Influence of annealing ambience on the formation of cobalt silicides

Article abstract of DOI:10.1016/S0009-2614(03)00330-0

By using two sets of Raman systems with excitation lines of 514.5 and 632.8 nm, the influence of annealing ambience was investigated on the formation of cobalt silicides. The results show that a more uniform, compact and thermal stable cobalt silicide film can be formed in the hydrogen annealing ambience than that in the Ar annealing ambience. Two characteristic bands located at 305 and 325 cm^-1, which may be assigned to the CoSi and CoSi₂, respectively, were found during the phase transition processes. The strong band which appeared at 325 cm^-1 that can only be detected with the excitation line of 632.8 nm was found to be due to the resonant Raman effect.

Full text of DOI:10.1016/S0009-2614(03)00330-0

Chemical Physics Letters 372 (2003) 15–21
www.elsevier.com/locate/cplett
Influence of annealing ambience on the formation
of cobalt silicides
a
Feng-ming Liu , Bin Ren , Yu-xiong Jiang , Jian-hui Ye , Zhong-qun Tian
a,
a
b,1
a
*
a
State Key Laboratory for Physical Chemistry of Solid Surfaces, Department of Chemistry, Xiamen University, Xiamen 361005, China
b
Institute of Materials Research and Engineering, 3 Research Link, Singapore 117602, Singapore
Received 27 January 2003
Abstract
By using two sets of Raman systems with excitation lines of 514.5 and 632.8 nm, the influence of annealing ambience
was investigated on the formation of cobalt silicides. The results show that a more uniform, compact and thermal stable
cobalt silicide film can be formed in the hydrogen annealing ambience than that in the Ar annealing ambience. Two
ꢀ1
2
characteristic bands located at 305 and 325 cm , which may be assigned to the CoSi and CoSi , respectively, were
ꢀ1
found during the phase transition processes. The strong band which appeared at 325 cm that can only be detected
with the excitation line of 632.8 nm was found to be due to the resonant Raman effect.
Ó 2003 Elsevier Science B.V. All rights reserved.
1
. Introduction
of Ti silicide processes [1–3,7], but its application
to sub-0.25 lm technologies has been questioned
due to the difficulty in maintaining a low sheet
As the dimensions of very-large-scale integra-
tion (VLSI) are scaled down, self-aligned silicides
play a critical role in high-performance CMOS
technologies, serving both as a means of achieving
low sheet resistance of gate, source and drain ar-
eas, minimizing RC delay and providing a low
resistance contacting structure [1–6]. Among the
various silicides, most of present works have been
2
resistance on small geometries [1,8]. CoSi has
been receiving increasing attention due to its low
resistance, high-thermal stability, and low lattice
defect [1,4,6]. Furthermore, the formation of Co
silicide is not sensitive to the scaling of the latter
feature size [1]. Thus, great effort has been made in
industry to develop manufacturing-worthy Co
silicide processes for the 0.25/0.18 lm technology
[1,3,4]. However, the formation of Co silicide is
sensitive to the oxidation and contamination of the
Si surface and the annealing ambience, which may
result in the formation of films of high-sheet re-
sistance [1–3]. Two approaches have been explored
to overcome these processes: one is to use a cap-
ping layer, typically Ti or TiN [4,9,10], another
2 2
concentrated on TiSi and CoSi . Ti silicide has
been used for several generations and recent efforts
were quite successful in improving the scalability
*
Corresponding author. Fax: +86-592-208-5349.
E-mail address: bren@jingxian.xmu.edu.cn (B. Ren).
Also corresponding author.
1
0
009-2614/03/$ - see front matter Ó 2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0009-2614(03)00330-0

16
F. Liu et al. / Chemical Physics Letters 372 (2003) 15–21
approach is called no-capping process, in which
Co is deposited onto the silicon surface at an ele-
vated temperature [3].
Raman system (Dilor) [19], which has a built-in
He–Ne laser (632.8 nm) with a power of 5 mW on
sample. Its microscope attachment is based on an
Olympus BX40 system and it uses a 50Â long
working-length objective (8 mm). Another is
R1000 Raman spectrometer from Renishaw op-
Raman spectroscopy has several advantages
compared with other analytical methods [11]. It
does not require special sample preparations and
is nondestructive under normal circumstance. It
can provide the information on the vibrational
properties of a material in a wide frequency
range, especially in the low frequency region.
