Imagine a security wall so high that no software exploit could possibly scale it. Now imagine that instead of climbing the wall, an attacker simply shakes the ground until the wall collapses. That is essentially what is happening with DRAM read disturbances. By manipulating how memory is accessed at the hardware level, attackers can flip bits—changing 0s to 1s—to bypass the most stringent software security layers.
The Art of the Hammer and the Press
For years, the industry has known about "RowHammer." This occurs when a specific row of memory is repeatedly opened and closed (hammered), causing electrical leakage that flips bits in adjacent rows. It's a chaotic, high-frequency attack.
But there is a quieter, equally dangerous sibling: "RowPress." Instead of the rapid-fire hammering, RowPress involves keeping a single DRAM row open for an extended period. While RowHammer is about the frequency of access, RowPress is about duration. Both result in the same nightmare for security: unauthorized data modification without ever "writing" to the victim row.
Bridging the Modeling Gap
Recent research is finally bridging the gap between how these attacks look in a lab and how they function at a device level. We now understand that these aren't just random glitches; they are predictable physical phenomena. For instance, RowHammer often induces specific leakage patterns, while RowPress can draw electrons in a way that favors different types of bitflips (like "1" to "0").
As DRAM cells shrink to pack more gigabytes into our devices, they become more sensitive. The closer these cells get, the easier it is for an "aggressor" row to disturb its neighbor, making modern hardware inherently more vulnerable to these electrical leaks.
Sources
Media
