Powder Technology 292 (2016) 242250

Contents lists available at ScienceDirect

Powder Technology

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

Modeling the hydrochlorination reaction in a laboratory-scale uidized

bed reactor
Matthias Colomb a, Srinivas Palanki b,, Nicholas D. Sylvester c
a
Mitsubishi Polysilicon America Corp., Theodore, AL 36582, United States
b
College of Engineering, Lamar University, Beaumont, TX 77710, United States
c
Department of Chemical and Biomolecular Engineering, University of South Alabama, Mobile, AL 36688, United States

a r t i c l e i n f o a b s t r a c t

Article history: In the manufacturing process of high purity, electronics-grade silicon, the hydrochlorination reaction step for
Received 17 August 2015 converting silicon tetrachloride (STC) to trichlorosilane (TCS) is usually carried out in a uidized bed reactor.
Received in revised form 30 November 2015 However, the design and operation of industrial-scale uidized bed reactors for the conversion of STC to TCS is
Accepted 29 December 2015
currently based on ad hoc rules of thumb and operator experience rather than on reaction engineering principles.
Available online 28 January 2016
In this paper, a uidized bed model was developed by using the KuniiLevenspiel uidization framework. The
Keywords:
predictive capabilities of this model were tested on laboratory-scale experimental data from the literature. It is
Hydrochlorination shown that the modied KuniiLevenspiel model accurately predicts the no catalyst addition and copper catalyst
Fluidized bed reactor addition experiments of the hydrochlorination uidized bed with an average percent difference of less than 6%.
Kinetics By making a suitable adjustment to catalytically active iron content, the model prediction is in excellent
Hydrodynamic model agreement with experimental data with hydrochloric acid in the feed stream. The modied KuniiLevenspiel
model is well suited for use in scale-up calculations for an industrial reactor as all the parameters are physically
reasonable and can be predicted for larger reactors.
2016 Elsevier B.V. All rights reserved.

1. Introduction Simg 3HClSiHCl3 H2 2

Silicon is the second-most abundant element on the Earth's crust.

However, it rarely occurs in pure form and most silicon is used commer-
cially with little or no processing in the building industry (clays, silica
sand and stone, Portland cement) and the ceramics industry.
Metallurgical-grade silicon (98% pure) is also used in steel rening
and aluminum-casting. However, the electronics industry requires a
very highly puried form of silicon and the most common approach to
purifying silicon involves the conversion of metallurgical-grade silicon
into volatile liquids, such as trichlorosilane (TCS) and silicon tetrachlo-
ride (STC). TCS is separated by distillation and the byproduct, STC, can
be converted to TCS by a hydrochlorination process. Gaseous TCS is
blown over the surface of a pre-existing pure silicon seed rod heated
to a high temperature and high-purity silicon crystallites are grown di-
rectly on the surface via chemical vapor deposition [1].
Mui and Seyferth [1] proposed that the conversion of STC to TCS is a
two-step process as shown below:
SiCL4 H2 SiHCL3 HCl
Corresponding author.
E-mail address: srinivas.palanki@lamar.edu (S. Palanki).
OverallSimg 3HSiCl4 2H2 4SiHCl3 3
The hydrochlorination reaction step for converting STC to TCS is usu-
ally carried out in a uidized bed reactor. The use of a uidized bed re-
actor is advantageous due to the large surface area available for reaction
and the removal of inert solids by carry-over. However, the design and
operation of industrial-scale uidized bed reactors for the conversion
of STC to TCS is currently based on ad hoc rules of thumb and operator
experience rather than on reaction engineering principles. Developing
a scalable model from the laboratory uidized bed is particularly impor-
tant because of the potential for inefciencies in larger reactors [2]. In
this paper, a uidized bed model is developed by using the Kunii
Levenspiel uidization framework [2]. The predictive capabilities of
this model are tested on laboratory-scale experimental data from the
literature.
Experimental results from Cygon [3] and Sill et al. [4] were used to
1 test the model developed in this paper. These are the only published re-
sults suitable for testing the hydrochlorination model. The quality of the
results is very good and the tests judiciously spanned the relevant in-
dustrial reaction conditions. In addition, many of the results intentional-
ly did not reach equilibrium and are therefore useful for testing the
hydrodynamic aspects of a uidized bed hydrochlorination reaction

http://dx.doi.org/10.1016/j.powtec.2015.12.044
0032-5910/ 2016 Elsevier B.V. All rights reserved.
 M. Colomb et al. / Powder Technology 292 (2016) 242250 243

