Fiber Optic Fourier Transform Infrared Spectroscopic
Techniques for Advanced On-Line Chemical Analysis in
Semiconductor Fabrication Tools
Michael Kester*, Marc Trygstad*, and Paul Chabot1
* ABB, Analytical and Advanced Solutions, 433 Northpark Central Drive, Suite 100, Houston, TX 77073, USA
, Analytical and Advanced Solutions, 585 Char est Blvd. East, Suite 300, Quebec, Quebec G1K9H4, Canada
Abstract. A unique analytical methodology has recently been developed to perform real-time, on-line chemical analysis
of bath solutions in semiconductor fabrication tools. A novel, patented fiber optic sensor is used to transmit infrared
light directly through the tube walls of the circulating bath solutions within the fabrication tool in a completely noninvasive, non-extractive way. The sensor simply "clips" onto the tubing, thus permitting immediate analysis of the bath
composition by Fourier Transform infrared (FTIR) spectroscopy. The infrared spectrometer is capable of multiplexing
up to eight "Clippir™" sensor heads to a single interferometer using fiber optic cables. The instrument can analyze
almost any bath solution utilized today.
The analysis is performed using the near-infrared (NIR) portion of the electromagnetic spectrum, where absorption
bands related to molecular vibrations can be found. The Fourier Transform infrared spectrometer gives access to
absorption bands over a wide range of frequencies (or wavelengths), and the absorptions are correlated to concentrations
using a chemometric approach employing a partial least-squares algorithm. Models are generated from this approach for
each chemistry to be analyzed.
This paper will review the analytical technology necessary to make such measurements, and discuss the instrument
performance criteria required to achieve accurate and precise measurements of bath chemistries. The ability to measure
non-infrared absorbing compounds will be discussed, as will the nature of the influence of sample temperature on
measurement. Issues critical to the development of robust models and their direct implementation on multiple channels
and even different instruments will be considered.
INTRODUCTION
Semiconductor fabrication tools use several
different aqueous and organic solutions to effect
change on silicon wafers during the chip
manufacturing process. The concentration of these
chemicals over the lifetime of a bath is important to
determine and control for the production of consistent,
high quality products.
Current methods of analysis can be categorized into
two broad categories: in-situ and extractive. In-situ
measurements, such as conductivity, offer rapid realtime measurement of the chemical bath, but suffer
from a lack of selectivity between ionic or organic
species within the bath, and often require daily
recalibration due to instrument drift over time. There
are also contamination issues associated with inserting
a measurement device into the chemical bath. More
selective extractive methods such as infrared
spectroscopy, can be used, but suffer from lag-time
and bath contamination issues associated with the
extraction process.
A new technique for measuring chemical bath
concentrations, which combines the advantages of insitu, real-time, non-invasive measurement with the
necessary level of component selectively, has recently
been developed.
The measurement concept is
relatively straightforward: near infrared light is
delivered to the bath solution from a centralized
Fourier
Transform
Near
Infrared
(FT-NIR)
spectrometer using spectroscopy-grade fiber-optic
cable; with the use of a Clippir™ sensor head, the
CP683, Characterization and Metrology for VLSI Technology: 2003 International Conference,
edited by D. G. Seiler, A. C. Diebold, T. J. Shaffner, R. McDonald, S. Zollner, R. P. Khosla, and E. M. Secula
© 2003 American Institute of Physics 0-7354-0152-7/03/$20.00
764
infrared light passes directly through the circulation
tubing wall, through the circulating bath solution, and
exits the solution through the tubing wall; the
transmitted light is collected and returned to the
spectrometer by means of a second fiber-optic cable.
The individual chemical components in the bath are
then selectively determined by applying a Partial Least
Squares (PLS) chemometric algorithm to the
transmitted infrared spectrum.
APPARATUS
The ABB model FTPA2000-260 spectrometer is
used in this application. It has eight SMA-905 fiber
optic connectors that simultaneously launch the
modulated infrared light into a maximum of eight
separate 300 jim diameter, ultra-low OH,
spectroscopic-grade fiber optic cables. The infrared
light travels through the fiber optic cable, through the
Clippir™ sensor head and back to the spectrometer
through a second fiber optic cable as shown in
Figure 1. The transmitted light is received at the
spectrometer by eight separate germanium detectors,
which operate efficiently at room temperature. The
gains on the operation amplifiers can be adjusted
electronically to account for transmission losses
through the fiber, sample and Clippir™ sensor head.
The advantages of this technique over traditional
methods are numerous:
(1) installation of the
Clippir™ sensor head is rapid with little or no process
interruption - it simply clips onto the existing
recirculating bath tubing, (2) the installed sensor head
does not come into contact with the solution at any
point, thereby eliminating any possibility of bath
contamination, (3) the spectrometer can electronically
multiplex up to eight sensor heads simultaneously,
(4) the rack-mountable spectrometer can be located up
to 100 meters away from the wet stations, (5) the
hardware is identical for all measured chemistries - a
chemistry-specific software calibration model is all
that is required to measure the components of almost
every bath solution in use today, (6) sample
preparation or sample conditioning is not required,
(7) no additional reagents or solvents are necessary,
and (8) the measurements are precise, accurate,
obtained quickly, drift-free, and operator-independent.
