MLC NAND Flash Quantizer Using Deep Learning for Polar Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The reliability of MLC NAND flash systems is compromised by channel distortions due to unpredictable noises and offsets, making accurate log-likelihood ratio (LLR) calculation and channel modeling challenging, which affects the decoding performance of polar codes.
Innovation Solution
A deep learning-based method for optimizing a Polar-RNNA quantizer that transforms MLC flash detection into a neural network task, expands read thresholds, constructs an LLR mapping table, symmetrizes the channel, and uses a genetic algorithm to optimize quantization intervals, eliminating the need for complex channel modeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If accurate LLR calculation is performed using traditional methods with multiple read thresholds, then decoding performance is improved, but computational complexity and memory access overhead increase significantly
Solution Approach 1:
The patent transforms the continuous LLR calculation into a discrete quantization process by changing the parameter representation from continuous voltage values to discrete threshold indices. The neural network learns optimal quantization thresholds that map continuous read voltages to discrete LLR values, reducing computational complexity while maintaining decoding performance.
Solution Approach 2:
The patent replaces the traditional mechanical/computational LLR calculation system with a neural network-based system. Instead of using complex lookup tables and multiple read thresholds, the neural network directly maps read voltages to quantized LLR values, simplifying the computational mechanism.
2Measurement precision
If MLC flash channel modeling is performed to optimize quantization intervals, then quantization accuracy is improved, but the complexity of channel modeling increases due to unknown offsets and noises
Solution Approach 1:
The patent creates a simplified copy of the MLC flash channel characteristics through neural network training. Instead of modeling the complex physical channel with unknown parameters, the neural network learns an empirical mapping that captures the essential channel behavior, avoiding the need for explicit channel modeling.
Solution Approach 2:
The neural network acts as an intermediary between the raw read voltages and the quantization process. It learns to compensate for channel distortions, offsets, and noises internally, eliminating the need for explicit channel modeling and complex preprocessing steps.
3Reliability
If more read thresholds are used to quantize LLR, then decoding reliability is improved, but read time and system delay increase
Solution Approach 1:
The patent uses a dynamic quantization approach where the neural network adapts the quantization thresholds based on the learned channel characteristics. Instead of using a fixed large number of thresholds, the network dynamically determines the optimal quantization levels, achieving high reliability with fewer thresholds and reduced read time.
Data Source
AI summary
A method for optimizing a Polar-RNNA quantizer of MLC NAND flash based on deep learning comprises the following steps: Step S1: transforming an MLC flash detection task into a deep learning task, and obtaining three hard-decision read thresholds based on a neural network; Step S2: expanding six soft-decision read thresholds based on the three hard-decision read thresholds; Step S3: constructing an LLR mapping table, and obtaining new LLR information of MLC flash based on the LLR mapping table; Step S4: symmetrizing an MLC flash channel, and performing density evolution; and Step S5: optimizing the soft-decision read thresholds based on a genetic algorithm to obtain an optimal quantization interval. According to the invention, polar codes can be directly used for MLC flash channels without the arduous work of MLC flash channel modeling, so that the reliability of MLC flash is effectively improved.