Cloud Based FPGA Hardware Security

Abstract

Cloud compute is an opportunity that has been in the public consciousness for some time now, however more recently we have tried to introduce FPGA’s into cloud-based systems. This offers the advantage of parallelizing a workload and results in substantially higher output per voltage. However, FPGA’s have several security vulnerabilities as they can be exploited to shut down systems or to gain unauthorized access to models of other clients running on the same platform. This research is important in the cloud compute space as FPGA’s have the potential to substantially speed up server performance per Watt but currently, the security risks are vast. By removing these exploits, clients may be more willing to send sensitive tasks into the cloud thus reducing client hardware costs and hosts stand to increase revenue stream. In this project, we will attempt to replicate a cloud-based FPGA system using a Stratis V FPGA. We will use this system to run an ML model and then run an attack to extract the model or to disrupt the function of the FPGA itself. Finally, we will attempt to create a defense of the FPGA system.