Recovery of tungsten from the exhaust of a tungsten chemical vapor deposition tool

Article abstract of DOI:10.1149/1.1394120

An ETC DryScrub plasma scrubber has been successfully tested for the capturing and recovery of metallic tungsten from the exhaust of a W chemical vapor deposition (CVD) tool. The scrubber operation was completely transparent to the upstream CVD process. The WF₆ destruction removal efficiency of the scrubber was determined with a quadrupole mass spectrometer. The system had been tested with two different plasma frequencies: 100 and 40 kHz. With the 100 kHz frequency, the destruction efficiency of WF₆ reached an initial value of 98% at a nominal dissipated power of 1200 W. However, the layer of W deposited on the scrubber electrodes contained hydrates of tungsten oxides. Moreover, the destruction efficiency of WF₆ dropped to less than 70% after eight consecutively processed wafers (`memory effect'). Introducing an intermediate H₂ plasma treatment ensured a continuously high efficiency, and improved significantly the purity of the deposited W layer in the scrubber. With the 40 kHz power supply, the maximum efficiency reached is more than 99% from a nominal dissipated power of 1100 W on. The purity of the deposited W layer is high (>99%). No memory effect was observed. Successful marathon runs have been performed with each tested frequency.

Full text of DOI:10.1149/1.1394120

Journal of The Electrochemical Society, 147 (12) 4665-4670 (2000)
4665
S0013-4651(99)12-050-0 CCC: $7.00 © The Electrochemical Society, Inc.
Recovery of Tungsten from the Exhaust of a Tungsten Chemical
Vapor Deposition Tool
Nausikaä B. H. Van Hoornick, Jan A. B. Van Hoeymissen, and Marc M. Heyns
IMEC, SPT/ESH Group, B-3001 Heverlee, Belgium
An ETC DryScrub plasma scrubber has been successfully tested for the capturing and recovery of metallic tungsten from the exhaust
of a W chemical vapor deposition (CVD) tool. The scrubber operation was completely transparent to the upstream CVD process.
The WF destruction removal efficiency of the scrubber was determined with a quadrupole mass spectrometer. The system had been
6
tested with two different plasma frequencies: 100 and 40 kHz. With the 100 kHz frequency, the destruction efficiency of WF6
reached an initial value of 98% at a nominal dissipated power of 1200 W. However, the layer of W deposited on the scrubber elec-
trodes contained hydrates of tungsten oxides. Moreover, the destruction efficiency of WF dropped to less than 70% after eight con-
6
secutively processed wafers (“memory effect”). Introducing an intermediate H plasma treatment ensured a continuously high effi-
2
ciency, and improved significantly the purity of the deposited W layer in the scrubber. With the 40 kHz power supply, the maximum
efficiency reached is more than 99% from a nominal dissipated power of 1100 W on. The purity of the deposited W layer is high
(>99%). No memory effect was observed. Successful marathon runs have been performed with each tested frequency.
©
2000 The Electrochemical Society. S0013-4651(99)12-050-0. All rights reserved.
Manuscript submitted December 13, 1999; revised manuscript received July 18, 2000.
Chemical vapor deposition of tungsten (W-CVD) is a commonly
used processing technique in the semiconductor industry. The effi-
ciency of this deposition process however is low: as much as 75% of
the scrubber resulting in N , H , and ammonia as gaseous reaction
2 2
products; the nonconductive materials ammonium chloride and sili-
con nitride are deposited on the scrubber electrodes. The Si deposi-
2
the source gas tungsten hexafluoride (WF ) remains unused in the
tion treatment was performed with the ETC DryScrub system be-
6
tool exhaust. To date, only destructive methods can be used to treat
the waste gases, e.g., burners and wet scrubbers. The disadvantage of
these systems is that the waste gas stream is transferred to a waste-
water stream. It is known that W-contaminated water is difficult to
treat, due to the presence of multiple chemical states of W-contain-
ing species. No material is recovered or reused.
Environmental benign integrated circuit (IC) processing is be-
coming increasingly important. In addition, it is expected that the
legislation on discharge limits will become more stringent in the
future. The purpose of the work presented here is to assess the pos-
sibility of capturing and recovering (metallic) tungsten from the
exhaust gas of a W-CVD tool in a dry way, using an ETC DryScrub
plasma scrubber. By using this technique, a relatively large amount
of the expensive W can be reused or sold which should make this
treatment system economically feasible.
tween the roots blower pump and the rotation vacuum pump.
Experimental setup.—The complete experimental setup is dis-
played in Fig. 2. The ETC DryScrub system is installed in bypass
mode in the exhaust line of the W-CVD tool (type Applied Materials
P5000). It has been set up in prepump position because at this loca-
tion, the pressure is always lower than 10 Torr, which is the maximal
allowable operating pressure for the ETC DryScrub system.
