Memory System Read Level Quantization via Hyperplane Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In NAND flash memory systems, vector quantization methods for reading multi-value data face inaccuracies due to potential selection of incorrect representative points, especially when threshold voltage distributions change, leading to overlapping regions and incorrect data correction.
Innovation Solution
A memory system that divides an n-dimensional space into regions using hyperplanes, assigns representative points to each region, and employs a binary tree structure to trace branch nodes and determine a second read level by comparing the first read level with voltage levels, ensuring accurate data reading by applying the appropriate voltage to the word line.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If vector quantization is used to compress and record multiple read levels, then the number of read levels can be reduced and storage efficiency improved, but the accuracy of representative point selection deteriorates, leading to potential selection of incorrect representative points
Solution Approach 1:
The patent divides the n-dimensional space into multiple subspaces using hyperplanes, and further segments each subspace into regions with assigned representative points. This hierarchical segmentation allows efficient compression while maintaining accurate representative point selection by ensuring each data point is mapped to the correct region.
Solution Approach 2:
The patent assigns different representative points to different regions within the n-dimensional space, allowing each local region to have its own optimized representative point. This local quality approach ensures that representative points are accurately selected for local data distributions while maintaining overall compression efficiency.
2Reliability
If the threshold voltage distribution shifts due to memory cell degradation or environmental factors, then the original read levels become inaccurate, but re-calibrating all read levels increases system complexity and time consumption
Solution Approach 1:
The patent pre-calculates and stores multiple candidate representative points for each region in the n-dimensional space. When threshold voltage shifts occur, the system can quickly switch between pre-computed representative points without performing complex real-time recalibration, thus maintaining reliability while reducing complexity.
Solution Approach 2:
The patent allows dynamic selection of representative points based on changing threshold voltage distributions. By monitoring voltage shifts and adjusting which representative points are active, the system adapts to degradation without requiring complete re-calibration, reducing both complexity and time consumption.
3Productivity
If a binary tree structure is used to trace branch nodes for representative point selection, then the search process becomes systematic and efficient, but the computational overhead increases compared to direct lookup
Solution Approach 1:
The patent uses a binary tree structure to segment the search space into manageable branches. Each node in the tree represents a subspace division, allowing the search process to systematically eliminate large portions of the search space at each level, improving efficiency while keeping computational overhead manageable through hierarchical organization.
Data Source
AI summary
A memory system includes a non-volatile memory and a controller configured to divides an n-dimensional space into a plurality of regions by a plurality of hyperplanes, assign a representative point of a read level for reading data from a plurality of memory cells to each region, trace a branch node in the binary tree by determining whether a first read level is higher or lower than a voltage level at the branch node of the binary tree, determine a read level of a representative point assigned to a region correlated with a leaf node among the plurality of divided regions as a second read level corresponding to the first read level when reaching the leaf node of the binary tree by tracing the branch node in the binary tree, and cause the memory to read data of the cells by applying a voltage of the second read level.