Professional Documents
Culture Documents
Fitouri 2005
Fitouri 2005
Fitouri 2005
Paper
phys. stat. sol. (a) 202, No. 13, 2467 – 2473 (2005) / DOI 10.1002/pssa.200520077
1 Introduction
The low-temperature buffer layer plays a determinant role in the success of the GaN heteroepitaxy on
sapphire, but leads to a high density of dislocations. However, for efficient device operation, high mate-
rial quality is needed. The origin of the defect formation is found to be related to the density of islands in
the two-dimensional (2D) buffer layer. The optimization of the density and the size of these islands can
be achieved by using a three-dimensional (3D) growth process, thus reducing the dislocation density
from 1010 to about 4 × 108 cm–2. The 3D growth mode is induced either by a high-temperature annealing
of the buffer layer under H2 carrier gas [1, 2] or by the use of “antisurfactant” atom like silicon [3–6].
Vennéguès et al. [7] reported for the first time a new 3D growth process called “SiN treatment” for GaN
on sapphire substrate. This technique consists of an in situ exposure of sapphire substrate to a mixture of
ammonia (NH3) and silane (SiH4) at high temperature before the buffer layer deposition. Later, this pro-
cess was successfully applied on the GaN film [3, 8]. This technique is of great interest because it pro-
ceeds like a kind of randomly nanoELOG without the multiple complicated steps of the ELOG technol-
ogy. The success of the SiN treatment depends on the duration of the treatment and the flow rate of SiH4
[5, 6]. The SiN treatment induces spectacular changes in the in situ reflectivity signal and different
growth modes can be observed, i.e. (I) three dimensional growth (3D), (II) a coalescence process and
(III) bidimensional growth (2D). The evolution modeling of the reflectivity signal versus time can
be seen from two points of view. In the first approach, the decrease of the reflectivity signal is associated
*
Corresponding author: e-mail: Tarek.boufaden@fsm.rnu.tn, Fax: +216 73 500 278
with the appearance of a macroscopic surface roughness (rms) and the scattering theory is used [9, 10].
The refractive index of the deposited layer is assumed constant. By the point-by-point analysis, this route
gives the rms versus growth time. In the case of the SiN treatment, especially in the region of the 3D
growth, an extreme rms appears to limit the validity of the scattering theory approximation. The reflec-
tivity can be simulated using the effective medium approximation. The deposited film that presents a
rough surface can be regarded as a perfectly smooth surface with an apparent refractive index n(t). In this
paper, we developed the two points of view and we discussed the validity limit of the first approxima-
tion. For the second model, we extracted the refractive index and the growth rate versus the process time
by fitting the experimental reflectivity signal. Particular interest is also given to the effect of the SiN
treatment on the rapidity of the coalescence process. The recovery time was deduced and a simple model
was proposed to understand the behavior of this parameter within the treatment duration and the SiH4
flow rate.
2.5
Reflectivity (a.u)
2.0
1.5
Fig. 1 Reflectivity signal versus the growth
1.0 time of GaN layer elaborated on sapphire sub-
strate using SiN treatment (SiH4 flow: 10 sccm
0.5 (3.7 nmol/s), SiN treatment duration: 75 s). The
different stages are: (I) 3D growth, (II) coales-
0.0 cence process and (III) is the 2D growth mode.
The black circle indicates the recovery time.
0 1000 2000 3000 4000
Time (s)
In a first attempt, we used the simplified scattering theory [9, 11, 12] to understand the reflectivity
signal evolution. The rough surface reflectivity (Rr) is related to the smooth surface reflectivity (Rs) by:
È 4πσ ˆ 2 ˘
Rr = Rs exp Í - Ê ˙,
Î Ë λ ¯ ˚
where λ (632.8 nm) is the wavelength used and σ is the surface roughness (rms). The rms must satisfy
the Rayleigh criteria (σ ≤ 50 nm), i.e.
4πσ
£ 1.
λ
The theoretical reflectivity of the smooth surface (Rs) was calculated taking into account the experi-
mental growth rate profile (particularly in the region II). The growth rate is deduced from the period of
the reflectivity oscillations (T) by [13]:
λ
V= .
