Advanced Methods for the Analysis of Semiconductor Manufacturing Process Data - Advanced Techniques in Knowledge Discovery and Data Mining - page 36

sification, dimensionality reduction, and interactive data visualization tech-
niques can be employed.
2.2.4 Electrical Test Data
After fabrication, electrical tests (ET) on the wafer level are carried out to
assess that all single devices defined by the process are within their specified
range. The devices tested are separate from the actual ICs on the wafer, often
placed into the space between individual chips that is needed to singulate
them later. These test structures are laid out carefully to isolate the layers
needed to process them as much as possible from other layers. These tests are
also called parametric tests as the results are actual measurement values for
device parameters, like the threshold voltage or saturation currenrt of some
specific transistor.
Later the actual ICs on the wafer are subjected to functional and para-
metric tests (FT and PT) on the wafer to decide which devices should be
packaged after singulation. These tests are usually performed on a multitude
of devices to save time. A sequence of tests is performed on each chip, and
the first test that fails is recorded. The failed chips continue to be tested, but
as the fail may have put it into an undefined state, the results of these tests
cannot be relied on.
Both electrical and functional test data are stored in databases (s. Fig. 2.6)
and is often preprocessed to facilitate analysis. Such preprocessing routinely
includes the removal of spurious faults, calculation of derived values for pa-
rameteric data, and binning for functional data. Binning collects several in-
dividual tests that are associated with the same failure mechanism into a
so-called fail bin.
Often the IC will again be tested after being fully packaged. When relia-
bility is of utmost concern, a burn-in procedure may be performed to weed
out early fails, necessitating further tests.
2.2.5 Data Analysis
Standard data analysis concentrates on keeping the process within specifica-
tion limits, thus ensuring the quality of the final product. Typically a normal
distribution of the measured parameter is assumed and parameters of the
distribution like median and sigma are reported. In conjunction with the
process specification limits, the so-called process capability cp and process
centering cpk can be calculated. These methods and their application are
widely accepted and mandated by various quality management methods and
standards like ISO 9000.
However, their application to process specification, process trouble shoot-
ing, and process optimization often does not yield the desired results. Due
to their univariate nature, complex interactions between parameters are not