Advanced Methods for the Analysis of Semiconductor Manufacturing Process Data - Advanced Techniques in Knowledge Discovery and Data Mining - page 30

also been employed in the smart fabrication project from 1995 to 2000 by
a consortium including TEMIC, Siemens, and the University of Tubingen
(Rosenstiel et al.). The detection of characteristic failure patterns [2.36] and
yield prediction [2.37] were some of the pursued goals in this project.
In these and similar efforts, Kohonen's SOM has been employed with
static visualization techniques. The advanced methods investigated here,
however, differ in many ways and especially target on bringing improvements
with regard to mapping speed, mapping error reduction, user convenience,
and interactivity in the analysis process. The respective methods briefly
browsed in Section 2.3.2 can serve to project data in a lower-dimensional
space to make it amenable for interactive human perception-based analysis
as well as automatic variable or variable group selection and pattern clus-
tering. The objective of the current phase of the work and this chapter is
to demonstrate the viability of the addressed methods for real process data
extracted from a modern CMOS process. As a feasibility study, data with
known but nonobvious information content prove that the methods can in-
deed help in rapidly detecting the desired information. In the second phase
of the feasibility study, novel information and knowledge shall be extracted
from additional process data by applying the proposed methods, e.g., in-
teractive multivariate data visualization. In this regard, the chapter is as
organized as follows. In the next section, the general data acquisition process
and the chosen instance data for the conducted experiments are described. In
the following section, the spectrum of applied methods and their tool imple-
mentations are covered. Then the conducted experiments and the achieved
results are presented and discussed. Before concluding, the envisioned per-
spective of the work and the related information processing architecture for
manufacturing process monitoring and optimization are introduced.
2.2 Semiconductor Manufacturing and Data Acquisition
2.2.1 Brief History of the IC
Semiconductor devices had a slow start as a curiosity that was not well under-
stood. Still, they had important niche applications in radio communications,
when vacuum tubes could not be used. As the understanding of their princi-
ples of operation grew, refinements to the manufacturing process first enabled
military applications and then delivered the first commercially available de-
vices in the form of single-pn-junction diodes and transistors in the early
1950s. The year 1958 marked the birth of the monolithic integrated circuit,
now commonly just called IC. The invention of the IC is attributed to TI en-
gineer Jack Kilby, but without the planar manufacturing process developed
in the same year by Jean Hoerni and advanced by Robert Noyce and Gordon
Moore at Fairchild, 3 it would likely have taken quite a bit longer for the idea
3 R. Noyce and G. Moore left Fairchild to cofound Intel in 1968.