ML-Assisted Hard Decision Decoding for NAND Flash Read Power
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern non-volatile memory systems, such as NAND flash memories, face significant power consumption challenges due to the high energy expenditure of error correction decoders, which is exacerbated by the need to correct bit errors in multi-level memory cells.
Innovation Solution
The implementation of a memory system that utilizes a machine learning algorithm, specifically a neural network, to generate reliability information for hard decision decoding, reducing the number of decoder iterations and thus lowering energy consumption by identifying reliable bits across adjacent pages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional error correction decoding is used on multi-level memory cells, then bit errors can be corrected, but power consumption increases significantly
Solution Approach 1:
The patent applies preliminary action by performing hard decision decoding on adjacent pages before the main decoding operation. This pre-processing step identifies and corrects obvious errors in neighboring pages, reducing the burden on the power-consuming iterative decoder for the current page. The decoder uses information from previously decoded adjacent pages to improve the efficiency of the current decoding operation.
Solution Approach 2:
The patent utilizes copying by leveraging the decoded data from adjacent pages as reference information for the current page decoding. The hard decision results from neighboring pages are copied and used to inform the decoding process of the target page, reducing the need for repeated iterative decoding operations and thereby lowering power consumption while maintaining error correction capability.
2Reliability
If iterative decoding is performed to correct errors in multi-level memory cells, then error correction efficiency improves, but the number of iterations increases power consumption
Solution Approach 1:
The patent applies partial action by performing hard decision decoding on only the adjacent pages rather than all pages, and using this partial information to assist the main decoding operation. This selective approach provides sufficient error correction capability while avoiding the excessive energy consumption that would result from processing all pages through the same intensive decoding process.
3Productivity
If hard decision decoding is performed on all pages, then decoding speed increases, but error correction capability decreases
Solution Approach 1:
The patent merges two different decoding approaches: hard decision decoding is applied to adjacent pages for speed, while the main page benefits from both hard decision results and soft information from the iterative decoder. This combination maintains high error correction capability for the critical data while achieving fast processing for adjacent pages, resolving the contradiction between speed and reliability.
Data Source
AI summary
A memory system includes a plurality of memory cells each storing multiple bits and a memory controller having a processor. The memory controller is configured to read outputs from the memory cells in response to a read command from a host to generate first raw data of a first page and second raw data of a second page adjacent to the first page. The memory controller is further configured to perform a hard decision (HD) decoding on the first raw data to generate first decoded data. The processor is configured to apply the first decoded data and the second raw data as input features to a machine learning algorithm to generate reliability information. The memory controller is further configured to perform a HD decoding on the second raw data using the reliability information to generate second decoded data.