Machine Learning Nonvolatile Memory Read Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Nonvolatile memory devices face challenges in maintaining optimal operating conditions due to changes in threshold voltage distributions and read latency, leading to increased read latency and performance degradation.
Innovation Solution
A method using machine learning to determine optimal operating conditions by collecting training data based on feature information and error correction code decoding results, updating a machine learning model, and selecting the appropriate read level sets and sequences to adapt to changing conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a different read sequence is executed to account for altered threshold voltage distributions, then read accuracy is improved, but read latency increases
Solution Approach 1:
The patent applies preliminary action by training the machine learning model in advance using historical read data and threshold voltage distribution information. The model is prepared beforehand to quickly predict optimal read sequences for current operating conditions, avoiding the need for time-consuming assessment during actual read operations. This allows the system to switch to appropriate read sequences immediately when distribution alterations are detected.
Solution Approach 2:
The patent implements feedback by continuously monitoring read results and using successful read operations to update and refine the machine learning model. The model learns from actual read outcomes and threshold voltage distribution changes, adjusting its predictions to optimize future read sequences. This closed-loop feedback mechanism ensures the model adapts to changing conditions while maintaining low latency through accurate predictions.
2Adaptability or versatility
If multiple read sequences are prepared to account for various operating conditions, then adaptability is improved, but device complexity increases
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between the memory device and the controller. This model acts as a smart mediator that automatically selects appropriate read sequences based on input features such as threshold voltage distribution characteristics, temperature, and wear level. The intermediary handles the complexity of managing multiple read sequences, freeing the controller from complex decision-making logic while maintaining high adaptability to various operating conditions.
Solution Approach 2:
The patent applies parameter changes by using the machine learning model to dynamically adjust read sequence parameters based on operating conditions. Instead of manually configuring multiple fixed read sequences, the model continuously optimizes read voltage levels, pulse widths, and sequence timing parameters in response to changing threshold voltage distributions and environmental factors. This dynamic parameter adjustment achieves high adaptability without requiring complex manual configuration.
3Productivity
If read operations are optimized for current conditions, then productivity is improved, but measurement precision of operating conditions is required
Solution Approach 1:
The patent applies partial action by using a limited set of key features (such as threshold voltage distribution moments, temperature ranges, and wear indicators) rather than attempting to measure all possible operating parameters with high precision. The machine learning model is trained on these selected features to predict optimal read sequences, achieving good productivity improvement without requiring exhaustive measurement of all operating conditions. This selective feature approach balances measurement precision requirements with productivity gains.
Data Source
AI summary
According to a method of controlling an operation of a nonvolatile memory device using machine learning, operating conditions of the nonvolatile memory device are determined by performing an inferring operation using a machine learning model. Training data that are generated based on feature information and error information are collected, where the error information indicate results of error correction code (ECC) decoding of the nonvolatile memory device. The machine learning model is updated by performing a learning operation based on the training data. Optimized operating conditions for individual user environments are provided by collecting training data in the storage system and performing the learning operation and the inferring operation based on the training data.