Therefore, Raman spectroscopy has been consid-
ered as a useful tool in the study of the formation
mechanism of silicides [12]. However, up to now,
there are only few Raman studies on Co silicides
þ
erating at 514.5 nm provided by an external Ar
ion laser. The laser power on the sample is 6 mW.
Its microscope attachment is based on a Leica
system and also uses a 50Â long working-length
objective (8 mm).
The cobalt-deposited silicon wafers were an-
nealed in a homemade high-temperature cell
[17,18], which can be easily adapted to a confocal
microprobe Raman system. The Raman cell can
work at temperatures as high as 1000 °C. It takes
only about 1 min to heat up the surface from room
temperature to 1000 °C by proper control of the
heater (platinum wire) and the thermocouple (K-
type) installed underneath the sample. The cooling
of the system is realized by flowing cold water
through the shell of the cell to keep it at the room
temperature. When the heater is turned off, the
temperature in the sample chamber can quickly
decrease from 1000 to 100 °C in 1 min. However, it
takes several minutes to cool the system from 100
°C to room temperature.
[
13–15]. The reaction of cobalt films on silicon has
been reported to proceed through the phase se-
quence of Co [13,16]. In Ref.
15], a very weak and narrow Raman band at
2
Si ! CoSi ! CoSi
2
[
2
ꢀ
1
67 cm attributed to the k ꢁ 0 scattering was
2
reported on the CoSi film. Surprisingly, the fol-
lowed investigations did not mention this band
any more even by using a more sensitive Raman
spectrometer [13,15]. A micro-Raman system was
only used to measure the uniformity of thin CoSi2
films by comparing the first-order Si Raman line
ꢀ
1
at 520 cm in different areas [15]. Recently, Zhao
et al. investigated the thickness and the unifor-
mity of the CoSi thin films annealed in the tem-
perature range of 450–490 °C. They detected for
Co films of 30 nm in thickness were evaporated
by electron beam onto the commercially prepared
and polished Si(1 1 1) single crystal substrates at a
ꢀ
1
the first time Raman shifts at 204 and 220 cm
ꢀ
1
ꢀ1
due to CoSi and at 150 cm due to Co Si [13].
2
rate of 1 nm min . The silicon wafers were sub-
There is still no report on the CoSi . It revealed
2
jected to the accepted cleaning procedure of RCA
solution immediately prior to loading into the
electron-beam evaporation system. The cobalt-
deposited silicon wafer was then cut into a size of
2 Â 2 mm and directly put into the annealing cell.
To remove the oxygen, the cell was flowed with
the difficulty in obtaining the Raman signal of
2
CoSi .
In this Letter, with a well-designed Raman cell
that is capable of working at high temperatures
[
17,18], the formation of Co silicide by annealing
in either Ar or H was investigated. Two excitation
2
2
high-purity Ar or H for 5 min at a rate of 10
lines were used in order to employ possibly the
resonance Raman effect to detect very weak Ra-
man signal of cobalt silicide.
bubbles/s. Finally, the sample was annealed in the
Ar or H ambience for the desired time and then
cooled down to the room temperature before the
Raman characterization.
2
Since the roughened silicon surfaces are very
nonuniform and the Raman signal intensities from
different regions will be slightly different, 3–5 spots
in different regions were measured on a sample and
the spectrum with a moderate intensity was pre-
sented.
2
. Experimental
Raman experiments were performed on two sets
of confocal microprobe Raman systems operating
at different wavelengths. One is the LabRam I

F. Liu et al. / Chemical Physics Letters 372 (2003) 15–21
17
3
. Results and discussion
to the TA phonon vibration from the silicon [20],
also changes its intensity during the annealing
processes. Besides these bands, we did not detect
3
.1. Annealing in Ar ambience
2
any bands coming from CoSi , coinciding with
Fig. 1 shows the Raman spectra from the
samples annealed in the Ar ambience for 1 min at
previous reports [13,15]. By correlating the tem-
perature-dependent Raman measurement and the
phase diagram, no signal from CoSi may be due
4
50, 500, 550, 600, 650, and 750 °C, respectively.