2nT
The GaN refractive index (n) was fixed, in all regions, at 2.37. The point-by-point calculation of σ
variation as a function of the growth time is reported in Fig. 2. Simultaneously with the decrease of the
reflected intensity, the surface roughness increases, indicating the formation of hexagonal and 3D hill-
ocks (region (I)). The rms attains its maximum value at the end of the island formation. In the region (II),
the rms decreases as a result of the beginning of the coalescence of the formed hillocks. It reaches a
lowest but not a zero value when the reflectivity mean value is quite constant. The not-zero value indi-
cates that the contrast between the maximum and the minimum of the reflectivity signal is not at its
highest value and the quality of the sample can be ameliorated for higher thicknesses. In region I, the rms
value is not in accordance with the Rayleigh criterion. Einfeldt et al. [14] reported that in this region, the
rms is about 100 nm. At this growth stage, the surface exhibits a grainy structure where grains are mostly
separated from each other. In this region of extreme roughness, the GaN deposited film at the instant (t),
is assumed to be formed by disconnected islands separated by the bare sapphire substrate. This inhomo-
geneous system is equivalent to a homogenous medium with an effective refractive index n varying with
the growth time (t). The film is divided into adequately large numbers of sublayers. Each layer corre-
sponds to a growth time axis interval where the refractive index and the growth rate were assumed con-
stant. This approach is justified. In fact, in situ ellipsometry monitoring of the GaN growth revealed an
apparent refractive index that varies as a function of growth time and growth mode (2D or 3D) [15–17].
A similar approach was also successfully applied by Leycuras et al. [18, 19] to monitor the evolution of
100
(I) (II) (III)
80
60
rms (nm)
40
2.0
1.5
1.0
0.5
0.0
both the refractive index and growth rate during the surface transformation of As-doped Ge layer depos-
ited on GaAs.
We have applied this formulation to fit the whole measured reflectivity. We clearly observe that this
model gives a good fit of experimental reflectivity (Fig. 3) over the entire growth regions. The n profile
versus growth time is given in Fig. 4. It gradually decreases from the refractive index of the buffer layer
(2.37) to a minimum that is less than that of the sapphire substrate at the end of region I where the sur-
face roughness is at its maximum. When the coalescence process takes place in region II, n increases and
tends to the refractive index of 2D GaN in region III. The zone where the refractive index is lower than
that of sapphire (1.76) corresponds to the extreme roughness region delimited in the rms plot (Fig. 2). In
this region, the ellipsometry revealed a minimum of the refractive index that coincides with that of the
sapphire. These measurements were done for the GaN growth without SiN treatment. The calculated
growth rate profile is reported in Fig. 5. We also added the experimental growth rate measured from the
reflectivity signal. We observed that the growth rate increased drastically in the beginning of region I.
For the two plots, we notice that the maximum growth rate value coincides with the extreme roughness.
The growth is three-dimensional and it is dominated by the formation of 3D islands so the growth rate is
essentially along the C vertical axis. At the end of region I, the hillocks delimited initially only by the R
faces expand laterally and their C top faces appear to induce a decrease of the vertical growth rate. This
is in accordance with observations made by Beaumont et al. [20] in the growth of GaN by lateral over-
2.5
2.0
Refractive Index
1.5
Time (s)
30
(I) (II) (III) Fig. 5 (online colour at: www.pss-a.com) Growth
rate profile versus growth time. For comparison we
25 added the experimental growth rate profile (line +
symbol) calculated using a GaN constant refractive
index of about 2.37.
Growth rate (Å/s)
20
15
10
0
0 1000 2000 3000 4000
Time (s)
growth using a SiN mask. The decrease of the growth rate is probably reached when the hillocks come
into contact with the neighboring ones. Afterwards (region III), the growth rate stabilizes around a mean
value until the end of growth.
When the mean value of the growth rate is first reached, we measure a growth time close to that de-
fined by Koleske et al. [21] as the recovery time. This is indicated by a black circle in Fig. 1. The recov-
ery time is defined as the time required to reach a reflectance signal mean value of about 75% of the
smooth GaN reflectivity signal (region III).
