| |
 |
| |
| |
Wednesday, November 26,2008
11F Ballroom D, The Ambassador Hotel Hsin Chu |
| |
| |
Instructor: MKS Instruments |
| |
| |
■ Abstract: |
| |
In today’s advanced semiconductor processes, fault detection is typically implemented by monitoring twenty to fifty parameters, or more. Univariate analysis (UVA) based fault detection methodologies are plagued with high false alarm rates by the very nature of the statistics of fault detection. This tutorial will present the inherent advantages of multivariate analysis (MVA) fault detection methodologies, and show how a significant reduction in false alarms can be achieved using MVA as compared to UVA techniques. |
|
| |
First we will discuss the basic principals of UVA and MVA, defining the working parameters and statistics. Then we will focus on decision theory, and show a direct comparison between multivariate and univariate statistics, and the implications on false alarm rates. This comparison will show the inherent advantage of MVA in false alarm rate reduction. The analytical results will be supported by a comparison of the frequency of UVA alarms to MVA alarms based on actual wafer data. |
|
| |
Additionally, we will discuss the statistics of downstream yield loss and analyze relative strengths of MVA and UVA in reducing the probability of yield loss later in the process. When the false alarm rate is reduced, the production rate increases as a result of fewer interruptions due to operator intervention in the process. When the probability of yield loss decreases, yield increases. MVA based fault detection is a simple solution for both productivity improvement and yield enhancement. |
|
| |
This tutorial is intended for process and tool engineers and managers who want to understand the motivation for and conceptual foundations of MVA based fault detection and advanced process control. While some mathematics is used to derive the results presented, mathematical proficiency in statistics and decision theory is not required. Attendees will leave the tutorial with an understanding of the benefits of multivariate data analysis relative to fault detection, the reduction of false alarms, and yield improvement. |
| |
| |
■ About the Instructor: |
|
|
Home > Tutorial