model. These experiments were conducted in a laboratory-scale uid- The power law kinetics indicates that the reaction is based on the
ized bed reactor to study the conversion of STC to TCS. The reactor chemisorption of the reactant species. The thermodynamic equilibrium
was glass-lined to preclude possible catalytic effects of the wall. All ex- constant Kp was derived from the work of Mui [8].
periments were carried out at atmospheric pressure. The laboratory  
scale reactor had a 5 cm internal diameter. Cygon and Sill varied impor- 36; 200
K p 0:329 exp 5
RT
tant reaction parameters such as temperature, supercial feed velocity,
hydrogen/STC ratio, bed height (through initial metallurgical silicon
The reaction quotient is dened by Eq. (6).
charge), copper catalyst addition, and the addition of hydrochloric
acid to the feed stream. Supercial velocity was varied from 4 to
p4SiHCl3 patm
12 cm/s. Temperature ranged between 500 and 600 C. Molar ratio of Qp 6
hydrogen to STC was varied from 0.67 to 4. HCl was added from 1.7 p3SiCl4 p2H2
to 6% on a total volume basis. Most experiments were executed
with a 400 g charge of metallurgical silicon resulting in a static bed Becker [7] theorized that the copper catalyzed hydrochlorination
height of 16.5 cm. Some tests decreased the metallurgical silicon to reaction occurs in parallel with the iron catalyzed reaction. The best
100 g resulting in a static bed height of 4.12 cm. A majority of exper- kinetic model for this reaction was also a power law with inhibition.
iments utilized metallurgical silicon with a mean Sauter diameter of
182 m although some tests used 278 m particles. Pure liquid STC xSiCl4;Cux Siy xH ;CuSiy

pSiCl4 pH22 Qp
was sent to an evaporator and thereafter mixed with hydrogen. The r Cux Siy aCu kCux Siy xSiHCl3 ;Cux Siy 1 7
pSiHCl3 Kp
short-term experiments of interest for this work used metallurgical
silicon, which was reacted for less than 10 h, representing about
10% silicon conversion. Again, the specic activity and reaction rate constant, the activation
Lehnen's [5] analytical chemistry work provided a foundation for the energy and the exponential factors were taken as free parameters and
kinetic data developed by Becker [7]. Lehnen showed that initially, met- adjusted to minimize the sum of square error between experimental
allurgical silicon is coated with a silicon dioxide (SiO2) layer that inhibits data and the kinetic model. The resulting parameters are shown in
reaction. The hydrochlorination reaction without catalyst addition oc- Table 2.
curred due to iron impurities in the silicon. When the reactor was heat- These reaction mechanisms are incorporated in a uidized-bed
ed, aluminum impurities were rapidly discharged as aluminum model in the next section.
trichloride (AlCl3) thereby exposing a greater surface area of catalytical-
ly active iron. Sill [6] reported that large areas in the surface on the sili- 2. Model development
con did not react without catalyst addition. Sill [6] also reported that
with copper catalyst the reaction proceeded at the surface in a struc- Levenspiel developed a hydrodynamic model for uidized bed reac-
tured orientation. He reported nding reaction sites typically found in tors with ne uidized particles [9]. In this model, the bubbles rise faster
the iron catalyzed reaction that produced holes in the silicon; however, than the emulsion and are surrounded by a cloud of particles that move
the surface reaction by copper dominated. Becker [7] suggested that the at the same rate as the bubble. In addition, a wake of solids is dragged
kinetic mechanism of the hydrochlorination reaction was a result of behind the bubble. The process gas in the bubble forms a vortex and is
catalytically active iron inherent to metallurgical silicon along with segregated from the gas in the cloud-wake and emulsion phase. Within
a parallel reaction with copper when it is added as a catalyst. Becker this gas are sparsely distributed solids where some reaction occurs. A
[7] developed separate kinetic models for the iron reaction and the modication of this model was developed for larger particles (Geldart
copper catalyzed reaction. He tested several kinetic models, and A and Geldart AB classication) by Kunii and Levenspiel [2]. In this
for each case chose a kinetic rate expression based on power laws model, the bubble and a cloud of solids around the bubble rise up rapid-
as this provided the least error when compared to Lehnen's [5] ly through an emulsion that is at minimum uidization conditions.
experimentation. Since the cloud volume is small, gas ow through the cloud is ignored.
Becker [7] proposed the following kinetic equation for the In addition, the emulsion phase is considered to be stagnant. The
hydrochlorination reaction as catalyzed by the iron impurities inherent model is composed of three phases: the bubble containing a small
to the silicon. amount of entrained solids, the cloud-wake of solids surrounding and
rising with the bubble, and the emulsion phase with solids at minimum
uidization conditions. These assumptions are valid when the super-
xSiCl; Fex Siy xH2 Fex Siy
 cial velocity over the minimum uidization velocity is much greater
pSiCl44 pH2 Qp
r Fex Siy aFe kFex Siy xSiHCL3 ;Fex Siy 1 4 than 1 [2].
pSiHCL3 Kp
u0
1 8
umf
Because the specic activity of the metallurgical silicon was un-
known, the initial product of the specic activity and the reaction rate The KuniiLevenspiel model is combined with the reaction kinetics
constant pre-exponential factor was interpreted as a free parameter. to give the following equation for a given reactant, intermediate,
In this factor, the activation energy and the exponential powers were byproduct, or product.
chosen to minimize the sum squared error of experimentation. The
values of the resulting parameters are shown in Table 1.



overall disappearance reaction transfer to
9
in bubble in bubble cloudwake

Table 1 Table 2
Parameters of the iron catalyzed kinetic equation. Parameters of the copper catalyzed kinetic equation.

