Inference Engine for NAND Read Threshold Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of maintaining process uniformity in NAND process shrinking and three-dimensional stacking, combined with the variability in operational conditions such as program/erase cycles, retention times, and temperatures, leads to inaccurate read thresholds in memory systems, resulting in higher bit error rates and performance degradation.
Innovation Solution
A storage system and method that utilize an inference engine to infer optimal read thresholds based on multiple memory parameters and conditions, including time and temperature groups, bit error rates, program/erase counts, and physical page locations, using machine-learning methodologies to learn non-linear dependencies and adjust read thresholds accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If read thresholds are fixed for all memory locations and conditions, then device complexity is reduced, but measurement precision and reliability deteriorate due to variability in process uniformity and operational conditions
Solution Approach 1:
The patent segments the memory device into multiple groups (e.g., by physical location, program/erase cycle count, retention time, temperature) and assigns different read thresholds to each segment. This allows the system to maintain high measurement precision for each segment while managing complexity through structured organization of threshold data.
Solution Approach 2:
The patent implements dynamic read threshold adjustment based on operational conditions such as program/erase cycles, retention times, and temperatures. The read thresholds are not fixed but change dynamically according to the current state of the memory device, improving reliability without requiring complete redesign of the reading mechanism.
2Reliability
If read thresholds are dynamically adjusted for different operational conditions, then reliability and measurement precision improve, but device complexity and difficulty of detecting and measuring increase
Solution Approach 1:
The patent performs preliminary characterization of the memory device to establish baseline read thresholds and operational parameters before actual read operations occur. This preliminary action includes measuring voltage distributions, determining program/erase cycle counts, and establishing retention time characteristics, which are then used to set appropriate read thresholds in advance.
Solution Approach 2:
The patent incorporates feedback mechanisms where the controller monitors read error rates and operational conditions, then adjusts read thresholds accordingly. The system continuously refines threshold values based on observed performance, creating a closed-loop system that improves reliability while managing complexity through intelligent control.
3Measurement precision
If comprehensive read threshold calibration is performed for all conditions, then measurement precision improves, but loss of time increases due to high-latency recovery flows
Solution Approach 1:
The patent applies partial calibration by determining read thresholds for representative samples of memory locations and conditions rather than exhaustively calibrating every possible scenario. The controller uses statistical models and inferred characteristics to extend calibration results to unmeasured conditions, achieving sufficient precision without the time cost of complete calibration.
Solution Approach 2:
The patent creates copies of calibration data and threshold values that can be quickly retrieved and applied to different memory locations and conditions. Instead of performing full calibration operations each time, the system uses pre-computed and stored threshold information that can be rapidly accessed, reducing time loss while maintaining measurement precision.
Data Source
AI summary
A storage system has an inference engine that can infer a read threshold based on a plurality of parameters of the memory. The read threshold can be used in reading a wordline in the memory during a regular read operation or as part of an error handling process. Using this machine-learning-based approach to infer a read threshold can provide significant improvement in read threshold accuracy, which can reduce bit error rate and improve latency, throughput, power consumption, and quality of service. In another embodiment, a storage system is configured to use a binary full-depth symmetrically-sorted tree to infer a read threshold based on a plurality of parameters of the memory.