AI-Integrated Non-Volatile Memory for Preemptive Block Defect Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Non-volatile memory storage systems face challenges in reliably predicting defects before they lead to data loss, as existing methods lack efficient and timely detection mechanisms.
Innovation Solution
A machine learning-based memory health management system is implemented, utilizing two pre-trained models and an inference circuit to predict defects by analyzing current operation metrics, with a faster, less reliable initial prediction and a more intensive, accurate secondary prediction when needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single intensive prediction model is used to accurately predict defects, then measurement precision is improved, but loss of time increases due to intensive data collection and processing
Solution Approach 1:
The prediction system is segmented into two distinct models: a first prediction model that provides quick, less accurate predictions using minimal data collection, and a second prediction model that provides accurate predictions through intensive data collection and processing. This segmentation allows the system to balance speed and accuracy by using the appropriate model for different operational contexts.
Solution Approach 2:
The first prediction model performs partial action by using reduced data collection and simpler processing to provide rapid predictions. The second prediction model performs excessive action by collecting more data and conducting more intensive analysis to ensure high accuracy. This approach allows the system to use partial action when speed is prioritized and excessive action when accuracy is critical.
2Reliability
If extensive data collection and intensive processing are performed continuously, then reliability of defect prediction is improved, but use of energy increases
Solution Approach 1:
The system segments the prediction process into two modes: a first mode using the first prediction model with minimal data collection and processing that consumes less energy, and a second mode using the second prediction model with extensive data collection and intensive processing that consumes more energy but provides higher reliability. This segmentation enables the system to optimize energy consumption by selecting the appropriate prediction mode based on operational requirements.
Solution Approach 2:
The first prediction model performs partial data collection and processing to provide reliable predictions with lower energy consumption. The second prediction model performs excessive data collection and intensive processing to ensure high reliability when needed. This approach allows the system to use partial action for routine predictions and excessive action for critical assessments, optimizing the balance between reliability and energy consumption.
Data Source
AI summary
A non-volatile storage apparatus comprises a plurality of memory cells that store host data and two models, a control circuit for writing to and reading from the memory cells, and an inference circuit. The inference circuit uses the first model with a first set of one or more metrics describing current operation of the non-volatile storage apparatus to make a first level prediction about defects and uses the second model with a second set of one or more metrics describing current operation of the non-volatile storage apparatus to make a second level prediction about defects. In one embodiment, the first level prediction is faster to make and uses less data collection, but is not as reliable, as the second level prediction. While second level prediction is more reliable, it takes more time to perform and requires a more intensive data collection, so it is only used when needed.