ImPress: Securing DRAM Against Data-Disturbance Errors via Implicit Row-Press Mitigation

TL;DR

ImPress addresses the security risk posed by Row-Press (RP), a newer Data-Disturbance Error mechanism in DRAM that can flip bits with far fewer activations than Rowhammer (RH). The authors introduce a Unified Charge-Leakage Model to quantify the combined damage from RH and RP and propose two implementations: ImPress-N (naive, integer-based) and ImPress-P (precise, fractional). ImPress-N converts RP activity within tRC windows into equivalent RH activations, preserving existing RH thresholds but hampering in-DRAM trackers; ImPress-P measures the actual RP open time to compute an accurate Equivalent Activation Count (EACT) and updates trackers with fractional activations, preserving TRH and providing compatibility with both MC-based and in-DRAM trackers. Across Graphene, PARA, Mithril, and MINT, ImPress-P achieves comparable or better reliability with substantially lower performance and storage overheads than ExPress and ImPress-N, while avoiding JEDEC changes. Overall, ImPress enables pragmatic, transparent protection against RP without sacrificing RH tolerance or requiring changes to memory specifications.

Abstract

DRAM cells are susceptible to Data-Disturbance Errors (DDE), which can be exploited by an attacker to compromise system security. Rowhammer is a well-known DDE vulnerability that occurs when a row is repeatedly activated. Rowhammer can be mitigated by tracking aggressor rows inside DRAM (in-DRAM) or at the Memory Controller (MC). Row-Press (RP) is a new DDE vulnerability that occurs when a row is kept open for a long time. RP significantly reduces the number of activations required to induce an error, thus breaking existing RH solutions. Prior work on Explicit Row-Press mitigation, ExPress, requires the memory controller to limit the maximum row-open-time, and redesign existing Rowhammer solutions with reduced Rowhammer threshold. Unfortunately, ExPress incurs significant performance and storage overheads, and being a memory controller-based solution, it is incompatible with in-DRAM trackers. In this paper, we propose Implicit Row-Press mitigation (ImPress), which does not restrict row-open-time, is compatible with memory controller-based and in-DRAM solutions and does not reduce the tolerated Rowhammer threshold. ImPress treats a row open for a specified time as equivalent to an activation. We design ImPress by developing a Unified Charge-Loss Model, which combines the net effect of both Rowhammer and Row-Press for arbitrary patterns. We analyze both controller-based (Graphene and PARA) and in-DRAM trackers (Mithril and MINT). We show that ImPress makes Rowhammer solutions resilient to Row-Press transparently, without affecting the Rowhammer threshold.

Figures (19)

• Figure 1: Towards practical Row-Press solution: (a) Attack pattern for Rowhammer and Row-Press (b) Impact of RowPress on the Rowhammer Threshold (c) Explicit Row-Press (ExPress) rowpress mitigation, which limits the aggressor row-open time (tON) to tMRO, reduces the tolerated RH threshold (d) Our proposal, Implicit Row-Press (ImPress) mitigation, treats a row-open for tRC as equivalent to an activation (ACT) for RH-mitigation, retains the same RH threshold.
• Figure 2: Pattern for the Row-Press Attack (PRE denotes Precharge operation)
• Figure 3: Performance impact of limiting the time a row is open to a particular value, termed as tMRO (Maximum Row-Open Time). While SPEC workloads (low/medium spatial locality) are less sensitive to tMRO value, Stream workloads (high spatial locality) can suffer significant slowdown at low tMRO.
• Figure 4: Reduction in Tolerated TRH (T*) if the maximum tON is constrained to tMRO (Note: the data is obtained from Table-8 of rowpress)
• Figure 5: Performance of Graphene and PARA as tMRO is varied. Stream has slowdown at low tMRO. [Note: All values are geometric means.]