Fluidized Bed Process with Silane

Limin Jiang, Benjamin F. Fieselmann, Liguo Chen, and David Mixon

Abstract
Silane-based fluidized-bed technology is used to produce polysilicon for solar
cells by decomposing silane onto silicon particles suspended in a heated stream of
silane and hydrogen. Silane-based fluidized-bed reactors potentially provide a
lower cost method to produce polysilicon than the current Siemens reactors that
dominate the silicon market. Production of silicon in a fluidized bed requires
80–90% less electrical energy than the currently favored Siemens process and
converts a batch process into a more economical continuous process. The spher-
ical granular silicon product from fluidized-bed reactors is preferred to the
polysilicon rods produced by the Siemens process for downstream processing.
Production of silicon by fluidized beds has been carried on for over 20 years, but
the simpler Siemens process has dominated polysilicon production because of the
high purity of its polysilicon product and the availability of low-cost electricity.
The economics of the silane-based fluidized-bed technology has improved sig-
nificantly due to advances in reactor design, process modeling, and operational
experience. Fluidized-bed technology is the leading candidate to eventually
provide less expensive polysilicon for solar cells.

L. Jiang
GCL Solar Energy Trading Limited, Suzhou, Hong Kong
Jiangsu Zhongneng Polysilicon Technology Development Co., Ltd., (GCL), Xuzhou, Jiangsu,
PRC, China
e-mail: jianglimin@gclsolarenergy.com
B.F. Fieselmann (*) • L. Chen • D. Mixon
GCL Technology Research Center, LLC, Princeton, NJ, USA
e-mail: benfieselmann@gclsolarenergy.com; liguochen@gclsolarenergy.com;
davidmixon@gclsolarenergy.com

# Springer-Verlag GmbH Germany 2017 1

D. Yang (ed.), Handbook of Photovoltaic Silicon,
https://doi.org/10.1007/978-3-662-52735-1_5-1
2 L. Jiang et al.

Keywords
Silicon fluidized-bed • Silicon production with silane • Silane-based fluidized-
bed • Fluidized bed and Siemens silicon • FBR polysilicon • Granular silicon

Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Commercial Production of Polysilicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Fluidized-Bed Process for Producing Silicon from Silane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Background on Fluidized-Bed Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Industrial Applications of Fluidized-Bed Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Applying Fluidized-Bed Technology to the Production of Polysilicon . . . . . . . . . . . . . . . . . . . . . 11
Commercial Production of Silane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Total Silane-Based Process for Converting MGS to Granular Polysilicon . . . . . . . . . . . . . . . . . . 12
Siemens Process for Producing Polysilicon Rods from Trichlorosilane . . . . . . . . . . . . . . . . . . . . . 13
Comparison of the Fluidized-Bed Process and the Siemens Process for Making Polysilicon 15
Process of Polycrystalline Silicon Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Downside to the Use of Silane-Based Fluidize Bed Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Alternate Technologies for the Production of Polysilicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Alternate Approaches to Polysilicon Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Silane Reactions in a Fluidized-Bed Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Chemical Reaction of Silane in a Fluidized-Bed Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Homogeneous, Heterogenous, and Scavenging Reactions of Silicon Particles . . . . . . . . . . . . . 19
Modeling of Silane Fluidized-Bed Reactors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Operation of a Silane-Based Fluidized-Bed Reactor to Make Silicon . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Feed Gas to the Fluidized-Bed Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Preheating the Feed Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Distributor and Jet Nozzles to Introduce the Feed Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Different Designs for Heating Silane-Based Fluidized-Bed Reactors . . . . . . . . . . . . . . . . . . . . . . . 29
Reactor Dimensions, Pressure, and Silicon Production Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Agglomeration of Silicon Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Control of the Silicon Particle Size Distribution in the Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Factors to Consider When Optimizing Silicon Particle Size Distribution . . . . . . . . . . . . . . . . . . . 33
Silicon Powder Formation and Handling in a Fluidized-Bed Reactor . . . . . . . . . . . . . . . . . . . . . . . 33
Removing the Silicon Deposited on the Reactor Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Safety Issue with Silane and Hydrogen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Atmospheric Contamination of Silicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Current Status of Commercial Research and Development of Silane Fluidized-Bed Reactors 35
Specifications for Polycrystalline Silicon for PV Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Improving FB Technology with Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Introduction

The market price for purified polysilicon for solar applications has reached a critical
point such that new technology is needed to further reduce the production costs for
polysilicon. The price of polysilicon has decreased significantly over the last decade
with over building of Siemens production capacity (Bernreuter 2016). Further major
cost reductions using the current Siemens process are unlikely. In order for the price
Fluidized Bed Process with Silane 3

of electricity from solar cells to achieve grid parity with other sources of electricity in
more markets, the price of polysilicon must continue to drop. Silane (SiH4)-based
fluidized-bed (FB) technology is now the leading candidate to provide a significant
reduction in the price of polysilicon chiefly because FB reactors use 10–20% as
much electrical power as a Siemens reactor.
Fluidized-bed technology using silane has been applied to silicon preparation
since the early 1980s by Ethyl Corporation and MEMC (now SunEdison) (Roselund
2015). REC has produced FB silicon for many years (REC 2017). Despite the
decades of commercial experience, the silane-based fluidized-bed technology has
failed to achieve its expected production efficiencies and anticipated cost savings.
Currently silane-based fluidized-bed technology accounts for less than 10% of world
production of polysilicon (IHS 2014). The development of the silane-based fluid-
ized-bed process for silicon production still remains a major engineering challenge
to reach its full potential.
This article reviews the state of technology for polysilicon production from
silane-based fluidized-bed reactors. The presentation is industrially oriented with
an emphasis on the practical issues that are encountered when operating a fluidized-
bed reactor with silane. Much of the research and development on silane-based
fluidized-bed reactors has been carried out by industry. The result is that there are
few published articles on polysilicon production technology by those working with
the technology. There are a large number of patents on the subject and a few third-
party news reports. Two useful reviews on various aspects of silicon production
using silane in a fluidized bed are the articles by Filtvedt et al. (2010) and Jianlong
et al. (2011).
This review also provides a brief background on the current Siemens technology
for producing polysilicon. This article outlines why the Siemens process is currently
the preferred method for polysilicon production and presents the engineering chal-
lenges that must be overcome before FB technology can reach the point where it can
replace the Siemens process.

Commercial Production of Polysilicon

As the demand for polysilicon for photovoltaic applications increases, there has been
a search for more economical methods for producing polysilicon. Many competing
technologies have been examined over the last 50 years, but the Siemens process has
emerged as the most cost-effective method for producing polysilicon, despite its
many deficiencies. Of all the alternative methods for polysilicon production, the
silane fluidized-bed process has shown the greatest potential for replacing the
Siemens process (Filtvedt et al. 2010).
In order to understand how FB technology could eventually become the dominant
method for polysilicon production, one needs to understand how polysilicon is
currently produced. Although the Siemens process and fluidized-bed technology
appear to be different, they both produce high purity polysilicon and deal with many
4 L. Jiang et al.

of the same technical challenges. The chemistry of the two methods is shown in the
chart below:

Most of the commercial silicon purification methods start with metallurgical

grade silicon (MGS) (Mazumder 2000). High purity silicon dioxide ore is reduced
with carbon in an electric arc furnace at 1900
C as follows:

SiO2 þ C ! Si þ CO2 (3)

