Journal of Crystal Growth 404 (2014) 89–99

Contents lists available at ScienceDirect

Journal of Crystal GrowthContents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/jcrysgro

Modeling and simulation of silicon epitaxial growth in Siemens

CVD reactor
Haoyin Ni, Shijie Lu, Caixia Chen n
Key Laboratory of Coal Gasification and Energy Chemical Engineering of Ministry of Education, East China University of Science & Technology, China

art ic l e i nf o a b s t r a c t

Article history: Siemens CVD reactor is an important chemical device for the production of polysilicon. The chemical and
Received 16 April 2014 physical phenomenon involved in the reactor is very complex. Understanding the multispecies thermal
Received in revised form fluid transport and its interaction with the gas/surface reactions is crucial for an optimal design and
1 July 2014
operation of the reactor. In the present paper, a mathematical model was constructed to describe the
Accepted 2 July 2014
fluid dynamics, the heat and mass transfer and the reaction kinetics of the epitaxial growth process in
Communicated by Chung wen Lan
Available online 9 July 2014 industrial CVD reactors. A modified reaction kinetics model was used to represent the gas phase and
surface reactions. The kinetics model was validated using the published experimental data obtained in a
Keywords: temperature range similar to the industrial CVD processes of silicon productions. The epitaxial growth of
A1. Computer simulations
silicon in a Siemens reactor was simulated using commercial Computational Fluid Dynamics (CFD)
A1. Growth models
software ANSYS FLUENT. The distributions of gas velocity, temperature and species concentrations in the
A1. Fluid flows
A3. Chemical vapor deposition processes reactor were predicted numerically. Based on the numerical simulation results, a sensitivity analysis was
B1. Semiconducting silicon carried out to determine the key factors influencing the growth rate in industrial CVD reactors. Under
the conditions of fixed heating power applied to three different rod diameters of 50 mm, 80 mm and
100 mm, the simulated results show, when the rods' diameter is 50 mm, the surface temperature is high
and the gas temperature is low, the growth rate of silicon is determined by the transport of gas species.
When the rods' diameter increases to 80 mm, the averaged surface temperature decreases to 1361 K, the
surface reaction rate and transport of gas species control the growth rate of Si together. When the rods'
diameter is 100 mm, the surface temperature decreases further, the rates of surface reactions become
the control factor of deposition rate of Si.
& 2014 Elsevier B.V. All rights reserved.

1. Introduction silicon hydrides and the deposition of hyper pure silicon [3]. The
final step, which is the most capital and energy intensive step, is
Photovoltaic (PV) power generation has experienced rapid usually done in a chemical vapor deposition (CVD) reactor. In the
growth in the past decade due to increased demand for renewable CVD reactor, a substrate is heated and exposed to one or more
energy sources, and the wishes for reductions of greenhouse gases volatile precursors, which react and/or decompose on the sub-
and dependency on fossil fuels. However, the market share of PV strate surface to produce high purity crystalline. Siemens reactors
power is still less than 1% of total energy supply due to high costs and the fluidized bed reactors (FBR) are the two types of most
of manufacturing solar cells. About 90% of world's solar cells are commonly used CVD processes in today's industrial production of
produced by using mono- or multi-crystalline silicon wafer tech- polysilicon [4]. Although the FBR process has obvious advantages
nology. The solar grade silicon production accounts for nearly half including low energy costs and continuous production of grain
of the total costs of solar energy production [1,2]. Therefore, a silicon particles, it is still immature in the control of fine dusts
decrease in costs of solar grade silicon production is essential in from undesired homogeneous decomposition in the gas phase.
enabling large scale PV power generation. Currently, over 90% of Industrialization of the FBR process is also limited by security
solar grade silicon is produced by using chemical routes which problems of monosilane gas [5]. Due to these limitations, Siemens
consist of three steps, namely, (1) the synthesis of silicon hydrides process currently dominates the solar grade silicon industry, and
from metallurgical grade silicon (MG-Si), (2) the purification of the preference will be likely to continue in the future.
synthesized silicon hydrides, and (3) the decomposition of purified In a Siemens process, high purity silicon rods are exposed to a
chemical gas compound known as trichlorosilane (TCS) in a bell
reactor. The TCS gas decomposes at a high temperature around
n
Corresponding author. Tel.: þ 86 21 64252057. 1100 1C and deposits high purity polysilicon on the surface of
E-mail address: [email protected] (C. Chen). silicon rods. Byproducts of remaining toxic chlorine are recycled

http://dx.doi.org/10.1016/j.jcrysgro.2014.07.006
0022-0248/& 2014 Elsevier B.V. All rights reserved.
90 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99

