Rethinking FPGA Security in the New Era of Artificial Intelligence

With various possible attacks against commercial electronic devices reported over the past few decades, the security of hardware devices and systems has become an urgent problem. Accordingly, a large number of solutions and countermeasures have been explored to mitigate these attacks. Artificial intelligence, as one of the fastest-growing research areas, also makes a unique impact on the landscape of vulnerabilities and countermeasures of hardware. As a vital subset of artificial intelligence, machine learning algorithms are found of great use in hardware security from both constructive and destructive perspectives. In this paper, we provide a survey of such double-edged sword impact of machine learning techniques on the security of hardware. In particular, we focus on the discussion of FPGA security. We enumerate both countermeasures and attacks based on pure machine learning algorithms, as well as the integration of machine learning and other methods, such as side-channel analysis. In addition, we also discuss the security concerns of FPGAs when they are used as carriers or accelerators for machine learning algorithms. Specifically, we present the security issues of FPGAs in two different application scenarios: 1) as a standalone computing resource and 2) as a public-leased computing resource shared by multiple users.