Liquid silicon (melting point 1414
C) is drained out of the furnace and collected.
This arc furnace process produces metallurgical grade silicon that is
98.5–99.5% pure.
For photovoltaic and semiconductor applications, further purification of the
silicon is needed to reduce impurity levels to the part per million (ppm) or to the
part per billion (ppb) by mass level. (These levels of impurity are referred to as 6 N
(99.9999%) or 9 N (99.9999999%), respectively.) Roughly photovoltaic grade
silicon must be at or purer than the 6 N level. Semiconductor grade silicon is
typically in the 9 N or better range of purity. For photovoltaic applications, the
highest purity of polysilicon possible is favored because it yields more efficient solar
cells. Industrially, one is continually faced with the tradeoff between producing
higher efficient solar cells with the increased cost of higher purity polysilicon.
The most common method for purification of silicon requires conversion of the
MGS to trichlorosilane (HSiCl3). Purification occurs by two sequential steps. Under
reaction conditions to produce HSiCl3 from silicon most other elements present as
impurities are removed because others elements do not form volatile compounds
under the conditions where trichlorosilane is formed. Secondly, the HSiCl3 that is
produced can be further purified by distillation to remove residual impurities with
different vapor pressures.
The highly purified trichlorosilane can be directly converted back to polysilicon
by the Siemens process, or it can be reacted to form silane. The silane can be
thermally decomposed to polysilicon by a FB process. In both cases, metallurgical
grade silicon forms a volatile compound or compounds which are purified and then
decomposed back to silicon for further processing. This conversion through multiple
Fluidized Bed Process with Silane 5

Fig. 1 Simplified basic fluidized-bed reactor with silicon particles suspended in a stream of silane
and hydrogen. The reactor is heated to 650–850
C so that the silane decomposes and deposits
silicon onto the suspended particles. Over time larger particles of silicon are removed as product and
smaller silicon particles are added to maintain the particle size distribution of the suspended silicon
particles. Hydrogen, unreacted silane, and silicon dust are removed from the top of the reactor

steps makes possible the production of ultrapure silicon necessary for photovoltaic
and semiconductor applications. In the purification process to form polysilicon the
critical steps are the conversion of either trichlorosilane or silane back to silicon.
In the following sections the operation of a fluidized-bed process with silane to
form polysilicon will be presented. The use of trichlorosilane in the Siemens process
will be presented later for contrast.

Fluidized-Bed Process for Producing Silicon from Silane

Fluidized-bed technology works well with silane as the feed gas to make silicon. The
FB reaction involves the deposition of silicon from silane onto small, heated
particles of silicon that are suspended in a flow of feed gas as shown in Fig. 1.
When describing the environment in a fluidized-bed reactor, it has become
common to refer to the bubble phase and the emulsion phase. The bubble phase is
the gas contained within a bubble. For a silane-based reactor the emulsion phase
refers to the dense mixture of silicon particles and the surrounding gas.
6 L. Jiang et al.

The reactor presented above is a simple type of fluidized bed presented as an

introduction. Later in this review, more modern and sophisticated reactors will be
discussed. In the simple model, the reactor is heated through the walls. The feed gas
enters the reactor from the bottom with sufficient flow gas to fluidize the silicon
particles. During the course of the fluidized-bed operation, small seed particles of
silicon are periodically added and larger particles are removed permitting continuous
production of silicon. The silane decomposes on the silicon particles and deposits
silicon.

Background on Fluidized-Bed Engineering

It is worthwhile to cover some of the basics of fluidized-bed engineering since they

are vital to the understanding of the process that produces silicon in a fluidized bed.
For a complete discussion of fluidized-bed engineering, please refer to the books by
Kunii and Levenspiel (1991) and by Yang (2003).

Minimum Fluidization Gas Velocity

The minimum fluidization velocity is the gas superficial fluid velocity at the onset of
fluidization during which the drag force on the particles by the upward moving fluid
is balanced by the weight of the particles. The minimum fluidization velocity can
normally be calculated by the Ergun equation as below
 


1:75 2 150 1  emf d3p ρg ρs  ρg g
Re þ Remf ¼ (4)
e3mf ϕs mf e3mf ϕ2s μ2

and

ρg umf dp
Remf ¼ , (5)
μ
where emf is void fraction at minimum fluidizations, φs is the sphericity of the
particles, ρs is the particle density, ρg is the gas density, dp is the diameter of the
particle, μ is the viscosity of the gas and g is the acceleration of gravity.

Minimum Bubbling Gas Velocity

The minimum bubbling velocity is the superficial fluid velocity at which bubbles
start to form. This bubbling behavior normally occurs after the onset of the fluidi-
zation, so the minimum bubbling velocity should be at least equal to or larger than
the minimum fluidization velocity.
Fluidized Bed Process with Silane 7

Fig. 2 Geldart classification of particles for fluidization based on particle size and density. Geldart
B and D group are those encountered in a silane-based fluidized-bed reactor. The approximate
location of the silicon particles is noted in the chart (Geldart 1986)

Geldart Classification of Particles

By studying the fluidization behavior of different solids, Geldart (1986), classified
them into four distinguishable groups with clearly recognizable fluidization behav-
ior. The classification is described as follows:

• Geldart C: The particles are small (<100 microns) and cohesive. They are very
difficult to fluidize because the interparticle forces are large and the particles tend
to agglomerate together.
• Geldart A: The particles fluidize easily. They exhibit smooth fluidization behavior
after minimum fluidization velocity. They have controlled bubbling behavior with
recognizable minimum bubbling velocity.
• Geldart B: The particles fluidize well with bubbling action and gas bubbles appear
right after minimum fluidization velocity.
• Geldart B and D: The particles are very hard to fluidize, since they normally are
large, dense solids. Stable spouted bed can be formed for this type of solids.

Geldart’s classification has since become the standard and can be readily
displayed by Fig. 2:

Bed Pressure Drop

Once the bed reaches the minimum fluidization state, the pressure drop across the
bed is equal to the weight of the bed per unit area. Therefore,
8 L. Jiang et al.

Wg
ΔP ¼ , (6)
A
where ΔP is the pressure drop across the bed, Wg is the total weight of the bed, A is
the cross-sectional area of the bed.
We can conclude that once the bed is in the fluidization state, varying the gas flow
rate will not change the total pressure drop across the bed as long as the total weight
of the particles is fixed. The pressure drop across the bed can also be formulated as
follows:

ΔP  
¼ ð1  eÞ ρs  ρg g, (7)
H
where H is the bed height, and e is the void fraction of the bed. For gas fluidized bed,
since ρs ρg, so

ΔP
¼ ð1  eÞρs g: (8)
H
The fluidized bed density ρb = (1  e)ρs can be estimated from the bed pressure
drop

ΔP
ρb ¼ : (9)
gH

Slugging Beds
In fluidized beds, with increasing gas velocity the bubbles can increase in size to the
bed diameter and form slugging beds. The gas slugs rise at regular intervals and
cause large pressure fluctuations inside the beds. The pressure fluctuation inside the
bed can result in sever vibration of the reactor and its associated piping, hence
causing damage. Slugging should be avoided in those cases if possible. The follow-
ing correlation (Baeyens and Geldart 1974) can be used to calculate the minimum
slugging velocity Ums
pffiffiffiffiffiffi
U ms ¼ U mf þ 0:07 gD, (10)

where D is the reactor diameter.

