AVS 49th International Symposium
    Plasma Science Tuesday Sessions
       Session PS-TuP

Paper PS-TuP6
Introduction of a Powerful New Method of Generating Accurate Endpoint Traces Combined with Process Fault Classification for Low Dimensional Open Areas

Tuesday, November 5, 2002, 5:30 pm, Room Exhibit Hall B2

Session: Plasma Applications
Presenter: D. Knobloch, Infineon Technologies AG, Germany
Authors: D. Knobloch, Infineon Technologies AG, Germany
F.H. Bell, Infineon Technologies AG, Germany
K. Voigtlaender, ADP GmbH, Germany
J. Zimpel, ADP GmbH, Germany
Correspondent: Click to Email
A powerful new method of generating accurate endpoint traces based on full range optical emission spectroscopy (OES), e.g. for contact hole and VIA etch, is presented. Conventional endpoint systems derive endpoint signals by monitoring one or two discrete wavelengths. However, this technique is not accurate enough for logic and advanced memory products with low open areas and highly topographical structures. Consequently, more sophisticated and robust endpoint control of plasma processes requires the use of the full optical spectral information and powerful analysis methods. Statistical methods based on algorithms, such as evolving windows factor analysis (EWFA), principal component orientation (PCO), Hotelling's T2 and others have been used to find suitable endpoint traces. However these methods do not tend to be robust enough, because all changes in plasma emission - not only endpoint relevant changes - will be traced. To overcome these problems, a newly developed algorithm that separates endpoint information from process faults and variations is presented. The algorithm uses the complete individual weighted spectral endpoint information for a long-term robust and sensitive endpoint detection. The algorithm is based on modelling techniques constructing an optimal superposition of spectral eigenvectors to generate a spectral software filter. The application of this method in the production works like a black box for the engineer. The only requirement is the predefinition of the shape of the expected endpoint trace. Accurate and robust endpoint detection of open areas beyond 0.5% and highly topographical structures is achieved and presented.