The upstream W-CVD deposition process cycle contains four
major steps: (i) nucleation, (ii) via fill, (iii) purge, and (iv) chamber
clean. During the nucleation step, H , SiH , and WF are used as
2
4
6
source gases (with Ar as carrier gas); only H and WF are used dur-
2
6
ing the via fill. During the clean step consecutively a nitrogen triflu-
oride (NF ) plasma and a H plasma are applied. The wafer size used
3
2
in this process was 150 mm. Typically a W layer of 1 ␮m was de-
posited on the wafer per process cycle.
During the via fill step of the W-CVD deposition, the total gas
flow was 5600 standard cubic centimeters per minute (sccm) (Ar,
Experimental
Plasma scrubber.—The ETC DryScrub system is in essence a
plasma enhanced (PE) CVD tool. Inside this scrubber, PE reactions
can take place. Figure 1 displays a cross section of the scrubber. It
contains two concentric spiral electrodes, labeled 3 and 4 [one
grounded (anode) and one connected to the radio frequency (rf)
feed-through (cathode)]. Three ceramic isolators (7) are used to sep-
arate the electrodes from each other. Both electrodes are placed in a
grounded cylindrical Al inner housing (5), which is in turn sur-
rounded by a vacuum-sealed outer housing (8). In this way, a plas-
ma can be created between the electrodes. The rf power supply can
have different output frequencies. Two frequencies have been tested
in these experiments: 100 kHz Ϯ 100 Hz and 40 kHz Ϯ 40 Hz.
Some scrubber operation parameters can be varied: power applied to
the plasma, orientation of the electrodes, pressure in the scrubber,
concentrations of gases coming into the scrubber, and electrode ma-
terial. The electrodes are typically made of Al but are also available
in Fe and stainless steel. The volume of the scrubber is about 40 L.
The mean flow length of the gas through the system is about 1.4 m.
The spacing between the spiral electrodes is 1.3 cm. The total elec-
2
trode surface is 42,000 cm .
The scrubber already has been used successfully to treat exhaust
1
2
gases of hot wall nitride deposition and Si deposition processes
with conversion efficiencies of more than 95%. In the hot wall
Figure 1. Cross section of the ETC DryScrub® system; 1: rf connector, 2: rf
feed-through, 3: anode, 4: cathode, 5: inner housing, 6: spiral sets, 7: isola-
tor, 8: outer housing.
1
nitride deposition process treatment, dichlorosilane is destroyed in
Downloaded on 2015-06-07 to IP 129.100.58.76 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

4
666
Journal of The Electrochemical Society, 147 (12) 4665-4670 (2000)
S0013-4651(99)12-050-0 CCC: $7.00 © The Electrochemical Society, Inc.
Results
Influence of the ETC DryScrub system on the W-CVD pro-
cess.—A paramount requirement for the scrubber system is that it
must be completely transparent to the upstream W-CVD process.
Therefore, initial experiments have been performed to monitor the
quality of the deposited W layer on the wafers during scrubber
operation.
The pressure in the exhaust has been measured (at position 1,
Fig. 2). As could be expected (see previous section), a 25% increase
to 1.2 Torr was observed when the gases were led through the scrub-
ber. About the same pressure was observed when the scrubber was
active. This pressure increase however had no influence on the
W-CVD process.
The weight and the resistivity of the deposited W layer on the
wafers in the W-CVD tool during scrubber operation have been
determined and were very well within the specifications needed. The
standard deviation on the resistivity was very low: 0.64% of a resis-
tivity value of 0.0962 ⍀/ᮀ, indicating a very good uniformity of the
deposited W layer on the wafer. The values were all comparable with
the ones obtained before the installation of the scrubber. Particle
addition was also investigated. No additional particles larger then
0
.3 ␮m were generated using the ETC DryScrub system.
It can safely be concluded that the presence and operation of the
Figure 2. Schematic of the complete experimental setup.
ETC DryScrub system had no influence on the upstream deposition
process in the W-CVD tool. Consequently, no process requalifica-
tion is required upon installing the scrubber, given the currently used
vacuum pump configuration.
N , H , and WF ). The relative concentration of WF in this gas mix-
2
2
6
6
ture is only 1.34%. The mean residence time of the gases in the
exhaust line is about 6 s; in the scrubber itself it is less than 1 s. Ide-
ally, the recovered metallic tungsten should be as pure as possible.
Therefore, only during the via fill step of the W-CVD process the
plasma scrubber is activated. In this way, the codeposition of Si on
the electrodes of the plasma scrubber is minimized.
Gas stream analysis.—A mass spectrometric analysis of the
gases present in the exhaust was performed with the Balzers QMS.
All measurements were done using an ionization energy of 40 eV to
avoid the creation of too many double ionized species in the analyz-
er. Secondary electron multiplier detection was used for sensitivity
reasons.