back to the system and used to reproduce TCS. Once the silicon conversion of HCl to SiCl4. These findings reiterated the impor-
rods are grown to a certain diameter (150–200 mm), the silicon tance of a detailed gas phase reaction mechanism for a compre-
rods are cooled down and removed off the reactor. Since the hensive simulation model. Ho et al. [18] developed a reaction
temperature of TCS decomposition is high and the deposition of Si mechanism including the gas-phase and surface reactions for a
is slow (24–80 h), the power consumptions of the Siemens process TCS–H2 CVD reactor operated in a temperature range of 1025–
are extremely high (about 60 kW/h for one kilogram silicon) [6]. 1175 1C. Although a well-stirred 0-D flow and a 2-D boundary-
Increasing the volume of a CVD reactor is considered an efficient layer flow codes were employed, the simulation results showed a
way of decreasing the energy consumption of unit silicon output. good agreement with experimental data for a single wafer reactor.
However, a large reactor with more silicon rods may pose complex In the present paper, a reaction model based on Ho's gas and
inner structures and result in more complicated gas fluid surface reaction mechanisms was used in a comprehensive simu-
dynamics and larger temperature gradients. A non-uniform sur- lation model describing the reactive flow in an industrial Siemens
face temperature is often blamed of being a major reason for an reactor. The reaction kinetics was modified by comparing and
inferior quality of silicon products such as popcorn-like surface. To fitting with the experimental data sets obtained by Habuka et al.
help in the design and scale-up of very large scale Siemens [8] and Alba et al. [19]. A series of simulations were carried out for
reactors, a thorough understanding of the physical–chemical an industrial CVD reactor under the conditions of fixed heating
phenomena is critically important in the high temperature, high power applied to three different rod diameters of 50 mm, 80 mm
pressure reactors. and 100 mm. Sensitivity analysis was performed to determine the
Due to the fact that the decomposition of TCS involves key controlling factors of the epitaxial growth process in Siemens
flammable and toxic gas intermediates at high pressures and reactors.
temperatures, experimental studies of the reaction process in a
Siemens reactor are challenging and expensive. In recent years,
advancements in computational fluid dynamics (CFD) have pro- 2. Model descriptions and simulation methods
vided a viable alternative approach to the study of the multi-
species reacting fluid dynamics, the gas, and/or surface transport 2.1. Governing equations of TCS–H2 reacting flow
phenomena and gas/surface reactions. Applications of CFD tech-
nique in the simulations of the distributions of the gas flow The chemical and physical phenomenon involved in the
velocity, gas and surface temperature, gas species concentrations Siemens reactor is very complex, which includes turbulent gas-
and the deposition rate on the rods surface have been reported in eous flow mainly due to buoyancy and high speed jet flow; gas
the literature. Cheng and Hsiao [7] numerically investigated the heat up and reactions containing a great amount of immediate
effects of reactor geometry on the flow and temperature fields in a products; surface depositions and the crystal growth of silicon.
horizontal CVD reactor with a circular disk embedded at the The traditional CFD approach of describing a turbulent reacting
bottom of the reactor; the chemical reactions were omitted in flow is employed. The Favre averaged equations for a steady-state
their computational model. Habuka et al. [8–10] performed a gas flow are written as follows.
series of CFD studies on a single wafer CVD reactor using Mass:
commercial CFD software FLUENT. Surface reaction rates were
∂
assumed to be controlled by the rate of TCS chemisorption on the ðρui Þ ¼ Sm ð1Þ
∂xi
surface and the rate of decomposition of chemisorbed species into
Si. The deposition rate coefficients were obtained by fitting with where ui is the gas velocity components. Sm is a source term
their experimental measurements. Coso et al. [11,12] incorporated accounting for the mass consumptions or productions due to
the Habuka reaction model in an analytical model, and evaluated reactions.
the growth rate, deposition efficiency, and the power-loss depen- Momentum:
dence on the gas velocity, the mixture gas compositions and the

∂ ∂p ∂ ∂ui ∂uj ∂
rods' surface temperatures. Analytical solutions agreed with ðρui uj Þ ¼  þ μ þ  ðρui''uj''Þ þ F s ð2Þ
∂xi ∂xj ∂xj ∂xj ∂xi ∂xi
the experimental data and computational results reported in the
literature [12]. Huang et al. [13] employed the same simplistic where p is the pressure. Fs is a sink source term of momentum
Habuka reaction model and performed CFD simulations for an corresponding to mass deposition on the surface.
industrial Siemens CVD reactor with 12-rods. Presently, most of Enthalpy:
the numerical studies focused on the flow and thermal field in a
∂ ∂ ''
simplified CVD reactor; lumped kinetic models were used for the ðρui hÞ ¼ ½  ρui''h  ð3Þ
∂xi ∂xi
descriptions of surface depositions, and the gas phase reactions
were often ignored in either analytical or computational models where h is the enthalpy of gas mixture.
[14]. Gas species:
Regardless a long history of commercial use of the Siemens ! !
process, there is still no agreement on whether the deposition rate ∇ Uðρ v Y i Þ ¼  ∇ U J i þ Ri ð4Þ
is controlled by the surface reactions or by the transport of gas !
where J i is the diffusion flux of species i, which arises due to
phase precursors [14]. Cavallotti and Masi [15] used a detailed concentration gradients
surface reaction mechanism in the predictions of deposition and

! μ
etching rates obtained from two horizontal reactors of Angermeier J i ¼  ρDi;m þ t ∇Y i ð5Þ
et al. [16] and Habuka et al. [8], respectively. In combination with a Sct
simple gas phase reaction mechanism, the surface mechanism was where Sct is the turbulent Schmidt number, and Di;m is the
further applied to the simulation of a small scale Siemens reactor diffusion flux of species i.
based on a published patent by Reuschel and Kersting [17]. The ρ is defined as ρ ¼ ρϕ =ϕ, and related to the pressure and
mechanisms can reasonably interpret the competition of deposi- temperature via the equation of state:
tion and etching processes. The simulation results also suggested
that the deposition rates were highly dependent on the gas phase pM
ρ¼ ð6Þ
temperature, because an increase in gas temperature led to a fast RT
 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99 91

Table 1
Gas phase and surface reactions.

No. Reaction A Ea (cal/mol) Comments Ref.

