Storage Controller Machine Learning Access Classifier
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Storage devices with nonvolatile memory often fail to provide optimum operating performance due to varying user data usage patterns and environments, as manufacturers set algorithms based on average usage patterns and environments, which may not align with individual user needs.
Innovation Solution
Incorporating a controller with a buffer memory and processing unit that performs machine learning to collect access result and environment information, generating an access classifier to predict and optimize future access operations, allowing for selective performance of read or reprogram operations based on predicted results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manufacturers set algorithms based on average usage patterns and environments, then the storage device can operate with standardized control logic, but it fails to provide optimum operating performance for individual users with varying usage patterns
Solution Approach 1:
The storage device performs machine learning autonomously by itself, collecting access information and generating access classifiers without external intervention. The controller automatically trains the machine learning model using accumulated access patterns, enabling the system to adapt to individual user behaviors self-service style, resolving the contradiction between adaptability and complexity by making the system self-configuring rather than requiring complex pre-programming for all scenarios
Solution Approach 2:
The system collects access information in advance during normal operation and performs machine learning training preliminarily to generate access classifiers before they are needed for optimization. By accumulating access patterns and training models in advance, the system prepares optimized access strategies proactively, allowing it to adapt to user patterns without requiring complex real-time decision-making logic
2Productivity
If the storage device uses fixed access algorithms, then the device structure remains simple, but the operating performance varies when usage patterns differ from averages
Solution Approach 1:
The controller serves multiple functions: it acts as both the storage management unit and the machine learning processing unit. The same controller that manages normal storage operations also collects access information, trains machine learning models, and generates access classifiers. This multi-functionality approach improves productivity by enabling adaptive optimization without adding separate dedicated hardware components, thus avoiding excessive complexity
Solution Approach 2:
The patent combines the access control function with machine learning capabilities into a unified system. The access information collection, machine learning training, and access classifier generation are merged into the existing controller architecture rather than being implemented as separate systems. This merging approach enables improved operating performance through adaptive algorithms while avoiding the complexity overhead of separate dedicated hardware or software modules
3Reliability
If the storage device collects and processes access information for machine learning, then it can predict and optimize future accesses, but the processing time and computational resources increase
Solution Approach 1:
The system performs machine learning training periodically rather than continuously. Access information is collected during normal operation, and the controller trains machine learning models at periodic intervals or when certain conditions are met (e.g., when sufficient data is accumulated). This periodic approach maintains prediction accuracy by regularly updating access classifiers while minimizing the time computational processing impacts normal storage operations
Solution Approach 2:
Access information is collected in advance during normal storage operations without interrupting productivity. The system accumulates access patterns preliminarily in the background, and machine learning training is performed on this pre-collected data. By preparing training data in advance and performing computations during idle or low-load periods, the system achieves reliable prediction accuracy while minimizing the time loss impact on normal storage operations
Data Source
AI summary
A storage device may include a nonvolatile memory device, a buffer memory, and a controller. The controller may perform first accesses on the nonvolatile memory device using the buffer memory, collect access result information and access environment information of the first accesses in the buffer memory, and generate an access classifier that predicts a result of a second access to the nonvolatile memory device by performing machine learning based on the access result information and the access environment information collected in the buffer memory.