Multi-Level Memory Controller Using Probability-Based Soft Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-level solid state non-volatile memory systems face increased errors due to reduced signal distance between voltage levels, leading to reduced noise margin and higher susceptibility to erroneous readouts, especially in low-level signals with larger noise distributions.
Innovation Solution
The implementation of an analog-to-digital converter and advanced encoding and modulation techniques that increase the number of digital levels in the output signal beyond the original multi-level memory array, allowing for improved error correction and increased storage density by utilizing soft information and iterative decoding methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multi-level encoding is used to increase storage density, then storage density is improved, but signal distance between voltage levels is reduced leading to higher error rates
Solution Approach 1:
The patent divides the error correction task into multiple stages using a multi-level decoding architecture. The first decoder performs initial error correction on hard decision data, while a second decoder performs iterative soft decision decoding. This segmentation allows the system to handle the increased error rates from multi-level encoding through a structured, multi-phase correction process, thereby maintaining reliability while preserving the storage density benefits of multi-level cells.
Solution Approach 2:
The patent introduces soft information as an intermediary between the analog voltage levels and the final digital data. The soft decision decoder uses probabilistic information about the likelihood of each voltage level to make more accurate decoding decisions. This intermediary layer of soft information allows the system to recover data more reliably from the reduced signal distances inherent in multi-level encoding, thus improving error rates without sacrificing storage density.
2Device complexity
If conventional hard decision decoding is used, then device complexity is low, but error correction capability is insufficient for multi-level memories
Solution Approach 1:
The patent segments the decoding function into two distinct decoders with different complexity levels. The first decoder uses simple hard decision decoding for initial error correction, while the second decoder implements more complex iterative soft decision decoding. This segmentation allows the system to achieve high error correction capability through the second decoder while keeping the overall architecture manageable by dividing the complex task into smaller, more manageable components.
Solution Approach 2:
The patent applies preliminary hard decision decoding before the more complex soft decision decoding. This preliminary action of the first decoder reduces the error burden on the second decoder, allowing it to focus computational resources on the most difficult error cases. This staged approach enables high error correction capability while managing device complexity by performing simpler operations first.
3Quantity of substance
If more voltage levels are used in memory cells, then storage density increases, but noise margin decreases making readings more susceptible to errors
Solution Approach 1:
The patent introduces soft information as an intermediary that captures the probabilistic nature of voltage level detection. Instead of directly converting analog voltages to digital bits, the system uses soft decision decoding to interpret the likelihood of each voltage level. This intermediary process allows the system to distinguish between noise-induced voltage variations and genuine data signals, thereby reducing noise susceptibility while maintaining the high storage density enabled by multiple voltage levels.
Solution Approach 2:
The patent implements iterative decoding with feedback loops in the soft decision decoder. The decoder repeatedly refines its estimates of the stored data by comparing expected voltage levels with actual readings and adjusting its decisions accordingly. This feedback mechanism allows the system to overcome noise-induced errors by iteratively converging on the most likely original data, thereby reducing noise susceptibility while preserving the storage density benefits of multi-level voltage states.
Data Source
AI summary
A memory controller includes an encoder, a modulator, and a demodulator. A nonvolatile memory includes memory cells, each programmable to one of three or more levels. According to first encoded data, the modulator programs a first subset of the memory cells to a first of the levels and a second subset of the memory cells to a second of the levels. Measurable values of the first subset are characterized by a first probability density function having a first width. Measurable values of the second subset are characterized by a second probability density function having a second width. The first width is greater than the second width. The encoder generates the first encoded data based on input data such that the first subset is smaller than the second subset. The demodulator is configured to output second encoded data in response to measurable values of the memory cells.