G1 SiCl3H ¼HCl þSiCl2 3.162  1014 72,900 [16]

G2 SiCl2H2 ¼ SiCl2 þH2 3.162  1013 69,300 [16]
G3 SiCl2H2 ¼ HSiCl þ HCl 6.918  1014 75,800 [16]
G4 H2ClSiSiCl3 ¼SiCl4 þ SiH2 1.585  1013 55,500 .[16]
G5 H2ClSiSiCl3 ¼SiCl3H þ HSiCl 3.162  1013 49,800 [16]
G6 H2ClSiSiCl3 ¼SiCl2H2 þ SiCl2 6.310  1013 44,300 [16]
G7 Si2Cl5H ¼ SiCl4 þHSiCl 5.012  1013 52,300 [16]
G8 Si2Cl5H ¼ SiCl3Hþ SiCl2 7.943  1013 45,900 [16]
G9 Si2Cl6 ¼SiCl4 þ SiCl2 1.585  1014 48,800 [16]
S1 SiCl3H þ4Si(S) ) Si(B) þSiH(S) þ 3SiCl(S) 4.1  10  4  3800 STICK, COV [16]
S2 2SiH(S) ) 2Si(S) þ H2 8.606  1023 57,100 COV [16]
S3 SiH(S) þ SiCl(S) ) 2Si(S) þ HCl 7.0  1025 65,000 [19]
S4 2SiCl(S) þ Si(B) ) SiCl2 þ2Si(S) 4.758  1022 67,000 [16]
S5 SiCl2 þ 2Si(S) ) Si(B) þ 2SiCl(S) 2.2 15,000 STICK [16]
S6 HSiCl þ2Si(S) ) Si(B) þ SiH(S) þSiCl(S) 0.06 5000 STICK [16]
S7 H2 þ 2Si(S) ) 2SiH(S) 0.1 17,300 STICK, COV [16]
S8 HCl þ2Si(S) ) SiH(S) þ SiCl(S) 0.07 5000 STICK [16]
S9 SiCl2H2 þ 4Si(S) ) Si(B) þ2SiH(S) þ 2SiCl(S) 1.20  10  3  3800 STICK, COV [16]
S10 SiCl4 þ 4Si(S) ) Si(B) þ 4SiCl(S) 3.0  10  5  3800 STICK, COV [16]

Table 2 ~ ¼ Ω  2ε ω
Ω ij ij ijk k
Numerical method used to perform the simulation.
Ωij ¼ Ωij  εijk ωk
Variables Operative conditions pffiffiffi
A0 ¼ 4:04; As ¼ 6 cos ϕ
pffiffiffi qffiffiffiffiffiffiffiffiffiffi

Pressure–velocity Simple 1 S S S 1 ∂uj ∂ui
ϕ ¼ cos  1 ð 6WÞ; W ¼ ij jk3 ki ; S~ ¼ Sij Sij ; Sij ¼ þ
coupling: 3 S~ 2 ∂xi ∂xj
Pressure Standard
discretization: A realizable k–ε model is used to model these terms. The turbulent
Momentum Second order upwind
discretization:
kinetics energy k and the turbulence dissipation rate ε are
Energy Second order upwind determined by solving the transport equations
discretization:

∂ðρkui Þ ∂ μt ∂k
Reacting species: Second order upwind for k : ¼ þ Gk þ Gb  ρε  Y M ð9Þ
Near-wall Standard wall functions ∂xi ∂xi σ k ∂xi
treatment:


Under-relaxation Pressure: 0.3, density: 0.8, momentum: 0.5, turbulent ∂ðρεui Þ ∂ μt ∂ε ε2 ε
factor: kinetic energy:0.8, turbulent dissipation rate:0.8, for ε : ¼ þ ρC 1 S ε  ρC 2 pffiffiffiffiffiffi þ C 1ε C 3ε Gb ð10Þ
turbulent viscosity: 0.8, reacting species: 0.8, energy: 0.9
∂xi ∂xi σ ε ∂xj k þ νε k

where
  qffiffiffiffiffiffiffiffiffiffiffiffi
The above set of conservation equations is closed by physical η k
C 1 ¼ max 0:43; ; η¼S ; S¼ 2Sij Sij ; C 1ε ¼ 1:44;
models discussed below. ηþ5 ε
C 2 ¼ 1:9; σ k ¼ 1:0; σ ε ¼ 1:2:
2.2. Turbulence closure model

The reactant gases are introduced into the reactor by a number 2.3. Gaseous and surface reaction mechanisms
of nozzles distributed on the bottom wall. The inlet gas velocity of
each nozzle is 32 m/s. The jet flow in the reactor is highly A reaction mechanism which is based on Ho's detailed gas
turbulent, which is characterized by fluctuations in time and space phase reaction and a set of semi-empirical surface reactions are
of the velocity components and every scalar. To follow the used in the present simulation model. The equations of reactions
conventions, Favre-averaging is applied to every conservation and the kinetic parameter are summarized in Table 1. Note that
equation. Two types of additional turbulent flux terms are pro- the parameters of the reaction SiH(S)þ SiCl(S) ) 2Si(S) þHCl are
duced in the averaged Eqs. (2) and (3). These terms are the taken from Cavallotti and Masi [15]. Validations of the mechanism
Reynolds stresses (7) in the momentum equation and the Reynolds are performed and discussed in Section 3.1.
fluxes (8) in the enthalpy equation.


