Flash Memory Decoding Status Prediction for Faster Read Retry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Flash memories experience read data errors due to degradation, leading to read failures despite multiple retries with varying read parameters, as existing methods fail to accurately predict decoding success.
Innovation Solution
A decoding status prediction system is built using machine learning to classify read records into groups, train prediction models, and select models based on distances to central points, predicting decoding success with a log-likelihood ratio.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If read retry operations are performed multiple times with varying read parameters, then the probability of successful decoding is improved, but the read failure problem still occurs when data errors persist
Solution Approach 1:
The system performs preliminary classification of read records into different groups based on error characteristics before actual read operations. By pre-training prediction models on these classified groups, the system can predict decoding success probability in advance, avoiding unnecessary retry operations and reducing time loss while maintaining high decoding success rates.
Solution Approach 2:
The patent replaces the traditional mechanical retry approach with a machine learning-based prediction system. Instead of blindly retrying read operations multiple times, the system uses trained prediction models to evaluate decoding success probability and make informed decisions, substituting mechanical repetition with intelligent prediction.
2Measurement precision
If read parameters are changed frequently during read retry process, then decoding accuracy is improved, but the complexity of parameter management increases
Solution Approach 1:
The patent segments read records into different groups based on their error characteristics and classification results. By dividing the data into distinct groups (e.g., first portion and second portion of read records), the system can apply different prediction models to different segments, simplifying parameter management while maintaining high decoding accuracy for each specific case.
Solution Approach 2:
The system dynamically changes prediction parameters based on the classification results and error characteristics of different read record groups. Instead of using fixed parameters, the prediction models adapt their parameters according to the specific characteristics of each data group, improving accuracy while managing complexity through systematic parameter adaptation.
3Measurement precision
If prediction models are trained with one to one combination of read records from K0 and K1 groups, then prediction accuracy is improved, but the number of models and training complexity increases
Solution Approach 1:
The patent applies local quality by training specific prediction models for different local characteristics of read records. Each prediction model is specialized for a particular combination of K0 and K1 groups, capturing the unique error patterns and characteristics of each local data segment. This localized approach improves prediction accuracy for specific cases while maintaining overall system effectiveness.
Data Source
AI summary
A method of building a decoding status prediction system is provided. Firstly, plural read records are collected during read cycles of a flash memory. Then, the plural read records are classified into read records with a first read result and read records with a second read result. Then, a first portion of the read records with the first read result are divided into K0 groups according to a clustering algorithm, and a second portion of the read records with the second read result are divided into K1 groups according to the clustering algorithm. Then, the read records of the K0 groups and the K1 groups are used to train prediction models. Consequently, K0×K1 prediction models are generated. Then, the prediction models are combined as a prediction database.