aFe kFexSiy EA, FexSiy xSiCl4, FexSiy xH2, FexSiy xSiHCl3, FexSiy aCu kCuxSiy EA, CuxSiy xSiCl4, CuxSiy xH2, CuxSiy xSiHCl3, CuxSiy
mol1 kgcat-1 s-1 Pa-1003 J/mol mol1 kgcat-1 s-1 Pa-1086 J/mol

1.59810-1 120,073 2.521 0.501 2.019 1.387810-2 109,557 2.532 0.629 2.075
244 M. Colomb et al. / Powder Technology 292 (2016) 242250




transfer to reaction in transfer to it was occasionally reduced to 100 g. For particles with a mean Sauter di-
10
cloudwake cloudwake emulsion ameter of 182 m umf was set equal to 0.0276 m/s and for the 278 m
particles umf was set equal to 0.058 m/s. The void fraction at minimum
uidization conditions (mf) was calculated as per recommendations



transfer to reaction in in [2] and was set to 0.571. All of the parameters are physically reason-
11
emulsion emulsion able and can be predicted for reactors of various sizes and the model
may be integrated along the reactor height. The model in [9] suggests
The rate of disappearance of a gas component in the bubble phase is selection of a representative bubble diameter for the reactor. The taller
equal to the change due to reaction in the bubble phase plus the mass the reactor, the larger this representative bubble size becomes because,
transfer of that component from the bubble to cloud phase. Eqs. (10) as the bubbles rise through a bed, the bubbles expand in size and in-
and (11) are algebraic and solve for the mass transfer in the rate expres- crease in velocity. The lack of solids in the bubbles results in bypassing
sion. The rate of mass transfer from the bubble to the cloud equals the of the reactant gas thereby greatly decreasing conversion rate with
reaction in the cloud plus the mass transfer of the gas to the emulsion. vertical height.
Finally, the mass transfer to the emulsion equals the reaction in the The KuniiLevenspiel model presented in Eqs. (12), (13), and (14) is
emulsion. designed for the Geldart A and Geldart AB group particles. The reaction
For rst order kinetics of reactant gas (A) the model is solved with conditions of the hydrochlorination reaction may change the particle
the following equations. classication from the Geldart AB group to the Geldart B group. In this
reaction regime, the chemical concentrations in the cloud phase, con-
dC Ab dC verges to the solution of the emulsion phase and the KuniiLevenspiel
 u0 Ab f b K r C Ab K bc C Ab  C Ac 12
dt dz model is no longer valid. A modied KuniiLevenspiel model is devel-
oped in this section that is valid for the Geldart AB group and particles
K bc C Ab  C Ac f c K r C Ac K ce C Ac  C Ae 13
that have just transitioned into the Geldart B classication. This model
eliminates the reaction in the cloud-wake phase as shown in Eqs. (23)
K ce C Ac  C Ae f e K r C Ae 14
and (24).

The rise velocity of a single bubble is given by Eq. (15) as




overall disappearance reaction transfer to
recommended by Davidson and Harrison [15]. 23
in bubble in bubble emulsion
1=2


ubr 0:711gdb 15 transfer to reaction in
24
emulsion emulsion
The velocity of the bubbles in the bed is calculated from
The modied KuniiLevenspiel model was used for the hydrochlo-
ub u0 umf ubr : 16
rination reaction by including the kinetic equations provided by Becker
[7]. For the solution of the model, it was assumed that the reaction was
The fraction of the bed occupied by bubbles is isothermal and that gas density was constant. Molecular concentrations
u0  umf were converted to partial pressures by assuming the ideal gas law. For
: 17 the hydrochlorination reaction, the modied KuniiLevenspiel model
ub  umf
takes the following form:
The mass transfer rates are calculated from dC ib  
u0 f b Si vi Fe r Fex Siy Cu r Cux Siy K be C ib C ie 25

 1=2
! dz b
umf Di g 1=4
K bc 4:50 5:85 5=4
18  
db db K be C ib C ie f e Si vi Fe r Fex Siy Cu r Cux Siy : 26
e
!1=2
mf Di ubr The two model Eqs. (25) and (26) were solved simultaneously for
K ce 6:77 : 19
3
db reactant species hydrogen and STC, and the product chemical TCS. The
denition of fe was modied to the following form.
The volume of solids per volume of bed fraction for each phase is  
f e 1mf 1f b : 27
given by

f b 0:001  0:01 0:005 for this work 20 The mass transfer resistance Kbe was set equal to the mass transfer
resistance from the bubble to the cloud as recommended in [2] for

