LDPC Decoding with Adaptive LLR Tables for NAND Memory Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LDPC decoding methods face performance degradation due to mismatches between assumed and actual memory error models, particularly influenced by temperature variations and program-erase cycles, leading to sub-optimal log-likelihood ratios (LLRs) and reduced error correction capabilities.
Innovation Solution
A method that involves creating and selecting LLR tables based on metadata indicating characteristics such as current read voltage levels, cross-temperature differences, and program-erase cycle status, using a-priori state-transition-matrices (STMs) to determine initial LLR tables, and iteratively updating these tables during decoding to improve reliability metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If predefined static LLR tables are used based on assumed error models, then decoding complexity is reduced and operation is simplified, but decoding accuracy and error correction capability deteriorate due to mismatch between assumed and actual memory error distributions
Solution Approach 1:
The patent transforms the static LLR table into a dynamic selection mechanism. Multiple LLR tables are pre-computed for different error models (Gaussian, Rayleigh, Rician, etc.), and the appropriate table is dynamically selected based on metadata indicating actual memory conditions such as temperature, program-erase cycle count, and read voltage level. This resolves the contradiction by maintaining operational simplicity through pre-computed tables while improving accuracy through adaptive selection matching actual error distributions.
Solution Approach 2:
The patent changes the parameter of LLR values based on memory operating conditions. By receiving metadata about temperature, P/E cycles, and voltage levels, the system selects LLR tables with parameters optimized for those specific conditions. For example, at higher temperatures or after many P/E cycles where error distributions shift, different LLR parameters are applied to match the degraded memory characteristics, thereby maintaining high error correction capability without complicating the decoding operation.
2Reliability
If multiple LLR tables are created and stored for different memory conditions, then error correction capability is improved through better matching of actual error models, but memory usage and device complexity increase
Solution Approach 1:
The patent segments the LLR tables based on distinct error models and memory conditions. Instead of using a single monolithic table or a complex adaptive algorithm, the system divides LLR tables into categories corresponding to different error distributions (Gaussian, Rayleigh, Rician) and memory states (temperature ranges, P/E cycle ranges, voltage levels). This segmentation allows for manageable table storage and simple metadata-based selection, improving error correction while controlling complexity.
Solution Approach 2:
The patent performs preliminary computation of multiple LLR tables for different error models before actual decoding operations. These pre-computed tables are stored in memory, eliminating the need for complex real-time calculations during decoding. The complexity is shifted to the initialization phase, where tables are generated offline based on theoretical error models, allowing fast and simple table selection during operation based on received metadata about actual memory conditions.
3Measurement precision
If LLR tables are updated iteratively during decoding, then decoding accuracy and convergence speed are improved, but decoding time and computational overhead increase
Solution Approach 1:
The patent performs preliminary selection of the most appropriate LLR table based on metadata about actual memory conditions before the decoding process begins. By matching metadata (temperature, P/E cycles, voltage) with pre-computed tables for different error models, the system starts with highly accurate initial LLR values. This preliminary action reduces or eliminates the need for iterative updates during decoding, thereby improving accuracy while minimizing additional decoding time and computational overhead.
Data Source
AI summary
A decoding system and method of a non-volatile memory are provided in which information regarding a characteristic of a non-volatile memory is used to determine an initial log-likelihood-ratio (LLR) table from among a number of LLR tables. The decoding is then performed using the determined initial LLR table.