Heat Transfer in Fluidized Beds

One of the major advantages of using fluidized-bed reactors is its capacity for heat
transfer features. Due to the rapid mixing of the fluidized particles, the entire bed is
nearly isothermal. Since hot spots can be minimized, the process operations can be
well controlled.
Fluidized Bed Process with Silane 9

Heat transfer between particles and gas in a fluidized bed is very fast and they can
be regarded as being at the same temperature in most cases. In rare transient cases,
the following correlation (Yang 2003) can be used to calculate the heat transfer
coefficient:
For 0.1  Rep  50

hp d p
Nup ¼ ¼ 0:0282Re1:4 0:33
p Prg (11)
kg

For 50  Rep  104

hp d p
Nup ¼ ¼ 0:0282Re1:4 0:33
p Prg , (12)
kg

where hp is the heat transfer coefficient between the particles and the gas, dp is the
particle diameter, kg is the gas thermal conductivity, and Prg is the Prandtl number of
the gas.
Rep is the particle Reynolds number

ρg d p u0
Rep ¼ : (13)
μg

There are many correlations in the literature for calculating the wall-to-bed heat
transfer coefficient hw. Leva’s formula (Yang 2003) has been used extensively for
large particle bubbling fluidized beds

hw d p
Nuw ¼ ¼ 0:525Re0:75
p : (14)
kg

Heat transfer between the fluidized bed and a gas feeding distributor plate has
been studied by Zhang and Quyang (1985):
!0:4 !0:4
hd d p ρ Cps d p ρg u0
Nud ¼ ¼ 0:09 s : (15)
kg ρg Cpg μg

With the condition,

d p ρg u0
0:3 < < 40, (16)
μg

where μg is the gas viscosity, u0 is the superficial gas velocity, Cps is the specific heat
of the solid, Cpg is the specific heat of the gas.
10 L. Jiang et al.

Industrial Applications of Fluidized-Bed Technology

There are a wide range of commercial processes that use fluidized-bed technology in
multiphase reactions where solids are reacted with gases. The basic concept of
fluidized-bed technology is that solids are suspended in a high velocity stream of
gas. A major advantage of fluidized-bed technology is the good heat and mass
transfer between materials in the solid phase and gas phase. Good heat transfer is
especially useful with very exothermic reactions where over heating can result in
“hot spots” in a fixed bed reactor. A second major advantage with fluidized-bed
reactors is that the processes can be made continuous by adding and removing solid
and gaseous reactants from the reactor.
There are several negative aspects to the use of FB technology. Fluidized-bed
reactors can be challenging to control. Parameters such as feed gas make up,
temperature, flow rate, and particle size have to be held within a small range of
settings. The bed level must be carefully monitored. The possibilities of collapsing
the bed or elutriating bed particles out of the top of the reactor continually exist.
Temperature, pressure, concentration, and many other parameters vary throughout
the reactor at any given moment. Understanding what is going on in a FB reactor is
difficult. Reaction process modeling may be required to interpret the behavior in a
reactor and to correctly respond to changing reactor conditions.
Scalability is difficult to achieve with a fluidized bed (Howard 1989). It is not
easy to predict the conditions in the reactor as the diameter and height of the unit
increase. As the diameter of the reactor increases the bed volume increases faster
than the wall heating surface area.
Fluidized beds are used extensively in the petrochemical and energy industries.
Catalytic cracking of high molecular weight petroleum is accomplished by passing
gaseous hydrocarbons over a suspended bed of catalyst. Catalytic reforming is
another process where fluidized-bed technology is used to upgrade petroleum.
In many fluidized-bed processes the contents of the reactor bed either do not
change, as when catalyst particles make up the bed, or decrease over time, as when
the suspended particles are consumed in a reaction. In contrast, in the silane-based
FB process the particles grow in diameter as silicon is deposited. The industrial
process that is superficially similar to the fluidized-bed deposition of silane on silicon
particles is the growth of polymer beads. The growth of silicon particles and the
growth of polymer beads in a FB reactor are similar in that the suspended particles
increase in size as the process is operated. Polymer beads are introduced into the
fluidized-bed reactor. The product beads are suspended and grow as monomer vapor
is added. The product beads are periodically screened and the larger beads are
removed and the small beads are returned to the reactor. Fluidized-bed technology
is used with ethylene, propylene, styrene, and other monomer feedstock to generate
flowable beads that can later be processed into plastics.
Fluidized beds are widely used in the silicon industry in many processes. Many
reactions involving silicon tend to be very exothermic and make use of the excellent
heat transfer characteristics of fluidized-bed reactors. The preparation of silicones
depends on the highly efficient production of the monomer dimethyldichlorosilane
Fluidized Bed Process with Silane 11

(Si(CH3)2Cl2) by the reaction of methyl chloride with silicon in a fluidized-bed

reactor. The reaction is referred as the Direct Process. Particles of silicon and a
copper catalyst are suspended in a stream of methyl chloride to carry out the
following high yield reaction:

2 CH3 Cl þ Si ! SiðCH3 Þ2 Cl2 : (17)

Some of the trichlorosilane used in the silicon industry is prepared in fluidized

beds using the following highly exothermic direct chlorination process:

Si þ 3 HCl ! HSiCl3 þ H2 : (18)

It is not surprising that the fluidized-bed hydrochlorination process reaction

shown below was developed to convert the Siemens byproduct silicon tetrachloride
back to trichlorosilane, for recycle back to the Siemens process and for the produc-
tion of silane.

Si þ 2 H2 þ 3 SiCl4 ! 4 HSiCl3 (19)

It is clear that the silicon industry makes use of fluidized-bed technology to carry
out many of the key reactions. Most of the silicon-based reactions that use a
fluidized-bed technology have been optimized and are routinely run to prepare
numerous chlorosilane and organosilane compounds. Despite decades of experience
optimizing and working successfully with silicon-based fluidized beds to make
chlorosilanes and organosilicones, it has remained a challenge to achieve the same
level of success with the deposition reaction to produce silicon. The potential
benefits of fluidized beds such as high selectivity, good yield, and increased
throughput have been achieved with many silicon and organosilicon reactions.
Difficulties remain in optimizing the process to deposit silane on fluidizedsilicon
particles. The next section presents the specifics of how FB technology is applied to
the case of silane deposition on suspended silicon particles.

Applying Fluidized-Bed Technology to the Production of Polysilicon

Polysilicon is produced in a fluidized bed by suspending a bed of granular silicon

particles in a flow of silane bearing gas that is usually diluted with hydrogen. The hot
reactor walls heat the silicon particles which then transfer heat to the silane and
hydrogen feed gases. The silane is decomposed in the FB reactor and deposits silicon
on the granular silicon particles. As indicated in Fig. 1, bubbles of feed gas can form.
The silane in the bubbles may have little opportunity to react with the silicon
particles. The rapidly moving and often violent actions of the silicon particles in
the bed lead to a wide range of differing conditions in the reactor over time.
The overall chemical reaction behind the silane-based FB reactor is the following
thermal decomposition:
12 L. Jiang et al.

n SiH4 ! Si þ Six þ 2n H2 : (20)

Deposited Si Si Dust

This reaction can go to completion and is not limited by equilibrium (Walch and
Dateo 2001). The silane decomposition reaction yields two main silicon products.
The desired process is the deposition of silicon onto the fluidized particles in the
emulsion phase to increase the size of the polysilicon granules. The silane also
decomposes to form fine silicon dust by the homogeneous thermal decomposition in
the bubble phase.

Commercial Production of Silane

The most economical method for silane production for use in a FB uses dispropor-
tionation of trichlorosilane (Breneman 1987, 2016). In this well-established route
trichlorosilane is converted to silane by using staged amine catalyzed reactions and
multiple distillations. In silane plants anion exchange styrene resins with amine
functional groups are used as the disproportionation catalyst, and the chlorosilanes
in the liquid phase are passed through several heated resin beds, equilibrated, and
distilled. The net reaction to produce silane is the following:

4 HSiCl3 ! SiH4 þ 3 SiCl4 : (21)

In the disproportionation process, there is no net stoichiometric consumption of

silicon or any element. There is only a rearrangement of hydrogen and chloride
bonded to silicon.