  3umf =mf their Geldart B model. In [2], it is recommended that the term for diffu-
f c 1mf 21 sion into the cloud phase be eliminated. This model retains the term for
ubr umf =mf
diffusion across the boundary of the bubble to the cloud phase originally
 
f e 1 mf 1 f c f b 22 derived by Davidson and Harrison [15]. Although chemical concen-
trations in the cloud phase are close or equal to the emulsion for these
Diffusion coefcients Di for each modeled chemical species were reaction conditions, the diffusion into a very thin cloud phase is still
estimated by the technique set forth in [10]. The supercial reactant required for an accurate model.
gas velocity (u0) was set by the experiment. The minimum uidization
 1=2
!
velocity was set according to the values found by Becker [7] and umf Di g 1=4
K be 4:50 5:85 5=4
28
changed based on the size of metallurgical silicon. The majority of ex- db db
periments performed by Cygon [3] and Sill [4] made use of metallurgical
grade silicon with a mean Sauter diameter of 182 m. Some of their ex- The solution of the KuniiLevenspiel model requires a bubble size to
periments were performed with silicon having a mean Sauter diameter be estimated [9]; however, no method to estimate a representative
of 278 m. The initial charge to the reactor was usually 400 g; however, bubble size is presented. In reality bubbles are formed at the distributor
 M. Colomb et al. / Powder Technology 292 (2016) 242250 245

Fig. 1. Effect of supercial velocity on bubble size.

Fig. 3. Reactant and product gas concentrations along vertical bed height at a supercial
velocity of 12 cm/s.

of a reactor and increase in size and velocity as they travel up the bed.
The bubble size correlation of Yasui et al., [11] is incorporated in the An initial bubble size estimation was investigated for the porous
modied KuniiLevenspiel model to model the increase in bubble size plate distributor of Cygon [3] and Sill [4]. Miwa et al. [12] proposed
along the bed height. Yasui and Johanson, [11] conducted the majority Eq. (31) for a porous plate distributor.
of their experiments in a 10 cm diameter reactor with particle sizes  2
ranging from 12 to 450 m. The experimental conditions correlate db0 0:376 u0 umf 31
well with the conditions of the experiments of Cygon [3] and Sill [4].
The following equation was proposed by Yasui and Johanson [11] to For a perforated plate distributor the following equation is
predict bubble diameter versus bed height. recommended in [2].

 0:63 1:30 u0 umf 0:4
u0 db0 32
db 0:16Si dSi 1 z 29 g 0:2 Nor
umf
The estimation of initial bubble size is not straightforward for
Yasui and Johanson [11] reported difculty in measuring bubbles the hydrochlorination reaction under consideration in this paper
less than 1.3 cm in diameter. Bubbles are usually smaller than this size [13]. Although Cygon [3] and Sill [4] used a porous distributor
above the distributor plate in laboratory scale reactors. Eq. (29) is zero plate, this work determined that the initial bubble size for these
at the distributor, which is not a realistic representation of reality. The
equation could be improved by the addition of an initial bubble size as
shown in Eq. (30).

 0:63
u0
db 0:16Si dSi 1 z db0 30
umf

Fig. 2. Reactant and product gas concentrations along vertical bed height at a supercial Fig. 4. Parity plot of laboratory hydrochlorination experimental yield vs predicted results
velocity of 4 cm/s. and no catalyst addition. Solid black lines show the boundary of 20% deviation.
246 M. Colomb et al. / Powder Technology 292 (2016) 242250
Fig. 5. Parity plot of laboratory hydrochlorination experimental yield vs predicted results
with Cu catalyst addition. Solid black lines show the boundary of 20% deviation.
experiments was best estimated with equation (32) and an orice
density of 20,000 m 2. The incorporation of the bubble size model
to the modied KuniiLevenspiel allows the parameters of , f e ,
and Kbe to change along the bed height as occurs in an actual uid-
ized bed.
For a subset of experiments performed by Cygon [3] and Sill [4],
hydrochloric acid was added to the feed stream. This addition was
only tested on particles with a mean Sauter diameter of 278 m. Ad-
justments for hydrochloric acid in the feed were made as per the
recommendations of Becker [7]. It is assumed that the hydrochloric
acid in the feed is quickly reacted by the chlorination reaction of
Eq. (1) to hydrogen and TCS without side reactions. The model
was therefore initialized with the actual inlet concentration of
STC, an additional amount of hydrogen according to the stoichio-
metric equation, and an initial concentration of TCS whose amount
was based on stoichiometry as well.
C H2 ;0 C H2 ;IN C HCl;IN =3
C SiHCl3 ;0 C HCl;IN =3
3. Results and discussion
The modied KuniiLevenspiel model for the hydrochlorination
process results in three ordinary differential equations coupled with
three algebraic expressions. This differential-algebraic system was
solved numerically in MATLAB. The predicted yield was calculated
from Eq. (35) and compared with the experimental results of Cygon
[3] and Sill [4].
3n_ SiHCl3 ;OUT  100
Y SiHCl3
4n_ SiCl4 ;IN n_ HCl;IN
Eq. (35) accounts for the TCS made directly by hydrochloric acid in
the feed stream. It is also adjusted by stoichiometric coefcients for a
good representation of the yield accomplished by the hydrochlorination
reaction. The experiments by Cygon [3] and Sill [4] show the yield as a
function of changing experimental conditions for the laboratory scale
uidized bed reactor for the hydrochlorination reaction. All experi-
ments were carried out at atmospheric pressure. Cygon and Sill used a
glass lined reactor with a 5 cm internal diameter. Cygon and Sill varied
important reaction parameters such as temperature, supercial feed
velocity, hydrogen/STC ratio, bed height (through initial metallurgical
silicon charge), copper catalyst addition, and addition of hydrochloric
acid to the feed stream. Most experiments were executed with a 400 g
charge of metallurgical silicon resulting in a static bed height of
16.5 cm. Some tests decreased the metallurgical silicon to 100 g
resulting in a static bed height of 4.12 cm. A majority of experiments
utilized metallurgical silicon with a mean Sauter diameter of 182 m
although some tests used 278 m particles.
Typical modeling results for the reactor were selected from the ex-
periments of Cygon [3] and Sill [4] to demonstrate the effect of super-
cial velocity on bubble size and changing chemical concentrations in the
bubble phase throughout the reactor. The selected experiments did not
include copper catalyst or HCl addition. The temperature was 550 C,
the molar ratio of hydrogen to STC was 2, and the particle size was
182 m. Supercial velocity was 4 and 12 cm/s. Fig. 1 shows the effect
of supercial velocity on the bubble size in the laboratory scale reactor.
Interestingly, Yasui and Johanson's, [11] model was one of the only
models available that predicted sufciently small bubbles for a laborato-
ry scale reactor. The bubble growth is nearly linear in the modeled
range. The velocities of 4 and 12 cm/s test the limits of a bubbling uid-
ized bed. At 4 cm/s the bed is in the slow bubble regime [2]. At 12 cm/s
the bed is in the regime of Geldart A particles appropriate for the origi-
nal KuniiLevenspiel model [2]. This work has predicted the yield from
the hydrochlorination reaction well in both regimes with the modied
model. Figs. 2 and 3 show the concentration of reactant and product
gases in the bubble phase along vertical reactor height.
Fig. 4 shows a comparison between model predictions and the ex-
perimental results of Cygon [3] and Sill [4] without additional catalyst.
This data set covers temperature ranges of 500600 C, supercial
feed rates of 4 to 12 cm/s, hydrogen/STC ratios of 0.67 to 4, and particles
sizes with mean Sauter diameters of 182 m and 278 m. All but one
predicted yield fell within 20% of the experimental yield. For tests
that Cygon [3] and Sill [4] repeated, the model prediction is graphically
compared with the average experimental yield. Percent difference is
33
34
35
Fig. 6. Parity plot of laboratory hydrochlorination experimental yield vs predicted results
with hydrochloric acid in the feed. Solid black lines show the boundary of 20% deviation.
 M. Colomb et al. / Powder Technology 292 (2016) 242250 247

