NAND Read Threshold Prediction with Condition-Specific Model Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of maintaining process uniformity in NAND process shrinking and three-dimensional stacking, coupled with varying operational conditions, leads to inconsistent read thresholds in data storage devices, causing increased variability and higher bit error rates, especially in less-mature memory nodes.
Innovation Solution
Implementing a multi-model system that selects and switches between prediction models optimized for specific conditions, using machine learning to adapt read thresholds based on parameters like program/erase cycles, temperature, and bit error rates, without increasing hardware resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single prediction model is used for read threshold generation, then device complexity is reduced, but reliability deteriorates due to inability to adapt to varying operational conditions
Solution Approach 1:
The system dynamically selects between multiple prediction models based on current operational conditions (temperature, program/erase cycle count, retention time). The model selection is not static but adapts in real-time to changing conditions, allowing the system to maintain high read threshold accuracy across diverse operational scenarios without requiring a single complex universal model
Solution Approach 2:
The system changes the parameter of model selection based on operational conditions. Different models are trained for different condition ranges (e.g., temperature ranges, P/E cycle ranges), and the system selects the appropriate model by matching current parameters to the training conditions of available models
2Adaptability or versatility
If multiple prediction models are implemented to handle different conditions, then adaptability improves, but device complexity increases
Solution Approach 1:
The system segments the operational space into distinct condition ranges (temperature ranges, P/E cycle ranges, retention time ranges) and assigns different prediction models to different segments. This segmentation allows each model to be optimized for specific conditions without requiring a single model to handle all scenarios, reducing the effective complexity of model management
Solution Approach 2:
The framework provides a universal mechanism for handling multiple models that can accommodate different prediction models for different conditions. The model selection and switching infrastructure is designed to be general-purpose, supporting various model types and condition parameters without requiring condition-specific control logic for each model
3Manufacturing precision
If read thresholds are kept fixed, then device complexity is minimized, but manufacturing precision deteriorates due to process variability in NAND shrinking and stacking
Solution Approach 1:
Multiple prediction models are pre-trained during manufacturing for different operational conditions and P/E cycle ranges. This preliminary action captures process variations and conditions-specific characteristics in advance, allowing the system to select the appropriate pre-trained model rather than adjusting thresholds dynamically during operation. The heavy lifting of adaptation is done during manufacturing/model training
Data Source
AI summary
A data storage device can generate a recommended read threshold value using a model that was trained under a plurality of conditions. However, such a model may result in an undesirable bit error rate or programming latency. If that should occur, the data storage device can use a different model trained under a condition similar to a current condition of the data storage device. In addition to avoiding an undesirable bit error rate or programming latency, this can result in improved throughput, improved quality of service, and reduced power consumption.