∂ui ∂uj
 ρui''uj'' ¼ μt þ ð7Þ 2.4. Turbulence-reaction interaction model
∂xj ∂xi
The turbulence–chemistry interactions are modeled by the
μt ∂ϕ
ρui''h'' ¼ ð8Þ Eddy-Dissipation-Concept (EDC) model [20]. The EDC model
σ ϕ ∂xi assumes that reactions occur in small turbulent structures, called
where fine scales [21]. The length fraction of the fine scales is modeled as

3=4
1 vε
μt ¼ ρC μ k2 =ε; Cμ ¼ ξn ¼ C ξ 2 ð11Þ
A0 þ As ðkU n =εÞ k
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where n denotes fine-scale quantities and C ξ is a volume fraction
Un  Sij Sij þ Ω ~ Ω ~
ij ij constant and takes a value of 2.1377.
92 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99

Fig. 1. Schematic diagram of single-wafer CVD reactor [8].

Fig. 2. Comparison of growth rate measured by different authors and calculated using different methods. (Habuka [8], Ho [18], Alba [19] mean the experimental deposition
rates obtained from their published papers. Simulation (Chemkin-model) means the simulated deposition rates approached by using Chemkin 4.1 coupled with our modified
reaction model (Table 1); simulation (Fluent-model) means the simulated deposition rates approached by using Fluent 14.0 coupled with our modified reaction model
(Table 1); and simulation (Chemkin–Ho) means the deposition rates approached by using Chemkin 4.1 coupled with Ho's [18] reaction model).

Gas species are assumed to react in the fine structures over a

time scale:

v1=2
τn ¼ C τ ð12Þ
ε
where C τ is a time scale constant and is equal to 0.4082.
The source term in the conservation Eq. (4) for the mean
species is modeled by

ρðξn Þ2
Ri ¼ ðY n Y i Þ ð13Þ
τn ½1  ðξn Þ3  i

where Y ni is the fine-scale species mass fraction after reacting over

the time τn .
In-situ adaptive tabulation (ISAT) [22] is used to accelerate the
chemistry calculations. It could accelerate the calculations by two
to three orders of magnitude than a direct integration, offering Fig. 3. Analysis of the sensitivity of the surface kinetic mechanism with respect to
substantial reductions in run-times. the growth rate of Si under a surface temperature of 1398 K.
 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99 93

Fig. 4. Comparison of growth rate along the wafer length computed using Fluent and Chemkin at an inlet hydrogen mole fraction 0.97 and a surface temperature 1398 K.

2.5. Radiation model Fig. 1(a) shows the geometric model of a planar shear flow
reactor used in the CHEMKIN simulations. The length in the flow
In a Siemens reactor, silicon rods are continuously heated to direction is 20 cm, and the height is 2 cm. The mixture gas was
maintain a surface temperature as high as 1100 1C; hence, the introduced into the reactor with an inlet velocity of 0.3 ms  1, and
radiation heat transfers between rods, rods–walls and wall–gases reacted on the substrate with a surface temperature of 1398 K.
are significant. About 60–75% energy consumption of Siemens A Boundary-layer Flow model was used for the simulations. The
process is due to radiation [19]. In order to simulate the surface-to- governing equations describing the boundary-layer flow are given
surface and surface-to-gases radiation, the discrete ordinates (DO) as follows:
radiation model [23] is used to solve the radiative transfer Momentum:
equation (RTE) for the spectral intensity.


∂u ρu dm _ dm_ ∂u dp ρu ∂ ∂u
Z 4π ρu  _ ζ  l þ ¼ 2 ρuμ þ gðρi  ρÞ ð15Þ
!!! !! σs ! !0 ! ! 0
∂x m dx dx ∂ζ dx m _ ∂ζ ∂ζ
∇ U ðI λ ð r ; s Þ s Þ þ ðaλ þ σ s ÞI λ ð r ; s Þ ¼ aλ n2 I bλ þ I λ ð r ; s ÞΦð s U s ÞdΩ
π 0
where m _ is the mass flux, the independent variables x and ζ
ð14Þ represent the axial coordinate and the normalized stream function
respectively, and the pressure is defined by
ρRT
2.6. Numerical methods p¼ ð16Þ
M
The conservations of mass, momentum and energy of fluid Species:


flows are expressed in terms of non-linear partial differential ∂Y k ρu dm _ dm _ l ∂Y k ρu ∂
ρu 
_
ζ  ¼ω
_ kW k 
_ ∂ζ
ðρY k V k;y Þ k ¼ 1; …; K g
equations. Those equations are solved iteratively using Ansys ∂x m dx dx ∂x m
FLUENT 14.0. Numerical method and schemes are listed in Table 2. ð17Þ
where ω _ k is the chemical production rate of the kth species due to
gas-phase reactions, and Wk is the molecular weight of the kth
3. Results and discussion species
Energy:
3.1. Validation of the reaction mechanism


