Pinpoint Rowhammer: Suppressing Unwanted Bit Flips on Rowhammer Attacks

Abstract

In recent studies, sophisticated attack vectors that use a Rowhammer bug have been developed. These attacks are dangerous, given that they can corrupt data stored in arbitrary memory rows without accessing them. Successful Rowhammer attacks require to flip data of the target cell. However, non-target cells are also corrupted by the attacks. Such unwanted bit flips can lead to unexpected consequences such as an attack failure and a system crash. We propose a novel Rowhammer method, namely, Pinpoint rowhammer, which flips the target bit while suppressing unwanted bit flips. The basic idea is the use of an effective data pattern for the target bit and ineffective data patterns for non-target bits. We evaluate the proposed method by conducting 107,965 attack instances on four different dynamic random-access memory (DRAM) modules. The proposed method increases the attack success rate from 28.9% to 72.4%, when compared with the state-of-the-art method (double-sided Rowhammer). In addition, the proposed method suppresses 99.7% of the unwanted vulnerable cells.