Memory Cell Fail Classification Using AI Threshold Voltage Images

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for analyzing threshold voltage distributions of memory cells are manual and inconsistent, leading to difficulty in setting accurate determination criteria and requiring significant resources, especially when cell types and memory device capacities vary.

Innovation Solution

A fail classification device that includes a memory to store threshold voltage distribution information and a processor to determine cell type, transform data into a two-dimensional image, and classify fails using a target artificial intelligence model optimized for the cell type, thereby accommodating variations in data and reducing resource requirements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If manual analysis of threshold voltage distributions is performed, then flexibility in handling different cell types is maintained, but analysis accuracy and consistency deteriorate

Engineering Contradiction:

Improveflexibility in handling different cell typesVSAvoidanalysis accuracy and consistency

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent transforms the one-dimensional threshold voltage distribution data into two-dimensional images, changing the data representation format. This allows AI models to process the data consistently across different cell types while maintaining the ability to handle variations through image-based feature extraction, thereby improving both accuracy and adaptability simultaneously

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces manual analysis with automated AI-based analysis. By substituting human operators with machine learning models that process two-dimensional images of threshold voltage distributions, the system achieves consistent and accurate analysis across different cell types without the inconsistencies inherent in manual methods

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Use of energy by moving object

If manual analysis methods are used, then resource requirements are reduced, but analysis efficiency and productivity deteriorate

Engineering Contradiction:

Improveresource requirementsVSAvoidanalysis efficiency

Core Design Contradiction:

Use of energy by moving objectVSProductivity

Solution Approach 1:

The patent replaces manual analysis with automated AI-based analysis using two-dimensional images. This substitution dramatically improves analysis efficiency and productivity while the resource consumption increases only moderately due to the efficient image-based processing approach, achieving a favorable balance between automation benefits and resource usage

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If AI models are used for fail classification, then analysis accuracy and consistency improve, but resource requirements and device complexity increase

Engineering Contradiction:

Improveanalysis accuracy and consistencyVSAvoiddevice complexity and resource requirements

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential features by transforming threshold voltage distributions into two-dimensional images and feeding them to AI models. This extraction approach achieves high analysis accuracy while keeping device complexity manageable by focusing on critical information rather than processing all raw data

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the data representation from one-dimensional voltage values to two-dimensional images, which enables efficient processing by AI models. This parameter transformation improves analysis accuracy while the standardized image format actually simplifies the overall system architecture compared to handling variable-format raw data

Inventive Principle:

Principle #35Parameter changes

4Measurement precision

If analysis methods are optimized for specific cell types, then accuracy for those cell types improves, but adaptability to other cell types deteriorates

Engineering Contradiction:

Improveaccuracy for specific cell typesVSAvoidadaptability to different cell types

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal analysis system where two-dimensional images of threshold voltage distributions can be processed by the same AI model across different cell types. The image-based approach and AI model are designed to be cell-type-agnostic, automatically adapting to different memory cell configurations while maintaining high accuracy for each specific type

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12488855B2Fail classification device for plurality of memory cells and method thereof

Publication Date: 2025.12.02 SK HYNIX INC

AI summary

A fail classification device determines a cell type of a plurality of memory cells on the basis of threshold voltage distribution information or determines that the threshold voltage distribution information is abnormal. The fail classification device transforms the threshold voltage distribution information into a two-dimensional image, inputs the two-dimensional image to a target artificial intelligence model corresponding to the cell type, and classifies the fail type of the plurality of memory cells on the basis of the output result of the target artificial intelligence model.