A set of electropneumatic valves and a valve controller system
make it possible to set the scrubber in bypass during the NF cham-
3
ber clean of the W-CVD chamber, preventing the removal of the de-
posited material on the scrubber electrodes. Another possibility is to
switch the controller to the scrubber purge mode in order to be able
On the mass spectra obtained during processing, one could easi-
ly assign some peaks to the products of the reactions occurring in the
W-CVD tool, as well as to the various products of the plasma
process occurring in the scrubber. The most prominent peaks are
those coming from the source gases WF , H , and SiH . In addition,
to pump/purge the scrubber with pure N . It is also possible to add
2
an additional gas (like H ) to the scrubber inlet during the scrubber
2
6
2
4
active operation (at position 1, Fig. 2).
HF/F, WO , and WOF could be identified as the most important by-
3 4
The real-time, continuous gas analysis was carried out with a
quadrupole mass spectrometer (QMS). A Balzers QMG-421,
equipped with a differential pumping system, was used to sample at
different pressure regions. This mass spectrometer allows quantita-
tive as well as semiqualitative measurements. The sampling tubing
is made of Teflon-like material because it has a higher resistance
products. Evidence for the presence of minor amounts of air and
moisture in the dry scrubber itself were found as well.
Since W has isotopes at masses 182 [relative abundance (R.A.),
26.3%], 183 (R.A., 4.3%), 184 (R.A., 30.7%), and 186 (R.A.,
28.6%), a complex mass spectrum is obtained when measuring the
effluent stream. Therefore, a lot of attention went to the correct iden-
tification and assignment of the individual mass peaks. Since the
parent ion WF ϩ6 does not exist, WF ϩ5 was found to be the most im-
toward aggressive gases. It also reduces WF -metal catalyzed reac-
6
tions. The sampling lines have a length of approximately 50 cm
each, and an outer diam of 0.3 cm. Sampling can be done at the inlet
and outlet of the plasma scrubber (at position 1 and 2, Fig. 2). The
signals in the mass spectrometer are detected with a delay of about
portant fragment of WF after electron impact dissociation and ion-
6
ization at 40 eV. This was confirmed by the National Institute of
3
Standards and Technology reference mass spectrum at 70 eV.
1
5 s compared to the moment the gases are introduced in the
Therefore, the highest isotope peak of the WF ϩ5 fragment (mass
upstream W-CVD chamber. The W-CVD tool is equipped with a dry
279), has been selected to monitor WF . Moreover, this mass does
6
3
ϩ
roots blower pump with a capacity of 250 m /h and a wet rotation
not interfere with a peak of WOF , the largest fragment of WOF .
3
4
3
pump with Fomblin oil and a capacity of 65 m /h.
When other fragments of WF were monitored, mass peaks were
6
The pressure was monitored during the via fill deposition step on
different locations throughout the setup. Pressure values were 80 Torr
in the W-deposition tool, 1.5 Torr at the sample point before the
always chosen in such a way that no mass interference of other
species was present. Table I gives a concise overview of possible
species present with corresponding masses.
scrubber (both measured with Baratron pressure gauges), and about
A linear calibration curve of WF over four concentration dec-
6
Ϫ5
10
Torr in the mass spectrometer chamber (measured with a cold
ades was obtained. The detection limit of WF was about 5 ppm.
6
cathode pressure gauge). Over the scrubber, a pressure drop of
.5 Torr was measured. The conductance of the exhaust piping, both
Figure 3 displays a time resolved spectrum of the representative
peaks of H (mass, 2), HF (mass, 20), SiH (mass, 30), and WF
6
0
2
4
without and with the scrubber, was calculated for air at a mean pres-
sure of 2 Torr. The flow in the exhaust lines was in both cases in the
laminar regime. A conductance of 11.212 m /s was calculated for the
(mass, 279) during consecutive nucleation, via fill, and purge steps
of the W-CVD process. Sampling was done here at the scrubber out-
let (position 2, Fig. 2). The different masses were measured with dif-
ferent QMS sensitivities. The two measurement conditions are (A)
scrubber not active and (B) scrubber active.
3
original piping before installing the scrubber, and a conductance of
3
0
.025 m /s was obtained when the gases were going through the
scrubber. This higher flow resistance prompted only an adjustment of
Figure 4 gives an overview of the ion intensities obtained for the
most important peaks during the via fill step of the deposition for
the N ballast at pump level. No other changes had to be made.
2
Downloaded on 2015-06-07 to IP 129.100.58.76 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

Journal of The Electrochemical Society, 147 (12) 4665-4670 (2000)
4667
S0013-4651(99)12-050-0 CCC: $7.00 © The Electrochemical Society, Inc.
Table I. Mass peaks interpretation table.