Total Silane-Based Process for Converting MGS to Granular

Polysilicon

In order to make the silane-based fluidized-bed process possible, three chemical

units described below are required. The first two reactions are well understood
equilibrium processes that have been optimized for high performance. The third
reaction used to deposit silane on silicon in a FB reactor is just the final step in the
overall process to produce a highly purified granular silicon product.

1. Metallurgical grade silicon (MGS) is converted to trichlorosilane by

hydrochlorination with silicon tetrachloride and hydrogen

Si þ 3 SiCl4 þ 2 H2 ! 4 HSiCl4 : (22)

2. Trichlorosilane is disproportionated to silane and silicon tetrachloride

Fluidized Bed Process with Silane 13

Fig. 3 Schematic diagram of a Siemens reactor showing how silicon is deposited via thermal
deposition of trichlorosilane on heated, shaped silicon rods. The reaction is run in a bell jar where
the reactor walls are water cooled to prevent silicon deposition

4 HSiCl3 ! SiH4 þ 3 SiCl4 : (23)

3. Silane is deposited in a fluidized bed onto granular silicon seed particles and
releases hydrogen

SiH4 ! Si þ 2 H2 :

When the three chemical reactions are added, the net result is the following:

Si ðMGS, 98:5-99:5%pureÞ ! Si ðpolysilicon, 6N to 9N purityÞ: (24)

There is theoretically no net consumption of hydrogen and chloride in the three-

step process. In reality small amounts of hydrogen (as H2) and chlorine (as HSiCl3 or
SiCl4) must be added to replace process losses due to required purification steps.
To understand how silane-based fluidized-bed technology can compete with the
Siemens process, it is important to understand how the Siemens process operates.
Both technologies produce polysilicon and have many characteristics in common.

Siemens Process for Producing Polysilicon Rods from Trichlorosilane

The Siemens process depends on converting metallurgical grade silicon to tri-

chlorosilane (HSiCl3). After one has obtained highly purified trichlorosilane by
distillation, the final purification step is the conversion to polysilicon by thermal
decomposition.

The Siemens Process

Siemens Process deposits trichlorosilane onto a seed rod of polysilicon via the
following reaction (H. Gutschel 1962):
14 L. Jiang et al.

HSiCl3 ! Si þ HCl þ SiCl4 þ H2 : (25)

Equation 25 is not a balanced chemical equation but instead represents an

equilibrium. This Siemens reaction is a chemical equilibrium process that does not
go to completion but produces a number of compounds. The reaction produces
hydrogen, silicon tetrachloride, and hydrogen chloride. Depending on reaction
conditions the Siemens process typically deposits less than 25% of the tri-
chlorosilane feed in a given pass through the reactor (Filtvedt et al. 2012). Extensive
recycling of feed gases is required during the operation of the Siemens process. The
Siemens reactor is schematically shown in Fig. 3.
The Siemens process involves feeding a gas mixture of hydrogen and tri-
chlorosilane into a bell jar-shaped chamber containing U-shaped silicon rods and
depositing silicon on the hot silicon surfaces. An electric current is passed through
the rods to bring the silicon rods to a temperature near 1150 C. The trichlorosilane is
heated to the reaction temperature as it approaches the hot silicon rods. The tri-
chlorosilane reacts on the hot silicon surface to deposit silicon and to release silicon
tetrachloride, hydrogen, and unreacted trichlorosilane. The diameter of the rods
increases as silicon is deposited.
The Siemens process has several operational difficulties that were of little concern
when the price of polysilicon was high and the price of electricity to power the
Siemens reactors was low. As a cold wall reactor process, much of the heat generated
to deposit silicon is removed by cooling of the reactor walls resulting in low energy
efficiency. Furthermore, the Siemens process is a batch process that must be shut
down in each production cycle when the rods have grown sufficiently large. The
thick product rods are replaced with new seed rods and the deposition process is
repeated.
The success of the Siemens reactors has been largely credited to the decoupling of
the heating of the silicon rods from the heating of the feed gas (Filtvedt et al. 2010).
The feed gas is only heated when the trichlorosilane and hydrogen are near the hot
silicon rods where the deposition reaction occurs. The walls of the Siemens reactor
are vigorously cooled to prevent silicon deposition on the walls and to hold down the
temperature of the feed gas throughout the bell jar except near the heated
silicon rods.
The major advantage of the Siemens process is that it is a mature technology that
can produce high purity silicon (often greater than 9 N). The Siemens process is
relatively easy to operate and has become very reliable. Siemens reactors are run
worldwide by many companies. The Siemens process is made commercially viable
by a separate fluidized-bed hydrochlorination process shown in the summary below.
In the hydrochlorination process metallurgical grade silicon is reacted with the
Siemens by-product silicon tetrachloride (SiCl4) and hydrogen to produce tri-
chlorosilane feed for the Siemens reactor.
Almost all commercial Siemens reactors use trichlorosilane as the feed material.
The process produces very little silicon powder, as the reaction intermediate hydro-
gen chloride reacts with silicon powder to produce more trichlorosilane according to
the following reaction:
Fluidized Bed Process with Silane 15

Si þ 3 HCl ! HSiCl3 þ H2 : (26)

In summary, the chemistry involved in the Siemens process is based on the

following two reactions:

1. Metallurgical grade silicon, silicon tetrachloride, and hydrogen are converted to

trichlorosilane in a heated hydrochlorination equilibrium reaction

Si þ 3 SiCl4 þ 2 H2 ! 4 HSiCl3 : (27)

2. In a Siemens reactor trichlorosilane undergoes an equilibration reaction to deposit

silicon and produce silicon tetrachloride, hydrogen chloride, and hydrogen

HSiCl3 ! Si þ HCl þ SiCl4 þ HSiCl3 þ H2 : (28)

Over the last 50 years the Siemens process has been optimized by using increas-
ingly larger reactors and achieving high conversion rates (Fu et al. 2015). The
Siemen reactors are readily scalable for commercial operation. Production levels
are increased by adding more reactors. The economics of the Siemens process
depends on the recycling of all forms of the off-gas products or their use in other
processes within a chemical complex. One needs to reduce losses of silicon or
chloride from the process. Even though Siemens technology has been brought to a
high level of development, the major deficiency remains that the process is a batch
process that is a high consumer of electric power.
The Siemens process remains the dominant process for the production of poly-
silicon due to the high purity of the product silicon and the high reliability of reactor
operations. As the demand for polysilicon continues to increase, the question
remains whether the rising cost of electricity and the added costs for running a
batch process will make the FB route to make polysilicon production more cost-
effective.

Comparison of the Fluidized-Bed Process and the Siemens Process

for Making Polysilicon

There are both advantages and disadvantages to the operation of each type of
polysilicon production process. With increasing demand for polysilicon and with
the high electrical power consumption of the Siemens process, there are renewed
incentives to investigate new production techniques. Fluidized-bed technology has
the potential to bring significant cost reduction specifically to polysilicon production
as it has done in many other industrial processes.
The following table compares various aspects of the silane-based FB production
of polysilicon with the Siemens process:
16 L. Jiang et al.

