Flash Controller Outlier Block Management Using DNN TVSO Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Solid State Drives (SSDs) do not effectively manage outlier blocks, which diverge from performance as they approach end-of-life, leading to increased raw bit error rates and reduced storage capacity, resulting in premature retirement of blocks and increased system latency.
Innovation Solution
A method and apparatus at the flash controller level that uses deep neural networks to predict Threshold Voltage Shift Offset (TVSO) values for outlier blocks, allowing them to continue being used by performing multiple test reads, training the neural network, and adjusting TVSO values until a training error threshold is met, thereby extending the life of these blocks without compromising Uncorrectable Bit Error Rate (UBER).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional PE-based reliability states are used to manage flash memory blocks, then the majority of blocks can be effectively managed, but outlier blocks diverge from performance and experience higher raw bit error rates, leading to premature retirement
Solution Approach 1:
The patent applies local quality by transitioning from uniform PE-based reliability management to block-specific TVSO-based management. Each outlier block receives customized TVSO values determined by deep neural network predictions, allowing localized optimization of read parameters for each block's specific characteristics rather than applying general PE-cycle-based parameters to all blocks.
Solution Approach 2:
The patent implements parameter changes by introducing TVSO (threshold voltage shift offset) as a dynamic parameter that varies for each outlier block based on DNN predictions. Instead of relying solely on PE cycle counts, the system adjusts TVSO values to compensate for block-specific degradation patterns, thereby maintaining reliability and extending lifespan.
2Reliability
If outlier blocks are retired early to maintain performance, then system reliability is maintained, but storage capacity is reduced
Solution Approach 1:
The patent implements feedback through deep neural networks that continuously predict TVSO values based on block performance characteristics. The system monitors outlier blocks, uses DNN to predict optimal TVSO parameters, applies these parameters, and maintains UBER within acceptable thresholds, creating a closed-loop feedback system that prevents premature retirement.
Solution Approach 2:
The patent replaces the conventional mechanical/threshold-based retirement approach with an intelligent DNN-based prediction system. Instead of using fixed PE-cycle thresholds or simple error rate thresholds for retirement decisions, the system uses deep neural networks to predict TVSO values and maintain block usability, substituting intelligent algorithms for traditional rule-based management.
3Duration of action of moving object
If deep neural networks are used to predict TVSO values for outlier blocks, then block lifespan is extended, but device complexity increases
Solution Approach 1:
The patent applies self-service by enabling the flash memory controller to automatically train and update deep neural networks using data from its own operations. The system collects performance data from outlier blocks, uses this data to train DNN models, and continuously improves its own prediction accuracy without external intervention, thereby managing complexity through self-learning.
Solution Approach 2:
The patent implements preliminary action by pre-training deep neural networks using characterization data collected during manufacturing or initial operation. The DNN models are prepared in advance with learned parameters that can be quickly applied to predict TVSO values for outlier blocks, reducing the computational complexity during actual operation.
Data Source
AI summary
A method for outlier management at a flash controller includes testing a flash memory device to identify one or more outlier blocks of the flash memory device. Hyperparameters for a DNN are loaded into a training circuit of the flash controller. Test reads of the one or more outlier blocks are performed and a number of errors in the test reads is identified. The DNN is trained using a mini-batch training process and using the identified number of errors in the test reads and is tested to determine whether the trained DNN meets a training error threshold. The performing, the identifying, the training and the testing are repeated until the trained DNN meets the training error threshold to identify parameters of an outlier-block DNN. A neural network operation is performed using the identified parameters to predict a set of TVSO values. A read is performed using the set of predicted TVSO values.