∂T ρuc dm_ dm_ ∂T ρu ∂ ∂T Kg
ρucp  _ p ζ  l ¼ 2 ρuλ  ∑ ω _ k M k hk
The gas and surface reaction kinetics were validated using ∂x m dx dx ∂ζ m _ ∂ζ ∂ζ k¼1
three sets of experimental data published in the literature, ρ2 u Kg
∂T
 ∑ Y k V k;y cpk ð18Þ
including those of the deposition rate given by Ho et al. [18] and _
m k¼1 ∂ζ
Habuka et al. [8] from a single wafer CVD reactor, and Alba [19]
from a single rod CVD prototype. The validation simulations are As the multicomponent transport is used, the diffusion velocity
divided into two steps: (1) simulation of deposition rates using V k;y is given by
CHEMKIN 4.1 and (2) simulations of deposition rates using ρuyα ∂xk DTk ρuyα ∂T
FLUENT 14.0. V k;y ¼ K g ∑ M j Dk;j  ð19Þ
_
xk M m jak ∂ ζ ρY k T m
_ ∂ζ
The experimental setups of the single wafer CVD reactor were
similar. Referring to Habuka's research, [8] silicon depositions took 100 grid points were used in the vertical direction normal to the
place in a horizontal reactor as shown in Fig. 1. The gas mixture substrate face. Note that the grid point's number below 100
(TCS, H2) was introduced into the reactor under atmospheric resulted in much different deposition rates due to numerical
pressure with an inlet velocity of 0.67 ms  1 and reacted the errors. Further increase of the grid points beyond 100 did not
surface of a disk substrate (8 in. of diameter) held horizontally make a significant difference. The absolute and relative tolerances
on a fixed susceptor. The substrate was heated to 1398 K using an use the default settings of 1.0E  8 and 0.0001, respectively. The
infrared furnace through a quartz wall reactor. The epitaxial thin- simulations were first performed using the reaction mechanism of
film thicknesses were measured using Fourier-transform infrared Ho et al. [18]. As shown in Fig. 2, the predicted deposition rates
spectra. (Simulation (Chemkin–Ho)) were much lower than Habuka's
94 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99

Fig. 5. (a) Geometrical model of Siemens CVD reactor and (b) computational grids.

Table 3 Despite the fact that CHEMKIN is a robust and mature chem-
Boundary conditions. istry simulation tool, it could not incorporate complex fluid
dynamic of a practical CVD process. To further confirm the validity
Variables Boundary conditions
of the modified reaction model, 2-D steady state simulations were
Operating pressure: 0.3 MPa conducted using FLUENT 14.0 for the same single wafer CVD
Inlet mass flow rate: 0.1624 kg/s reactor. In the FLUENT simulations, the real geometry and bound-
Inlet H2 mole fraction: 0.737 ary conditions which are identical to Habuka's experiment were
Inlet gas temperature: 573 K
Heating power applied to the rods surface: 9.136  106 w/m3
used (shown in Fig. 1(b)). 104,480 Cells were used based on a grid
Outlets: Pressure outlet (473 K) dependence test, and a surface temperature of substrate 1398 K
Temperature of the side and top reactor walls: 673 K was used in all simulations. Numerical methods used in the
Temperature of the bottom wall: 523 K simulation are listed in Table 2. The simulation results at various
Temperature of electrode supports: 873 K
inlet hydrogen mole fractions are shown Fig. 2 (Simulation
Internal emissivity: Rod: 0.7, others: 0.9
(Fluent-model)) and compared with that obtained in CHEMKIN
simulations. As is shown in the figure, the deposition rates
experimental data, but close to Ho's experimental data. However, predicted by the two simulation approaches are close. Fig. 4
deposition rates from Ho's experiment were much lower than the compares the growth rate along the wafer length at an inlet
deposition rates obtained by Habuka et al. [8] and Alba et al. [19] hydrogen mole fraction of 0.97, and only minor discrepancies are
under similar conditions. Because no details of Ho's experiment can be observed due to the differences of SiHCl3 concentrations on the
found in published literature, other two sets of experimental data surface. While 100 grid points were used to resolve the boundary
were used. A sensitivity analysis was performed using perfectly stirred layer in the CHEMKIN simulations, only 40 grids were used in the
reactor (PSR) model with the gas phase and surface kinetic mechan- vertical direction of the computational domain in the FLUENT
isms. The surface temperature of the reactor is 1398 K and the gas simulations, and a wall function method was used in the estima-
temperature is 1100 K. Sensitivity coefficients were calculated with tion of boundary values adjacent to the wall surface.
respect to the growth rate of Si as S¼ ln(GSi)/ln(ai), where ai is the
kinetic constant of reaction i. The sensitivity analysis results (Fig. 3) 3.2. Simulations of an industrial CVD reactor
showed that the reactions whose rate is limiting for the growth rate of
Si are S1, S2, S3, S4, S7 and S8. As mentioned in Ho's paper, for reaction The turbulent reacting gas fluid dynamics of a 12-rods Siemens
S3, they used an intermediate value from reported papers for a reason CVD reactor were computationally simulated in a 3-dimensional
of fitting with their experimental deposition rate. Referring to Caval- Eulerian frame using FLUENT 14.0. The geometric diagram of the
lotti and Masi's [15] research, the kinetic constants of reaction S3 reactor is illustrated in Fig. 5(a). The bell-shaped reactor was
(Table 1) were modified to improve the fitting of experimental 3500 mm high with an inner diameter of 1720 mm. The vertical
deposition rates [8]. By decreasing the activation energy and increas- height of each silicon rod was 2400 mm. TCS–H2 mixture was
ing the pre-exponential factor of reaction S3, the SiCl(S) concentration introduced into the reactor from 9 inlets distributed on the bottom
on the surface was decreased. Consequently, the Si-etching reaction S4 of the reactor. As there is only one outlet on the bottom of the
was suppressed, and the overall Si deposition rate was increased reactor, the geometry of the reactor is center-plane symmetrical.
substantially. As shown in Fig. 2 (Simulation (Chemkin-model)), the Only half part of the reactor was computed for the purpose of
predicted deposition rates using the modified surface reaction model reducing the run-times. The computational domain was divided by
increase significantly and agree with the experimental data of Habuka hexahedral elements using ICEM CFD 14.0 as shown in Fig. 5(b).
et al. and Alba et al. Three cases corresponding to three different rods' diameters of
 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99 95

Fig. 6. Gas velocity vector distributions on the symmetry plane for 50 mm, 80 mm and 100 mm cases.