Table 1 Comparison of several factors between Siemens reactors and fluidized-bed reactors in the
production of polysilicon
Fluidized-bed process (Based on SiH4
Feed) Siemens process (Based on HSiCl3 Feed)
Potential for reduced Si cost Limited future cost reduction likely
Continuous operation Batch process
Flowable, spherical Si product Crushing step required to convert rods to chunk Si
Noncorrosive feed gas Corrosive feed gas
Reaction goes to completion Limited equilibrium reaction producing HCl and
SiCl4
Produces higher amounts of powder Minimal Si powder produced
80–90% less energy required High electric power consumption
Minimal cooling required (hot wall) Large cooling load to cool reactor walls (cold wall)
Developing technology Mature optimized technology
Reactor operates at 650–850
C High temperature reactor operates near 1150 C
Chloride-free feed product Residual chloride in product from HSiCl3 feed
Violent fluidized-bed environment Stable, fixed reactor environment

Based on Table 1, the comparison between the Siemens and FB processes is quite
complex. Each technology has advantages and disadvantages. The following dis-
cussion expands on the information presented in the chart.
There are several key issues encountered when comparing the two silicon pro-
duction methods. As a batch process, there is extensive time and effort required to
remove the silicon product rods from a Siemens reactor and set up the reactor for
each new cycle. Because silicon is harvested from a Siemens reactor as a thick rod
that is broken into large chunks, there is little surface contamination from the
environment. In contrast the production and handling of granular silicon from a
fluidized-bed reactor is prone to contamination. Both during production and during
handling of the final granular polysilicon product the vastly larger surface area of the
silicon granules can lead to contamination from other elements that come into
contact with the product. In general, silicon from Siemens reactors tends to have
higher purity than silicon from fluidized-bed reactors, even when considering the
crushing process to reduce the size of Siemens rods (Bernreuter 2014). Solar cell
efficiency is dependent on the purity of the silicon going into the solar cell. Even
though FB silicon does not normally have the purity found in silicon produced by the
Siemens process, it has been argued that FB reactors offer a viable, low-cost source
of polysilicon to manufacture solar cells (Steemann et al. 2012).
One of the major differences between FB reactors and the Siemens process is the
feed gases. These differences in chemical and reaction properties result in many of
the differences in the two types of processes. The fluidized-bed reactor uses silane,
which is a chloride-free silicon hydride gas. Silane can decompose completely under
the reaction conditions in a noncorrosive environment. With a silane feed another
advantage is that the reaction can be run at a lower temperature because of silane’s
lower decomposition temperature (see section “Chemical Reaction of Silane in a
Fluidized Bed Process with Silane 17

Fluidized-Bed Reactor”). The main downside consideration with silane as a feed gas
is that a disproportionation plant is required to convert trichlorosilane to silane.
A major industrial factor is that the Siemens process produces only rods of
polysilicon that have to be crushed into smaller particles for downstream crystalli-
zation processing.

Process of Polycrystalline Silicon Fabrication

Spherical silicon particles are the favored form for handling polysilicon for down-
stream continuous crystallization processing (Filtvedt et al. 2010). Round particles
of silicon have greater flowability for feeding polysilicon. The spherical, granular
product silicon from a fluidized-bed reactor is uniquely suited as feed for continuous
Czochralski (Cz) crystal pullers in the manufacture of monocrystalline silicon. The
specific need for flowability in polysilicon is yet another cost-reducing factor
encouraging the development of fluidized-bed technology.
Research continues on fluidized-bed reactors to improve their benefits. As men-
tioned fluidized-bed reactors can be run in a continuous mode with far less electrical
power demands. On the down side, fluidized-bed silicon reactors must be periodi-
cally cleaned and repaired.
As mentioned above, there are many industrial processes that successfully utilize
fluidized beds in various chemical processes. Typically, such processes are devel-
oped and optimized over a number of years to reach a stage where the process is
stable in commercial operation. The use of fluidized-bed technology for the depo-
sition of silane to silicon has been slow to follow this pattern. Even though the
theoretical advantages of fluidized bed for silane production are well recognized, the
actual development of the technology in commercial-sized plants has proven to be
challenging. The difficulties result from the nature of the reaction that is being
carried out and the need to resolve conflicting factors. Many of these issues will
be expanded later in this article.

Downside to the Use of Silane-Based Fluidize Bed Technology

Table 1 points out the main advantages and disadvantages of fluidized-bed technol-
ogy to produce polysilicon, but it does not prioritize them. One major disadvantage
of FB reactors is that they may produce less pure polysilicon than the Siemens rods
at this time, but the potential exists for improvement in the purity of FB polysilicon
product. Another disadvantage is that FB process is more complicated to operate.
Problems typically encountered with all FB reactors such as defluidization and
plugging may occur. The possibility of slugging, wall deposition, and agglomeration
all complicate the operation of a FB reactor. The production of silane alone requires
the operation of a disproportionation plant before the actual silane deposition unit
can be run. The production of silicon dust as a by-product of using silane in a FB
reactor is a disadvantage because of product loss and because of the need to remove
18 L. Jiang et al.

the silicon dust. The numerous disadvantages of the FB process have allowed the
Siemens process to dominate as the preferred method for polysilicon production.
Progress has been made in overcoming or reducing many of the current disadvan-
tages encountered with the FB silane process. A more systematic review of the
challenges of the silane-based FB process will be presented later.

Alternate Technologies for the Production of Polysilicon

Although this article focuses on fluidized silane-bed technology to produce poly-

silicon, it is useful to briefly mention other related methods currently being studied to
produce polysilicon. By recognizing other methods for silicon purification, it is
possible to put some context into why the silane-based fluidized-bed method is
currently considered as the most likely to succeed and to replace the Siemens
process.

Use of Trichlorosilane in Fluidized-Bed Production of Polysilicon

The use of silane in a fluidized bed has been mentioned above as the most favored
alternative for fluidized-bed production of silicon. Industry has explored many other
feed compounds for deposition to produce polysilicon. In 1961 DuPont patented the
use of trichlorosilane in a fluidized-bed reactor to purify silicon (Bertrand and Olsen
1961). Wacker has operated a plant and has developed fluidized-bed technology
using trichlorosilane directly as the feed material (Wacker 2016). Wacker claims to
have eliminated silicon dust formation and to have produced high purity polysilicon
(Weidhaus et al. 2005).
The trichlorosilane-based FB process produces spherical polysilicon but the
plants have so far proven not to be economical. The major difficulty is that the
deposition reaction has to be run at the higher temperature range of 850–1000
C,
which is far above that of a silane-based FB reactor. The reactor has to withstand a
corrosive chloride environment at high temperatures. One encounters difficulties
with materials of construction that can withstand the highly corrosive environment
and can achieve high levels of purity in the product polysilicon. There is also the
problem of chloride incorporation into the silicon. As with the Siemens process,
deposition with trichlorosilane runs as an equilibrium reaction that does not go to
completion so that extensive recycling is required. Silane remains the favored silicon
compound for use in a FB reactor for polysilicon production.

Use of Silane in Siemens Reactors to Make Polysilicon

One novel approach to silicon production uses a Siemens type reactor, with silane
replacing trichlorosilaneas the feed gas (Revankar and Lohoti 2014). With silane
feed the Siemens reactors can be run at a lower temperature. The major problem
using a Siemens reactor with silane appears to be excessive silicon powder forma-
tion. To reduce the powder problem the Siemens reactors have to be run with lower
silane conversions. A high proportion of the silane must be recovered from the off
gas and recycled.
Fluidized Bed Process with Silane 19

Alternate Approaches to Polysilicon Production

There are many other approaches to polysilicon production that have been explored
in the patent literature. One interesting approach deals with the problem of creating
silicon dust. The technique is to convert all of the silane directly to silicon powder in
a free space reactor. The silicon powder is compressed into briquets and melted to
form an ingot (Iya 1987).

