Wafer Virtual Metrology Using Process Path Moving Averages
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in semiconductor manufacturing is predicting wafer quality accurately without measuring physical characteristics of all wafers, as historical data from specific devices is insufficient due to the influence of numerous unit processes, leading to inconsistent and time-consuming virtual metrology predictions.
Innovation Solution
A virtual metrology method that collects log data and measured values of sample wafers, classifies them by process paths, calculates moving average values, and uses these to predict physical characteristics of unmeasured target wafers, improving accuracy and reducing prediction time by focusing on process path information rather than entire historical data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If virtual metrology is performed using historical data from a specific device, then measurement values can be predicted without actual measurement, but prediction accuracy is insufficient because the measured value is affected by numerous unit processes
Solution Approach 1:
The patent segments the historical data by process paths, dividing the comprehensive manufacturing process into distinct pathways. By classifying measured values according to their specific process paths and performing virtual metrology within each path segment, the system achieves more accurate predictions while using limited sampling measurement data effectively.
2Productivity
If sampling measurements are performed on only a portion of manufactured wafers, then wafer manufacturing time is reduced, but measurement data is insufficient for accurate quality prediction
Solution Approach 1:
The patent changes the parameter of data organization by introducing process path classification. Instead of treating all measurement data uniformly, the system transforms the data structure to group measurements by their specific process paths, enabling accurate quality prediction even with reduced measurement quantities through effective data utilization.
3Reliability
If virtual metrology uses comprehensive historical data from all devices, then more process variations are covered, but prediction time increases and consistency decreases
Solution Approach 1:
The patent extracts and isolates the relevant process path information from comprehensive historical data. By taking out only the data corresponding to the specific process path of the target wafer and excluding unrelated data from other paths, the system achieves both high prediction consistency and reduced processing time through focused data analysis.
Data Source
AI summary
A virtual metrology method for a wafer includes collecting log data including process path information of wafers manufactured in a semiconductor process; collecting measured values of sample wafers of which physical characteristics are measured in the semiconductor process, the sample wafers being a group of wafers selected from among the manufactured wafers; classifying measured values of the sample wafers according to process paths based on the process path information; calculating a moving average value of measured values classified for each process path; and determining a moving average value corresponding to a process path of an unmeasured target wafer from among the manufactured wafers as a predicted value of the physical characteristics of the target wafer.