The collected spectral data is transferred from the
spectrometer to a computer via an Ethernet
connection. The ABB FTSW2000 process software is
used for continuous collection of spectra, prediction of
bath concentrations using the PLS models, and
exporting of digital and analog results to a local DCS
or LAN server via industry-standard protocols.
FIGURE 1. Diagram of FT-NIR spectrometer and Clippir™ hardware connected by fiber optic cables, with the Clippir™
sensor head mounted on a bath recirculation line.
765
usually temperature dependent. Typically, multiple
spectra of the same sample are recorded over a
temperature range of ± 5 C. When these spectra are
included into the PLS calibration model, the sensitivity
to temperature is eliminated.
METHOD DEVELOPMENT
The PLS calibration model is typically generated in
a laboratory setting by carefully preparing mixtures of
bath solutions gravimetrically, such that the
concentration of each component in the bath varies
independently over the typical concentration range of
operation. For one-component chemistries (e.g., HF in
water), a univariate experimental design is set up,
whereas for two- or three-component chemistries (e.g.,
HF and HC1 in water), a full factorial experimental
design is devised. Anywhere from 20 to 40 individual
solutions are prepared for calibration development.
An example experimental design for development of a
BOB calibration is shown in Figure 2.
Usually, 128 individual near infrared spectral scans
are co-added into one spectrum at a resolution of
32 cm"1, over a period of approximately one minute.
Therefore, the refresh frequency of the measurements
made on each Clippir™ is approximately one minute
multiplied by the number of Clippirs installed on a
given spectrometer.
A background reference spectrum is periodically
collected on the circulation tube when it is empty. The
sample absorbance spectrum is then calculated by
taking the negative log of the ratio of the sample
spectrum divided by the reference spectrum.
Therefore, the sample absorbance spectrum is
independent of most of the instrument function, and is
only related to infrared absorbance due to the sample.
O
I 24
Once all of the calibration absorbance spectra are
collected (at various temperatures if necessary), they
are loaded into standard, commercial PLS modeling
software (e.g., PLSplus/IQ by Thermo Galactic), along
with their gravimetrically determined concentrations.
The spectra are pre-processed by the software before
calculating the PLS calibration model and its statistical
parameters. The statistical parameters of the resulting
calibration model are then used to determine the
expected standard error of covariance (SECV) for the
model, and to eliminate potential outliers.
20
2
3
HF (wt %)
» »
'
»
»
9
®&
»
*
$
2> 24
»
«» *
2
»>
g
»
•»
»
»
®
<
^|
I 2 2 ,!
*
»
»
ff
20 ——————————————————————————————
0
1
2
3
4
5
§
»
RESULTS AND DISCUSSION
HCI (wt %)
Calibration Performance
O
As an example of calibration performance, the HFHC1 bath chemistry will be considered. As shown in
Figure 2, this calibration was constructed where HF
ranged from 0 to 5 % (by weight), the HCI ranged
from 0 to 5 % (by weight), and the temperature ranged
from 21.0-25.5 C.
2
0
1
2
3
4
5
HF(wt%)
FIGURE 2. Calibration solution concentrations (in weight
%) for an HF-HC1 bath.
Temperature variations of the sample must be taken
into account since near infrared absorbance bands are
766
The results show that excellent repeatability
repeal
(standard deviation of measurement of the same
sample repeatedly) and low uncertainty (standard
deviation of the agreement with the gravimetric value)
can be obtained. The uncertainty associated with
monitoring HF is 0.06 wt %, the uncertainty associated
with monitoring HCI is 0.10 wt %, and the uncertainty
associated with monitoring H2O is 0.10 wt %. A
repeatability study performed by measuring the same
sample 20 times indicates that the standard deviation
associated with monitoring HF is 0.02 wt %, and HC1
is 0.04 wt %.
Calibration Model Transfer
Due to a slight optical shift inherent in the nonuniformity of the Clippir™ sensor heads and/or the
sample tube wall, it is sometimes necessary to make
minor adjustments to the initial factory calibration.
There are two ways to do this: (1) new spectral data,
acquired after the system is installed, is added to the
original calibration model, resulting in the Clippirspecific optical shift being directly incorporated into
the calibration model, or (2) a minor slope and bias
correction is determined and applied to the predicted
results from the original calibration model. Once the
new spectral data are incorporated into the PLS model,
or the slope and bias are determined for a given
installation of a particular Clippir™ sensor head,
further adjustment of the calibration model is usually
not necessary.
FT-NIR measurements plotted against gravimetric
measurements for the calibration sample set are shown
in Figure 3. R2 values are all greater than or equal to
0.995 and the standard error of covariance (SECV,
equivalent to the calibration uncertainty) for each
component was less than 0.10 %.