Mass number
amu)
Species
(
2
81, 279
WFϩ5
WF ϩ5 and WOF
ϩ
2
2
78
77
4
ϩ
WF5
ϩ
WOF
4
2
76, 275,274
62, 260
ϩ
2
WF4
WOF ϩ3 and WF
ϩ
4
2
2
59
58
ϩ
WF4
ϩ
WOF
3
2
2
57, 256, 255
43, 241
ϩ
2
WF3
WF ϩ3 and WOF
ϩ
2
2
2
40
39
ϩ
WF3
Figure 4. Ion intensities of selected masses for two conditions, scrubber
active and scrubber not active, sampled at the scrubber outlet.
ϩ
WOF
2
38, 237, 236
24, 223
ϩ
2
WF2
Another determining parameter for the DRE is the total pressure
in the scrubber. Typically, the pressure in the exhaust line of the
CVD tool during processing is about 1 Torr, staying well below the
recommended operating pressure for the scrubber of 10 Torr.
Changing the pressure in the scrubber to lower values could have a
positive effect on the scrubber efficiency. However, this could not be
verified since the minimum pressure in the exhaust line is fixed by
the upstream W-CVD process and the limited capacity of the down-
stream vacuum pump. Increasing the pressure by adding additional
carrier gas to the exhaust stream resulted in lower efficiencies.
Tests have shown that the [WF ]/[H ] ratio in the exhaust also in-
both scrubber active and scrubber not active conditions (sampled at
position 2, Fig. 2).
Determination of the WF destruction efficiency.—An important
6
property regarding the scrubber performance is the destruction
removal efficiency (DRE) of WF . It is defined by the formula
6

iM279,scrubber active


WF6 DRE ϭ 1 Ϫ
ϫ 100
[1]

i

M279,scrubber not active 
6
2
The i in the equation denotes the area under the time resolved inten-
sity of mass 279 during the via fill step (see Fig. 3). The uncertainty
on the efficiency values is Ϯ1% absolute.
The DRE of the scrubber is sensitive to several parameters. Some
of them can be changed, others are more or less fixed in the setup
presently used. An important parameter is the frequency of the plas-
ma generator. The results reported below are grouped as function of
two frequencies which have been tested: 100 and 40 kHz.
fluences significantly the WF DRE of the scrubber. The smaller the
6
ratio is, the higher the efficiency will be. However, in normal condi-
tions, this ratio is determined by the upstream W-CVD process and
cannot be changed ([WF ]/[H ] Ϸ 0.15). Adding up to 4 standard
6
2
liters per minute (slpm) forming gas (10% H in N ) directly to the
2
2
scrubber inlet (at location 1 in Fig. 2) during scrubber operation in
order to decrease the mixing ratio, causes a decrease of the scrubber
efficiency. The positive effect of increasing the [H ] is suppressed
2
here by the negative effect of the pressure increase.
Results with the 100 kHz plasma frequency.—Dissociation effi-
ciency optimization.—The scrubber performance is very sensitive to
Initial experiments showed that the orientation of the electrodes in
the scrubber had also an effect on the WF DRE of the scrubber. The
the dissipated power to the plasma. The WF DRE of the scrubber is
6
6
destruction efficiency was about 10% lower when the gases were
flowing from the outside of the electrode to the inside, than when the
flow was from the inside to the outside. No clear explanation could
be found for this observation, but it is certainly related to the asym-
metric radial distribution of the plasma between the electrodes. In the
experiments described below, Fe electrodes with inside to outside ori-
entation and a nominal dissipated power of 1200 W have been used.
increasing with increasing power (see Fig. 5). Initial experiments
were done with an Al electrode. Since the maximum applicable
power using the Al electrode was limited to 800 W, further experi-
ments were done with an Fe electrode. The maximum power that
2
could be applied here was 1200 W (power density, 0.06 W/cm ). It
has to be noted that no information could be obtained about the
amount of power reflected from the electrodes. Consequently, all
power-values mentioned in this article are “nominal”.
Amount of W that can be recovered.—Given the total inlet flow of
WF process gas and the amount of W deposited on the wafer in the
6
Figure 3. Time resolved mass spectrum during nucleation, via fill, and purge
steps of the W-CVD process with the scrubber not active (A) and with the
scrubber active (B) for several representative peaks, sampled at the scrubber
outlet.
Figure 5. WF destruction efficiency as a function of dissipated power given
to the plasma (100 kHz power supply experiments).
6
Downloaded on 2015-06-07 to IP 129.100.58.76 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

4
668
Journal of The Electrochemical Society, 147 (12) 4665-4670 (2000)
S0013-4651(99)12-050-0 CCC: $7.00 © The Electrochemical Society, Inc.
upstream CVD process, the maximum amount of W that can be
deposited on the scrubber electrodes can be calculated. Theoretical-
ly, with a DRE of 99%, about 1.8 g pure W per W-CVD cycle can
be deposited on the electrodes. An amount of 1.9 g deposit has been
found on the electrodes, indicating that the layer is not pure W.
