Storage Device Machine Learning Model Selection for Feature Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As semiconductor memory devices become increasingly miniaturized, they experience variations in feature characteristics between adjacent chips, blocks, and wordlines, leading to inefficiencies in operation and data storage, which existing technologies fail to adequately address.
Innovation Solution
A storage device incorporating a feature information database and a machine learning module that selects and tunes machine learning models based on feature information to optimize operation, using a model pool, selector, and tuner to minimize errors between predicted and measured values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of moving object
If process miniaturization is implemented to reduce storage device size, then storage device dimensions are reduced, but feature variations between adjacent chips, blocks, and wordlines increase
Solution Approach 1:
The patent divides the storage device into multiple segments (chips, blocks, wordlines) and applies individual machine learning models to each segment. The controller selects specific models based on the segment's unique feature information, allowing customized handling of variations in each segmented unit while maintaining overall device miniaturization.
Solution Approach 2:
The patent changes operational parameters dynamically by selecting different machine learning models based on measured feature information. The system measures actual feature values (such as threshold voltage distributions) and adjusts the selected ML model accordingly, enabling adaptation to manufacturing variations without increasing physical device size.
2Measurement precision
If machine learning models are selected and tuned based on feature information, then operational accuracy is improved, but device complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-training multiple machine learning models offline before device operation. During actual operation, the controller only needs to measure feature information and select from the pre-trained models, avoiding the complexity of real-time model training while maintaining high operational accuracy.
Solution Approach 2:
The patent introduces machine learning models as intermediary components between the raw feature information and the control decisions. These models act as mediators that process the complex feature data and provide optimized control parameters, simplifying the overall control architecture while improving accuracy.
3Adaptability or versatility
If multiple machine learning models are maintained for different memory devices, then adaptability to feature variations is improved, but memory and processing requirements increase
Solution Approach 1:
The patent implements a dynamic model selection mechanism where the controller chooses from multiple pre-trained machine learning models based on real-time feature measurements. This dynamic approach allows the system to adapt to different memory device characteristics without maintaining all models simultaneously active, optimizing resource utilization while preserving adaptability.
Data Source
AI summary
A storage device, including a feature information database configured to store feature information about a memory device; and a machine learning module configured to select a machine learning model from a plurality of machine learning models the corresponding to an operation of the memory device based on the feature information, wherein the memory device is configured to operate according to the selected machine learning model.