Neural Network Read Threshold Estimation for Memory Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory systems face challenges in determining an optimal read threshold voltage, leading to sub-optimal read operations that result in increased bit errors and decreased reliability due to distorted or overlapping threshold voltage distributions, especially in multi-level cell technologies like MLC and TLC.
Innovation Solution
A deep neural network is employed in the memory system's controller to estimate an optimal read threshold voltage based on input information such as read threshold sets, checksum values, and asymmetric ratios of ones and zeros counts, using a combination of multiple matrices and bias vectors, thereby eliminating the need for additional read operations and improving read accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional read threshold determination schemes are used, then the memory system can perform read operations, but the reliability decreases due to distorted or overlapping threshold voltage distributions leading to increased bit errors
Solution Approach 1:
The patent replaces traditional mechanical/electrical threshold determination methods with a neural network-based computational approach. The neural network processes input features (read thresholds, checksums, asymmetric ratios) to predict optimal read threshold voltages, substituting conventional voltage sensing and adjustment mechanisms with intelligent algorithmic determination, thereby improving both reliability and precision simultaneously
Solution Approach 2:
The patent changes the parameters used for threshold determination from simple voltage measurements to a comprehensive set of features including read threshold voltages, checksum values, and asymmetric ratios of ones and zeros counts. This parameter transformation enables the neural network to capture complex distribution characteristics and determine optimal thresholds more accurately, resolving the contradiction between reliability and measurement precision
2Measurement precision
If additional read operations are performed to determine optimal threshold, then the accuracy improves, but the productivity decreases due to increased operation time
Solution Approach 1:
The patent performs preliminary actions by collecting features (read thresholds, checksums, asymmetric ratios) during normal read operations and using them as input to the neural network. This preliminary data collection and processing enables the system to determine optimal thresholds without requiring additional read operations, thus maintaining both high accuracy and productivity
Solution Approach 2:
The patent creates a computational model (neural network) that copies and simulates the complex relationship between read conditions and optimal thresholds. Instead of performing multiple physical read operations to find the optimal threshold, the system uses the trained neural network model to predict the optimal threshold directly from feature inputs, eliminating time-consuming iterative reads while maintaining accuracy
3Reliability
If multiple read threshold voltages are tested, then the reliability improves, but the loss of time increases due to multiple sensing operations
Solution Approach 1:
The patent substitutes multiple physical sensing operations with a single neural network inference operation. The neural network has been trained to evaluate multiple threshold scenarios computationally and directly output the optimal threshold, replacing the need for iterative voltage testing and multiple sensing cycles, thereby reducing time loss while maintaining reliability
Solution Approach 2:
The neural network acts as an intermediary between the raw read features and the optimal threshold determination. Instead of directly testing multiple thresholds through repeated sensing, the system uses the neural network as a mediator that processes feature inputs and predicts the optimal threshold in a single operation, eliminating time-consuming intermediate sensing steps while ensuring reliable bit error reduction
Data Source
AI summary
A scheme for non-parametric optimal read threshold estimation of a memory system. The memory system includes a memory device including pages and a controller including a neural network. The controller performs read operations on a selected page using a read threshold set; obtain the read threshold set, a checksum value and an asymmetric ratio of ones count and zeros count which are associated with decoding of the selected page according to each of the read operations; provide the obtained read threshold set, the checksum value and the asymmetric ratio as input information to the neural network; and estimate, by the neural network, an optimal read threshold voltage based on the input information and weights including a combination of multiple matrices and bias vectors.