Fig. 7. Gas temperature distributions on the symmetry plane for 50 mm, 80 mm and 100 mm cases (K).

50 mm, 80 mm and 100 mm were considered in the present paper. heating power (9.136  106 W/m3) was provided by Jiangsu
Grids numbers for the three cases were 757,238, 771,964 and Zhongneng Polysilicon Technology Development Co., Ltd. The
788,302, correspondingly. The numerical schemes used in the heating flux is calculated by the following equation:
simulations are listed in Table 2 and the boundary conditions are
qf lux ¼ Q  V rod =Srod ð20Þ
listed in Table 3. The geometry and boundary conditions are
provided by Jiangsu Zhongneng Polysilicon Technology Develop- where Q is the heating power, V rod is the volume of the silicon rod and
ment Co., Ltd. The default convergence criterion of 0.001 was used Srod is the surface area of the silicon rod. Through conversion, three
for all quantities except energy which used a value of 1  10  6. As fixed heating fluxes of 118,768, 77,199, and 61,759 W/m2 were applied
it is difficult to measure the heat flux of rods' surface, a fixed to the rods surfaces of 50 mm, 80 mm, and 100 mm, respectively.
96 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99

Fig. 8. Rod surface and gas temperature distributions for 50 mm, 80 mm and 100 mm cases (K).

Fig. 9. Simulated Si growth rate along the rod length.

Table 4
Averaged results for three computation cases.

50 mm 80 mm 100 mm

2
Averaged quantities of rods' surface GR (kg/m s) 0.002253 0.001785 0.001315
Ts (K) 1437.8 1360.5 1338.1

Averaged quantities of space Tg (K) 865.4 1024.0 1035.6

Averaged quantities of outlet HCl 0.245 0.269 0.231

H2 0.59 0.562 0.598
SiHCl3 0.163 0.133 0.155
SiCl4 4.57E  04 0.0064 0.01
SiH2Cl2 1.42E  04 0.0018 0.005

About 48 h were required for obtaining a convergent solution of each due to the improvement in heat convection. Fig. 8 shows the
case on a SUPERMICRO Workstation with 4 Intel Xeon E5-2680 CPUs. temperature distributions on the rods'’ surfaces. The temperature
3-Dimensional fields of gas velocity, temperature and species on the rods' surface decreases with the increase of rods' diameter
concentrations in the reactors were simulated, and the distribu- due to the decrease in heat flux. Since the flow and transport on
tions of surface species and the Si deposition rates were predicted. the rod's surface vary during the real process, the Jiangsu Zhong-
Fig. 6 shows the vector distributions of the three simulation cases neng Polysilicon Technology Development Co., Ltd. estimates that
scaled by gas velocity magnitude on the symmetry plane. The the growth rate is of the orders between 10  3 and 10  4 kg/m2 s.
velocity magnitude in the reactor increases with rods' diameter Fig. 9 shows the simulated deposition rate along the rod length;
due to decrease in volume spaces. Fig. 7 shows the gas tempera- the simulation results are in the same range. Comparing with the
ture distributions of the three cases on the same symmetry plane. vector distribution on the symmetry plane, it is interesting to find
As the velocity magnitude increases, the gas temperature increases the low deposition rates regions which appear at the places where
 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99 97

Fig. 10. Simulated Si growth rate on individual rod as a function of surface temperature (18 points are shown to represents 6  3 ¼18 Si rods in three simulation cases, the
number is area-weighted average Si growth rate of one rod).

Fig. 11. Contributions of 4 major surface reactions directly leading to Si growth rates.

Fig. 12. Analysis of the sensitivity of the surface kinetic mechanism with respect to the growth rate of Si under different surface temperatures.

eddies form. In order to determine the key factors influencing the (Table 1), formations of SiCl4 and SiH2Cl2 have positive impacts on
Si deposition rates, the space and surface distributed data were Si deposition. The more SiCl4 and SiH2Cl2 are formed, the less HCl
averaged in the post-processing. Table 4 lists the averaged quan- will be produced in the gas phase. HCl is considered the most
tities for the three cases. Those are the surface averaged growth active etching agent; hence, the formation of SiCl4 favors the film
rate GR, space averaged gas temperature Tg, surface averaged growth significantly [14].
temperature of 12 rods Ts, and the concentrations of species Surface averaged growth rates of each rod were computed for
(HCl, H2, SiHCl3, SiCl4, SiH2Cl2) at exit. three computation cases. Fig. 10 shows the variation of the averaged
As shown in Table 4, with the increase of rod diameter, the growth rates as a function of surface temperature for each rod. It
rod's surface temperature decreases, and the averaged gas tem- obviously shows that when the surface temperature is low (under
perature increases. The mole fractions of SiCl4 and SiH2Cl2 in the 1350 K), the growth rates show a strong dependence on the surface
outlet increase with the gas temperature, because their production temperatures, and fluctuate around the average values.
rates in the gas phase are higher than the consumption rates on Overall deposition is composed of two portions, one is the
the surface. Also as shown in the gas phase reaction mechanism production of Si and the other is the consumption of Si. As is shown
98 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99

