Prediction of wafer homogeneity maps using a virtual metrology scheme consisting of time resolved OES measurements and a neural network

摘要

With the further scaling of device dimensions, the plasma fabrication processes become more and more complex. A permanent growing parameter field of the production tools on the one hand side and a steady decrease of the local optima inside these parameter fields on the other hand side make the long time used empirical process optimization nearly impossible. To overcome these problems plasma diagnostic methods will be used to analyze chemical and physical conditions directly during the running process. But most of the plasma diagnostic tools measure conditions directly in the plasma glow zone and deliver no information of the mechanism occurring in the device area on the wafer surface. Consequently, the benefit of plasma diagnostic data for process engineers is often more or less limited. Virtual metrology (VM) approaches have been developed in the last years to compile complex tool data sets into interesting process outputs. This enables a wafer to wafer monitoring of process results without cost-intensive and time consuming real metrology [1]. The capability of such virtual metrology schemes for prediction of process outputs was shown in the past for chemical vapour deposition (CVD) [2] and reactive ion etch (RIE) [3] processes. The paper of Kim [4] shows the use of such a scheme to predict etch geometries of RIE processes. The analysis of such geometries using real metrology is a time consuming process, which is often destructive and needs dedicated test or monitor wafers.