Silane Reactions in a Fluidized-Bed Reactor

The reaction of silane in a FB reactor consists of a complex series of chemical steps.

The thermal decomposition of silane has been extensively studied (Walch and Dateo
2001). The low pressure (often far below one atmosphere) epitaxial deposition of
silicon from silane plays a major part in the semiconductor industry and is well
understood. The higher-pressure deposition of silicon at above one atmosphere
encountered in the preparation of polysilicon from silane has had much more limited
study. A high-pressure (well above one atmosphere) FB silane-based fluidized-bed
silicon plant has been constructed and operated in a joint venture between SunEdison
and Samsung (Roselund 2015). After a startup and initial operation, this plant has
not restarted due to business conditions. GCL has also developed high pressure FB
technology using silane feedstock.

Chemical Reaction of Silane in a Fluidized-Bed Reactor

The decomposition of silane is known to occur between 350
C and 480
C

depending on conditions such as hydrogen pressure and the availability of surfaces
that can support the deposition reaction (Filtvedt et al. 2010). In a free space reaction
in the absence of silicon particles it is known that the pyrolysis of silane produces
mostly silicon dust and not crystalline silicon. It is also well known that silicon
particles must be heated to the range of 650–850
C for the deposition of silicon
compounds to form polysilicon on silicon particles. Several attempts have been
made to combine the information on the lower temperature decomposition and the
higher temperature deposition reaction to provide a coherent mechanism that
explains the operating chemistry of a silane fluidized-bed reactor (Lai et al. 1986).

Homogeneous, Heterogenous, and Scavenging Reactions of Silicon

Particles

To understand the mechanism of how a silane-based fluidized bed converts silane to

silicon one needs to consider the two phases present in the reactor. The bubble phase
consists of the space between the silicon particles existing as bubbles. The emulsion
20 L. Jiang et al.

Fig. 4 Three main types of processes found in a silane-based fluidized-bed reactor. They include
(1) a heterogenous reaction on the surface of a silicon particle, (2) a homogeneous reaction in the
gas phase forming silicon powder, and (3) surface scavenging of silicon powder on a silicon particle
followed by deposition

phase is made up of the silicon particles and the nearby gases. The small silicon
particles provide a large surface area for reaction with the gas phase.
The three main types of reactions taking place in a silane-based fluidized bed are
shown in Fig. 4.
The chemistry of the bed can be broken down into three types of reactions. The
preferred reaction is the heterogenous decomposition of silane on a silicon particle.
The undesirable reaction occurs with the homogeneous thermal decomposition of
silane to produce fine silicon dust (Kerner et al. 2003). The favorable scavenging
mechanism removes some of the silicon dust by the accretion of dust in the
submicron size range onto fluidized silicon particles. The silicon dust then chemi-
cally sinters onto the granular particles aided by the heterogeneous reaction.
It is possible to understand how this proposed mechanism can explain the
observed FB reactor behavior. The heterogeneous reaction benefits from the large
surface area provided by the silicon particles. The bubbles and space occupied by the
hot gas phase provide a place where the homogeneous reaction produces silicon
dust. Some of the dust could migrate to the surface of hot silicon particles that then
scavenge the silicon dust.
A typical silane fluidized-bed reactor is run at a temperature between 650
C and
850
C to allow the growth of silicon particles by deposition. The ultimate goal is to
find operating conditions where powder production is limited so that the rate of
granular silicon preparation is commercially viable. Powder formation cannot be
completely suppressed and some silicon is always formed as a fine dust by-product
at the current level of silane FB reactor development.
The nature of the fluidized-bed reactor further complicates the optimization of
granular silicon production. Fluidized-bed reactors provide a highly variable
Fluidized Bed Process with Silane 21

reaction environment. Temperature, pressure, concentration, silicon surface area,

and many other parameters vary throughout the reactor in a complex way that has
to be modeled analytically. There are a large number of physical parameters that
must be optimized to control the complex reactions involved in the operation of a
fluidized bed to convert silane to polysilicon.

Modeling of Silane Fluidized-Bed Reactors

The complexity of designing and operating a silane-based fluidized-bed reactor is a

daunting task. There are so many challenging issues and complex factors in play that
simple intuition and engineering tools have not proven sufficient to solve the
problems. One area that shows promise in improving understanding of the process
involved is computational fluid dynamics (CFD) modeling (Kunii and Levenspiel
1991). The silane fluidized-bed system represents a complex case that is being
developed.

Advantages of Using Numerical Simulation

The steady improvement in the speed and memory capacity of computers has led to
the wide spread use of numerical simulation. This branch of physics and engineering
complements experimental and theoretical work by providing an alternative cost-
effective means of solving problems. As such it offers the means of testing theoret-
ical advances for conditions unavailable experimentally. By using numerical simu-
lation, the lead time in design and development is significantly reduced. Simulation
can provide more detailed and comprehensive information. More importantly,
numerical simulation is much cheaper than real-time experimental testing and will
become even more so in the future.

Classification of Models for Fluid Bed

Generally speaking, there are two main categories of models for fluid beds: 1D
phenomenological models and detailed CFD models (Filtvedt et al. 2010).
Historically, two classes of phenomenological models have been proposed to
model fluid beds: the pseudo homogeneous approach and the two-phase approach
(Yang 2003). The conventional pseudo homogeneous model, where only one phase
is taken into account, may include ideal flow models, dispersion models, and
residence time distribution models. Due to lack of accuracy, the use of this approach
has been slowly dropped.
The two-phase model assumes that the bed can be divided into a bubble phase and
an emulsion phase. The bubble phase contains a large gas voidance and very little
solids. The emulsion phase contains most of the bed particles. The two-phase model
was originally proposed by Toomy (1952).
Detailed CFD models solve the mass, momentum, and energy equations on a
point-to-point approach to compute local variables. There are several approaches
used in detailed CFD models, each making a tradeoff between the modeling effort
and the computational cost. Greater physical resolution reduces the modeling effort,
22 L. Jiang et al.

but increases the computational cost. Generally speaking, there are two main
categories: the Euler-Euler model and the Euler Lagrangian model. In the Euler-
Euler model, both the gas phase and solid phase are treated as an interpenetrating
continuum. The conservation equations are solved for each phase with phase
interaction terms and constitutive relations. In the Euler-Lagrangian model, the
solids are treated as a discrete phase. This model is more realistic, but is computa-
tionally more intensive and assumptions are made (Pannala et al. 2010).