Table 3
Model predictions for experiments by Cygon [3] and Sill [4] with no catalyst or HCl addition.

Temp. Molar ratio (H2/STC) xHCl (mol%) dp (m) WCu (wt.%) mSi u0 (cm1 s1) YTCS, EXP (mol%) YTCS, PRED (mol%) Difference %
( C) (g)

500 0.67 0 182 0 400 9 4.20 4.60 9.46

550 0.67 0 182 0 400 9 5.87 6.51 11.0
600 0.67 0 182 0 400 9 7.37 8.43 14.4
500 1 0 182 0 400 9 4.84 4.93 2.05
550 1 0 182 0 400 9 6.65 7.05 6.08
600 1 0 182 0 400 9 8.32 9.29 11.7
500 2 0 182 0 400 6 7.04 6.71 4.76
500 2 0 182 0 400 9 6.55 5.55 10.4
500 2 0 182 0 400 12 4.43 4.77 7.60
550 2 0 182 0 400 4 10.58 10.98 4.07
550 2 0 182 0 400 6 9.56 9.51 0.70
550 2 0 182 0 400 9 8.44 7.98 5.45
550 2 0 182 0 400 12 7.16 6.90 3.60
600 2 0 182 0 400 4 11.90 13.75 15.5
600 2 0 182 0 400 6 11.26 12.41 10.1
600 2 0 182 0 400 9 10.56 10.71 1.45
600 2 0 182 0 400 12 9.78 9.40 3.94
500 4 0 182 0 400 9 8.69 6.22 28.4
550 4 0 182 0 400 9 10.20 8.96 12.1
600 4 0 182 0 400 9 12.57 12.13 3.53
500 2 0 278 0 400 9 6.82 6.02 11.6
550 2 0 278 0 400 6 10.80 10.28 4.78
550 2 0 278 0 400 9 9.69 8.66 10.6
550 2 0 278 0 400 12 8.48 7.59 10.5
600 2 0 278 0 400 6 12.27 13.24 7.9
600 2 0 278 0 400 9 11.61 11.56 0.392
600 2 0 278 0 400 12 11.18 10.31 7.79

Table 4
Model predictions for experiments by Cygon [3] and Sill [4] with copper catalyst and no HCl addition.

