Modular Read Threshold Models for NAND Memory Corner Cases
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of maintaining process uniformity in NAND process shrinking and three-dimensional stacking, coupled with varying operational conditions, leads to inconsistent read thresholds in data storage devices, especially for less-mature memory nodes, resulting in higher bit error rates and performance degradation.
Innovation Solution
A modular-model system is employed in data storage devices, comprising a base model for typical conditions and add-on models for corner cases, using machine learning techniques to infer read thresholds accurately and efficiently, with hardware implementations that handle both common and rare conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single base model is used for inferring read thresholds, then hardware complexity is reduced, but read threshold accuracy deteriorates under corner case conditions
Solution Approach 1:
The model is segmented into a base model for typical conditions and add-on models for corner cases. The base model handles common scenarios with simple hardware, while add-on models are selectively activated for specific corner cases (e.g., open wordline, open block, bad block conditions), achieving high accuracy without permanently increasing hardware complexity.
Solution Approach 2:
The system dynamically selects which add-on models to activate based on the current operational conditions. The controller determines whether corner case conditions exist and selectively applies add-on models only when needed, allowing the system to adapt between simple base model operation and enhanced add-on model operation as conditions change.
2Measurement precision
If add-on models are used for corner cases, then read threshold accuracy is improved, but computational resources increase
Solution Approach 1:
Instead of deploying complete add-on models for all possible corner cases, the system uses partial models or simplified versions that provide sufficient accuracy for each specific corner case. This reduces the total computational burden while maintaining adequate performance across all conditions.
Solution Approach 2:
Different levels of model complexity are applied to different operational conditions. The base model provides general coverage, while targeted add-on models provide enhanced accuracy only for specific corner cases where needed, rather than uniformly increasing complexity across all operations.
3Device complexity
If read thresholds are fixed, then device complexity is minimized, but bit error rate increases due to process variability
Solution Approach 1:
The system changes the read threshold parameter dynamically based on operational conditions and physical location. Instead of using fixed read thresholds, the base model infers appropriate thresholds for typical conditions, and add-on models adjust thresholds for corner cases, maintaining low bit error rates without requiring complex fixed threshold structures.
4Measurement precision
If comprehensive models are used for all conditions, then read threshold accuracy is maintained, but power consumption increases
Solution Approach 1:
The system uses periodic or conditional activation of add-on models rather than continuous operation. The controller periodically assesses whether corner case conditions exist and activates add-on models only when necessary, reducing overall power consumption while maintaining accuracy when needed.
Solution Approach 2:
The power-intensive add-on model computations are extracted and separated from the continuous operation. Instead of running comprehensive models continuously, the system extracts and applies only the necessary add-on model computations for specific corner cases, reducing overall power consumption while maintaining accuracy.
Data Source
AI summary
A model can be used to infer a read threshold for reading a memory of a data storage device. In some situations, such as when the memory has an open wordline or an open block, the model may not provide an accurate read threshold. In such situations, one or more additional models can be used as modular add-ons to the original model to provide a more-accurate read threshold, which can result in a reduced bit error rate, as well as improved throughput, quality of service, and power consumption.