Neural Soft Decoding of BCH Codes for Noisy Flash Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Flash memory devices, such as NAND flash memory, experience errors due to noise and interference during programming and read operations, which existing error correction codes struggle to address effectively, especially under high stress conditions and with low-complexity digital signal processing.
Innovation Solution
A memory system that includes an encoding/decoding system configured to process log-likelihood ratios and a syndrome vector to obtain confidence values for each bit of a codeword, estimate an error vector, and apply soft decision decoding, using a fully-connected deep neural network to correct errors and improve decoding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional algebraic solvers are used for error correction decoding, then the decoding process is straightforward to implement, but the computational complexity is high and error correction capability is limited under high stress conditions
Solution Approach 1:
The patent replaces traditional algebraic decoding methods (mechanical/systematic approach) with a deep neural network-based soft decoding approach. The DNN model learns optimal decoding strategies from training data, substituting complex algebraic computations with trained neural network inference that achieves better error correction with lower computational complexity.
Solution Approach 2:
The patent changes the decoding paradigm from hard decision algebraic solving to soft decision neural network processing. By using log-likelihood ratios (LLRs) as inputs and confidence values as outputs, the system transforms the problem parameters to enable probabilistic error correction that is more effective under high stress conditions while reducing enumeration complexity.
2Reliability
If soft decision decoding with deep neural network is used, then error correction capability and robustness to channel noise are improved, but the implementation complexity increases
Solution Approach 1:
The patent applies preliminary action by training the deep neural network model offline before deployment. The DNN is trained on synthetic training data to learn optimal decoding patterns, so that during actual operation, the pre-trained model can perform robust error correction with simple forward propagation computations, avoiding the need for complex real-time calculations.
Solution Approach 2:
The patent uses soft information (log-likelihood ratios) as copies of the hard decision data to enable probabilistic processing. By working with LLRs instead of binary decisions, the system preserves channel information and enables gradient-based optimization during training, improving robustness while the trained model simplifies implementation during deployment.
3Productivity
If hard decision decoding is used, then the decoding process is computationally simple, but the error correction capability is insufficient under high stress conditions
Solution Approach 1:
The patent introduces dynamics by transitioning from static hard decision thresholds to adaptive soft decision processing. The deep neural network dynamically adjusts decision boundaries based on learned patterns from training data, allowing the system to adapt to different channel conditions and error patterns, thereby improving error correction capability while maintaining efficient inference.
Data Source
AI summary
Systems, methods, non-transitory computer-readable media to perform operations associated with the storage medium. One system includes a storage medium and an encoding/decoding (ED) system to perform operations associated with the storage medium, the ED system being configured to process a set of log-likelihood ratios (LLRs) and a syndrome vector to obtain a set of confidence values for each bit of a codeword, estimate an error vector based on selecting one or more bit locations with confidence values from the set of confidence values above threshold value and applying hard decision decoding to the selected one or more bit locations, calculate a sum LLR score for the estimated error vector, and output a decoded codeword based on the estimated error vector and the sum LLR score.