Quality of the deposited W layer.—The quality of the deposited W
layer on the scrubber electrodes, obtained after about 350 deposition
cycles with the above described optimized process parameters, is not
satisfactory. The layer has a deep blue color and thus is not made of
pure metallic W. The crystals obtained were identified by X-ray dif-
fraction (XRD) analysis as H WO иyH O, a form of tungsten blue,
x
3
2
one of the products of the oxidation of W with moisture (see the Dis-
cussion section). In addition, a significant concentration of F-con-
taining species was present in these crystals. Total reflection X-ray
fluorescence spectrometry (T-XRF) analysis of the layer could not
detect the presence of other metals. Microscopic inspection revealed
that the deposit is made of several individual layers on top of each
other. Furthermore, it can be seen that the coverage of the electrodes
is local and not homogeneous.
Figure 7. Optimization tests of the intermediate H plasma treatment as func-
2
tion of duration and H flow. Data always taken at the ninth wafer of a batch
2
(100 kHz power supply experiments).
Memory effect and intermediate H plasma treatment.—Using the
2
optimized conditions, an initial WF6 DRE of approx. 99% was
reached. However, it was found that the within batch efficiency
decreased with each consecutive wafer processed in the W-CVD
after the initial dip). The most optimal treatment is a trade off be-
tween the H gas consumption and the time necessary (or available)
2
tool. As can be seen on Fig. 6 (data set of no H plasma treatment),
2
to perform the treatment.
A marathon run has been performed using this intermediate H2
treatment. As can be seen on Fig. 8, after an initial dip, a continuous
after processing of eight wafers, the efficiency dropped to only 81%.
In order to cope with this memory effect and to improve the effi-
ciency of the scrubber, an H plasma treatment of the electrodes was
2
WF destruction efficiency of Ն98% is achieved. Visual inspection
6
introduced after each wafer processed. Moreover, this should result in
an improved quality of the deposited layer. The justification for this H2
plasma treatment introduction can be found in the Discussion section.
of the deposited W-containing layer on the scrubber electrodes
revealed that the quality obtained when using an intermediate H2
treatment is very different from the deposited layer obtained when
This intermediate H treatment consists of striking a plasma in
2
no H treatment was used. The layer has now a homogeneous metal-
2
the scrubber at the same nominal power used during the deposition
lic color and is homogeneous in structure as well. The coverage on
the electrode is complete and seems to be uniform over all the elec-
trode surface. Only the layer in the region near the rf feed-through is
bluish, indicating the presence of WO иyH O. T-XRF and XRD
step of the process, while flowing H gas and some additional carri-
2
er gas (Ar) through the scrubber. The H comes from the upstream
2
W-CVD chamber, where it is used during the cleaning of the depo-
sition chamber. During this step, the pressure in the scrubber is
x
2
analysis revealed that the layer on the electrodes consists of pure W,
of both crystalline and amorphous phases. No evidence was found
for the presence of other metals.
<
1 Torr. Initial experiments using this intermediate H treatment
2
demonstrated that it was indeed effective in regaining a WF destruc-
6
tion efficiency of more than 97%, after an initial small dip. Figure 6
displays the WF DRE of a run where an H plasma treatment is
Results with the 40 kHz power supply.—The same experimental
setup was used as with the 100 kHz experiments (see Fig. 2). The in-
stallation of this plasma power supply required only an additional
6
2
applied of 1 slpm for 120 s.
Investigations have been carried out to optimize the conditions of
this H treatment by defining the optimal H flow rate and flow time.
water cooling circuit. The WF DRE was also determined by meas-
2
2
6
Since the introduction of the H treatment has a negative effect on
uring time-resolved mass signals with the Balzers mass spectrometer.
As in the 100 kHz experiments, Fe electrodes with inside to out-
side orientation have been used at an operation pressure of 2 Torr.
2
the process throughput of the W-CVD tool, the time necessary for
the treatment should be as short as possible. Figure 7 gives an
overview of WF DRE results for several H treatment conditions
6
2
WF destruction efficiency.—The influence of the dissipated power
6
(with data obtained each time at the ninth wafer of a batch, thus well
on the WF DRE has been assessed. Again, the scrubber efficiency
6
Figure 6. Comparison of the WF destruction efficiency as function of wafer
6
Figure 8. WF destruction efficiency as function of wafer number during a
marathon run, performed with the 100 kHz power supply and with an inter-
number without and with intermediate H plasma treatment. Pressure during
6
2
H treatment was 0.7 Torr with a H flow of 1000 sccm (100 kHz power
2
2
mediate H treatment of 120 s.
supply experiments).
