Advanced Methods for the Analysis of Semiconductor Manufacturing Process Data - Advanced Techniques in Knowledge Discovery and Data Mining - page 34

Fig. 2.4. A 300-mm wafer at so-called floodlight inspection to check for correct
printing of the mask).
data collected about each lot is therefore scattered about various databases.
Lately, single-wafer processing has become more important. Often the exact
sequence of wafers through a single-wafer process or the position of wafers (re-
spectively lots in batch tools) will be needed to pinpoint problems found with
specific wafers. For so-called single-wafer tracking, this information needs to
be fully recorded, which is only possible if all tools can read the wafer ID
and lot information automatically and are connected to a database system.
Additionally, a vast amount of (often temporary) data is produced and eval-
uated for inline process monitoring and closed-loop process control. It can
be estimated that a typical semiconductor manufacturing line produces such
data in excess of 1 TByte per day. It is therefore essential to evaluate, prune,
and compact much of this data directly at the source. Routine reports are
extracted for common purposes like maintenance, documentation, process
control and optimization, and quality management. Process data that are
actually stored, whether on the process tool itself or in a database, are usu-
ally kept only for a limited time or in a rolling log file to limit the storage
requirements. This is far from an optimal solution as most of the data will be
completely normal and therefore uninteresting, while crucial data needed to