Memory Row Hammer Detection With Dynamic Refresh Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Memory devices are vulnerable to data degradation due to increased access rates, particularly from 'row hammer' attacks, which can cause nearby memory cells to decay before they can be refreshed, necessitating targeted refresh operations to prevent data loss and system malfunctions.
Innovation Solution
Implementing a dynamic threshold system based on random number generation to detect aggressor rows, which reduces the likelihood of multiple word lines being identified as aggressors simultaneously, thereby minimizing the duration of targeted refresh operations and maintaining system availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a static threshold is used to detect aggressor rows, then the detection process is simple and fast, but multiple word lines may be incorrectly identified as aggressors simultaneously, causing extended refresh operations and increased device unavailability
Solution Approach 1:
The patent implements a dynamic threshold that changes over time based on a pseudo-random sequence, replacing the static threshold. This dynamic approach ensures that the threshold value varies across different detection cycles, preventing systematic errors where multiple word lines consistently exceed the same fixed threshold. The dynamic nature maintains detection simplicity while significantly improving accuracy.
Solution Approach 2:
The patent changes the threshold parameter from a fixed value to a time-varying value generated by a pseudo-random sequence. This parameter change allows the detection system to adapt its sensitivity over time, reducing false positives where multiple aggressor rows are simultaneously detected. The pseudo-random variation in threshold values ensures diverse detection patterns without complicating the overall detection mechanism.
2Reliability
If targeted refresh operations are performed frequently to prevent data degradation, then data reliability is improved, but device unavailability increases and system productivity decreases
Solution Approach 1:
The patent employs feedback by using detection results from previous cycles to guide refresh operations. The dynamic threshold detection provides accurate identification of actual aggressor rows, and this information feeds into targeted refresh operations only for affected word lines. This feedback mechanism ensures refresh operations are performed only when and where needed, maintaining data integrity while minimizing disruption to system productivity.
Solution Approach 2:
Instead of performing blanket refresh operations across all memory, the patent applies partial action by targeting only the specific word lines identified as victims of aggressor row attacks. This selective approach ensures sufficient refresh coverage for affected areas while avoiding unnecessary refresh operations on unaffected memory regions, thereby maintaining system availability.
3Measurement precision
If the detection threshold is lowered to reduce false positives, then detection accuracy improves, but the detection becomes more sensitive to normal access patterns, potentially triggering unnecessary refresh operations
Solution Approach 1:
The patent uses a dynamic threshold that varies over time according to a pseudo-random sequence, replacing the need for manual adjustment of fixed threshold values. This dynamic approach naturally balances sensitivity and specificity by ensuring the threshold changes in a predictable yet non-repeating pattern, reducing false positives from normal access patterns while maintaining high detection accuracy for actual attacks.
Data Source
AI summary
Aggressor rows may be detected by comparing access count values of word lines to a threshold value. Based on the comparison, a word line may be determined to be an aggressor row. The threshold value may be dynamically generated, such as a random number generated by a random number generator. In some examples, a random number may be generated each time an activation command is received. Responsive to detecting an aggressor row, a targeted refresh operation may be performed.