Hybrid Row Activation Tracking for DRAM Row Hammer Mitigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing severity of row hammer interference in DRAM due to relentless scaling poses a reliability and security threat, necessitating effective mitigation strategies that are efficient and scalable for future systems.
Innovation Solution
A hybrid tracker system combining SRAM-based group-count tables and DRAM-based per-row trackers to monitor and mitigate row hammer effects, utilizing SRAM for efficient counting and DRAM for precise tracking of high-activation rows, with performance optimizations through caching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If per-row tracking is implemented for all rows, then row hammer mitigation precision is improved, but resource overhead and performance penalty increase significantly
Solution Approach 1:
The patent divides the tracking structure into two segments: an SRAM-based group-count table that tracks aggregated counts for groups of rows, and a DRAM-based per-row tracker that tracks individual row counts. This segmentation allows the system to achieve precise per-row tracking only when necessary, while using less resource-intensive group-level tracking for the majority of rows, thereby reducing overall resource overhead while maintaining mitigation precision.
2Speed
If SRAM is used for all tracking, then access speed is improved, but area and cost increase significantly
Solution Approach 1:
The patent applies different memory technologies to different tracking needs: SRAM is used for the group-count table where fast access is critical for filtering the majority of rows, while DRAM is used for the per-row tracker where larger capacity is needed but access speed requirements are lower. This local differentiation of memory quality optimizes the balance between access speed and area utilization.
3Reliability
If tracking threshold is reduced to address future row hammer severity, then mitigation effectiveness is improved, but false positives and performance overhead increase
Solution Approach 1:
The patent performs preliminary filtering using the SRAM-based group-count table before detailed per-row tracking. By first identifying groups of rows that exceed aggregation thresholds, the system can then apply more intensive per-row tracking only to those specific groups, thereby reducing false positives and performance overhead while maintaining effectiveness against future row hammer attacks with lower thresholds.
Data Source
AI summary
A hybrid tracker system and method are disclosed for RH mitigation, which combines the best of both SRAM and DRAM to enable low-cost mitigation of RH at ultra-low thresholds. The system consists of two structures. First, an SRAM-based structure is provided that tracks aggregated counts at the granularity of a group of rows and is sufficient for the vast majority of rows, which receive only a few activations. Second, a per-row tracker stored in the DRAM-array is provided which can track an arbitrary number of rows; however, to limit performance overheads, this tracker is used only for the small number of rows that exceed the tracking capability of the SRAM-based structure.