Advanced Methods for the Analysis of Semiconductor Manufacturing Process Data - Advanced Techniques in Knowledge Discovery and Data Mining - page 63

C118

C063

C119 C071

Fig. 2.21. Visualization of SPLIT6 by four selected component plots.

opposed to the existing split. Parameters also might be redundant with regard
to this issue. From the available supervised methods, automatic selection of
features has been employed to find an answer to this question for the regarded
application data. The SBS selection method delivered the best results for
the higher-order methods in the conducted experiments. In Table 2.4 the
results for lot and split discrimination (SPLIT6) and only split discrimination
(SPLIT3) are documented for the three regarded cost functions and the best
obtained results.
For instance, application of SBS with q si reduced the SPLIT6 database to
just nine parameters. Figure 2.22 shows the resulting projection with nearly
linear separability of the data. From the resulting projection in Fig. 2.22, as
well as the later Fig. 2.23, the existing asymmetry of the split can clearly be
observed, which is a very significant achievement of the regarded visualization
method. The expectation, of course, is that the selected parameters are dom-