Defect Classification System for Semiconductor Wafer Yield
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Semiconductor device fabrication is hindered by particle and metal contamination on process wafers, leading to product failures, yield decreases, and resource inefficiencies due to the complexity of defect classification and source identification in advanced manufacturing processes.
Innovation Solution
A computational system employing machine learning and deep learning techniques for defect classification, combining metrology data with manufacturing information to predict refined posterior distributions and identify defect sources, with optional Bayesian analysis for further classification and suggested corrective actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional manual defect classification methods are used, then resource expenditure and analysis time increase, but defect classification accuracy and source identification capability remain insufficient
Solution Approach 1:
The patent replaces manual mechanical defect classification with an automated machine learning system that uses metrology data, manufacturing information, and Bayesian analysis to classify defects and identify sources automatically, eliminating the need for manual analysis while improving accuracy
Solution Approach 2:
The defect classification system performs self-service by automatically analyzing metrology data, comparing it with manufacturing information, and generating defect classifications and source identifications without requiring external manual intervention, thereby reducing both time and resource expenditure
2Measurement precision
If comprehensive defect analysis with multiple data inputs is performed, then defect classification accuracy improves, but system complexity increases
Solution Approach 1:
The patent segments the complex defect analysis system into distinct functional modules: a defect classification engine that processes metrology data, a Bayesian analysis component that integrates manufacturing information, and an output generation module, allowing each segment to handle specific tasks independently while maintaining overall system accuracy
Solution Approach 2:
The patent introduces a Bayesian analysis intermediary that acts as a mediator between raw manufacturing information and the defect classification process, synthesizing multiple data inputs (metrology data, manufacturing information, prior distributions) into refined posterior distributions that guide accurate defect classification without requiring direct complex interactions between all input variables
3Productivity
If automated defect classification systems are implemented, then analysis speed and productivity improve, but the complexity of implementing and maintaining the system increases
Solution Approach 1:
The defect classification system is designed with multi-functionality to handle various defect types, data formats, and manufacturing processes through a unified Bayesian analysis framework, allowing the same system architecture to serve multiple purposes and reducing the complexity of implementing separate specialized systems for different scenarios
Data Source
AI summary
Defects on a substrate comprising electronic components can be classified with a computational defect analysis system that may be implemented in multiple stages. For example, a first stage classification engine may process metrology data to produce an initial classification of defects. A second stage classification engine may use the initial classification, along with manufacturing information and/or prior defect knowledge to output probabilities that the defects are caused by one or more potential sources.