Advanced Methods for the Analysis of Semiconductor Manufacturing Process Data - Advanced Techniques in Knowledge Discovery and Data Mining - page 28

structure of the data. This procedure of knowledge discovery and the online
decision making serve to control the respective complex processes, e.g., for
quality assurance purposes, keeping the process in a multivariate window of
allowed parameter tolerances.
One important instance of this general problem class with a significant
commercial impact and stringent information processing demands is repre-
sented by the analysis, control, and optimization of manufacturing processes
in the semiconductor industry. Typical aims are the centering of the process
in a so-called process window and the assurance of an optimum yield based
on functional and electrical tests. For instance, in [2.53] a good general intro-
duction to the topic can be found. In this particular work, decision trees are
applied to determine significant individual variables or groups of variables. A
more focused example is given in [2.3], where data mining and various classi-
fication techniques are applied to a single processing step dealing with wafer
cleaning. Leading-edge technology and the corresponding manufacturing lines
have reached an unprecedented complexity in terms of both required machin-
ery and the required process monitoring, control, and optimization demands.
Thus, modern semiconductor manufacturing processes feature an increasing
number of processing steps with an increasing complexity of the steps them-
selves from initial wafer preparation to final passivation. Due to the continued
validity of Moore's exponential growth law (see, e.g., the SIA ITRS roadmap
[2.2]) the complexity of the processes will continue to increase at a rapid pace.
In Section 2.2.2, a brief introduction to this part of the presented work will
be given. Consequently, a tremendous amount of monitoring data are gener-
ated by the manufacturing line. The generated data have to be analyzed with
regard to the required process specification or qualification, i.e., whether the
process remains in the process window (see Fig 2.1). In simple models, the
process window can be described, e.g., by a multiparameter or multivariate
bounding box with thresholds in each parametric dimension. Exceeding the
threshold makes overt that the process is going out of specification for one
or several of the involved parameters. This approach neglects multivariate
dependencies and higher-order correlations of variable groups. Figure 2.2 de-
picts typical problems occurring, such as the process being off-centered or
showing correlated parameters or multimodality. The same holds for the typ-

Product specification limit Product specification limit

Process specification limit

Process specification limit

Process deviation Process deviation

6 sigma

Idealized process window Real process window

Fig. 2.1. Illustration of a process window.