Yield Loss Root Cause Analysis for High-Dimensional Process Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor and TFT-LCD manufacturing, identifying the root causes of yield loss is challenging due to the high-dimensional variable selection problem, where the number of process parameters exceeds the number of samples, making it difficult to find key process parameters causing yield loss among numerous production tools with limited workpiece samples.
Innovation Solution
A method and system using a triple phase orthogonal greedy algorithm (TPOGA), least absolute shrinkage and selection operator (LASSO), or sample-efficient regression trees (SERT) to identify key devices and process parameters likely to cause yield test failures by analyzing in-line metrology and final inspection values, with a reliance index to gauge the reliability of search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional big data analysis is performed on all production-related data to find root causes, then comprehensive analysis coverage is achieved, but the analysis time and computational complexity become excessively long and high due to the enormous and complicated amount of data
Solution Approach 1:
The patent extracts only the relevant subset of production data that is actually needed for yield loss analysis, rather than analyzing all available production-related data. This is achieved through selective data collection from specific process stages and tools that are most likely to contribute to yield loss, significantly reducing the data volume while maintaining analysis effectiveness
Solution Approach 2:
The patent segments the production data into different categories based on process stages, tool types, and parameter relevance. By dividing the data into manageable segments and analyzing only the critical segments, the system reduces overall analysis complexity and time while preserving the ability to identify root causes
2Reliability
If all production-related data is collected for yield analysis, then complete data coverage is achieved, but the data complexity and difficulty in searching for root causes increase significantly
Solution Approach 1:
The patent extracts only the essential data elements needed for yield analysis from the complex production environment. It identifies and extracts key process parameters, metrology data, and tool performance metrics that are directly relevant to yield loss, filtering out unnecessary data to reduce complexity
Solution Approach 2:
The patent applies different data collection and analysis strategies to different process stages and tool types based on their specific characteristics and contribution to yield loss. Rather than uniformly analyzing all data, it focuses analysis resources on the specific areas where yield loss is most likely to occur
3Measurement precision
If yield management system analyzes all production tools to find root causes, then comprehensive tool coverage is achieved, but the system becomes ineffective when the number of workpieces is small at RD and ramp-up phases
Solution Approach 1:
The patent changes the analysis parameters and statistical methods based on the available sample size. When workpiece numbers are small (at RD and ramp-up phases), it employs different analytical approaches compared to mass production, adjusting the sensitivity and threshold parameters to maintain effective root cause identification with limited data
Solution Approach 2:
The patent performs preliminary analysis and identifies potential root causes early in the development process even with small sample sizes. By establishing baseline data and early warning indicators during RD and ramp-up phases, it prepares for more effective yield management before full-scale production begins
Data Source
AI summary
Embodiments of the present invention provide a two-phase process for searching the root causes of the yield loss in the production line 100. In a first phase, process tools and their process tool types that are likely to cause the yield loss are identified, and in a second phase, the process parameters that are likely to cause the yield loss within the process tool types found in the first phase are identified. In each phase, two different algorithms can be used to generate a reliance index (RIk) for gauge the reliance levels of their search results.