Memory Scan Frequency Control Using Read Margin Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory sub-systems employ static scan frequencies and thresholds based on end-of-life specifications, leading to over-scanning and over-folding, which reduce performance and affect quality of service due to inaccuracies in predicting memory device capabilities at different stages of life.
Innovation Solution
Adaptive scanning of memory devices using supervised learning to generate page type and wordline group-based margin predictions during the middle-of-life stage, allowing for accurate bin assignments and adjusted scan frequencies based on data retention capability predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If static scan frequencies and thresholds based on end-of-life specifications are used, then memory device capabilities are ensured at end-of-life, but over-scanning and over-folding occur reducing performance and quality of service
Solution Approach 1:
The patent implements dynamic scan frequency adjustment based on the memory device's actual age and measured performance characteristics. Instead of using a fixed static scan frequency for all devices regardless of their life stage, the system adapts the scan frequency to match the device's current reliability profile, performing more scans on younger devices and fewer scans on older devices while maintaining data integrity thresholds.
Solution Approach 2:
The system changes the scan frequency parameter based on measured read margin characteristics and device age. By monitoring actual performance metrics and adjusting the scan frequency parameter dynamically, the system optimizes the balance between reliability assurance and performance, avoiding both over-scanning and under-scanning conditions.
2Reliability
If static scan frequencies based on end-of-life specifications are used, then data integrity is maintained at end-of-life, but unnecessary scanning operations increase time consumption and reduce efficiency
Solution Approach 1:
The patent applies partial scanning actions based on the device's current risk profile. Instead of performing excessive scans on all devices, the system performs scans only when necessary based on measured performance margins and age-related degradation patterns, reducing time consumption while maintaining data integrity for devices that actually need monitoring.
Solution Approach 2:
The scan frequency is made dynamic rather than static, allowing the system to reduce scanning operations on devices showing stable performance characteristics while increasing scans on devices exhibiting degradation signs, thereby optimizing the time-integrity tradeoff adaptively.
3Measurement precision
If supervised learning-based adaptive scanning is implemented, then scan frequency accuracy is improved, but device complexity increases
Solution Approach 1:
The patent introduces a supervised learning model as an intermediary component that processes measured read margin data and device metadata to predict optimal scan frequencies. This intermediary layer translates complex performance measurements into actionable scan frequency recommendations, improving accuracy while isolating the complexity within a dedicated prediction module rather than distributing it throughout the entire memory subsystem.
Data Source
AI summary
A processing device in a memory sub-system determines one or more read margin levels associated with the memory device. A machine learning model is applied to the one or more read margin levels to generate a read margin prediction value associated with the memory device. Based on the margin prediction value, the memory device is assigned to a selected bin of a set of bins. A media scan operation is executed on the memory device in accordance with a scan frequency associated with the selected bin.