Parametric PV-Level Modeling for Read Threshold Voltage Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory systems face challenges in determining an optimal read threshold voltage, leading to read errors due to distorted or overlapping threshold voltage distributions, which existing algorithms fail to accurately address.
Innovation Solution
A system and method that utilize parametric PV-level modeling and deep learning to estimate an optimal read threshold voltage by generating cumulative mass function samples, receiving cumulative distribution function values representing skew normal distributions, and determining probability density function values to find the cross-point of these distributions as the optimal read threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing algorithms are used to determine read threshold voltage, then the process is simple, but read errors occur due to distorted or overlapping threshold voltage distributions
Solution Approach 1:
The system performs preliminary parametric PV-level modeling to characterize threshold voltage distributions before actual read operations. By pre-establishing the statistical models (Gaussian, skew-normal, or asymmetric Laplace distributions) and their parameters based on program voltage levels, the system prepares accurate reference frameworks in advance, enabling reliable read threshold determination even when distributions are distorted or overlapping.
Solution Approach 2:
The patent introduces probability distribution models as intermediary representations between the physical threshold voltage distributions and the read threshold determination process. These models act as mediators that translate complex, distorted voltage distributions into manageable statistical parameters, allowing the controller to estimate optimal read thresholds through cross-point calculations without directly dealing with the complexity of distorted distributions.
2Measurement precision
If parametric PV-level modeling with deep learning is used, then read threshold estimation accuracy improves, but computational complexity increases
Solution Approach 1:
The system changes the parameters of probability distribution models (mean, standard deviation, skewness, kurtosis) based on detected PV levels and their frequencies. By dynamically adjusting these statistical parameters according to the actual distribution characteristics observed in the memory cells, the system achieves high estimation accuracy while maintaining computational efficiency through parameterized models rather than full-scale deep learning computations.
Solution Approach 2:
The patent creates simplified statistical copies (probability distribution models) of the complex threshold voltage distributions. Instead of performing computationally intensive deep learning inference on raw voltage data, the system works with compressed statistical representations that capture the essential characteristics of the distributions, significantly reducing computational complexity while preserving estimation accuracy.
3Measurement precision
If multiple probability distribution models are considered, then estimation accuracy for asymmetric distributions improves, but processing time increases
Solution Approach 1:
The system dynamically selects and switches between different probability distribution models (Gaussian, skew-normal, asymmetric Laplace) based on the detected characteristics of the threshold voltage distributions. By adapting the model choice to the actual data characteristics rather than using a fixed model, the system achieves high accuracy for various distribution types including asymmetric cases, while avoiding the computational overhead of evaluating all models simultaneously through conditional selection logic.
Data Source
AI summary
Embodiments provide a scheme for parametric PV-level modeling and an optimal read threshold voltage estimation in a memory system. A controller performs read operations on cells using read threshold voltages; generates CMF samples based on the read operations; and receives first and second CDF values, which correspond to CMF samples, each CDF value representing a skew normal distribution. The controller estimates first and second probability distribution parameter sets corresponding to the first and second CDF values, respectively; determines first and second PDF values using the first and second probability distribution parameter sets, respectively; and estimates, as an optimal read threshold voltage, a read threshold voltage corresponding to a cross-point of the first and second PDF values.