The problem: "The use of an FPGA by several users at the same time is a gateway for malicious attacks," says Gnad. Tricky hackers can use the versatility of FPGAs to perform side-channel attacks. In such a scenario, attackers use the chip's energy consumption to extract information with which they can crack its encryption. Such chip-internal measurements allow one customer of the cloud service to spy on another. In addition, hackers could not only spy out treacherous fluctuations in power consumption, they could also generate them themselves. "This could falsify the calculations of other customers or even cause the entire chip to crash, which could result in data loss," explains Krautter. Similar dangers exist with other IC architectures, Gnad continues. For example, SoCs, which are frequently used in Internet of Things applications.
Gnad and Krautter intend to solve the problem by restricting users' direct access to FPGAs. But this is not an easy task: the difficulty is to filter out malicious users without restricting benevolent users too much.
Original publication: https://doi.org/10.13154/tches.v2019.i3.305-339