Wafer Defect Prediction by Life Distribution Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing semiconductor devices face defects due to time-dependent dielectric breakdown (TDDB) leading to gate oxide film deterioration and malfunction, necessitating improved methods for predicting and classifying wafer defects by life distribution type.

Innovation Solution

A method and device for predicting wafer defects by estimating defect occurrence time data, distinguishing defect distribution types, and outputting wafer defect information using a prediction model, which includes a processor with distribution estimation and machine learning units to classify wafer grades based on defect characteristics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If wafer defects are predicted without classifying defect distribution types, then the prediction process is simpler, but the prediction accuracy and reliability are reduced

Engineering Contradiction:

Improvedefect prediction accuracyVSAvoidprediction process complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the defect prediction process by classifying defects into different distribution types (Weibull distribution with shape parameter β < 1, β > 1, or β = 1). This segmentation allows the system to apply different prediction models and parameters for each defect type, thereby improving prediction accuracy while managing complexity through structured classification.

Inventive Principle:

Principle #1Segmentation

2Reliability

If a single defect prediction model is used for all defect types, then the model is simpler to implement, but it cannot accurately predict different life distribution types

Engineering Contradiction:

Improveprediction reliability for different life distributionsVSAvoidmodel complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent changes the parameters of the prediction model based on the classified defect distribution type. Specifically, it uses different shape parameters (β) for the Weibull distribution model depending on the defect type: β < 1 for early failures, β > 1 for wear-out failures, and β = 1 for random failures. This parameter adaptation improves reliability for different life distributions while keeping the overall model structure manageable.

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If defect occurrence time data is not analyzed for distribution patterns, then data processing is faster, but defect classification and prediction accuracy are reduced

Engineering Contradiction:

Improvedefect distribution classification accuracyVSAvoiddata processing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary classification of defect occurrence time data into different distribution types before detailed prediction analysis. By pre-categorizing data based on distribution patterns (using statistical methods to identify Weibull distribution characteristics), the system prepares the data structure in advance, enabling more efficient and accurate subsequent prediction processing.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12579343B2Wafer defect prediction device and operating method thereof

Publication Date: 2026.03.17 SAMSUNG ELECTRONICS CO LTD

AI summary

A method of predicting wafer defect information includes estimating a distribution with respect to defect occurrence time data, the defect occurrence time data including information about a time associated with a wafer defect occurrence, distinguishing a defect distribution type according to a result of the estimating the distribution, and outputting wafer defect information predicted according to the distinguished defect distribution type.