in Table 1, the surface reactions S1, S5, S6, S9 and S10 contribute to temperature. When the rods' diameter is 50 mm, the heating flux
the total production of Si, and the consumption of Si is contributed by providing to the rods' surface is 118768 W/m2; the rod surface
the surface reaction S4. Fig. 11 shows the contributions of 5 major temperature is high and the gas temperature is low, the deposition
surface reactions directly leading to Si growth or etching under three rate of Si is controlled by the transport of gas species. When the
surface temperatures of 1438 K, 1361 K and 1338 K. As the reacting rods' diameter increases to 80 mm, the heating flux providing to
rate of S6 is much lower than the other five reactions, the contribu- the rods' surface decreases to 77,199 W/m2; the rates of surface
tion of S6 is negligible. It is interesting to find that the kinetic rates of reactions and the transport of gas species control the deposition
S1, S5 and S4 decrease with the surface temperatures, but the kinetic rate of Si together. When the rods' diameter is 100 mm, the
rate of S9 shows different changing tendencies for a reason of the heating flux providing to the rods' surface decreases further to
change of gas temperature. When the rod diameter is 50 mm, the 61,759 W/m2. The rates of surface reactions become the control
surface temperature is high (1438 K) and the gas temperature is low factor of deposition rate of Si.
(865 K), the mole fraction of SiH2Cl2 around the rods' surfaces is
much lower than the other two cases due to a slow decomposing rate
of TCS (the average value is 0.000138–50 mm compared with 0.0017– Notation
80 mm and 0.0045–100 mm). It is concluded that the reaction S1
contributes the most to production of Si due to a high concentration cp specific heat at constant pressure of gas mixture, J/(kg K)
of TCS around the rods' surface, and the total production and C μ , C ε1 ,
C ε2 turbulent model constant
consumption rates of Si decrease with the surface temperatures. Cξ volume fraction constant
A sensitivity analysis is applied to find the most sensitive Cτ time scale constant
elementary reactions affecting the deposition rate. The sensitivity Fs source term of momentum corresponding to mass
analysis was performed in PSR models with analogous conditions deposition source
to the Siemens reactors. As shown in Fig. 12, the reaction rates of Gb turbulent kinetic energy due to buoyancy
S1, S2, S3, S7 and S8 are sensitive to the deposition rate of Si. As Gk turbulent kinetic energy due to the mean velocity
reaction S1 provides the most productivity for Si (Fig. 11), the A gradients
factor (pre-exponential factor) of reaction S1 was changed from h thermal enthalpy, J kg  1
4.1  10  4 to 2.05  10  4 for the reason of finding the control k turbulence kinetic energy, m2 s  2
factor of deposition rate. Comparing with the deposition rates M mean molecular weight of a mixture, (kg/mol)
using the former A factor, it is interesting to find that the average Mk molecular weight of the kth species, (kg/mol)
deposition rates of Si change in different ways for different rods' m_ mass flux, kg/m s
diameters of 50 mm, 80 mm and 100 mm. When the rods' dia- m_l mass loss rate at the lower boundary, kg/m s
meter is 50 mm, the deposition rate does not change a lot (from ms source term of mass consumption or production due to
0.00225 kg/m2 s to 0.00229 kg/m2 s). Linking with the vector reactions
distribution on the symmetry plane (Fig. 6) and the deposition R gas-law constant, (J/mol K)
rates along the rod length (Fig. 9), the low deposition rate region is ui, uj, uk gas velocity components, m s  1
where the eddy forms. It means the transport of gas species xi, xj, xk coordinates of three directions, m
controls the deposition of Si. When the rods' diameter increases to Yi mass fraction of ith species
80 mm, the change of the deposition rate is significant (from YM dilatation dissipation
0.00179 kg/m2 s to 0.00153 kg/m2 s). As shown in Fig. 9, the ε dissipation rate of turbulent kinetic energy, m  2 s  3
fluctuation of deposition rate along the rod length is also sig- σk, σε turbulent Prandtl numbers for k and ε
nificant. It is concluded that the rates of surface reactions and the ρ gas density, (kg m  3)
transport of gas species control the deposition rate of Si together. λ thermal conductivity of the gas mixture, J/m s K
The change is more significant when the rods' diameter is 100 mm μ dynamic viscosity, kg m  1 s  1
(from 0.00132 kg/m2 s to 0.00111 kg/m2 s). Comparing the aver- μt turbulent viscosity, kg m  1 s  1
aged deposition rate along the rod length (Fig. 9), the deposition ξ length fraction
rate of Si is controlled by the rates of surface reactions. τ time scale
ω_ k chemical production rate of the kth species due to gas-
phase reactions, (mol/(cm3 s))
4. Conclusion ζ normalized stream function
n fine-scale quantity
A mathematical model was constructed to describe the multi- – mean of any quantity
species thermal fluid transport and its interactions with the gas/ " fluctuation of any variable
surface reactions. A modified reaction model based on Ho's
detailed gas phase reaction and a set of semi-empirical surface
reactions were employed in the present simulations. The reaction
Acknowledgments
mechanism was validated by published experimental data
obtained in a temperature range similar to Siemens reactors using
This work is supported in part by the China's National Science
CHEMKIN and Ansys FLUENT. Epitaxial growth rates of silicon in a
Foundation (NSFC) under Grants 21276085 and 20876049, and the
Siemens reactor with three different rod diameters of 50 mm,
Jiangsu Zhongneng Polysilicon Technology Development Co., Ltd.,
80 mm and 100 mm were simulated using CFD approach coupled
China.
with the modified reaction kinetics model. The distributions of gas
velocity, temperature and species concentrations in the reactor
were predicted numerically. Based on the simulation results, a References
sensitivity analysis was carried out to determine the key factors
influencing the growth rate in industrial CVD reactors. As the [1] K.H. Solangi, M.R. Islam, R. Saidur, N.A. Rahim, H. Fayaz, A review on global
solar energy policy, Renew. Sustain. Energy Rev. 15 (4) (2011) 2149–2163.
heating power provided to the silicon rods was fixed, an increase [2] S. Ranjan, S. Balaji, R.A. Panella, B.E. Ydstie, Silicon solar cell production,
of rod diameter resulted in a decrease of rods' surface Comput. Chem. Eng. 35 (8) (2011) 1439–1453.
 H. Ni et al. / Journal of Crystal Growth 404 (2014) 89–99 99

