RAM Fail Data Augmentation for Realistic Failure Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The decreasing frequency of failures in random access memories due to Error Correction Code (ECC) and product advancements makes it difficult to secure sufficient fail data for analyzing failures using artificial intelligence.
Innovation Solution
A fail data augmentation device and method that includes a memory to store fail data units and a data augmentation model, utilizing processors to input, generate, and selectively delete augmented fail data units, ensuring the data accurately reflects real-world memory conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Error Correction Code (ECC) and product advancements are implemented, then reliability of random access memory is improved, but frequency of failure decreases making it difficult to secure sufficient fail data
Solution Approach 1:
The patent uses a data augmentation model to generate synthetic fail data units that copy and transform existing fail data units. The model creates augmented fail data units by applying transformation parameters to original fail data, effectively copying the failure patterns while generating new variations. This resolves the contradiction by creating additional fail data through copying existing failure characteristics rather than waiting for natural failures to occur.
Solution Approach 2:
The patent applies parameter changes by transforming existing fail data units through a data augmentation model that modifies parameters such as error types, locations, and patterns. The model changes parameters of the fail data to generate diverse augmented fail data units while maintaining the essential failure characteristics. This allows generation of sufficient fail data quantity while preserving reliability information.
2Quantity of substance
If data augmentation model generates augmented fail data units, then quantity of fail data is increased, but accuracy of failure analysis may be compromised due to unrealistic generated data
Solution Approach 1:
The patent implements feedback by using a discriminator that evaluates augmented fail data units and provides feedback on their realism. The discriminator compares generated data against real failure patterns and returns feedback signals to the data augmentation model. This feedback loop allows the model to adjust generation parameters to improve accuracy while maintaining sufficient data quantity.
Solution Approach 2:
The patent applies local quality by selectively applying different transformation parameters to different parts of the fail data units. Instead of uniformly transforming all data, the model applies local transformations that preserve critical failure characteristics in specific regions while allowing flexibility in other areas. This ensures high accuracy in failure analysis by maintaining realistic failure patterns where they matter most.
Data Source
AI summary
A fail data augmentation device may input a plurality of fail data units to a data augmentation model, obtain a plurality of augmented fail data units outputted from the data augmentation model, and delete one or more of the augmented fail data units. The plurality of fail data units and the plurality of augmented fail data units includes a first parameter indicating one of a plurality of banks included in a random access memory, a second parameter indicating one of a plurality of matrices included in the bank corresponding to the first parameter, and a third parameter indicating one of a plurality of hex units included in the matrix corresponding to the second parameter respectively.