Temp. Molar ratio xHCl (mol%) dp (m) WCu (wt.%) mSi (g) u0 (cm1 s1) YTCS, EXP (mol%) YTCS, PRED (mol%) Difference %
( C) (H2/STC)

500 2 0 182 0.9 400 9 8.24 8.66 5.10

550 2 0 182 0.9 400 6 14.18 12.79 9.78
550 2 0 182 0.9 400 9 12.32 11.40 7.47
550 2 0 182 0.9 400 12 9.91 10.18 2.73
600 2 0 182 0.9 400 6 14.39 14.65 1.82
600 2 0 182 0.9 400 9 13.64 13.72 0.626
600 2 0 182 0.9 400 12 12.38 12.63 2.03
550 2 0 182 4.5 400 6 14.38 14.01 2.55
550 2 0 182 4.5 400 9 14.44 13.52 6.37
550 2 0 182 4.5 400 12 13.26 12.79 3.52
600 2 0 182 4.5 400 6 14.55 15.07 3.58
600 2 0 182 4.5 400 9 14.35 14.87 3.61
600 2 0 182 4.5 400 12 14.68 14.43 1.67

Table 5
Model predictions for experiments by Cygon [3] and Sill [4] with HCl addition and no catalyst (fe = 0.0066)

Temp. Molar ratio xHCl dp WCu mSi u0 YTCS, EXP YTCS, PRED Difference
( C) (H2/STC) (mol%) (m) (wt%) (g) (cm1 s1) (mol%) (mol%) %

500 2 2 278 0 100 9 4.90 4.68 4.46

500 2 2 278 0 200 9 6.30 5.96 5.96
500 2 2 278 0 300 9 7.20 6.84 5.05
500 2 2 278 0 400 9 7.75 7.50 3.17
500 2 3 278 0 100 9 4.80 4.86 1.33
500 2 3 278 0 400 9 7.50 7.58 1.05
500 2 3.7 278 0 400 9 9.30 7.66 17.7
500 2 4 278 0 400 9 8.40 7.69 8.34
500 2 6 278 0 400 9 8.60 8.05 6.34
550 2 3.7 278 0 400 9 11.40 10.55 7.47
600 2 1.7 278 0 400 9 12.80 13.29 3.84
600 2 2 278 0 400 9 12.10 13.30 9.89
600 2 3 278 0 400 9 11.85 13.32 12.4
600 2 3.1 278 0 400 9 13.00 13.32 2.45
600 2 3.7 278 0 400 9 13.30 13.33 0.25
600 2 4 278 0 400 9 12.40 13.35 7.64
600 2 4.8 278 0 400 9 13.20 13.38 1.34
600 2 6 278 0 400 9 12.20 13.43 10.1
248 M. Colomb et al. / Powder Technology 292 (2016) 242250

all the model predictions show less than n 20% percent difference
from experimental values with an average percent difference of 6.0%.
Tables 3 through 5 list the experiments of Cygon [3] and Sill [4] along
with the predicted values from the modied KuniiLevenspiel model.

4. Conclusions

The modied KuniiLevenspiel model accurately predicted the no-

catalyst addition and copper catalyst addition experiments of the
hydrochlorination uidized bed. Only one prediction in this experimen-
tal subset had a percent difference greater than 20% from the experi-
mental value. The average percent difference from experimentation is
less than 6%. Upon adding hydrochloric acid to the feed stream, a
model adjustment to the modied KuniiLevenspiel model must be
made by assuming the catalytically active iron content in the metallur-
gical silicon doubles. This is a possibility as Becker [7] proposed that HCl
removes an unreactive silicon dioxide layer from the metallurgical sili-
con. Although this model, with the modication for hydrochloric acid,
seems to slightly over predict yield at 600 C, it is unlikely that an indus-
trial scale reactor would be run at such a high temperature [7].
The modied KuniiLevenspiel model is well suited for use in scale-
up calculations for an industrial reactor. All the parameters are physical-
ly reasonable and can be predicted for larger reactors. In industrial reac-
tors, the growth of bubbles result in conversion along the bed height
Fig. 7. Parity plot of laboratory hydrochlorination experimental yield vs predicted results that is highly non-linear. The modied KuniiLevenspiel model devel-
with HCl in the feed and twice the amount of catalytically active iron. Solid black lines oped in this paper uses a bubble size prediction correlation to capture
show the boundary of 20% deviation. this non-linear behavior. In addition, this prevents the need to know a
representative bubble size a priori. For industrial reactors, other bubble
calculated by taking the quantity of experimental yield minus predicted correlations such as Werther's equation [14] might be more appropriate
yield multiplied by 100 and then divided by the experimental yield. The as it was developed to encompass the behavior of bubbles in larger sized
average difference is calculated by taking the sum of the absolute per- reactors. The laboratory scale reactor of Cygon [3] and Sill [4] would
cent difference of the individual experiments and dividing by the num- have been fairly easy to maintain at isothermal conditions. For this
ber of experiments. The average difference between predicted and model to be useful in an industrial scale reactor, the temperature prole
experimental results is only 8.2% for this set of experiments. along the bed height must be known and incorporated into the calcula-
Next, the predictions of the modied KuniiLevenspiel model were tion of the kinetic factors. In the future, this model will be modied and
compared to experiments with copper catalyst addition. Cygon [3] and compared to industrial data. The modied KuniiLevenspiel model is a
Sill [4] executed hydrochlorination experiments with 0.9% and 4.5% by useful tool to capture the hydrochlorination reaction and it has the po-
weight copper to metallurgical silicon. The results are shown in Fig. 5. tential to model the inefciencies of scale-up from laboratory reactors to
For this experiment set, Cygon [3] and Sill [4] varied temperature from commercial reactors with their greater bed length and diameter
500600 C, supercial feed rates of 6 to 12 cm/s, hydrogen/STC ratio resulting in larger and faster moving bubbles.
was xed at 2 with a particle size of 182 m. The modied Kunii
Levenspiel model performed very well on this experimental subset
with all predictions showing less than 10% percent difference from ex-
perimental values. The average difference between predicted and ex-
perimental results is 3.9%, which is much lower than the results of the
uncatalyzed reaction.
Model predictions were compared to experiments with hydrochlo-
ric acid gas in the feed. Hydrochloric acid was added in varying amounts
from 26% by volume. No additional catalyst was added for these exper-
iments. Temperature ranged from 500600 C. Supercial feed rate was
set to 9 cm/s and hydrogen/STC ratio was set to 2. These experiments
were performed on silicon with a mean Sauter diameter of 278 m.
The results of the modied KuniiLevenspiel model are shown in Fig.
6. The reactor bed height was changed by charging metallurgical silicon
in a range of 100400 g. The modied KuniiLevenspiel model fared
poorly for experiments executed at 500 and 500 C. The results were
generally good for experiments run at 600 C. Becker [7] proposed
that hydrochloric acid removes an unreactive silicon dioxide coating
present on the surface of the charged metallurgical silicon. For this rea-
son, it is speculated that the fraction of catalytically active iron in the
metallurgical silicon available for reaction doubled when hydrochloric
acid was added to the feed gas. To model the proposed mechanism
that hydrochloric acid in the feed doubles the amount of catalytically
active iron available for reaction, the modied KuniiLevenspiel
model parameter for fraction of iron ( Fe) was doubled from Fig. 8. Phases and mass transfer in the bubbling uidized bed for the hydrochlorination
0.0033 to 0.0066. The results are shown in Fig. 7. It is observed that reaction.
 M. Colomb et al. / Powder Technology 292 (2016) 242250 249