2
Downloaded on 2015-06-07 to IP 129.100.58.76 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

Journal of The Electrochemical Society, 147 (12) 4665-4670 (2000)
4669
S0013-4651(99)12-050-0 CCC: $7.00 © The Electrochemical Society, Inc.
is increasing with increasing dissipated power. The minimum nomi-
nal dissipated power that has to be applied to obtain an efficiency
Ն99% is about 1100 W. As opposed to the experiments with the
The chemistry becomes more complex when introducing mois-
ture and/or oxygen in the system, coming from air leaks and de-
gassing processes. There exist a variety of WOHF compounds in dif-
ferent ratios of oxidation states, each with its own physical charac-
1
00 kHz power supply, no memory effect has been observed.
A marathon run has been performed with a plasma power of
500 W. The results are displayed in Fig. 9. As can be expected, the
teristics. Among them the most important is WO , formed in Reac-
3
2
tion Eq. 5 as final oxidation product (taking place preferably at the
surface)
WF DRE is never lower than 99% during the complete run. No
6
overheating of the system has been observed, although the tempera-
ture of the electrodes was about 70ЊC, which is 20ЊC higher than in
the case of the 100 kHz power supply.
WF ϩ 3H O r WO3(s) ϩ 6HF(g)
[5]
6
2 (g)
A stoichiometric composition of tungsten trioxide is yellow. How-
ever, a very common form of tungsten oxides are the tungsten blues,
which are mixed valence compounds containing different ratios
Resulting deposited W-layer.—The quality of the deposited layer ob-
tained with the 40 kHz power supply is very good. It has a metallic
tungsten-like appearance, and seems to be very homogeneous in
structure and color. The coverage of the electrodes is complete and
uniform over all the electrode surface. No blue deposits have been
observed. No other metals besides W could be found with T-XRF
analysis. From this it was concluded that the purity of the layer is
very high.
W(VI)/W(V) (e.g., H0.1WO , and the corresponding hydrate
3
H0.1WO и1.06H O). It is reported that wet WO becomes blue on
3
2
3
exposure to ultraviolet light (e.g., from the plasma emission) as a
7
result of formation of tungsten blue.
8
WO can react with WF
3
6
WO3(s) ϩ 2WF6(g) r 3WOF4(s,g)
[6]
Discussion
WOF (tungsten oxyfluoride) in pure form is white, but in practice
4
In the upstream WCVD tool, the deposition chemistries are
SiH /WF and H /WF -based. Silane and H act as reducing agents
for the tungsten source. The main corresponding overall reaction
equations can be written as
it is usually contaminated with other tungsten oxides, which can give
it different colors (blue-yellow). It is known that those tungsten
oxides may cause reduced deposition of W, poor morphology, and
4
6
2
6
2
uneven growth. WOF has a vapor pressure of about 1 Torr at 21ЊC,
4
and can participate in the PECVD reactions. When it is oxidized
WF ϩ 3H
r W(s) ϩ 6HF(g)
[2]
[3]
6
2(g)
(
e.g., by the introduction of air), the volatile WOF is converted to
4
2
WF ϩ 3SiH
r 2W(s) ϩ 3SiF4(g) ϩ 6H2(g)
WO3 and tungsten blues which stay on the wall of the reactor.
As described earlier, during the scrubber experiments at 100 kHz
without intermediate H treatment, mass spectrometric WO signals
6
4(g)
Both reactions are thermally activated and proceed only at an accept-
able rate at higher temperatures (>300ЊC). Reaction Eq. 3 is thermo-
dynamically more favorable. Moreover, Reaction Eq. 2 is catalyzed
by a metallic substrate.
2
3
are observed. They increased as a function of lifetime of the elec-
trode. However, at the same time, only trace amounts of WOF
4
ϩ
(WOF3 ion) could be detected. Also, when opening the scrubber for
An alternative (nonthermal) activation technique to deposit tung-
visual inspection, the deposited layer exhibits a variety of colors,
ranging from metallic gray, over yellow to blue. The smell of HF
upon opening of the scrubber was observed and cracking of the de-
posited layer was heard.
sten is by means of a gas discharge plasma4,5 (i.e., PECVD), which
is also applied in the DryScrub system. Besides WF (with 1 Յ x Յ
) fragments and corresponding F radicals and F ions, the rf plasma
in the scrubber also produces atomic hydrogen
x
5
Regarding the chemical processes occurring in the ETC
DryScrub system, it can be concluded that the WF DRE is mainly
H2(g) ϩ eϪ r 2H• ϩ eϪ
6
[4]
(g)
dependent on the H dissociation efficiency, and thus on the absolute
2
•
concentration of H radicals (H ) and ions in the plasma. As described
which diffuses to the surface and reacts there in several consecutive
steps with WF fragments to form metallic W and HF (apparent acti-
vation energy is ϳ39 kJ/mol). Experiments described in literature
are done at elevated substrate temperature (350ЊC) and at pressures
above, the H concentration, the total pressure, and the plasma power
2
x
6
density play a determining role toward the H radical formation, but
could hardly be optimized in the present setup. The only [H]-deter-
mining parameter left which can be modified is the so-called surface
around 200 mTorr. The electrodes were powered at a frequency of
2
condition of the electrodes. It influences the overall H dissociation
4
.5 MHz. Power densities are ϳ0.06 W/cm . The DryScrub experi-
2
•
efficiency by catalyzing the surface H termination reactions like
ments described in present article are executed at distinctively dif-
ferent conditions of 100ЊC and 1.5 Torr. Power densities are ranging
from 0.04 to 0.25 W/cm at plasma frequencies in the kHz region.
