Paper 2020/1494

Nonintrusive and Adaptive Monitoring for Locating Voltage Attacks in Virtualized FPGAs

Seyedeh Sharareh Mirzargar and Gaiëtan Renault and Andrea Guerrieri and Mirjana Stojilović

Abstract

With every new generation, high-end FPGAs are becoming richer in features and resources, making the usage model of single-user per FPGA decreasingly cost-efficient. Although virtualized FPGAs enable multiple users to share the same FPGA, this multi-tenancy is not employed in practice because of potential security threats, such as voltage attacks. These attacks use power-wasting circuits to exercise excessive switching activity on the target FPGA to cause extreme voltage fluctuations, which produce timing faults in collocated circuits or, in extreme cases, reset target FPGA. In this work, we present the idea of automated embedding of the on-chip voltage sensors into the virtualized FPGAs and continuous monitoring of the core voltage for suspected fluctuations caused by a voltage attacker. Our sensors are nonintrusive and placement-adaptive because we implement them immediately after placing and routing the user design with resources that are left unused. We devise a novel measurement technique to continuously analyze the sensor outputs and locate the power-wasting circuits. Additionally, we are the first to use a synchronous power-wasting attacker, capable of producing timing faults, on Xilinx 7-series FPGAs and to successfully locate it. Hence, our proposed monitoring system enables the virtualized FPGA to identify the voltage attackers, at minimal cost, and prevent them from repeating the attack.