Security Threats in Approximate Computing Networks-on-Chip

P. Ampadu and S. Ma
Virginia Tech, United States

Keywords: Hardware security, NoC, approximate computing, SoC, AI

AI applications to enhance recognition and decision-making lend themselves to approximations, in order to enable compact, lightweight, high-speed, and energy-efficient compute, sensing, and actuation platforms. Networks-on-Chip (NoCs) have emerged as an effective hardware platform to process, communicate, and store large amounts of information in parallel and are well suited to implementation of AI algorithms. Unfortunately, the parallel and shared nature of NoCs also expose them to unique security attacks not common in bus-based and other traditional communication architectures. This research investigates those vulnerabilities and develops countermeasures to address them.