2
The spectra in Fig. 1a were acquired by using the
excitation line of 514.5 nm and that in Fig. 1b by
to the fact that it is Raman inactive or its Raman
cross-section might be too small.
6
32.8 nm. As shown in Fig. 1a, the signal of cobalt
When the above samples were characterized by
using the excitation line of 632.8 nm, as shown in
Fig. 1b, with increase of the temperature, a similar
trend to that in Fig. 1a can be found. For ex-
silicide cannot be detected at 450 °C, while a weak
band at 204 cm can be discerned at 500 °C and
keeps growing until the temperature reaches 550
ꢀ
1
ꢀ
1
°
C. At the same time, two bands at 224 and
ꢀ
ample, in both cases, the 204 cm band appears
at 500 °C, reaches its maximum intensity at 500
°C and disappears at 650 °C. However, a distinct
1
2
However, up to date, there is no report about the
40 cm appear, attributable to the CoSi [13].
ꢀ
1
ꢀ1
assignment of the 240 cm band. According to
the published temperature-dependent cobalt sili-
cide phase experiment, the abundant phase in
present experiment should be CoSi [13,16]. Al-
difference should be noted that the 204 cm band
is quite strong in intensity compared to that of the
ꢀ
1
224 cm band, as it can be seen in Fig. 1b. We
think the difference may originate from the cut-
edge margin effect of the notch filter [19]. The
edge margin (50% in intensity) of the specially
selected notch filter in Dilor Raman system can
though the signal from Co
2
Si was only observed
after the sample was annealed at 350 °C for 50 min
ꢀ
18], the increasing intensities of the 204 cm
band indicate a phase transition from Co Si to
1
[
ꢀ
1
2
achieve as low as 50 cm , and that of in Reni-
ꢀ
1
CoSi. Further increasing the annealing tempera-
shaw Raman system locates at ca. 100 cm .
Thus, a lower cut-edge margin effect on the
ꢀ
1
ture to 600 °C, the 204 cm band decreases in
intensity compared to that at 550 °C and finally
disappears when the temperature reaches 650–750
ꢀ
1
204 cm band could be in Dilor Raman system
than that in Renishaw Raman system and there-
ꢀ
1
ꢀ1
°C, indicating a phase transition from CoSi to
ꢀ
CoSi . In addition, the band at 301 cm , assigned
fore, the ratio of the 204 cm band to 224 cm
band in Fig. 1a is far larger than that of in Fig.
b. Another difference between Figs. 1a and b can
be found at 750 °C. In Fig. 1b, one can observe a
1
2
1
ꢀ
1
shoulder centered at 325 cm
01 cm , which cannot be detected in Fig. 1a.
to the peak at
ꢀ
1
3
According to the temperature-dependent phase
change of cobalt silicides, this band possibly
2
comes from CoSi . Since the cut-edge in both
ꢀ
1
Raman systems being lower than 100 cm , the
influence of the cut-edge margin can be ignored
on the bands with frequency higher than
ꢀ
1
3
00 cm , and therefore, only the influences from
the CCD quantum efficiency or a resonant Raman
effect might account for the appearance of the
ꢀ
1
3
25 cm band under the excitation line of 632.8
nm. However, since this band is very weak, it is
difficult to give a definite explanation by the two
models, which will be discussed in the latter
section by using higher quality spectra.
Fig. 1. Temperature-dependent Raman spectra from samples
annealed for 1 min in the Ar ambience. Excitation lines: (a)
514.5 nm, (b) 632.8 nm.

18
F. Liu et al. / Chemical Physics Letters 372 (2003) 15–21
Fig. 2. Temperature-dependent Raman spectra from samples
annealed for 3 min in the Ar ambience. Excitation line: 514.5 nm.
Fig. 3. Annealing time-dependent Raman spectra from the
sample annealed in the Ar ambience at 850 °C. Excitation line:
5
14.5 nm.
