Database Reference
In-Depth Information
Semiconductor production process
Semiconductor production process
Process control
& optimization
Monitored data
Relevant data detection: OCC/novelty
Relevant data detection: OCC/novelty
Observation
refinement
Condensed
data
Interactive exploratory
analysis & knowledge
acquisition
Interactive exploratory
analysis & knowledge
acquisition
Database
storage
Database
storage
Fig. 2.26.
Proposed system architecture for semiconductor manufacturing process
analysis.
phisticated approach of employing and combining the regarded methods will
be pursued next. A rough sketch of the envisioned information-processing
architecture is given in Fig. 2.26. Similar to other applications, e.g., event
classification in high-energy physics [2.39], a real-time classification stage is
included in the proposed architecture. This module shall assess locally and
in realtime whether interesting and relevant, i.e., novel, data occurred that
should be stored for ensuing interactive analysis by human experts. OCC
and the NOVCLASS model are first-choice candidates for this module. After
storing in the database, dimensionality-reduction methods and interactive
visualization will be undertaken for the analysis of the novel or abnormal
data. Resulting understanding and knowledge extraction provide the base-
line for potential actions as, e.g., classifier stage refinement or process control
and optimization activities. Especially the interactive data visualization mod-
ule can be significantly improved to the benefit of the regarded application.
This has already been demonstrated for a different application domain in
psychoacoustics, where an enhanced tool, denoted Acoustic Navigator (AN),
was devised [2.25]. AN has been equipped with improved display features,
such as multiple- and single-radar plots and practical search functions, which
effortlessly direct the analyst to data entries of interest in the map visualiza-
tion. These and numerous other convenience functions will allow transparent,
fast, consistent, and thus, productive work on large, high-dimensional, and
abstract databases. Figure 2.27 shows a first adaptation of the AN to the re-
garded application. Radar plots and the search function are illustrated. The
focus of the follow-up research shall be put on this crucial system compo-
Search WWH ::
Custom Search