Overlay Error Prediction From Tool Signals in Fan-Out WLP

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In semiconductor manufacturing, particularly in fan-out wafer level packaging (WLP), managing vertical alignment of interconnection features is challenging due to tool degradation, leading to overlay alignment errors and reduced capacity in forming features at expected locations.

Innovation Solution

The use of machine learning techniques to analyze tool signals from alignment tools, correlating them with overlay metrology measurements to predict and correct overlay errors by adjusting tool settings, thereby improving alignment accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If traditional overlay metrology processes are used to monitor vertical alignments, then alignment accuracy can be maintained, but tool degradation leads to overlay alignment errors and reduced capacity in forming features

Engineering Contradiction:

Improveoverlay alignment accuracyVSAvoidtool capacity

Core Design Contradiction:

Manufacturing precisionVSReliability

Solution Approach 1:

The system implements feedback by continuously monitoring tool signals during wafer processing and using machine learning to correlate these signals with overlay metrology measurements. The predicted overlay errors are fed back to adjust tool settings, creating a closed-loop control system that maintains alignment accuracy despite tool degradation.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent replaces traditional mechanical overlay measurement and correction systems with a machine learning-based predictive system. Instead of relying solely on physical metrology tools to detect and correct alignment errors, the system uses statistical models to predict overlay errors from tool signals and automatically adjusts tool parameters.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If machine learning techniques are used to predict overlay errors, then tool capacity is maintained despite degradation, but the system complexity increases

Engineering Contradiction:

Improvetool capacityVSAvoidsystem complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system changes parameters by transforming raw tool signals into predicted overlay error values through machine learning models. The system dynamically adjusts tool settings based on predicted errors, changing operational parameters to compensate for tool degradation and maintain productivity.

Inventive Principle:

Principle #35Parameter changes

3Manufacturing precision

If overlay errors are corrected by adjusting tool settings, then alignment precision is enhanced, but the measurement and control complexity increases

Engineering Contradiction:

Improveinterconnection feature alignmentVSAvoidoverlay metrology measurement

Core Design Contradiction:

Manufacturing precisionVSDifficulty of detecting and measuring

Solution Approach 1:

The machine learning model acts as an intermediary between raw tool signals and overlay metrology measurements. Instead of directly measuring overlay errors, the system uses the ML model to translate tool signals into predicted error values, simplifying the measurement process while maintaining precision.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12197138B2Machine learning on overlay management

Publication Date: 2025.01.14 TAIWAN SEMICONDUCTOR MANUFACTURING CO LTD

AI summary

The current disclosure describes techniques for managing vertical alignment or overlay in semiconductor manufacturing using machine learning. Alignments of interconnection features in a fan-out WLP process are evaluated and managed through the disclosed techniques. Big data and machine learning are used to train a classification that correlates the overlay error source factors with overlay metrology categories. The overlay error source factors include tool signals. The trained classification includes a base classification and a Meta classification.