My coauthors and I will be presenting the paper "FlipShield: Self-Improving Software Mitigation Against Rowhammer" at the Sixth Workshop on DRAM Security (DRAMSec) this weekend. Below is a preview of the abstract:

Secure memory is built on the assumption of correct hardware execution, yet the Rowhammer vulnerability demonstrates that repeated memory accesses can induce unintended bit flips in DRAM. Although many hardware mitigations have been proposed, recent Rowhammer attacks continue to bypass existing protections, making practical software-based defenses an immediate priority. However, current software mitigations often struggle to balance efficiency and accuracy, leading to substantial performance overhead or elevated false positive rates.

In this paper, we propose the idea of a two-level self-improving software-based mitigation that utilizes the synergy between hardware-assisted detection with software-based sampling to efficiently identify and mitigate potential Rowhammer attacks. Furthermore, our system leverages attack attempts as opportunities for online profiling and vulnerability discovery, continuously improving its protection capacity over time. We present and evaluate a proof-of-concept prototype of the Rowhammer early detection stage, demonstrating higher accuracy than existing PMC-based detection approaches while incurring low performance overhead on the SPEC2006 benchmark.