Notation Levenspiel model has shown that the emulsion and cloud-wake, show
a specic activity nearly or exactly the same chemical concentration. For the
C component concentration, mol1 m3 hydrochlorination reaction, it has been assumed that the cloud
d diameter, m phase is very thin and differences in chemical concentration be-
D component diffusivity, m2 s1 tween the cloud and emulsion phase was considered to be negligi-
db bubble diameter, m ble. The model is developed from hydrodynamic considerations
db0 initial bubble diameter, m and material balances.
dt internal reactor diameter, m
EA activation energy, J1mol 1 A.1. Hydrodynamics of a uidized bed
f volume of solids per volume of bed
g gravitational constant, m1 s2 Considering the hydrodynamics of a bubbling uidized bed, Da-
k reaction rate constant, mol1 kg1
cat s
1
Pa1.003 for Fe vidson and Harrison [15] have derived the following equations for
mol1 kg1cat s
1
Pa1.086 for Cu single bubble rise velocity and average bubble rise velocity in the
K gas interchange coefcient, s1 bed.
Kp equilibrium constant
1=2
Kr reaction rate constant (rst order reaction), s1 ubr 0:711gdb A:1
n_ molar ow rate, mol1s1
Nor orice density, m2 ub u0 umf ubr A:2
p partial pressure, Pa
patm atmospheric pressure, Pa Assuming that the bed velocity of the emulsion phase is the same at
Qp reaction quotient minimum uidization conditions and that the upward ow of solids is
R ideal gas constant, J1 mol1 K1 the same as the downward ow, the velocity of the emulsion phase is
r reaction rate, mol1 s1 kg1cat calculated by the following equation.
T temperature, K
u0 supercial velocity, m1 s1 umf
ue A:3
ub average bubble velocity, m1 s1 mf
ubr single bubble rise, m1 s1
umf minimum uidization velocity, m1 s1 Here mf is the fraction of voids in the bed at minimum uidiza-
v stoichiometric coefcient tion conditions. This term increases the value of the emulsion
x exponential power phase velocity by accounting for the volume occupied by solids as
YSiHCl3 yield of TCS, see Eq. (35), % u mf is based on an empty reactor. Assuming negligible change in
z reactor height, m gas phase density, a material balance was performed around the
gas entering the reactor.
Greek letters Total Gas Flow Gas Flow in Bubbles Gas Flow in Emulstion
volume of wake over volume of bubble
bed fraction in bubbles u0 Arx ub Arx ue Arx mf 1
fraction of voids in the bed
A:4
fractional weight of catalyst
density, kg1 m3
By substituting Eq. (A.3) into Eq. (A.4) the term can be calcu-
lated which is required to convert bubble phase volumes to a reac-
Subscripts tor volume for the molar balance see Eq. (17). The resulting
A component A (rst order reaction) equation for is in agreement with [2] for reactions with thin
b bubble phase cloud phases.
c cloud phase The fraction of solid particles in each phase available for reaction
CuxSiy copper catalyzed reaction must be considered. Levenspiel has reported that the typical volume
e emulsion phase of solids in the bubble phase to the volume of solids in the bed is be-
FexSiy iron catalyzed reaction tween 0.001 and 0.01 see Eq. (20). The solids' fraction in the emul-
i component (STC, H2, TCS) sion is calculated by the rst two expressions in Eq. (27). Since the
IN actual feed stream solids in the cloud phase have been grouped into the emulsion
mf minimum uidization phase for this model, only the solids in bubble must be subtracted
mg metallurgical grade silicon from the solids fraction in the bed to determine the solids fraction
OUT product stream in the emulsion.
rx reactor
Si silicon A.2. Molar balance
w wake phase
0 model initialization value The mass transfer of the reactant gas from the bubble phase to the
emulsion must be considered. A bubbling uidized bed is unique in
Appendix A. Derivation of the modied KuniiLevenspiel that mass transfer of the gas to the individual solids is typically
model for BFB neglected as it is negligible compared to the mass transfer resistance
between phases. For the molar balance on the bubbling uidized bed
The modied KuniiLevenspiel model was developed from rst reactor, it is assumed that a majority of the feed gas passes through
principles. The residence time of the gas in the reactor was calculated the reactor in the bubble phase. The chemical constituents also
assuming that the bed was in the bubbling uidized bed regime. Since pass into the emulsion phase by convective and diffusive mass trans-
u0 umf and the solid phase was found in the Geldart AB to Geldart B fer, see Fig. 8. Eq. (28) is used as developed in [2] for calculation
regime this was a good assumption. Modeling with the full Kunii of the resistance to mass transfer from the bubble phase to the
250 M. Colomb et al. / Powder Technology 292 (2016) 242250