Modeling of a Silane Fluid Bed

Several research groups have investigated modeling of a silane-based FBR for
silicon production. The main models are reviewed below:
Historically, there are many variations of two-phase models for fluidized bed and
the Kunii and Levenspiel (1991) model has wide spread use for the bubbling
fluidized bed. In this model, the bubble diameter can be kept invariant through the
bed and is treated as a fitting parameter. An alternative is to allow the bubble size to
vary along the bed axis. The initial bubble diameter formed at the surface of a
perforated and of a porous gas distribution plate can be calculated from the following
equations, respectively:
 
1:3 u0  umf
d b0 ¼ (29)
g0:2 N or
2:78
2
d b0 ¼ u0  umf , (30)
g

where umf is the fluidization velocity of the bed, u0 is the superficial gas velocity, g is
the gravity acceleration, Nor is the number of orifices per unit area. The bubble
growth at any level z in the bed can be correlated as:


1=3
db ¼ 0:0085 1 þ 27 u0  umf ð1 þ 6:8zÞ1:2 : (31)

The rise velocity of the bubbles can be calculated as

pffiffiffiffiffiffiffi
Ub ¼ 0:71 gdb þ u0  umf : (32)

The volume fraction occupied by bubbles can be calculated as

u0  umf
δ¼ : (33)
ub
The interchange coefficient of gas between bubble and the emulsion can be
calculated as
Fluidized Bed Process with Silane 23

umf
kbe ¼ 4:5 : (34)
db
umf can be calculated by the Ergun equation as below
 


1:75 2 150 1  emf d3p ρg ρs  ρg g
Re þ Remf ¼ (35)
e3mf ϕs mf e3mf ϕ2s μ2

and

ρg umf dp
Remf ¼ , (36)
μ
where emf is void fraction of the fluidized bed at a minimum fluidization, φs is the
sphericity of the particles, ρs is the particle density, ρg is the gas density, dp is the
diameter of the particle, and μ is the viscosity of the gas.
By making a simple mass balance and noting the reactant gas A (in our case
silane), we only consider the homogenous reaction in the bubble phase

dCAb
δUb ¼ δkhom CAb þ kbe ðCAb  CAe Þ: (37)
dz
And in the emulsion phase, only the heterogeneous reaction is considered

dCAe

ð1  δÞumf ¼ ð1  δÞ 1  emf khet CAe  δkbe ðCAb  CAe Þ, (38)
dz
where CAb means the concentration of reactant A in the bubble phase and CAe means
the concentration of reactant A in the emulsion phase.
Where the rise velocity of the bubble gas, not just the bubble is

U b ¼ ub þ 3umf : (39)

Several researchers have used the two-phase model to study silane pyrolysis in a
fluidized bed. Caussat et al. (1995) used four different models to compare with his
experiment data. He concluded that the fines are mainly formed by wall deposition
rather than homogeneous decomposition in the bed, which is in disagreement with
other studies (Hsu et al. 1987), but he used nitrogen as carrying gas, rather than
hydrogen, which can be responsible for the different results.
Lai et al. (1986) applied both an ideal back mix reactor model and the two-phase
fluidized model to simulate a silane FBR. In his model, similar to Hsu’s proposal,
there are nine different reaction pathways. He also used the method of moments to
count the population balance for fines. His model predictions show reasonable
agreement with experimental results. The continuous stirred tank reactor (CSTR)
model yields an upper estimate on the production rate and the fluidized-bed bubbling
24 L. Jiang et al.

reactor (FBBR) model overpredicts the formation of fines. It concludes that in order
to suppress fines formation, a reactor needs to have good gas-solid contacting in the
gas inlet distributor region and suppress bubbles formation.
Later Pina et al. (2006) used the two-phase model approach to simulate a spouted
fluidized-bed reactor. In a spouted fluidized bed, the inlet gas is introduced through a
spout nozzle at the bottom of the bed rather than a distributor plate. In his model, two
regions in series are taken into account, a spout zone in the bottom followed by a
fluidized-bed region. A new set of equations is used to model the spout region
exchanging gas between the spout and surrounding dense emulsion. He also used a
detailed population balance equation to model particle size distribution in the reactor.
Two adjustable parameters, a scavenging coefficient and an agglomeration constant,
have to be fitted to the experimental data for each case and then correlated as a
function of the silane concentration. Then the model was tested for longer experi-
mental runs and found to have good agreement with the data.
Kimura and Kojima (1991) used a simplified two-phase model for a silane FBR,
where the bubble diameter is invariant throughout the bed and is treated as a fitting
parameter. They only considered two reaction pathways: the homogenous reaction to
form fines and the heterogeneous reaction to deposit on the particles. Their simula-
tion results are compared with experimental data by Hsu et al. (1987). The experi-
mental tendency of increasing of fines elutriation with increasing silane
concentration or increasing bed temperature can be explained well. However, at
higher bed temperature, the numerical results of fines elutriation are much higher
than the experimental data. As the author pointed out, it can be due to the lack of a
scavenging effect in their model.

Eulerian-Eulerian CFD Model

Eulerian-Eulerian modeling is the most commonly used approach for fluidized-bed
simulations because it can handle large industrial-scale geometries. The general idea
is to formulate a multifluid model to treat both solid phase and gas phase as an
interpenetrating continuum and therefore to construct integral balances of mass,
momentum, and energy for both phases. Appropriate boundary conditions and
jump conditions will be applied at the interfaces to account for the discontinuity
effect.
Mass conservation equations of gas and solid phases:

@   ! 
αg ρg þ ∇  αg ρg V g ¼ 0 (40)
@t
@  ! 
ðαs ρs Þ þ ∇  αs ρs V s ¼ 0: (41)
@t
Momentum conservation equations of gas and solid phases:
Fluidized Bed Process with Silane 25

@ !  @ ! !  !
α g ρg V g þ ∇  αg ρg V g V g ¼ αg ∇p þ ∇  τ þ αg ρg g
@t @t
 ! ! 
þ K gs V g  V s (42)

@ !  @ ! ! 
αs ρs V s þ ∇  αs ρs V s V s ¼ αs ∇p  ∇Ps þ ∇  τ þ αs ρs
@t @t
 ! ! 
!
g K gs V g  V s : (43)

The drag between gas and solid phase is one of the dominant forces in gas-solid
fluidized-bed applications. There are many correlations in the literature for calculat-
ing the momentum exchange coefficient Kgs which may have different ranges of
application. The solid phase momentum equation contains an additional term called
solid pressure, Ps to account for particle-particle collisions. The stress term in solid
phase momentum transfer is normally computed by using kinetic theory.
MFIX (Multiphase Flow with Interphase Exchange) is an open-source CFD
(computational fluid dynamics) code developed and maintained by NETL (National
Energy Technology Laboratory), and it is a powerful tool for the simulation of
fluidized beds using the Eulerian-Eulerian method. Cadoret et al. (2007) performed
silane FBR experiments, and the MFIX code was used for simulation. They used
nitrogen as carrier gas and dense alumina powders as deposition seeds, both of which
are rarely used in industry. They did both 2D and 3D simulations and found that 2D
simulation results are very inaccurate and that 2D simulations were not suitable for
fluidized-bed modeling. Later, they performed another group of 3D simulations
(Reuge et al. 2009), in which they used detailed silane reaction kinetics. Besides
determining which kinetic model is more appropriate, they concluded that there is a
strong interaction between bed hydrodynamics and reaction chemistry. Also, silicon
deposition from silane mainly occurs in the dense zones of the bed where silylene
(SiH2) does much of the work within the bubble phase and at its periphery.

