Inspection of Partial Bitstreams for FPGAs Using Artificial Neural Networks

Abstract

Incorporating FPGAs in embedded designs, both for research and industry related applications, is getting increasingly common. Due to the inherent capability of an FPGA to reconfigure itself during run-time, entirely or partially, it has become a very cost effective and time efficient solution for end-users with ever-changing needs for their embedded and custom hardware designs. This capability allowing dynamic reconfiguration of FPGAs, unfortunately also poses a threat to hardware security in terms of malicious bitstream manipulation that can include attacks through intended hardware changes by insertion of hardware trojans, spy-wares or even energy thirsty hardware modules which eventually have adverse effects on energy critical applications. In this paper, we introduce a novel approach to tackle this problem using machine learning techniques for FPGA bitstream analysis. By making use of different Neural Networks, we present how it paves a way to analyze partial FPGA bistreams to trace a certain module, or to find inconsistencies which can be malicious to the target hardware. In contrast to traditional methods to inspect bitstreams, our method saves a significant amount of time.