•
ϩ H•
(ad)
r H2
H
[7]
[8]
2
(ad)
•
H
ϩ F• r HF
(ad)
(ad)
and
ϩ
Ϫ
(ad)
H
ϩ F
r HF
[9]
(
ad)
The electrode surface conditions are determined by the possible
pre-) treatment of the electrodes, and by the kind of deposits on the
(
electrodes (see previous paragraph). The ideal deposit consists of
pure W, not poisoned with any form of tungsten oxides. This can be
achieved by making the complete prepump exhaust system as gas-
tight as possible and by degassing the scrubber (bakeout) before oper-
ation, in order to minimize the presence of moisture and oxygen. On
the other hand, reduction of tungsten oxides, already present on the
electrodes, to metallic W is possible by treating the oxides in a H2
plasma. The corresponding overall reaction is
•
WO3(s) ϩ 6H r W ϩ 3H O
[10]
(s)
2
(g)
Of course, the water vapor must be pumped off. Treatment with H2
gas (without plasma) is only effective at temperatures >900ЊC.
Figure 9. WF destruction efficiency as function of wafer number during a
marathon run, performed with the 40 kHz power supply. No intermediate H2
treatment is used.
6
Moreover, an additional advantage of a H plasma is that via Reac-
2
Downloaded on 2015-06-07 to IP 129.100.58.76 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

4
670
Journal of The Electrochemical Society, 147 (12) 4665-4670 (2000)
S0013-4651(99)12-050-0 CCC: $7.00 © The Electrochemical Society, Inc.
tion Eq. 8 and 9, the adsorbed F atoms will be removed from the sur-
face. When the scrubber is active over a longer period (>5 min) or
at higher effective plasma powers, the increased temperature of the
electrodes will also increase desorption of F.
The observed memory effect during continuous scrubber opera-
tion at 100 kHz is thus determined by the scrubber electrode surface
condition, and can have several causes. One is the continuous build-
hydroxide solution. The blue tungsten oxide compounds can be
removed by a hydrogen peroxide solution. After the complete
removal of the deposits, the electrodes can be reused in the scrubber.
For reasons of completeness, it should be mentioned that some
safety precautions should be taken during handling of the electrode
after processing. It is recommended to wear skin and respiratory pro-
tection, since some W-O(-F) compounds are irritating and/or toxic.
9
up of WOF (following Reaction Eq. 6), resulting in a decrease of
4
Conclusions
the H dissociation efficiency. Prevention of WOF buildup was de-
2
4
Ϫ
scribed previously. Another cause can be the buildup of F ions on
the electrode surface. F ions are very stable in the gas phase
It has successfully been demonstrated that it is possible to cap-
ture and recover pure W from the exhaust of the W-CVD tool by
means of a commercially available ETC DryScrub system. Two dif-
ferent frequencies of the plasma power supply have been tested: 100
and 40 kHz. With the 100 kHz frequency, an initial WF6 DRE of
98% has been obtained at a nominal dissipated power of 1200 W.
However, a memory effect has been observed, causing the DRE to
drop from 98% to less than 70% after eight consecutively processed
wafers. This memory effect is believed to be caused by a continuous
buildup of FϪ ions in the scrubber. Under these conditions, the re-
sulting deposited W layer on the scrubber electrodes is of poor
chemical and physical quality. The memory effect has been resolved
by using an intermediate H2 plasma treatment of the scrubber, result-
ing in a continuous WF6 destruction efficiencies of at least 99%. The
introduction of the H2 treatment resulted also to an excellent deposit-
ed W layer on the scrubber electrodes.
With the 40 kHz plasma frequency experiments, a continuous
WF6 DRE efficiency of more than 99% has been reached with a
nominal plasma power of 1100 W or higher. No memory effect has
been observed and consequently no intermediate H2 treatment was
necessary. The quality of the deposited layer was very good and was
identified as pure W. No evidence was found of other metals within
the deposit.
Ϫ
(⌬H298 ϭ Ϫ61 kcal/mol) due to the high electronegativity of F. This
may result in a shielding of the plasma.
