NAND Read Threshold Calibration Using Combined Prediction Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of maintaining process uniformity and variability in read thresholds due to NAND process shrinking and three-dimensional stacking, especially under varying operational conditions, leads to increased bit error rates and degraded performance in data storage devices.
Innovation Solution
A data storage device that combines a linear regression model with a binary tree model to generate recommended read thresholds, leveraging the strengths of both models to provide continuous output while minimizing hardware complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If read thresholds are fixed to simplify operations, then ease of operation is improved, but measurement precision deteriorates due to inability to adapt to process variations and operational conditions
Solution Approach 1:
The patent implements dynamic read threshold adjustment by combining multiple prediction models (linear regression and tree-based models) that adapt thresholds based on operational conditions such as temperature, program/erase cycles, and retention time. This allows the system to transition from fixed to dynamic thresholds, resolving the contradiction between operational simplicity and measurement precision.
Solution Approach 2:
The system changes the parameter of read threshold values based on operational conditions. By using machine learning models to predict optimal thresholds under different conditions (temperature, PEC, retention), the system adjusts thresholds dynamically without requiring complex manual intervention, thus maintaining ease of operation while improving precision.
2Measurement precision
If multiple prediction models are combined to improve read threshold calibration accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent merges multiple prediction models (linear regression model and tree-based models) into a unified system that combines their strengths. The linear regression model provides continuous output for smooth threshold adjustments, while tree-based models handle non-linear relationships, achieving high accuracy without requiring an overly complex single model.
Solution Approach 2:
The combined prediction system serves multiple functions: it handles different operational conditions (temperature, PEC, retention), provides both continuous and discrete threshold adjustments, and adapts to various memory states. This multi-functionality justifies the added complexity by delivering comprehensive threshold calibration across all operating scenarios.
3Reliability
If read thresholds are dynamically adjusted to accommodate process variations, then reliability is improved, but use of energy increases due to additional calibration operations
Solution Approach 1:
The system performs preliminary threshold calibration using machine learning models during manufacturing or initialization. By pre-computing and storing prediction models that capture process variations, the system avoids the need for expensive real-time calibration operations during normal read operations, thus improving reliability while minimizing energy consumption during actual data reading.
Data Source
AI summary
A data storage device generates a recommended read threshold value by combining outputs of a binary tree model and a linear regression model. This provide a best-of-both-worlds result, as a binary tree model can describe complex output functions but can be limited to a finite set of output values, whereas the linear regression model has continuous value outputs. Other embodiments are provided.