NAND Flash Bad Block Prediction Using BGMS Error Rate Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing SSD technologies face challenges in predicting and addressing developed bad blocks in NAND flash memory devices, which can lead to premature failure and data loss, often requiring costly features like 'chipkill' for data recovery.
Innovation Solution
An improved Background Media Scan (BGMS) method that detects developed bad blocks by calculating and comparing average and maximum error rates across pages, predicting block failure based on thresholds, and retiring blocks before they become unusable, thereby eliminating the need for 'chipkill'.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional BGMS only compares maximum error rate with a pre-determined threshold, then the detection process is simple, but it cannot detect the sign of a developed bad block and predict its failure
Solution Approach 1:
The patent segments the error rate analysis into two distinct components: average error rate across all word-lines and maximum error rate among individual word-lines. This segmentation allows the system to compare not just the peak error but also the distribution pattern, enabling early detection of developed bad blocks before they cause complete failure.
Solution Approach 2:
The patent changes the detection parameters from a single threshold-based maximum error rate check to a dual-parameter approach comparing both average and maximum error rates. This parameter change enables the system to identify the characteristic signature of developing bad blocks, which manifest as increasing disparity between maximum and average error rates before catastrophic failure.
2Reliability
If chipkill method is used to rescue data from bad blocks, then data recovery is possible, but it requires additional NAND flash devices and increases system cost
Solution Approach 1:
The patent applies preliminary action by detecting and retiring blocks that show signs of becoming bad blocks, before actual data loss occurs. By monitoring the ratio of maximum to average error rates and comparing against thresholds, the system proactively identifies at-risk blocks and migrates their data to safe locations in advance, eliminating the need for complex post-failure recovery mechanisms like chipkill.
Solution Approach 2:
The patent converts the harmful effect of developing bad blocks into a beneficial early warning signal. The increasing disparity between maximum and average error rates, which indicates deteriorating block health, is transformed into a predictive metric that triggers preventive data migration, turning a potential failure mode into an actionable alert system.
3Measurement precision
If blocks are monitored continuously for bad block development, then failure prediction accuracy improves, but the monitoring process consumes more power and time
Solution Approach 1:
The patent implements periodic monitoring through the Background Media Scan (BGMS) mechanism, which periodically reads and evaluates error rates in idle or low-activity periods. This periodic approach balances detection accuracy with time efficiency by performing comprehensive error rate analysis at scheduled intervals rather than continuously, reducing time overhead while maintaining effective failure prediction.
Data Source
AI summary
Systems and methods are provided to detect a developed bad word-line of a flash memory. Embodiments provide an improved Background media scan (BGMS) process that can predict at the end of a block read if a word-line will potentially become bad with the use of the flash memory. Accordingly, data from the potentially bad block can be recovered and the block can be retired. The embodiments can minimize the need for the expensive chip-kill method.