emulsion phase. A molar balance on the bubble phase yields the fol-
lowing equation for a chemical constituent denoted A.
Flow in Flow outGenerationOut by Mass Transfer
0 fub Arx C Ab jz ub Arx C Ab jzz r b Arx z f b K be C Ab C Ae Arx z
0g
The equation is divided through by Arxz and the limit is taken as
z 0. The following equation results for the hydrochlorination reac-
tion.
dC ib  
ub f b Si vi Fe r Fex Siy Cu r Cux Siy K be C ib C ie
dz b [9] O. Levenspiel, Chemical Reaction Engineering, third ed. John Wiley & Sons, Hoboken,
For this equation the term ub is replaced by u0 as recommended
in [9] resulting in Eq. (25).
A mass balance is also performed on the emulsion phase.
Flow inFlow outGenerationOut by Mass Transfer
0 00r e Arx z f e K be C Ab C Ae Arx z 0g
The equation is divided through by Arxz and results in Eq. (26) for
the hydrochlorination reaction.
[2] D. Kunii, O. Levenspiel, Fluidization Engineering, second ed. Butterworth-
Heinemann, Newton, MA, 1991.
[3] P. Cygon, Reaction Studies for the Hydrochlorination of Silicon in a Laboratory Fluid-
ized Bed System(Diss.) U of Cologne, 1999.
[4] T. Sill, S. Buchholz, R. Weber, L. Mleczko, Thermodynamic, Mechanistic and Reaction
Engineering Aspects of Hydrochlorination of Silicon, Silicon for the Chemical Indus-
try V, Trondheim, Norway, 2000.
[5] R. Lehnen, Studies on the hydrochlorination of metallurgical silicon with SiCl4 and
A:5 H2 in a laboratory xed bed reactor, Final Report on the Project High-purity Silicon
for Photovoltaic Solar Energy Systems.", Bayer-AG, 2000.
[6] T. Sill, Studies of the Hydrochlorination of Silicon to Trichlorosilane in a Fluidized
Bed Reactor(Diss.) University of Bochum, 2001.
[7] F. Becker, Modeling and Simulations of Hydrochlorination of Silicon to
Trichlorosilane for the Development of a Technical Fluidized Bed Reactor(Diss.)
RWTH Aachen U, 2005.
[8] J. Mui, Investigation of the hydrochlorination of SiCl4, Final Report, JPL Contract
A:6 956061, 1983.
NJ, 1999.
[10] B.E. Poling, J.M. Prausnitz, J.P. O'Connell, The Properties of Gases and Liquids, fth ed.
McGraw-Hill, New York, NY, 2001.
[11] G. Yasui, L.N. Johanson, Characteristics of gas pockets in uidized beds, AIChE J 4 (4)
(1958) 445452.
[12] K. Miwa, S. Mori, T. Kato, I. Muchi, Behavior of bubbles in gaseous uidized beds, Int.
Chem. Eng. 12 (1972) 181.
[13] S. Mori, C.Y. Wen, Estimation of bubble diameter in gaseous uidized beds, AIChE J
A:7
21 (1) (1975) 109115.
[14] J. Werther, Scale-up modeling for uidized bed reactors, Chem. Eng. Sci. 47 (911)
(1992) 24572462.
[15] J.F. Davidson, D. Harrison, Fluidized Particles, Cambridge University Press, 1963.

References
[1] J. Mui, D. Seyferth, The hydrogenation of SiCl4, Final Report, JPL Contract 955382,
1981.