[3] G.D. Coso Sánchez, Chemical decomposition of silanes for the production of [13] Z. Huang, C. Liu, X. Yuan, S. Liu, F. Liu, Development of a polysilicon chemical
solar grade silicon (Doctoral dissertation, Telecomunicacion), 2010. vapor deposition reactor using the computational fluid dynamics method, ECS
[4] A.F.B. Braga, S.P. Moreira, P.R. Zampieri, J.M.G. Bacchin, P.R. Mei, New processes J. Solid State Sci. Technol. 2 (11) (2013) P457–P464.
for the production of solar-grade polycrystalline silicon: a review, Sol. Energy [14] S. Ravasio, M. Masi, C. Cavallotti, Analysis of the gas phase reactivity of
Mater. Solar Cells 92 (4) (2008) 418–424. chlorosilanes, J. Phys. Chem. A (2013).
[5] B.E. Ydstie, J. Du, Producing poly-silicon from silane in a fluidized bed reactor, [15] C. Cavallotti, M. Masi, Kinetics of SiHCl3 chemical vapor deposition and fluid
Solar Cells 2011, http://cdn.intechweb.org/pdfs/22416.pdf. dynamic simulations, J. Nanosci. Nanotechnol. 11 (9) (2011) 8054–8060.
[6] A. Müller, M. Ghosh, R. Sonnenschein, P. Woditsch, Silicon for photovoltaic [16] D. Angermeier, R. Monna, A. Slaoui, J.C. Muller, Modeling and analysis of the
applications, Mater. Sci. Eng.: B 134 (2) (2006) 257–262. silicon epitaxial growth with SiHCl3 in a horizontal rapid thermal chemical
[7] T.S. Cheng, M.C. Hsiao, Numerical investigations of geometric effects on flow vapor deposition reactor, J. Electrochem. Soc. 144 (9) (1997) 3256–3261.
and thermal fields in a horizontal CVD reactor, J. Cryst. Growth 310 (12) (2008) [17] Reuschei, K., U.S. Patent no. 3,200,009. U.S. Patent and Trademark Office,
Washington, DC, 1965.
3097–3106.
[18] A. Balakrishna, J.M. Chacin, P.B. Comita, B. Haas, P. Ho, A. Thilderkvist, Chemical
[8] H. Habuka, T. Nagoya, M. Mayusumi, M. Katayama, M. Shimada, K. Okuyama,
Kinetics for Modeling Silicon Epitaxy from Chlorosilanes, Sandia National Labora-
Model on transport phenomena and epitaxial growth of silicon thin film in
tories, Albuquerque, NM, and Livermore, CA., 1998 (No. SAND98-1874C).
SiHCl3–H2 system under atmospheric pressure, J. Cryst. Growth 169 (1) (1996)
[19] A. Ramos, C. del Cañizo, J. Valdehita, J.C. Zamorano, A. Rodríguez, A. Luque,
61–72.
Exploring polysilicon deposition conditions through a laboratory CVD proto-
[9] H. Habuka, M. Katayama, M. Shimada, K. Okuyama, Nonlinear increase in
type, Phys. Status Solidi C 9 (10–11) (2012) 2164–2168.
silicon epitaxial growth rate in a SiHCl3–H2 system under atmospheric [20] Magnussen, B.F., in: 19th AIAA Aerospace Science Meeting. Saint-Louis,
pressure, J. Cryst. Growth 182 (3) (1997) 352–362. Missouri, 1981.
[10] H. Habuka, Y. Aoyama, S. Akiyama, T. Otsuka, W.F. Qu, M. Shimada, [21] I.R. Gran, B.F. Magnussen, A numerical study of a bluff-body stabilized
K. Okuyama, Chemical process of silicon epitaxial growth in a SiHCl3–H2 diffusion flame. Part 2. Influence of combustion modeling and finite-rate
system, J. Cryst. Growth 207 (1) (1999) 77–86. chemistry, Combust. Sci. Technol. 119 (1–6) (1996) 191–217.
[11] G. Del Coso, I. Tobias, C. Canizo, A. Luque, Temperature homogeneity of [22] B. Yang, S.B. Pope, An investigation of the accuracy of manifold methods and
polysilicon rods in a Siemens reactor, J. Cryst. Growth 299 (1) (2007) 165–170. splitting schemes in the computational implementation of combustion
[12] G. Del Coso, C. del Canizo, A. Luque, Chemical vapor deposition model of chemistry, Combust. Fl*ame 112 (1) (1998) 16–32.
polysilicon in a trichlorosilane and hydrogen system, J. Electrochem. Soc. 155 [23] S.R. Mathur, J.Y. Murthy, Coupled ordinates method for multigrid acceleration
(6) (2008) D485–D491. of radiation calculations, J. Thermophys. Heat Transf. 13 (4) (1999) 467–473.