Infineon Technologies Dresden
Advanced Data Processing
Combining full IEP spectrum with an interferometric signal for
depth etch endpoint detection
Frank Hoffmann, Infineon Technologies Dresden GmbH & Co. OHG
Knut Voigtländer, Advanced Data Processing GmbH
Motivation
In absence of a stop layer, etch process and
equipment control is typically done by depth
monitoring using an interferometric endpoint
system.
The state-of-the-art method of a single
UV/VIS wavelength approach has to balance
between appropriate signal-to-noise ratio and
interference frequency.
Method Description
•
State-of-the-Art
Approach
Fig. 1: Process Scheme, shallow recess etch with no stop layer.
spectral Pattern
This approach uses full spectra
time signals to form the
interferometric reference trace
by combining the individual
wavelengths.
•
Model quality depending on target signal
period and phase:
- blue – very good modeling possible with
large signal period.
- green – sufficient modeling with fast
oscillating target
- red –too fast oscillating target – no good
modeling possible
associated time components
250
300
5
10 15 20 25
250
300
5
10 15 20 25
250
300
5
10 15 20 25
250
300
5
10 15 20 25
250
300
wavelength / nm
5
10 15 20 25
time / s
Approximation Loss
0
Signal phase
•
Comp.5 Comp.4 Comp.3 Comp.2 Comp.1
•
Fig. 2: Set of reference measurements
containing the full spectra; the time behavior
even of the shortest available wavelength has
long signal period and is not suited for exact
endpoint control.
Results
A PCA (principle component
analysis) based method is used to
decompose the full spectra into its
linear independent components. A
nonlinear optimization is used for a
suited superposition of these
components providing the final
combined interferometric time
signal.
2
1
1.5
2
1
Fig. 3: With PCA decomposition the independent
signals can be discovered. A Modeling procedure is
used to try to superimpose these signals to fit a
given fast oscillating target signal.
•
2.5
3
0
10
5s
1
20
30
Signal Periode /s
1
10s
1
-1
-1
5
10 15
time /s
20
25
modeling not
possible for 5s
target period
20s
0
0
0
0.5
40
-1
5
10 15
time /s
20
25
for 10s target
modeling is just
possible
5
10 15
time /s
20
25
very good
modeling for 20s
target period
Fig. 4
•
The full OES and IEP spectrum is collected by a standard EyeD endpoint system from Applied Materials, Inc.
(spectral range 200 … 800 nm, time resolution 0.1 sec).
• The incoming IEP spectra • The resulting interis multiplied by the
ferometric refeweighting vector during
rence signal has
etch.
shorter periods
• The final interferometric
(more frequent
time-based signal is
Min/Max features),
generated by summing up
less noise and
all weighted spectral
more clear interFig. 6: Interferometric reference signal obtained by single
intensities within a
ference
Fig. 5: Weighting vector for online
wavelengths and by the PCA method. Final IEP trace
calculation of the
relevant spectral range.
with pattern (t1 … t4) detection.
information.
red/green:
Interferometric reference signal
obtained by a typical single
wavelength (280/229 nm, smoothing
over 5 data points)
blue: Interferometric reference signal
provided by the pca method (smoothing
over 5 data points)
interferometric reference signal.
•
•
The final interferometric time
signal was used for in situ
depth calculation using
standard methods for
endpoint algorithms.
Using this method for
endpoint detection a more
reliable depth control for
these shallow recesses is
achieved in high volume
production.
Fig. 7: Depth distribution
(productive data) with
single wavelength (left)
and pca algorithm (right).
8th European Advanced Equipment Control / Advanced Process Control (AEC/APC) Conference Dresden - Germany, April 18 - 20, 2007