Modeless Read Threshold Voltage Estimation Using CDF Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory systems face challenges in accurately determining an optimal read threshold voltage due to distorted or overlapping threshold voltage distributions, leading to read errors, especially as program/erase cycles increase and cell-to-cell interference occurs.
Innovation Solution
A memory system incorporating a combined neural network that receives cumulative distribution function (CDF) values associated with program voltages, generates connection vectors, and estimates an optimal read threshold voltage using these vectors and weight values, employing deep learning for parametric framework modeling and optimal read threshold voltage estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional threshold voltage determination methods are used, then the process is simple, but read errors increase due to distorted or overlapping threshold voltage distributions
Solution Approach 1:
The patent introduces CDF values as an intermediary representation of threshold voltage distributions. Instead of directly working with distorted voltage distributions, the system converts them into CDF values that capture the essential characteristics in a standardized form, enabling more reliable optimal threshold voltage determination even when distributions are overlapping or distorted
Solution Approach 2:
The patent replaces conventional mechanical/mathematical threshold determination methods with a neural network-based approach. The neural network learns optimal threshold voltages from training data and generalizes to new conditions, substituting traditional algorithmic approaches with a data-driven model that handles distorted distributions more effectively
2Measurement precision
If neural network-based estimation is implemented, then read accuracy improves, but computational complexity increases
Solution Approach 1:
The patent performs preliminary training of the neural network offline using labeled data from threshold voltage distributions. This pre-computes the knowledge needed for accurate estimation, so that during actual operation, the network only requires forward propagation on new CDF values, significantly reducing online computational complexity while maintaining high accuracy
Solution Approach 2:
The patent transforms the input threshold voltage distributions into CDF values, changing the parameter representation to a standardized form. This transformation simplifies the input data structure and enables the neural network to focus on learning the mapping to optimal thresholds rather than handling raw distribution variations, reducing computational burden
3Quantity of substance
If multiple program voltage levels are handled, then storage capacity increases, but threshold voltage distortion increases
Solution Approach 1:
The patent segments the threshold voltage distribution into multiple distinct program voltage levels (e.g., PV0, PV1, PV2, PV3 for SLC/MLC/TLC). Each level is independently characterized by its CDF values, allowing the system to handle multiple storage states while maintaining clear separation between them through the neural network's learning capability
Solution Approach 2:
The patent uses CDF values as an intermediary that captures the statistical characteristics of each program voltage level's threshold distribution. This intermediary representation allows the system to handle multiple voltage levels with distorted or overlapping distributions by converting them into a common statistical framework that the neural network can process effectively
Data Source
AI summary
Embodiments provide a scheme for estimating an optimal read threshold voltage using a deep neural network (DNN) with a reduced number of processing. A controller includes a combined neural network, which receives first and second cumulative distribution function (CDF) values, each CDF value corresponding to a program voltage (PV) level associated with a read operation on the cells. The combined neural network generates first and second connection vectors based on the first and second CDF values and first weight values, and estimates an optimal read threshold voltage based on the first and second connection vectors and second weight values.