNs) to analyze the collected traffic to enhance data transmission security in the presence of thermal DoS attacks. Through comprehensive experimentation and evaluation, we demonstrate the effectiveness of NoCSNet in capturing the security profile of NoC architectures, which can be actively used in protecting NoCs’ data integrity and stability against thermal DoS attacks. The experimental results indicate that among the MLP, LSTM, and RNN deep neural networks, the RNN approach provides the highest attack detection accuracy of 93.8%. We anticipate that the collected dataset will help the community develop a deeper understanding of the susceptibility of NoCs against thermal DoS attacks.