Neural Polar Decoder for NAND Flash Non-AWGN Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing error correction decoding methods for memory devices, such as those using polar codes, face challenges with non-AWGN noise models and high computational intensity, particularly in NAND flash memory systems where data storage is stochastic and noisy.
Innovation Solution
A neural network decoder is employed to decode error correction codes, utilizing a feature extractor trained on probability data from the physical channel to extract features from voltage levels in memory devices, allowing for efficient decoding of information bits without assuming an AWGN model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a polar decoding scheme is used, then error correction capability is improved, but computational complexity increases
Solution Approach 1:
The patent replaces traditional sequential polar decoding algorithms (such as successive cancellation decoding) with a neural network-based decoder. The neural network is trained offline to learn the decoding function, and during operation, it performs parallel inference to decode polar codes. This substitution of mechanical/algorithmic decoding with a trained neural network model reduces computational complexity during actual decoding while maintaining error correction capability.
2Ease of operation
If traditional decoders assuming AWGN model are used, then decoding simplicity is maintained, but decoding accuracy deteriorates in Flash memory channels
Solution Approach 1:
The patent changes the fundamental parameters and assumptions of the decoding approach. Instead of assuming an AWGN channel model with simple probabilistic relationships, the system collects actual probability data from the Flash memory channel during training. The neural network learns the true channel characteristics including non-AWGN noise patterns, cell-to-cell variability, and program/read disturbances. This parameter change from theoretical assumptions to empirical channel characteristics improves decoding accuracy while the neural network structure maintains operational simplicity.
Solution Approach 2:
The patent performs preliminary training of the neural network decoder using collected probability data from the specific Flash memory channel before actual decoding operations. During this offline training phase, the system learns the channel characteristics and optimizes the decoder parameters. This preliminary action ensures that when the decoder is deployed for actual data decoding, it is already adapted to the specific channel conditions, achieving high accuracy without complex real-time adjustments.
Data Source
AI summary
A method, apparatus, non-transitory computer readable medium, and system for using an error correction code in a memory device with a neural network are described. Embodiments of the method, apparatus, non-transitory computer readable medium, and system may receive a signal from a physical channel, wherein the signal is based on a modulated symbol representing information bits encoded using an error correction coding scheme, extract features from the signal using a feature extractor trained using probability data collected from the physical channel, and decode the information bits with a neural network decoder taking the extracted features as input.