RowHammer

Rowhammer is a vulnerability that affects the design and implementation of software, particularly evident in DRAM memory structures. This flaw allows attackers to manipulate system memory by repeatedly accessing specific areas of RAM, causing read disturbance in neighboring cells where important operating system data might be stored. The vulnerability was first exposed in 2015 by Google researchers who demonstrated how RowHammer could be used to gain unlimited access to a computer's RAM. Subsequent studies have explored the structure and characteristics of DRAM to improve processing, enhance reliability, and mitigate the RowHammer vulnerability. More recently, these studies have shown that attackers can exploit system-level techniques using the RowHammer vulnerability to precisely flip bits in Deep Neural Networks (DNN) model weights, affecting inference accuracy. In 2021, researchers introduced the BlackSmith attack, demonstrating that under certain conditions, failures could occur even with RowHammer protections in place. Comparatively, a newer variant known as RowPress has been identified as more potent than RowHammer, requiring fewer "hammerings" to achieve its goal. This variant slightly modifies the memory access method, allowing attackers to bypass protections implemented by module manufacturers without needing to heat or "cheat". Despite the increased potency of RowPress, the likelihood of a home user experiencing a RowHammer, Blacksmith, or RowPress attack remains miniscule. The seriousness of the RowHammer vulnerability led to the issuance of CVE-2023-42465, an initiative to make sudo less vulnerable to RowHammer attacks. Austria-based security researchers Andreas Kogler and Jonas Juffinger have been guiding attendees through a decade of RowHammer attacks, educating on this notorious exploit hackers use to manipulate memory. Despite the ongoing research and mitigation efforts, RowHammer continues to pose a significant threat to computer systems, underscoring the need for continued vigilance and protective measures against such attacks.