A group of security researchers from the ETH Zurich university and Google have demonstrated a practical Rowhammer attack against DDR5.
Dubbed Phoenix and tracked as CVE-2025-6202, the DDR5 Rowhammer attack was found to be effective against 15 devices from SK Hynix, the largest DRAM manufacturer.
As part of a Rowhammer attack, a DRAM memory row is accessed repeatedly to cause electrical interference leading to bit flips in adjacent regions. This could lead to elevation of privileges, data corruption, data leakage, and in breaking memory isolation in virtual environments.
After more than a decade of known Rowhammer attacks targeting CPUs and CPU-based memory, a group of University of Toronto researchers this year demonstrated that such attacks are possible and practical against GPUs as well.
The newly devised Phoenix attack shows that, despite its more sophisticated in-DRAM Target Row Refresh (TRR) mechanisms meant to prevent Rowhammer attacks, DDR5 too is vulnerable.
To prove that, four ETH Zurich academics and two Google researchers reverse-engineered the TRR schemes in DDR5, discovering that a successful attack needs to “precisely track thousands of refresh operations”.
In their paper (PDF), the researchers explain that the protections DDR5 comes with require significantly longer Rowhammer patterns to be bypassed, and that these patterns need to remain in-sync with thousands of refresh commands.
Phoenix, however, was designed to resynchronize the pattern when missed refresh operations are detected, thus triggering bit flips that allowed the researchers to create a privilege escalation exploit and gain root on a commodity DDR5 system with default settings.
“We evaluate Phoenix on 15 DDR5 DIMMs from SK Hynix and show that it can trigger bit flips on all of them. We also demonstrate that the bit flips are exploitable by building the first Rowhammer privilege escalation exploit running in default settings on a PC in as little as 109 seconds,” the researchers note.
The researchers say they limited their work to SK Hynix devices due to the extensive effort of reverse engineering the implemented mitigations, and point out that DDR5 devices from other manufacturers should not be considered protected against Rowhammer attacks.
Tripling the refresh rate, the researchers say, prevents Phoenix from triggering bit flips, but incurs an overhead of 8.4%. More principled mitigations, such as per-row activation counters, should stop Rowhammer attacks completely, they say.
Phoenix was disclosed to SK Hynix, CPU vendors, and major cloud providers in early June. Last week, AMD released BIOS updates to address CVE-2025-6202 in client machines, the researchers note.
Related: VMScape: Academics Break Cloud Isolation With New Spectre Attack
Related: AI Systems Vulnerable to Prompt Injection via Image Scaling Attack
Related: Hybrid Networks Require an Integrated On-prem and Cloud Security Strategy
Related: Webcast Video: Rethinking Endpoint Hardening for Today’s Attack Landscape
