Randomized DRFM Bank Targeting for Row Hammer Mitigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing row hammer attack mitigation techniques in DRAM devices, such as pseudo target row refresh (pTRR), are predictable, allowing attackers to evade mitigation operations, and directing DRFM commands to all or same banks results in inefficiency and performance impact.
Innovation Solution
Randomize the sending of DRFM commands across multiple banks, using randomized address capture and DRFM selection based on Activate monitoring, to increase the likelihood of effective row hammer mitigation without performance loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If DRFM commands are directed to all banks or same bank in all bank groups, then row hammer mitigation coverage is improved, but bandwidth consumption increases and command efficiency decreases
Solution Approach 1:
The patent applies local quality by directing DRFM commands to specific banks rather than all banks or same bank in all bank groups. The memory controller selectively issues DRFM commands to banks based on their individual row hammer risk profiles, optimizing the balance between mitigation coverage and bandwidth efficiency.
Solution Approach 2:
The patent segments the memory subsystem into individual banks with independent row hammer risk assessment. Each bank is monitored separately for row hammer conditions, and DRFM commands are issued to specific banks only when needed, rather than applying a blanket approach to all banks.
2Device complexity
If pTRR technique is used with predictable address selection, then implementation complexity is reduced, but attack evasion capability increases
Solution Approach 1:
The patent introduces dynamics by making the DRFM command targeting adaptive rather than static. The memory controller dynamically adjusts which banks receive DRFM commands based on real-time row hammer risk assessment, transitioning from predictable static addressing to adaptive dynamic addressing that responds to actual attack conditions.
Solution Approach 2:
The patent implements feedback mechanisms where the memory controller monitors row hammer conditions in each bank and adjusts DRFM command distribution accordingly. This feedback loop allows the system to learn from actual attack patterns and modify its mitigation strategy to prevent attack evasion.
3Reliability
If DRFM commands are issued frequently to multiple banks, then row hammer mitigation effectiveness is improved, but performance impact increases
Solution Approach 1:
The patent applies partial action by issuing DRFM commands only to banks that actually exhibit row hammer conditions rather than all banks. This selective approach provides sufficient mitigation effectiveness for affected banks while avoiding the performance penalty of overly frequent commands to banks without row hammer risks.
Solution Approach 2:
The patent changes the parameter of command distribution from uniform across all banks to non-uniform based on risk profiles. By adjusting which banks receive commands and at what frequency based on their individual row hammer characteristics, the system optimizes the balance between mitigation effectiveness and performance impact.
Data Source
AI summary
In a memory subsystem, a controller can randomize the sending of directed refresh management (DRFM) commands for DRFM commands that hit multiple banks at a time. The controller can generate commands to indicate the memory device should capture addresses for various banks to use for pseudo target row refresh (pTRR) operations based on addresses associated with activate commands. The controller can randomize the indication of address capture for the memory device. With captured addresses and DRFM commands generated at random, the system can make better use of DRFM commands because the multiple banks are more likely to have addresses for pTRR, and randomization can reduce the ability for a row hammer attack to avoid the DRFM.