Storage Device Read Level Inference via Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing storage devices struggle to adapt to varying deterioration rates of non-volatile memory cells across different word lines due to shifts in threshold voltage distributions, leading to uncorrectable errors and data loss, as they rely on pre-set read voltages based on average usage patterns and environments.
Innovation Solution
A storage device that employs an artificial neural network model to infer optimal read levels by learning from on-cell counts and word line numbers, allowing dynamic adjustment of read levels based on actual usage patterns and environments, thereby improving data retrieval accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If pre-set read voltages based on average usage patterns are used, then device complexity is reduced, but read accuracy deteriorates due to varying deterioration rates across different word lines
Solution Approach 1:
The patent divides the memory device into multiple blocks, each with its own deterioration characteristics. By segmenting the read voltage management at the block level rather than using a uniform approach for the entire device, the system can apply tailored read voltages to each block based on its specific deterioration pattern, thereby improving read accuracy without excessively increasing overall device complexity.
Solution Approach 2:
The patent implements local quality by determining read voltages specifically for each block based on its individual deterioration characteristics. Instead of applying a single pre-set read voltage to all blocks, the system adjusts read voltages locally at the block level to match actual deterioration rates, which improves read accuracy while maintaining manageable complexity through localized rather than global optimization.
2Measurement precision
If block-level read voltage adjustment is implemented, then read accuracy is improved, but device complexity increases due to additional control mechanisms
Solution Approach 1:
The patent applies preliminary action by determining and storing optimal read voltages for each block in advance, before actual read operations occur. The controller pre-calculates read voltages based on block deterioration characteristics and stores them for later use, which improves read accuracy while reducing the complexity of real-time control decisions during actual read operations.
Solution Approach 2:
The patent implements feedback mechanisms where the controller monitors read operations and deterioration patterns across different blocks, using this information to adjust and optimize read voltages for each block. This feedback-driven approach improves read accuracy by adapting to actual deterioration while managing complexity through systematic rather than ad-hoc adjustments.
3Reliability
If individual word line deterioration is addressed, then data reliability is improved, but the complexity of managing multiple read levels increases
Solution Approach 1:
The patent segments the memory device into multiple blocks, each experiencing different deterioration rates. By managing read levels at the block level rather than attempting individual word line management, the system improves data reliability for each block while avoiding the excessive complexity that would arise from managing read levels for every individual word line separately.
Solution Approach 2:
The patent applies universality by using a common controller architecture that can manage multiple blocks with different deterioration characteristics. The controller serves multiple functions: monitoring deterioration across blocks, determining optimal read voltages for each block, and executing read operations with block-specific voltages, thereby improving data reliability while managing complexity through a multi-functional rather than specialized approach.
Data Source
AI summary
A storage device includes a non-volatile memory including a plurality of blocks, a buffer memory that stores a plurality of on-cell counts, which are generated by reading memory cells connected to a plurality of reference word lines of the plurality of blocks by using a read level, and an artificial neural network model, and a controller that inputs an on-cell count corresponding to a target block among the plurality of on-cell counts and a number of a target word line of the target block to the artificial neural network model, and infers a plurality of read levels for reading data of memory cells connected to the target word line using the artificial neural network model.