Non-Volatile Memory Chip Endurance Prediction Using Hidden Markov Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the endurance of non-volatile memory chips are inadequate due to random factors in endurance tests, lacking a reliable data-driven model for predicting various states, which hinders targeted repair and reduces the service life of these chips.
Innovation Solution
A method and apparatus using a machine learning model, specifically a Hidden Markov Model, to predict the endurance of non-volatile memory chips by building a dataset from endurance test data, training the model, and employing state transition and observed-state probability matrices to forecast future states, enabling targeted repair and improving chip longevity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical models at chip level are used for endurance prediction, then the prediction can be performed, but the prediction accuracy is insufficient due to random factors and noise in system and process manufacturing
Solution Approach 1:
The patent replaces traditional physical models with machine learning models that process statistical data from endurance tests. The machine learning model learns patterns from actual test data including random factors, substituting the deterministic physical modeling approach with a data-driven statistical approach that naturally handles variability and noise in the system.
Solution Approach 2:
The patent transforms the prediction approach by changing from fixed physical parameters to dynamic statistical parameters derived from endurance test data. The model uses probability distributions and statistical characteristics that adapt to the actual behavior of memory chips under various stress conditions, improving prediction accuracy despite manufacturing variations.
2Duration of action of stationary object
If traditional endurance testing is performed, then endurance data can be collected, but the test period is limited and cannot predict long-term reliability
Solution Approach 1:
The patent performs preliminary analysis of endurance test data to extract statistical patterns and characteristics that indicate future failure modes. By analyzing the evolution of error rates and failure patterns during the test period, the model extrapolates long-term reliability trends without requiring extended physical testing, thus predicting service life beyond the actual test duration.
Solution Approach 2:
The patent creates a virtual model that replicates the behavior of physical chips based on limited test data. This digital twin or virtual representation allows for long-term predictions by simulating chip behavior under various conditions without requiring actual long-term physical testing, effectively copying the chip's reliability characteristics from short-term tests to long-term predictions.
3Productivity
If repair measures are implemented without accurate prediction, then general maintenance can be performed, but targeted repair cannot be achieved reducing yield improvement
Solution Approach 1:
The patent implements a feedback mechanism where the machine learning model continuously learns from actual chip performance data and test results. The prediction results feed back into the repair process, allowing the system to identify which specific chips or memory blocks require repair and what type of repair is most effective. This closed-loop approach enables targeted repairs that improve yield by focusing resources on chips that will benefit most from intervention.
Data Source
AI summary
The present disclosure relates to a field of semiconductor technology, and in particular to a method and apparatus of predicting an endurance of a chip, a computer-readable medium, and an electronic device. The method includes: building a dataset according to endurance test data of a target chip, where the dataset includes a training set and a test set; acquiring a machine learning model, and training the machine learning model through the training set; and predicting an endurance of the target chip through the machine learning model.