In the same framework, the effect of the initial condition of the
electrode surface of the scrubber can be understood. It was found
that the WF DRE during processing of the first six to seven wafers
6
of a batch, can differ from batch to batch. This was related to the sur-
face condition of the electrodes before processing. When the treat-
ment preceding processing was an H2 plasma, the initial WF6
destruction efficiency was up to 6% higher than in the case where the
initial treatment was a simple WF flow (without plasma).
6
Note also that several analogous tests for destruction of NF in a
3
9
plasma reactor exhibited low efficiencies. A potential solution
could again be the use of an additional remote upstream H dis-
2
charge to treat the surface of the plasma scrubber. This prevents
buildup of fluorine in the scrubber.
The better performance of the 40 kHz plasma power supply can
tentatively be explained by the fact that, at this lower frequency,
there is a higher probability of electron loss to the scrubber elec-
trodes. Although the loss of negative ions (mainly the FϪ species
present in the plasma) may not be affected, it is expected that the
absolute FϪ concentration will also be lower. The overall plasma
Ϫ
then becomes more positive since the shielding effect of F decreas-
Regarding the implementation of the scrubber in an industrial
environment, the following can be concluded: The scrubber operation
itself was found to be completely transparent to the upstream CVD
process. No additional gasses are needed to operate the scrubber. Both
plasma frequencies tested (100 and 40 kHz) can be used. Preference
however should go to the lowest frequency for reasons of throughput
and gas consumption. The deposited W in the scrubber can easily be
recuperated. The problem with the limited lifetime of the scrubber
electrode due to arcing is still an issue that is subject for improvement.
es. Therefore, the positive ion energy will become higher and the
resistance heating of the electrode will increase accordingly. In addi-
tion, the maximum deliverable power will be higher. These benefits
would result in a higher dissociation efficiency of H and WF . On
2
6
the other side, the hotter electrode helps the rf adsorption and the
higher bias positive ions enhance the outgassing, the quality of the
deposit and the reaction rate of the gases in the plasma.
At the plasma power frequencies of 40 and 100 kHz (including in
the latter case an intermediate hydrogen plasma treatment), a high
quality W-layer on the scrubber electrodes is obtained. However, pure
W has good conductive properties. Therefore, the deposited layer can
cause short circuiting between the plasma electrodes. The electrodes
are isolated from each other by three ceramic isolators which are also
in contact with the exhaust gases. The isolators can become covered
by a tungsten layer as a function of processing time, thus creating a
conductive path causing short circuiting between the electrodes. This
results in a malfunctioning of the scrubber. This phenomenon occurs
already after processing of a few hundreds of wafers.
In principle, this arcing could be suppressed by using a lower
operating pressure in the scrubber or by lowering the electrode tem-
perature. However, both these parameters are more or less fixed in
the present test conditions. A preferable and more thorough solution
exists in redesigning the plasma electrodes and isolators.
Acknowledgment
The authors would like to thank R. Chiu from ETC for the inter-
esting technical discussions. This work was supported by the EC
through grant LIFE96ENV/B/484.
References
1
. D. Danielson, M. Carter, K. C. R. Chiu, and W. Snow, Semicond. Int., 12, 170
Oct 1989).
2. K. Stroup, Semicond. Int., 16, 74 (Sept 1993).
. IR and Mass Spectra, NIST Chemistry Webbook, NIST Standard Reference Database
No. 69, W. G. Mallard and P. J. Lindstrom, Editors, National Institute of Standards
and Technology, Gaithersburg, MD (March 1998).
. S. Tsuzuki, E. Nishitani, M. Nakatani, and A. Shintani, in Multilevel Metallization,
Interconnection, and Contact Technologies, L. B. Rothman and T. Herndon, Editors, PV
7-4, p. 24, The Electrochemical Society Proceedings Series, Pennington, NJ (1987).
(
3
4
8
5
6
. J. K. Chu, C. C. Tang, and D. W. Hess, Appl. Phys. Lett., 41, 75 (1982).
. J. Bailer, Comprehensive Inorganic Chemistry 3, p. 763, Pergamon, Oxford, U.K.
An additional issue that needs attention is the removal of the
deposit from the electrodes in order to recuperate the metallic W.
The bulk of the deposited W can be removed mechanically and col-
lected very easily. If necessary, it can even be scraped off manually.
The remaining part can be dissolved in a solvent to remove it from
the electrode surface. Metallic W can be dissolved in an ammonium
(
1973).
7. J. Schmitz, Chemical Vapour Deposition of Tungsten and Tungsten Silicides for
VLSI/ULSI Applications, p. 114, Noyes Publications, Park Ridge, NJ (1991).
. Technology Transfer 96013062A-TR, SEMATECH (1996).
. W. T. Elwell and D. F. Wood, Analytical Chemistry of Molybdenum and Tungsten
(
8
9
Including the Analysis of the Metals and their Alloys), Pergamon, Oxford, U.K.
(1971).
Downloaded on 2015-06-07 to IP 129.100.58.76 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).