AI Sensing Voltage Estimation for Non-Volatile Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current error correction technologies, such as BCH coding, are insufficient in maintaining the reliability of non-volatile memory due to advancements in manufacturing, leading to increased error probabilities and reduced product lifetime.
Innovation Solution
A method is developed to train artificial intelligence by supplying initial sensing voltages to memory units, classifying them into strong correct, weak correct, strong error, and weak error regions, and using machine learning to adjust these voltages based on strong correct and error ratios and histogram parameters to generate practical sensing voltages, thereby improving classification precision and reducing errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If BCH coding technology is used for error correction, then computation speed is fast, but correction capability is insufficient for modern manufacturing variations
Solution Approach 1:
The patent changes the parameter of error correction technology from BCH coding to LDPC coding. LDPC coding provides stronger correction capability for modern manufacturing variations while maintaining acceptable computation speed through optimized implementation, thus resolving the contradiction between speed and reliability.
2Reliability
If LDPC error correction technology is used, then correction capability is strong, but device complexity increases
Solution Approach 1:
The patent segments the error correction process into distinct phases: sensing voltage application, memory unit classification into strong correct/weak correct/strong error/weak error regions, and iterative decoding. This segmentation manages the complexity of LDPC coding by breaking it into manageable steps while maintaining strong correction capability.
Solution Approach 2:
The patent performs preliminary classification of memory units into four regions based on sensing voltages before executing the full LDPC decoding process. This preliminary action reduces the complexity of subsequent error correction by pre-identifying likely correct bits, allowing the system to focus computational resources on uncertain cases.
3Productivity
If initial sensing voltages are used without adjustment, then classification precision is low, but error probability increases
Solution Approach 1:
The patent implements feedback by using the results of initial classification and decoding attempts to adjust sensing voltages for subsequent operations. The system learns from classification errors and modifies voltages to improve precision, creating a closed-loop system that resolves the contradiction between speed and precision.
Solution Approach 2:
The patent makes the sensing voltages dynamic rather than fixed. Voltages are adjusted based on classification results and decoding performance, allowing the system to adapt to actual memory unit characteristics. This dynamic adjustment improves classification precision without permanently increasing system complexity.
Data Source
AI summary
A method of training artificial intelligence to estimate sensing voltages for a storage device is provided, which includes steps of: supplying initial sensing voltages to memory units; defining various storing states; comparing threshold voltages of the memory units with the initial sensing voltages to classify the memory units; calculating a ratio of the number of the memory units in a strong correct region to the number of in the strong correct region and a weak correct region; calculating a ratio of the number of the memory units in a strong error region to the number of in the strong error region and a weak error region; calculating the number of the memory units in the weak correct and error regions to obtain a histogram parameter; inputting the ratios and parameter to an artificial intelligence neural network; and using machine learning to analyze practical sensing voltages.