Per-Row DRAM Hammer Tracking for Dynamic Rowhammer Mitigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rowhammer mitigation methods in DRAM systems are inadequate in effectively managing the rowhammer phenomenon, which can lead to data corruption due to bit flips caused by repeated row activations.
Innovation Solution
Implementing per-row hammer tracking and mitigation systems that count row activation commands, identify aggressor and victim rows, and initiate targeted refresh operations to maintain data integrity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If row activation rate is increased to improve productivity, then data access speed improves, but rowhammer effects increase causing bit flips and data corruption
Solution Approach 1:
The system performs preliminary identification of aggressor rows by counting row activation commands before bit flips occur. The counter circuitry tracks activation frequencies and identifies rows exceeding thresholds, enabling preventive mitigation through targeted refresh operations on victim rows before data corruption happens.
Solution Approach 2:
The system implements feedback by continuously monitoring row activation rates and using this information to dynamically adjust refresh operations. The counter circuitry provides real-time feedback on activation frequencies, enabling the memory controller to adaptively identify aggressor rows and apply mitigative refreshes to protect victim rows, creating a closed-loop system that responds to actual usage patterns.
2Reliability
If per-row hammer tracking is implemented to improve data integrity, then bit flip probability decreases, but device complexity increases
Solution Approach 1:
The tracking system is segmented into dedicated counter circuitry for each row, allowing independent monitoring of activation frequencies. This segmentation enables precise identification of aggressor rows without requiring complex global analysis, as each row's activation count is tracked separately and compared against thresholds to determine victim rows for refresh operations.
Solution Approach 2:
The system uses the row activation data itself to identify aggressor rows, eliminating the need for external monitoring or complex analysis tools. The counter circuitry automatically tracks activations and the memory controller autonomously identifies aggressor-victim relationships based on activation patterns, allowing the system to self-diagnose and self-protect without additional complexity.
Data Source
AI summary
Managing row hammering in a DRAM device may include maintaining per-row activation command counts. A next aggressor row may be determined based on the counts. A victim queue may be maintained. A refresh operation may be directed to a row indicated by the victim queue when conditions include that the victim queue is not empty when the refresh command is received. The current aggressor row may be updated with the next aggressor row when conditions include that the victim queue is empty when the refresh command is received. Following updating the current aggressor row, the count of the next aggressor row may be updated. A victim row corresponding to the current aggressor row may be added to the victim queue if the victim queue is empty when the refresh command is received.