Center Read Reference Voltage Determination via PDF Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In non-volatile memory (NVM) systems, especially NAND-type Flash in solid state drives, the incorrect positioning of the center read reference voltage leads to increased raw bit error rates and LDPC decoding failures, particularly during hard data reads, and compromises confidence information in soft bit reads.
Innovation Solution
A method is introduced to calculate a new center read reference voltage by estimating cell values through hard or soft bit reads, decoding LDPC codewords, and modeling cumulative distribution functions (CDFs) and probability density functions (PDFs) to determine the optimal voltage for minimizing bit errors, which can be applied to both most significant bits (MSBs) and least significant bits (LSBs) of multi-bit cells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If BCH codes are used to reposition the center read reference voltage, then the reference voltage can be repositioned, but the ability to use soft bit reads and obtain confidence data is precluded
Solution Approach 1:
The patent segments the reference voltage adjustment process into multiple discrete steps: initial hard read, LDPC decoding, CDF estimation, and iterative PDF refinement. This allows the system to maintain soft bit read capability while achieving precise reference voltage positioning through incremental adjustments based on decoded information.
Solution Approach 2:
The patent implements a feedback mechanism where LDPC decoding results are used to estimate CDFs and refine PDFs, which then guide the adjustment of the center read reference voltage. This iterative feedback loop enables precise voltage positioning without sacrificing soft bit read capability, as the feedback is based on decoded information rather than requiring exclusive use of hard reads.
2Productivity
If the center read reference voltage is not correctly positioned, then hard reads can be performed, but the raw bit error rate increases and LDPC decoding fails
Solution Approach 1:
The patent performs preliminary actions by conducting hard reads to gather initial data for CDF estimation, then uses LDPC decoding to refine the reference voltage positioning. This preliminary hard read phase enables the system to maintain throughput while preparing accurate probability models that reduce bit error rates in subsequent reads.
Solution Approach 2:
The patent dynamically changes the reference voltage parameter based on estimated CDFs and PDFs derived from decoding results. By adjusting the reference voltage parameter iteratively based on decoded information, the system maintains high throughput while significantly reducing raw bit error rates through data-driven parameter optimization.
3Productivity
If the center read reference voltage is not correctly chosen, then hard reads can retrieve data, but confidence information from soft bit reads is not optimized
Solution Approach 1:
The patent makes the reference voltage determination process universal by using the same CDF estimation and PDF refinement mechanism for both hard reads and soft reads. This multi-functional approach allows the system to optimize confidence information for soft bit reads while maintaining efficient data retrieval through hard reads, as both operations benefit from the unified probability model refinement.
Data Source
AI summary
Embodiments include systems, methods, and apparatuses to estimate respective first and second cumulative density functions (CDFs) for values of a plurality of non-volatile memory (NVM) cells in a page of memory. The CDFs may be based at least in part on one or more decoder outputs of codewords for data stored in the page. Based at least in part on the CDFs, first and second probability density functions (PDFs) may be estimated for the values of the page of memory. A center read reference voltage may then be determined for reading a cell in the page. The center read reference voltage may be based at least in part on the first and second PDFs.