The recovery time evolution (tR) as a function of SiH4 flow rate and SiN treatment duration is plotted
in Fig. 6. The tR decreases with increase in both the silane flow rate and the SiN treatment duration. This
result is surprising. Indeed, several reports showed that this growth process leads to a formation of a thin
SiN layer acting as a nanomask on the sapphire substrate [3, 7, 8, 22–24]. The SiN thickness increases
with the treatment duration and the SiH4 flow rate. For longer treatment duration, the SiN layer thickness
was estimated to be about 5 nm from the AFM contrast. As a consequence, the sapphire surface coverage
by the SiN increases with the treatment duration and SiH4 flow rate. The free sapphire surface, where the
GaN growth starts, becomes more and more spaced. The island coalescence is therefore expected to take
a long time and the recovery time may increase, as reported by Frayssinet et al. [22] for long SiN treat-
1500 1600
1500
1400
1400
1300 1300
1200
1200
60 70 80 90 100 110 120 4 5 6 7 8 9 10
Fig. 6 (online colour at: www.pss-a.com) Reflectance recovery time versus silane flow and SiN treatment duration.
The recovery time, extracted from experimental reflectivity signals, is defined as the length of time to reach a reflec-
tance signal mean value of about 75% of that of the smooth GaN reflectivity signal. The lines are guides to the eye.
Fig. 7 Schematic mechanism of the growth using SiN layer. (a) Low SiN duration and low silane flow,
(b) long SiN duration and low silane flow. Long duration and high SiH4 flow rates are equivalent and in-
crease the sapphire surface coverage by SixNy that may play the role of a nanomask. It should be noted
that the period and dimension of opening are not the same.
ment. However, our results show opposite tendencies. Sakai et al. [23] and Park et al. [24] showed that
the sapphire treated surface was dominated by nanometer-sized holes from which the GaN growth starts.
For a longer SiN duration or a higher SiH4 flow rate, the distance between these openings increases and
their number and dimensions are reduced [23, 24]. Considering these observations, we tentatively sug-
gest in Fig. 7 schematic growth mechanism to explain the recovery-time behavior with both the treat-
ment duration and SiH4 flow rate. The exposed sapphire surface is the same from run to run. Therefore,
for a longer SiN treatment duration (or higher SiH4 flow rate), the distance between the nanoholes in-
creases and their number and dimension are reduced comparatively to the case of lower duration (or
lower SiH4 flow rate). For the former case, the 3D growth mode is more pronounced [3, 8]. Combined
with the low size of the openings, this leads to a rapid emergence of GaN islands over the SiN mask for a
longer treatment duration (or higher SiH4 flow rate). Then, because of the low sticking coefficients of
GaN on the SiN mask, the growth spreads laterally. Thus, the recovery time decreases. For a lower
treatment duration and lower SiH4 flow rates, the major adsorbed species have to diffuse on the large
sapphire opening. The large sticking coefficient on the Al2O3 substrate considerably reduces the surface
diffusion compared to highly SiN masked sapphire. The nucleation is enhanced on these openings and
their filling may take more time before the emergence of GaN islands over the mask, thus increasing the
recovery time.
3 Conclusion
The SiN treatment process used in GaN growth was monitored by in situ laser reflectometry. In the first
approach, the evolution of the reflectivity signal was simulated on the basis of the scattering theory ap-
proximation. The rms is determined by the point-by-point analysis. We observe that in the region of
extreme roughness, the Rayleigh criteria are not satisfied. To overcome this limitation, the effective
medium approximation was used. The growth rate and refractive index profile were varied to fit the three
regions of the reflectometry signal. An excellent fit of the experimental signal was obtained. Further
investigations will be carried out to explain the behavior of the refractive index especially in the region
of extreme roughness. The combination of the two approximations will be a refinement for a more accu-
rate description.
The coalescence duration, defined by the recovery time, was studied as a function of the SiH4 flow
rate and the SiN duration. Based on the random nano-epitaxial mask concept, we have proposed a sche-
matic model of the GaN growth explaining the recovery-time behavior.
