Semiconductor Yield Prediction From Defect Features and Wafer Coverage

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing production line defect monitoring technologies often underreport defects in semiconductor manufacturing, especially for lot wafers with tight production periods or short Q-Time, leading to inaccurate estimation of chip probing yield.

Innovation Solution

A method and apparatus using a pre-trained yield prediction model combining neural network and support vector machine structures to accurately predict semiconductor device defects and yield by extracting defect feature vectors and classifying qualified or unqualified yields, with steps including acquiring defect data, training the model with sample data, and optimizing using penalty coefficients and gamma values.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If existing production line defect monitoring technology is used to review trend charts daily, then defect notice highlight signals can be issued when defect rate is high, but defects may be underreported due to missed reviews or skipped lot wafers, affecting yield estimation accuracy

Engineering Contradiction:

Improvedefect monitoring efficiencyVSAvoiddefect detection accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary defect detection and classification on all lot wafers before final yield determination. By pre-processing defect data and identifying potential issues early in the production flow, the system ensures no defects are missed due to time constraints or skipped reviews, thereby improving both monitoring efficiency and detection accuracy

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

An intermediary AI model is introduced between raw defect data and final yield estimation. This model acts as a mediator that comprehensively analyzes all defect information without skipping any lot wafers, translating raw defect data into accurate yield predictions while maintaining both efficiency and precision

Inventive Principle:

Principle #24Intermediary (Mediator)

2Speed

If lot wafers with tight production periods or short Q-Time are skipped or not scanned, then production speed is maintained, but defect detection is incomplete, affecting chip probing yield estimation

Engineering Contradiction:

Improveproduction line speedVSAvoidyield estimation reliability

Core Design Contradiction:

SpeedVSReliability

Solution Approach 1:

Defect scanning and data collection is performed as a preliminary action for all lot wafers including those with tight production periods. By completing defect detection before production scheduling decisions, the system ensures no wafers are skipped, maintaining both production speed and yield estimation reliability

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The manual decision-making process for skipping lot wafers is replaced with an automated AI-based defect detection system. This substitution eliminates human judgment errors and consistent application of defect detection across all wafers regardless of production timing, ensuring reliable yield estimation while maintaining production speed

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Device complexity

If traditional defect monitoring methods are used with manual review, then system complexity is low, but defect detection coverage is incomplete and yield prediction accuracy is poor

Engineering Contradiction:

Improvemonitoring system complexityVSAvoidyield prediction accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The system employs self-service mechanisms where the AI model automatically learns from historical defect data and continuously improves its prediction accuracy without requiring complex manual configuration or intervention. This self-learning capability achieves high prediction accuracy while keeping the operational complexity manageable

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system transforms the monitoring approach by changing key parameters: transitioning from manual review to automated AI analysis, from sampling-based detection to comprehensive defect scanning, and from reactive defect notice to predictive yield estimation. These parameter changes dramatically improve accuracy while the modular architecture keeps complexity controlled

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12198061B2Method and apparatus for predicting yield of semiconductor devices

Publication Date: 2025.01.14 SHANGHAI HUALI INTEGRATED CIRCUIT CORP

AI summary

A method for predicting the yield of manufacturing semiconductor devices includes steps of: acquiring defect data of semiconductor devices to be predicted, wherein the semiconductor devices to be predicted include finished semiconductor devices and semi-finished semiconductor devices, and the defect data indicates a defect type and location of at least one defect of the semiconductor devices; inputting the defect data into a pre-trained yield prediction model, wherein the yield prediction model includes a neural network structure and a classification structure, the neural network structure is used to extract defect feature vectors from the defect data, and the classification structure is used to output classification results of qualified or unqualified yield according to the defect feature vectors; and determining, by the yield prediction model, classification results of qualified or unqualified yield of the semiconductor devices.