Hierarchical Row Access Tracking for Memory Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In memory devices, repeated access to certain memory cells can cause increased data degradation in neighboring cells, known as the 'row hammer' effect, making it difficult to efficiently identify and refresh affected cells due to space and power limitations, leading to excessive targeted refresh operations.
Innovation Solution
A method is introduced to track accesses to memory cells by dividing them into portions and sub-portions, where count values associated with these divisions are used to determine victim rows for targeted refresh operations, reducing the number of unnecessary refreshes by identifying only the sub-portions that are actually being accessed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all memory rows are tracked individually to identify victim rows for targeted refresh operations, then the accuracy of identifying affected cells is improved, but the space complexity and power consumption increase significantly
Solution Approach 1:
The patent divides the memory address space into multiple segments or groups, where each segment is tracked by a separate counter. Instead of tracking every individual row, the system tracks segments of rows (e.g., groups of 64 rows per counter). When a segment is identified as being repeatedly accessed, targeted refresh operations are performed on the victim rows within that segment. This segmentation approach reduces the number of counters needed from potentially thousands (one per row) to a manageable number (e.g., 64 counters for 4096 rows), thereby reducing space complexity while maintaining sufficient accuracy in identifying victim rows.
2Reliability
If comprehensive row access tracking is implemented to mitigate the row hammer effect, then data integrity is improved, but power consumption increases due to excessive refresh operations
Solution Approach 1:
The patent applies local quality by performing targeted refresh operations only on specific victim rows or segments that are actually affected by row hammer, rather than refreshing all rows uniformly. When a counter indicates that a particular segment has been repeatedly accessed, the system identifies and refreshes only the victim rows within that specific segment. This localized approach ensures data integrity in affected areas while avoiding unnecessary power consumption from refreshing rows that are not under attack, thereby reducing overall power consumption while maintaining reliability.
3Measurement precision
If the granularity of row tracking is increased to identify specific victim rows, then the precision of targeted refresh is improved, but the number of counters and tracking overhead increases
Solution Approach 1:
The patent implements a hierarchical segmentation approach where the memory address space is divided into multiple levels. The first level divides memory into large segments tracked by counters. When a counter threshold is exceeded, the system drills down to identify the specific victim rows within that segment. This multi-level segmentation allows the system to achieve high precision in identifying victim rows without requiring a counter for every single row, thus balancing precision with acceptable device complexity.
4Reliability
If frequent refresh operations are performed on all rows to prevent data degradation, then data integrity is maintained, but the operation time and power consumption increase
Solution Approach 1:
The patent implements periodic targeted refresh operations based on counter thresholds rather than continuous or uniform periodic refreshing of all rows. Each counter monitors the access frequency of its associated segment, and when a threshold is reached, a targeted refresh is triggered for the affected victim rows. This event-driven periodic action ensures that refresh operations are performed only when necessary, maintaining data integrity while minimizing unnecessary refresh operations that would consume time and power.
Data Source
AI summary
Apparatuses and methods for tracking all row accesses in a memory device over time may be used to identify rows which are being hammered so that ‘victim’ rows may be identified and refreshed. A register stack may include a number of count values, each of which may track a number of accesses to a portion of the word lines of the memory device. Anytime a row within a given portion is accessed, the associated count value may be incremented. When a count value exceeds a first threshold, a second stack with a second number of count values may be used to track numbers of accesses to sub-portions of the given portion. When a second count value exceeds a second threshold, victim addresses may be provided to refresh the victim word lines associated with any of the word lines within the sub-portion.