Besides the investigation of the composition of
the film, we also performed some experiment to
test the quality of the films. As shown in Fig. 2, the
surface with high resistance are always found to
be related to the contaminations of the silicon
surface, cobalt films and the annealing ambience
[1–3], among which the oxidation of the sample
ꢀ
1
intensity of the 520 cm band, assigned to the
first-order optical phonon Raman band, can reach
as high as 2500–5000 cps from the samples an-
nealed for 3 min from 450 to 850 °C. However,
even for a pure flat silicon wafer, the signal is only
about 6000 cps in intensity. It indicates that a large
amount of silicon atoms have transferred from the
bottom of the cobalt film to the top of surface,
which is a process that should be avoided in
CMOS technology since it will lead to not only a
nonuniform cobalt silicide film [2] but also to a
film with high resistance. Moreover, as shown in
Fig. 3, the cobalt silicide films formed in Ar am-
bience present a low thermal stability at 850 °C. A
and the existence of trace O in the annealing
2
ambience play the most crucial role in the an-
nealing processes. Normally, the use of rapid
thermal annealing method in forming silicide can
be of great value since only very short time is
needed. Therefore, the oxidation and possible
contamination do not likely continue to be a
problem [6]. Additionally, two alternative methods
have been explored to address this issue: one is to
use a capping layer, typically Ti or TiN [4,9,10],
another is a capless process in which Co is de-
posited at an elevated temperature [3]. In order to
minimize the contamination from the oxidation of
ꢀ
1
weak band at 400 cm gradually increases in in-
tensity from the initial state. After annealed for
ꢀ
silicon, we tried to introduce a reductive gas – H
to the annealing cell to explore the annealing
process.
2
1
3
30 s, the 400 cm band decreases in intensity but
many new bands appear in other frequency re-
gions. Although these bands cannot be assigned
definitely now, it reveals an obvious change in the
property of the cobalt silicide films. It also indi-
cates that the thermal stability of cobalt silicide
formed by such kind of treatment is lower com-
pared with that reported [6,21]. The origin of the
low thermal stability and the production of a
2
3.2. Annealing in the H ambience
The samples used here were annealed in H
ambience for 3 min at 450, 500, 550, 600, 650, 700
2
and 800 °C, respectively. As shown in Fig. 4, the
samples annealed in the H ambience have distinct
2

F. Liu et al. / Chemical Physics Letters 372 (2003) 15–21
19
2
Fig. 4. Temperature-dependent Raman spectra from samples annealed for 3 min in the H ambience. Excitation line: 514.5 nm.
difference in Raman spectra from that annealed in
the Ar ambience. First, the intensity of the
CoSi
than that for the formation of CoSi
of the 520 cm band is strong and almost keeps
constant with the increase of the temperature.
However, when the temperature reaches 600 °C or
higher, its intensity decreases remarkably accom-
2
. At low temperatures (450–550 °C), lower
2
, the intensity
ꢀ
1
ꢀ1
5
20 cm band is not more than 30 cps in the H2
ambience, compared to that of 2500–5000 cps in Ar
as shown in Fig. 2). This convincingly points out
that the cobalt silicide films formed in the H am-
(
2
ꢀ
1
bience are quite compact so that only a very small
amount of silicon might migrate through the film to
the surface or it is very difficult for the laser light to
penetrate underneath the silicide film. Second, one
can observe a new phenomenon in the frequency
panied by a decreasing signal at 204 cm . This
indicates not only the phase transition from CoSi
to CoSi
line is lower in the CoSi
2
but also the penetration of the 514.5 nm
films compared to that of
2
CoSi films. Additionally, it can be found that the
annealing reaction is delayed in the H ambience.
ꢀ
1
region between 300 and 400 cm . As the temper-
ature increases from 500 to 600 °C, two bands at
2
ꢀ
1
For example, as shown in Fig. 4, the 204 cm
band is only observed after annealing at 500 °C for
3 min in the H ambience and its maximum in-
ꢀ
1
3
05 and 361 cm can be found, whose intensity
almost changes in the same trend as that of the
2
ꢀ
ꢀ
1
2
3
04 cm
band, indicating that the 305 and
tensity appears at 600 °C. However, when a sample
is annealed in the Ar ambience for 3 min, a clear
1
61 cm bands might also come from CoSi. When
ꢀ
1
the temperature reaches 700 °C, the signals related
band at 204 cm can be found at 450 °C and
achieves its maximum intensity at 550 °C [18].
