Memory Device Row Hammer Refresh via Count Cells
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The row hammer phenomenon, where data stored in memory cells connected to an adjacent word line is lost due to aggressive access of a specific word line, poses a significant risk in volatile memory devices like DRAM.
Innovation Solution
A memory device is designed with a bank structure that includes count cells to store access count data for each row and a register to manage addresses and count data. The device performs a row hammer refresh operation based on the information stored in the register, effectively managing the risk of the row hammer phenomenon.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If count cells are used to store access count data for each row, then the ability to detect and manage row hammer attacks is improved, but the device complexity increases due to additional storage structures
Solution Approach 1:
The memory device is divided into functional segments: count cells for storing access counts, registers for managing critical row information, and a row hammer management circuit for coordinating refresh operations. This segmentation allows precise tracking of access patterns while distributing complexity across specialized components rather than requiring a completely new memory architecture.
Solution Approach 2:
The row hammer management circuit acts as an intermediary between the count cells/registers and the memory refresh system. It monitors access counts, identifies potential row hammer attacks, and triggers targeted refresh operations on adjacent rows, thereby preventing data loss without requiring complete memory system redesign.
2Reliability
If a register is added to manage addresses and count data, then the effectiveness of row hammer prevention is improved, but the device complexity increases
Solution Approach 1:
The register stores addresses of rows with high access counts and their associated count data in advance, before row hammer attacks can cause data loss. This preliminary tracking enables the row hammer management circuit to proactively identify vulnerable rows and perform preventive refresh operations on adjacent rows before data corruption occurs.
Solution Approach 2:
The system implements feedback through the row hammer management circuit, which continuously monitors access counts stored in count cells and register data, compares them against thresholds, and automatically triggers refresh operations when row hammer attack patterns are detected. This closed-loop feedback mechanism improves reliability by dynamically responding to actual access patterns.
3Reliability
If targeted row hammer refresh operations are performed, then data integrity is improved, but the use of energy increases due to additional refresh operations
Solution Approach 1:
Instead of performing uniform refresh operations across the entire memory array, the system applies refresh operations locally only to rows adjacent to those identified as potential aggressors. The row hammer management circuit calculates which specific rows need refreshing based on access count data, thereby maintaining data integrity in vulnerable regions while avoiding unnecessary energy expenditure in safe regions.
Solution Approach 2:
The system performs refresh operations on a selective basis rather than refreshing all rows. By using access count thresholds and register-based tracking, it applies refresh operations only when and where needed to prevent row hammer attacks, reducing overall energy consumption compared to comprehensive refresh strategies while maintaining sufficient data integrity protection.
Data Source
AI summary
A memory device includes at least one bank including memory cells, and a register corresponding to the at least one bank. The bank includes a plurality of rows that are arranged in a row direction, and a plurality of count cells, each of which stores count data associated with the number of times of access of each of the plurality of rows. The register stores an address and count data, which are associated with at least one row among the plurality of rows.