Enhancing Anonymity in NoC Communication to Counter Traffic Profiling by Hardware Trojans

Over time, interconnect systems have progressed from basic buses to more scalable networks-on-chip (NoC) IP. Since NoC is what drives SoC's performance, researchers pay close attention to its security needs. NoC can be used as a significant platform by adversaries from unreliable third-party IP providers to implant malicious circuits like hardware trojans (HT). Since it is directly related to the chip's functional specifications and floor plan design, any security breaches would malfunction the whole SoC itself. When packets are exchanged anonymously, it is more challenging for attackers to determine which nodes are communicating. This prevents decoding the nature and purpose of applications running on the cores. Only encryption cannot stop HTs from identifying communicating nodes in a tiny medium like NoC. We report the limitations of traditional anonymous communication in the presence of HTs that can predict the source node of NoC packets using lightweight traffic analysis. The HTs can even target certain nodes and leak packets from NoCs built with existing anonymous systems. In order to strengthen anonymity in NoC, both packet encryption and a routing method that explores path diversity are to be considered equally. We propose extended DyXY (E-DyXY) routing that introduces path diversity between nodes to reduce the source prediction accuracy of HTs at the expense of a marginal increase in packet latency. Experimental results show that we can reduce the rate of targeted packet leakage by 90% in anonymous systems embedded with the proposed E-DyXY algorithm when compared with baseline anonymous systems employing traditional XY routing.