R 2 = 0.998
SECV = 0.06 %
Process Control Trendlines
0
1
2
3
4
5
Even though this device measures non-invasively
through the sample tube wall, the repeatability of the
measurements has been shown to be more than
adequate for real-time, process control of the bath
concentrations.
Figure 4 shows an SCI bath
concentration over a period of approximately 4 hours.
While the H2O2 concentration remains relatively
constant, the NH4OH concentration is continuously
depleted, with the control system periodically injecting
more NH4OH to try and maintain a constant
concentration.
HF (Gravimetric wt%)
R 2 = 0.995
SECV = 0.10%
3
1
2
3
4
o.u
HCI (Gravimetric wt %)
100
z
t
2.5 ^
8
2.0 |
6
R 2 = 0.998
SECV = 0.10%
OCM
96 H
2
94
0
0: 00
92
1.5 i
4 ***/^»*«/\tf»*&^
1.0 £
0.5
Z
0.0
1:12
2:24
3:36
Time (hours)
90
90
92
94
96
98
100
FIGURE 4. H2O2 (upper trace) and NH4OH (lower trace)
concentrations of an SCI bath over approximately four
hours.
The NH4OH concentration is maintained by
periodically injecting more NH4OH as the concentration is
depleted, creating a saw-tooth trendline.
H2O (Gravimetric wt %)
FIGURE 3. Calibration solution concentrations (in weight
%) as determined by weight (gravimetric) and by FT-NIR
for an HF-HC1 bath.
767
It should also be noted that having an infraredactive bond in each measured component is not a
prerequisite for measurement. Ions that dissolve in
water often have an influence on the hydration sphere
of water, thereby creating an indirect means of
measuring the non-infrared active component. This
shift in energy of the strong water absorbance bands
can then be used to quantify non-absorbing ionic
species.
Chemistries and Analyzer Performance
Numerous bath components have been studied to
date. Table 1 provides a listing of the chemistry,
concentration range, repeatability, and temperature
range of each chemistry that has been calibrated. It
should be noted that the repeatability values listed in
the table are the upper limit of what can be achieved normally, a much lower (better) repeatability is
observed.
TABLE 1. Bath types that have been successfully monitored using FT-NIR spectroscopy. Several proprietary bath
types have been successfully monitored, but are not shown here.
Concentration
Chemical
Repeatability **
Temperature (C)
Type of Bath
Component
Range (wt %)
(wt %)
0-5
HF
0.
15-25
0-5
0.
15-25
HC1
H20
91 - 100
0.
15-25
0-10
HF
20-25
HF
0.
0-10
20-25
Buffered Oxide
HF
0.
NH4F
22-40
0.
20-25
56-72
0.2
H2O
20-25
NH4OH
0-3
Standard Clean 1
0.
20-70
H202
0-6
0.
20-70
(SCI)
91-100
H2O
0.2
20-70
0-3
Standard Clean 2
35-45
HC1
0.
0-6
H2O2
(SC2)
0.
35-45
91 - 100
0.2
H20
35-45
H2S04
0-10
0.1
20-25
Dilute Sulphuric
0-10
H202
20-25
(DSP)
0.1
80-100
H20
0.1
20-25
80-95
0.4
H2S04
130
Sulphuric Peroxide
H202
0-0.5
130
(SPM)
0-20
0.4
130
H2O
0-10
20-25
KOH / H2O2
KOH
0.1
0-10
20-25
H202
0.1
7-22
0.3
20
Mixed Acids
HF
22-40
20
HN03
0.3
5-35
CH3COOH
0.3
20
0.4
25-50
20
H2O
32
CH3COOH
25-35
0.3
Vapox Etch
32
NH4F
7-18
0.2
15-25
32
0.3
Ethylene Glycol
17-32
20-25
TMAH
TMAH
0.5
68-83
20-25
0.5
H2O
0-3
0.2
60-70
TMAH / H2O2
TMAH
0-3
60-70
H2O2
0.5
94-100
60-70
0.5
H2O
0-5
Citric Acid
0.1
20-25
Citric Acid
20-25
95-100
0.1
H2O
0-10
H3PO4
0.1
20-40
Cold Phosphoric
75-90
H3PO4
0.5
140-170
Hot Phosphoric
0-8
Cyantek CR-7S
HC104
0.3
20-25
0-10
0.4
20-25
Ce(NH4)2(N03)6
82 - 100
20-25
0.3
H20
** Achievable repeatability is typically better than the values listed in this table. For example, in the case of HF or NH4OH, online
performance of < 0.04 wt % is routinely achieved.
HF-HC1
wall has been described. The benefits of this noninvasive, multiplexing technique provide significant
advantages over traditional measurement methods.
The analytical performance of this infrared-based
method is more that adequate for real-time process
control and monitoring of most bath chemistries in use
today.
CONCLUSIONS
A unique analytical methodology for noninvasively
measuring
semiconductor
bath
concentrations directly through bath circulation tubing
768