Probabilistic Memory Row Tracking for Rowhammer Mitigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory attack mitigation techniques, such as targeted row refresh (TRR) and deterministic tracking algorithms, are vulnerable to thrashing-based attacks and incur high storage overheads or performance overheads, making them impractical for commodity DRAM, while trackerless solutions cannot be implemented transparently within DRAM.
Innovation Solution
Implement probabilistic tracker management by pseudo-randomly or randomly sampling a subset of memory access requests, updating a tracker with these requests, and performing mitigation actions based on the updated tracker to prevent memory attacks like rowhammer and rowpress.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deterministic tracking algorithms with larger number of counters are used, then security against thrashing-based attacks is improved, but storage overhead increases significantly
Solution Approach 1:
The patent changes the fundamental parameter of tracking from deterministic to probabilistic. Instead of maintaining a fixed set of counters that deterministically track specific rows, the system uses a probabilistic tracker that maintains a fixed number of counters but tracks rows based on probability proportional to their activation frequency. This parameter change resolves the contradiction by providing security against thrashing attacks (since attackers cannot predict which rows will be tracked) while maintaining a fixed, manageable counter count that limits storage overhead.
2Measurement precision
If one counter per row is maintained, then tracking accuracy is improved, but performance overhead increases up to 70%
Solution Approach 1:
The patent applies partial action by not tracking all rows uniformly. Instead of maintaining one counter per row (which would require fetching and updating thousands of counters), the probabilistic tracker maintains a fixed small number of counters and selectively tracks only a subset of rows based on probability. This partial tracking approach maintains sufficient accuracy for security purposes while dramatically reducing the performance overhead associated with fetching and updating counters.
Solution Approach 2:
The patent extracts the tracking function from the memory array itself, where counters would be stored in DRAM requiring additional accesses. Instead, the probabilistic tracker is implemented in the memory controller using a fixed small number of counters, separating the tracking functionality from the memory storage and eliminating the need for additional DRAM accesses to fetch and update counters.
3Adaptability or versatility
If counters are stored entirely within DRAM array, then integration is improved, but device complexity increases due to complex redesign
Solution Approach 1:
The patent introduces an intermediary structure - the probabilistic tracker in the memory controller - that mediates between the memory controller and the DRAM array. Instead of storing counters within the DRAM array (which would require complex MAT redesign), the tracker resides in the memory controller with a fixed small number of counters. This intermediary approach provides the necessary tracking functionality without requiring complex redesign of the DRAM memory arrays, making it suitable for commodity DRAM.
Data Source
AI summary
Rowhammer attacks, which are malicious processes that rapidly issue access requests to memory, can impose serious security threats including being used to tamper data, take control of entire systems, and even breach confidentiality. Current solutions to defend against these attacks are limited, as they typically employ a deterministic tracker to track the portions of memory accessed and to mitigate potential attacks accordingly. However, the deterministic nature of these trackers results in their own vulnerability. The present disclosure provides probabilistic tracker management for mitigation of rowhammer attacks and/or other memory attacks in which a row (or other defined portion of memory) is maliciously targeted to disturb contents of neighboring rows, which can prevent these types of attacks that otherwise take advantage of the determinism in prior used tracker designs.