Counter Memory Mat Layout for DRAM Row Hammer Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Certain access patterns to rows of memory cells cause an increased rate of decay in nearby rows, leading to a risk of information loss due to faster decay rates than assumed in auto-refresh operations, necessitating efficient tracking and management of access counts to identify aggressor rows for targeted refresh operations.
Innovation Solution
The memory array is divided into banks with a counter mat at the edge, using a folded digit line architecture and counter logic to store access counts, allowing for faster read-change-write cycles and flagging potential aggressor rows, which are then managed through targeted refresh operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If frequent accesses are made to a single row to improve data access speed, then data access efficiency is improved, but the decay rate of nearby memory cells increases causing information loss
Solution Approach 1:
The system performs preliminary tracking of row access patterns by maintaining counters for each row. Before information loss occurs, the system identifies aggressor rows through these counters and proactively performs targeted refresh operations on victim rows, preventing the decay that would otherwise lead to information loss.
Solution Approach 2:
The system implements a feedback mechanism where access counters continuously monitor row access frequencies. This feedback information is used to dynamically adjust refresh operations, identifying aggressor rows and targeting their neighboring victim rows for refresh, thereby adapting the refresh strategy based on actual access patterns to maintain data integrity.
2Reliability
If targeted refresh operations are performed on victim rows to prevent information loss, then data integrity is improved, but the complexity of refresh management increases
Solution Approach 1:
The memory system is segmented into aggressor rows and victim rows based on access pattern analysis. The counter mat is separately organized to track access counts for each row independently. This segmentation allows the refresh controller to selectively apply refresh operations only to victim rows neighboring aggressor rows, rather than managing refresh for the entire memory array, thereby reducing overall management complexity.
Solution Approach 2:
Different refresh strategies are applied to different regions of the memory array. Victim rows neighboring aggressor rows receive targeted refresh operations, while other rows follow standard refresh schedules. This localized approach to refresh management reduces the overall complexity by focusing computational resources only on the specific rows that require additional attention.
3Measurement precision
If access counters are maintained for all rows to identify aggressor rows, then the precision of aggressor identification is improved, but the memory resources required increase
Solution Approach 1:
The counter mat is segmented and organized separately from the main data storage array. Counters are arranged in a dedicated structure with specific word lines and bit lines allocated for counting purposes. This segmentation allows precise tracking of access patterns for all rows without consuming excessive resources from the main data storage capacity.
Solution Approach 2:
The counter mat utilizes a different dimensional organization compared to the data array. By arranging counters along specific word lines (e.g., word line N+1) and using bit lines for counter bits, the system creates a separate dimensional space for access pattern tracking. This dimensional separation enables precise measurement of access frequencies while maintaining efficient use of memory resources.
Data Source
AI summary
Apparatuses and methods for row hammer counter mat. A memory array may have a number of memory mats and a counter memory mat. The counter mat stores count values, each of which is associated with a row in one of the other memory mats. When a row is accessed, the count value is read out, changed, and written back to the counter mat. In some embodiments, the count value may be processed within access logic of the counter mat, and a row hammer flag may be provided to the bank logic. In some embodiments, the counter mat may have a folded architecture where each sense amplifier is coupled to multiple bit lines in the counter mat. The count value may be used to determine if the accessed row is an aggressor so that its victims can be refreshed as part of a targeted refresh.