Actually, a similar trend as that in Fig. 4 only oc-
curs on the sample annealed for 1 min in the Ar
to CoSi disappear, but another weak band at about
ꢀ
25 cm attributed to CoSi can be discerned. On
2
1
3
a closer inspection into the relative intensity of the
bands detected, one can find that the intensity of
ambience, as shown in Fig. 1a. This shows that H
might take part in the annealing reaction.
2
ꢀ
1
the 520 cm band changes with the formation of

20
F. Liu et al. / Chemical Physics Letters 372 (2003) 15–21
2
Fig. 5. Temperature-dependent Raman spectra from samples annealed for 3 min in the H ambience. Excitation line: 632.8 nm.
When the above samples were characterized by
using the excitation line of 632.8 nm, a dramatically
different phenomenon can be found. As shown in
(514.5 nm) is found to be two times as that of the
ꢀ
1
Dilor Raman system (632.8 nm) for the 520 cm
band from a single crystal silicon wafer. One can
also find that this band in Fig. 5 under 450 °C is
roughly two times in intensity as that of in Fig. 4
under the same temperature. However, the inten-
ꢀ
1
Fig. 5, two clear bands at 305 and 325 cm can be
observed. With the increase in temperature from
ꢀ
1
4
gradually increases at the beginning and then de-
50 to 700 °C, the intensity of the band at 305 cm
ꢀ
1
sity of the 325 cm band in Fig. 5 is far larger than
that in Fig. 4. Therefore, the appearance of this
band should be attributed to the resonant Raman
effect, which might be the main reason that the
other researchers did not observe this band by us-
ing other excitation lines [13,15]. Another distinct
ꢀ
1
creases. Meanwhile, a weak band at 325 cm can
be identified at 650 °C and increases in intensity
with the increasing temperature. These phenom-
ena, together with the fact that the two bands do
not shift in frequency with the temperature, indi-
ꢀ
1
ꢀ1
cate that the bands at 305 and 325 cm corre-
spond to the species of CoSi and CoSi
respectively. We think that the strong band of
difference in Fig. 5 and Fig. 4 is from the 520 cm
2
,
band. Unlike that in Fig. 4, the band in Fig. 5 does
not have a remarkable change in intensity with the
temperature, which may be due to the fact that
the red laser has a similar penetration capability in
all the three cobalt silicide phases and, so it is
not sensitive to the change of phase and film
thickness, different from that of the excitation line
of 514.5 nm.
ꢀ
25 cm shows not only the compact CoSi films
2
1
3
but also that the resonant Raman effect. In the
former section, two reasons were proposed to the
ꢀ
1
appearance of 325 cm under the excitation line of
32.8 nm. One is from the detection sensitivity of
6
the Raman instrument (major contribution is from
the difference in CCD quantum efficiency); another
is from the resonant Raman effect. However, the
Raman intensity in the Renishaw Raman system
More significantly, we found that the cobalt sil-
2
icide films annealed in the H ambience have a good
thermal stability. The intensities of the two bands at

F. Liu et al. / Chemical Physics Letters 372 (2003) 15–21
21
ꢀ
1
3
5
8
25 and 520 cm observed at 800 °C as seen in Fig.
almost keep constant with a long annealing time at
50 °C. It indicates that the annealing of cobalt
of China (29903009, 20021002, 90206039) and the
Ministry of Science and Technology.
silicide films in the H
a compact film but also a film with good thermal
stability. It shows that the annealing ambience of H
2
ambience can lead to not only
References
2
[
[
[
1] J.A. Kittl, W.T. Shiau, Q.Z. Hong, D. Miles, Microelec-
tron. Eng. 50 (2000) 87.
2] J.P. Gambino, E.G. Colgan, Mater. Chem. Phys. 52 (1998)
can largely reduce the negative influence of the ox-
idation and contamination from the sample and the
annealing chamber.
9
9.
