A Chiplet-Based High-Performance and Secure Hybrid Interconnection Network Design Against DoS and Sniffing Attacks
Md. Tareq Mahmud, Ke Wang · 2025
In the evolving domain of high-performance computing, chiplet-based architectures have emerged as a foundation for scaling system performance. To achieve significant performance scaling, recent research has explored hybrid inter-connection networks, consisting of silicone interposer, routers, wireless transmitters, and others. However, these architectures are vulnerable to various security threats, including active Denial-of-Service (DoS) and passive eavesdropping/sniffing attacks. To defend against these attacks, this research presents a complete security framework for chiplet-based hybrid interconnection networks for high-performance computing systems. First, we analyze the vulnerabilities present in modern chiplet-based hybrid interconnection networks and identify the attack vectors utilized by hybrid sniffing methods. To protect data integrity and confidentiality, we propose a security framework that includes collaborative detection of hybrid attacks and attackers with attacker isolation. The proposed approach establishes new standards for chiplet-based system security by reducing the possibility of data leakage and manipulation through comprehensive simulations using real-world application benchmarks. The results of this research provide a robust model for safeguarding future chiplet-based hybrid interconnection networks for high-performance computing systems, highlighting the vital need for security measures to evolve in conjunction with architectural advancements.