Cascade Model for Read Level Voltage Offset Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Slow Charge Loss (SCL) tracking methods for memory devices are inefficient in managing SCL variation due to their reliance on fixed look-up tables with limited bin numbers, making it challenging to optimize block scan trigger rates and handle cycling degradation, wordline groups, and temperature variations effectively.
Innovation Solution
A cascade model is employed to predict read level voltage offsets for memory cells, using a stacked structure of machine-learning models that consider attributes like wordline group, operating temperature, and programming cycles, allowing for adaptive adjustment of read level voltages and dynamic management of SCL.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional SCL tracking methods use fixed look-up tables with limited bin numbers, then device complexity is reduced, but measurement precision of read level voltage offsets deteriorates
Solution Approach 1:
The patent segments the SCL tracking problem into multiple independent linear regression models, each responsible for predicting voltage offsets for specific wordline groups. This segmentation allows the system to achieve high precision for each segment while keeping individual model complexity low, resolving the contradiction between overall system precision and structural complexity.
Solution Approach 2:
The patent transitions from a traditional tabular lookup structure to a dimensional regression model approach. By introducing multiple dimensions (different wordline groups, different read levels) and using linear regression functions in this multi-dimensional space, the system achieves continuous precision improvement without the discrete limitations of fixed bin numbers, thereby resolving the precision-complexity contradiction.
2Ease of operation
If conventional methods use fixed look-up tables, then ease of operation is improved, but adaptability to temperature variations and cycling degradation deteriorates
Solution Approach 1:
The patent replaces static fixed look-up tables with dynamic linear regression models that can adapt to changing conditions. The models are trained on historical data and continuously updated to reflect temperature variations and cycling degradation, enabling the system to maintain high adaptability while preserving operational simplicity through automated predictions.
Solution Approach 2:
The linear regression models perform self-updating by automatically learning from new data and improving their predictions over time. This self-service capability allows the system to adapt to environmental variations without requiring manual intervention or complex operational procedures, thus maintaining ease of operation while enhancing adaptability.
3Measurement precision
If cascade model with multiple machine-learning models is used, then measurement precision of read level voltage offsets is improved, but device complexity increases
Solution Approach 1:
The cascade model is segmented into multiple independent linear regression models, each handling a specific aspect of the prediction (different wordline groups, different read levels). This segmentation allows the system to achieve high overall precision through the combination of specialized models while keeping each individual model simple and computationally efficient.
Solution Approach 2:
The patent changes the parameters of the regression models dynamically based on operating conditions such as temperature and cycling history. By adjusting model parameters rather than changing the fundamental model structure, the system achieves high adaptability and precision while avoiding the complexity of multiple different model architectures.
Data Source
AI summary
Various embodiments use a cascade model to determine (e.g., predict or estimate) one or more read level voltage offsets used to read data from one or more memory cells of a memory device, which can be part of a memory sub-system.