AWARe-Wi: A jamming-aware reconfigurable wireless interconnection using adversarial learning for multichip systems

Abstract

Performance of the compute-intensive multichip platforms such as micro-servers and embedded systems are limited by the latency and power hungry chip-to-chip interconnections. Millimeter wave (mm-wave) wireless interconnection networks have emerged as an energy-efficient and low-latency solution for such multichip system communication. We refer such multichip systems with in-package mm-wave wireless interconnect as Wireless Network-in-Package (WiNiP). Despite providing performance enhancements, wireless channel, being an unguided medium, introduces potential security vulnerabilities inherited from traditional wireless networks such as jamming induced Denial-of-Service (DoS) and eavesdropping. Securing the systems against such induced threats often introduce large overheads and performance penalties. To address these challenges, we propose a WiNiP architecture that reuses the in-built Design for Testability (DFT) hardware for securing against external and Hardware Trojans (HT) induced internal attacks. The proposed architecture is capable of securing against adversaries with a reconfigurable wireless interconnection (AWARe-Wi). We deploy machine learning (ML) classifier to detect the threats. In addition, for a robust threat detection, we introduce an Adversarial ML (AML)-based approach in this work. To enable sustainable multichip communication in such systems even under jamming attack from both internal and external attackers, we design a reconfigurable Medium Access Control (MAC) and a suitable communication protocol. The simulation results show that, the ML and AML classifiers can achieve an accuracy of 99.87% and 95.95% respectively for attack detection while the proposed WiNiP can sustain chip-to-chip communication even under persistent jamming attack with an average 1.44× and 1.56× degradation in latency for internal and external attacks respectively for application-specific traffic patterns.