3] J.A. Kittl, Q.Z. Hong, H. Yang, N. Yu, S.B. Samavedam,
M.A. Gribelyuk, Thin Solid Films 332 (1998) 404.
[
[
4] Y.Z. Hu, S.P. Tay, J. Vac. Sci. Technol. B 17 (1999) 2284.
5] Z. Ma, L.H. Allen, D.D.J. Allman, Thin Solid Films 253
4
. Summary
(
6] J. Pelleg, S. Zalkind, L. Zevin, B.M. Ditchek, Thin Solid
1994) 451.
By using two sets of Raman systems with ex-
[
citation lines of 514.5 nm and 632.8 nm and two
kinds of annealing ambience, we investigated the
phase transition of the cobalt silicide. The results
show that when a cobalt-deposited sample was
annealed in the Ar annealing ambience, although a
rough phase transition can be observed, the ther-
mal stability and the uniformity of the annealed
cobalt silicide films are not good, in which the
silicon-related signals are very strong and the co-
balt silicide films under the temperature of 850 °C
were found unstable. On the other hand, when the
Films 249 (1994) 126.
[7] J.M.E. Harper, C. Cabral, C. Lavoie, Annu. Rev. Mater.
Sci. 30 (2000) 523.
[8] J.B. Lasky, J.S. Nakos, O.J. Cain, P.J. Geiss, IEEE Trans.
Electron Dev. 38 (1991) 262.
[
9] Q.F. Wang, K. Maex, S. Kubicek, R. Jonckheere, B.
Kerkwijk, R. Verbeek, S. Biesemans, K. De Meyer, VLSI
Tech. Digest 1995 (1995) 17.
[
[
10] T. Yamazaki, K. Goto, T. Fukano, Y. Nara, T. Sugii, T.
Ito, IEDM Tech. Digest 1993 (1993) 906.
11] Z.Q. Tian, J.S. Gao, X.Q. Li, B. Ren, Q.J. Huang, W.B.
Cai, F.M. Liu, B.W. Mao, J. Raman Spectrosc. 29 (1998)
7
03.
annealing ambience was changed from Ar to H
2
, a
[12] P.S. Lee, D. Mangelinck, K.L. Pey, Z.X. Shen, J. Ding, T.
Osipowicz, A. See, Electrochem. Solid-State Lett. 3 (2000)
much more uniform, compact and thermal stable
cobalt silicide film can be obtained. Through the
comparison of two excitation lines, a resonant
Raman process is assumed to occur under the ex-
citation line of 632.8 nm, and therefore, the
1
53.
[
[
[
[
13] J. Zhao, L.K. Ballast, T.Z. Hossain, R.E. Trostel, W.C.
Bridgman, J. Vac. Sci. Technol. A 18 (2000) 1690.
14] G. Guenin, M. Ignat, O. Thomas, J. Appl. Phys. 68 (1990)
6515.
ꢀ
1
15] A. Perez-Rodriguez, E. Roca, T. Jawhari, J.R. Morante,
R.J. Schreutelkamp, Thin Solid Films 251 (1994) 45.
16] F. Mahmood, H. Ahmed, C. Jeynes, W.P. Gillin, Appl.
Surf. Sci. 59 (1992) 55.
3
the CoSi , and other more detailed information
25 cm band, which is assumed to be related to
2
during the phase transition process of cobalt sili-
cide are found for the first time.
[
[
17] Y.Y. Liao, Chin. J. Light Scattering 10 (1998) 135.
18] F.M. Liu, B. Ren, Z.Q. Tian, J.H. Ye, A. See, L. Chan, J.
Appl. Phys., submitted.
[
19] S. Sharonov, I. Nabiev, I. Chourpa, A. Feofanov, P.
Valisa, M. Manfait, J. Raman Spectrosc. 25 (1994) 699.
20] P.A. Temple, C.E. Hathaway, Phys. Rev. B 7 (1973) 3685.
21] H.S. Rhee, B.T. Ahn, D.K. Sohn, J. Appl. Phys. 86 (1999)
3452.
Acknowledgements
[
[
The authors gratefully acknowledge the finan-
cial support from the Natural Science Foundation