Acknowledgements This work is supported by CMCU contract No. 1F1310 and DGRST.
References
[1] J. Han, T.-B. Ng, R. M. Biefeld, M. H. Crawford, and D. M. Follstaedt, Appl. Phys. Lett. 71, 3114 (1997).
[2] I. Halidou, T. Boufaden, A. Touhami, A. Rebey, and B. El Jani, phys. stat. sol. (a) 184, 263 (2001).
[3] S. Tanaka, M. Takeuchi, and Y. Aoyagi, Jpn. J. Appl. Phys. 39, L831 (2000).
[4] S. Haffouz, H. Lahrèche, P. Vennéguès, P. de Mierry, B. Beaumont, F. Omnès, and P. Gibart, Appl. Phys. Lett.
73, 1278 (1998).
[5] I. Halidou, Z. Benzarti, T. Boufaden, B. El Jani, S. Juillaguet, and M. Ramonda, Mater. Sci. Eng. B. 110, 251
(2004).
[6] Z. Benzarti, I. Halidou, T. Boufaden, B. El Jani, S. Juillaguet, and M. Ramonda, phys. stat. sol. (a) 201, 502
(2004).
[7] P. Vennéguès, B. Beaumont, S. Haffouz, M. Vaille, and P. Gibart, J. Cryst. Growth 187, 167 (1998).
[8] T. Böttcher, J. Dennemarck, R. Kröger, S. Figge, and D. Hommel, phys. stat. sol. (c) 0(7), 2039 (2003).
[9] C.-H. Wu, W. H. Weber, T. J. Potter, and M. A. Tamor, J. Appl. Phys. 73, 2977 (1993).
[10] R. S. Balmer, C. Pickering, A. J. Pidduck, and T. Martin, J. Cryst. Growth 245, 198 (2002).
[11] A. Rebey, M. M. Habchi, A. Bchetnia, and B. El Jani, J. Cryst. Growth 261, 450 (2004).
[12] A. Rebey, M. M. Habchi, Z. Benzarti, and B. El Jani, Microelectron. J. 35, 179 (2004).
[13] B. Beaumont, M. Vaille, T. Boufaden, B. El Jani, and P. Gibart, J. Cryst. Growth 170, 316 (1997).
[14] S. Einfeldt, T. Böttcher, S. Figge, and D. Hommel, J. Cryst. Growth 230, 357 (2001).
[15] A. Yoshikawa, K. Xu, Y. Taniyasu, and K. Takahashi, phys. stat. sol. (a) 190, 33 (2002).
[16] B. Cao, K. Xu, Y. Ishikani, and A. Yoshikawa, Thin Solid Films 455 – 456, 661 (2004).
[17] A. Bonanni, D. Stifter, A. Montaigne-Ramil, K. Schmidegg, K. Hingerl, and H. Sitter, J. Cryst. Growth 248,
211 (2003).
[18] A. Leycuras, M. G. Lee, and A. Hausmann, J. Appl. Phys. 78, 5680 (1995).
[19] A. Leycuras, Appl. Phys. Lett. 66, 1800 (1995).
[20] B. Beaumont, S. Haffouz, and P. Gibart, Appl. Phys. Lett. 72, 921 (1998).
[21] D. D. Koleske, A. J. Fischer, A. A. Allerman, C. C. Mitchell, K. C. Cross, S. R. Kurtz, J. J. Figiel, K. W. Full-
mer, and W. G. Breiland, Appl. Phys. Lett. 81, 1940 (2002).
[22] E. Frayssinet, B. Beaumont, J. P. Faurie, P. Gibart, Zs. Makkai, B. Pécz, P. Lefebvre, and P. Valvin, MRS
Internet J. Nitride Semicond. Res. 7, 8 (2002).
[23] S. Sakai, T. Wang, Y. Morishima, and Y. Naoi, J. Cryst. Growth 221, 334 (2000).
[24] S.-E. Park, S.-M. Lim, C.-R. Lee, and C. S. Kim, Byungsung O, J. Cryst. Growth 249, 487 (2003).