Lossy Row Access Counting for Row Hammer Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor memory devices, repeated access to a particular row of memory cells can cause increased data degradation in neighboring rows due to electromagnetic coupling, necessitating the identification and periodic refresh of affected memory cells to prevent information loss.
Innovation Solution
A system that monitors row access operations to identify aggressor rows by using a data storage unit with registers to count access frequencies, employing a filter mechanism to differentiate between true aggressor rows and noise rows, and periodically updating these counts to maintain accurate tracking of frequently accessed rows.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If row access counting is performed to identify aggressor rows, then the ability to identify and refresh victim rows is improved, but the complexity of the device increases due to additional counting and filtering mechanisms
Solution Approach 1:
The row address is segmented into a most significant portion and a least significant portion. The counting mechanism only tracks accesses to rows sharing the same most significant portion, effectively dividing the monitoring scope into manageable segments that reduce overall system complexity while maintaining reliability.
Solution Approach 2:
A filter mechanism acts as an intermediary between the row access signal and the counting mechanism. This filter pre-processes the address signals by comparing them against stored victim row addresses, allowing only relevant addresses to reach the counter, thereby simplifying the counting logic and reducing the complexity of the overall system.
2Measurement precision
If all row accesses are tracked continuously, then accurate identification of aggressor rows is improved, but the loss of time for monitoring and updating counts increases
Solution Approach 1:
Instead of continuously monitoring all row accesses, the system performs periodic updates of the count values at predetermined intervals. This periodic action maintains sufficient measurement precision for identifying aggressor rows while significantly reducing the time loss associated with constant monitoring and updating operations.
Solution Approach 2:
The system extracts only the necessary information from row accesses by using a filter that compares addresses against stored victim row patterns. This extraction approach focuses monitoring efforts only on relevant rows, improving identification accuracy while minimizing the time required for monitoring by ignoring irrelevant accesses.
3Measurement precision
If a filter mechanism is implemented to distinguish true aggressor rows from noise rows, then the precision of identification is improved, but the device complexity increases due to additional filtering components
Solution Approach 1:
The filter mechanism uses segmentation of the row address into most significant and least significant portions. By storing only the most significant portions of victim row addresses and comparing incoming addresses against these segments, the filter achieves high detection precision while keeping the filtering structure simple and manageable.
Solution Approach 2:
The filter stores simplified copies of victim row addresses (only the most significant portions) rather than complete address patterns. This copying approach maintains the ability to accurately identify true aggressor rows while reducing the complexity of the filtering mechanism by working with smaller, more manageable data representations.
4Reliability
If the memory device monitors and counts all row accesses, then the ability to prevent data degradation is improved, but the use of energy for monitoring and counting operations increases
Solution Approach 1:
The counting and monitoring operations are performed periodically at predetermined intervals rather than continuously. This periodic execution maintains effective data protection capability while significantly reducing energy consumption by keeping the monitoring logic inactive during intervals between updates.
Solution Approach 2:
The system extracts and processes only the essential information needed for detection by using a filter that compares addresses against stored patterns. This selective processing approach maintains reliable data protection by identifying true aggressor rows while minimizing energy use by avoiding comprehensive monitoring of all memory accesses.
Data Source
AI summary
Embodiments of the disclosure are drawn to apparatuses, systems, and methods for lossy row access counting. Row addresses along a to address bus may be sampled. When the row address is sampled it may be compared to a plurality of stored addresses in a data storage unit. If the sampled address matches one of the stored addresses, a count value associated with that address may be updated in a first direction (such as being increased). Periodically, all of the count values may also be updated in a second direction (for example, decreased).