Particle-in-Cell Method
The Eulerian-Eulerian approach has trouble modeling flows with a distribution of
particle types and sizes because separate governing equations must be solved for
each of them and it can be too computationally intensive. The traditional Eulerian-
Lagrangian approach will treat each particle as a tracking element and resolve the
fluid flow around each particle with a no-slip condition. This saves the efforts of
developing closure models, but is so computationally intense that only a small
number of particles and a small domain of fluid can be realistically done.
The particle-in-cell (PIC) method is based on the Eulerian-Lagrangian approach,
but also combines the good features of the Eulerian-Eulerian approach to make it
computationally affordable. Like the regular Eulerian-Langrangian model, PIC
employs a fixed Eulerian grid and Lagrangian particles, but rather than resolving
fluid flow around each particle, it averages the fluid flow over a spatial region
containing a number of particles. Also, the particles are not tracked one by one,
26 L. Jiang et al.

but a collection of particles called “numerical particles” are solved together, and both
the interactions of the particles and fluid-particle momentum transfer are calculated
through closure models on the Eulerian grid. Particle properties are interpolated to
the grid and the flow field on the grid is interpolated back to the particles.
Solid Phase Equation:

dus
 1 1
¼ Ds ug  us  ∇p þ g  ∇τs ¼ 0, (44)
dt ρs α s ρs

where Ds is the drag function and τs is the particle normal stress due to collisions, αs
is the particle volume fraction.
Mass and momentum conservation equations For the gas:

@   ! 
αg ρg þ ∇  αg ρg V g ¼ 0 (45)
@t
@ !  @ ! ! 
 !
αg ρg V g þ ∇  αg ρg V g V g ¼ ∇p þ ∇  αg τ þ αg ρg g þF, (46)
@t @t
where F is the rate of momentum exchange between the fluid and particles per unit
volume.
Barracuda is a commercial software owned by CPFD ⁣, https://cpfd-software.
com) that successfully implemented the PIC method to solve particle fluid interac-
tion problems, which has more and more wide spread use in industry. Parker (2011)
simulated JPL silane FBR experiments (Hsu et al. 1987) by using Barracuda. A
simple empirical first order silane reaction mechanism is adopted in his simulation
and both homogenous and heterogenous reaction rate data are taken from the
literature. A constant scavenging coefficient is used to capture the scavenging effect
and was found to reasonably capture the trend of the experiments. The simulation
results can predict the temperature gradient inside the bed fairly well in comparison
with experimental measurements. The simulation can also reproduce the fines
production rate with changing silane concentration in the feed gas. The effect of
the distributor type was examined and found that a nozzle type inlet would create a
spouted bed and a screen mesh inlet would result in much more uniform gas
distribution and a much better performing bubbling bed, which is also consistent
with the experimental results.

Multiscale Modeling
There are different time scales in the complicated physics of silane fluidized-bed
reactors. The chemical reaction is almost instantaneous, and the hydrodynamics can
take few seconds to gain equilibrium, while the particle size growth may take hours
and days to reach steady state. In order to capture all the time scales, Balaji et al.
(2010) developed a multiscale model for a silane FBR by coupling the CFD solver
COMSOL and the equation solver MATLAB. The hydrodynamics are solved by the
CFD module in COMSOL, which provides information to a CVD module in
Fluidized Bed Process with Silane 27

MATLAB. The CVD module calculates the overall production rate and feeds it to
population balance module in MATLAB. The population balance module calculates
the particle size distribution and gives an average particle diameter to feed the CFD
module to start another round of computation. The simulation results were validated
by experiments.

Operation of a Silane-Based Fluidized-Bed Reactor to Make

Silicon

The operation of a silane-based fluidized bed reactor is relatively complex and

provides unique difficulties when compared to the operation of most other industrial
FB reactors. In order to understand the state of this technology it is necessary to
review many of the operation steps that are encountered in a FB plant to make silicon
from silane.

Feed Gas to the Fluidized-Bed Reactor

One of the key factors in optimizing silicon deposition and reactor production rate is
the SiH4/H2 mole ratio of the feed gas. Ideally one would prefer to use 100% silane
as the feed gas in order to maximize reaction productivity, but this would result in the
formation of a large amount of silicon dust. Powder formation can be reduced by
diluting the silane feed gas with hydrogen. The optimum feed gasconcentration has
been found to be between 10% to 30% silane in hydrogen depending on the reactor
operating conditions and geometry (Filtvedt 2013). This amount of hydrogen does
not eliminate powder formation, but reduces it to acceptable levels. The hydrogen
acts primarily as a diluent to reduce the concentration of silane and thereby reduce
the reaction rate. As the major component in the fluidizing gas, the viscosity, thermal
conductivity, heat capacity, and other physical properties of hydrogen play a vital
part in maintaining the fluidization and reactions in the fluidized bed.
The hydrogen is recycled after passing through the fluidized-bed reactor. The
hydrogen, unreacted silane, and other reaction by-products can be cryogenically
separated. The hydrogen is returned to the reactor and mixed with silane to continue
the operation of the fluidized bed. The unreacted silane is purified and recycled back
to the reactor.

Preheating the Feed Gas

One of the most challenging aspects of operating a silane-based fluidized-bed reactor

is the control of heat and temperature. It is difficult to introduce enough heat through
the reactor walls into the reactor interior to achieve the optimum reaction tempera-
ture. In a typical reactor one would increase the temperature of the reactor walls to
obtain the desired amount of heat transfer to the reactants. This option must be
28 L. Jiang et al.

approached with caution for a silane-based process because it would greatly increase
silicon deposition on the reactor walls. The fluidized silicon particles need to be
heated from the reactor wall so that they can then heat the feed gas to the reaction
temperature. This heat transfer has to be accomplished in such a way as to minimize
the temperature of the reactor wall.
The use of a preheater to increase the temperature of the of the feed gas has
proved useful. The temperature of the preheater is limited by the reactivity of silane.
It is undesirable to decompose the silane in the preheater to form silicon dust or to
deposit silicon. Given the decomposition temperature of silane, the feed gas is
typically heated to less than 400
C depending on the operating conditions and
structure of the preheater (Filtvedt et al. 2012). The remainder of the heat energy to
bring the feed gases up reaction temperature of over 700 C must come from the
heated walls and heated fluidized silicon particles.
Many different arrangements have been studied for introducing heated feed gases
into the reactor. One patented approach involves heating hydrogen to over 1000
C
and using this hot hydrogen as the sole source of heat for the reactor. As the
hydrogen is mixed with the silane, the reaction gas is brought quickly to reaction
temperature (Spangler and Stucki 2015).

Distributor and Jet Nozzles to Introduce the Feed Gas

The feed gas is typically introduced into the bottom of the reactor through a
distributor or jet nozzles. The distributor consists of many orifices that spread the
feed gas out over the bottom surface of the reactor to make uniform fluidization of
the silicon particles possible. Jetnozzles consist of tubes with designed shapes and
diameters that can inject the feed gas stream further into the reactor bed. Jets offer the
ability to direct the flow of gases and increase the gas velocity into the bed of
granular silicon. One approach in the patent literature, which makes use of the ability
to aim the flow from the distributor proposes directing hydrogen towards the wall of
the reactor to reduce the amount of silicon deposition on the reactor wall. (Kulkarni
et al. 2013).
One major challenge is to ensure that the temperature of the distributor or jets is
kept below the reaction temperature of silane as feed gas passes through the
distributor to the reactor. Much of the heat that warms the distributor comes from
the heated silicon particles that migrate from hot parts of the reactor. The granular
particles can be at the reaction temperature which is well above the decomposition
temperature for silane. The objective is to keep the bottom of the reactor with the
distributor or jet nozzles cool enough to prevent the premature decomposition of
silane to silicon deposits or dust that will plug the jet nozzles or the distributor
orifices.
There have been several approaches in the patent literature for cooling of the
distributor. Most of the methods use cooling fluids to reduce the temperature of the
distributor to below the decomposition temperature of silane. The major differences
Fluidized Bed Process with Silane 29

Fig. 5 Reactor designs using

resistive heaters to provide
thermal energy to the reactor.
Type 1 reactor (Fig. 5a) has a
metal wall. In Type 2 (Fig. 5b)
the reactor has a liner that
helps to reduce metal
contamination to the fluidized
silicon particles

Fig. 6 Cold wall reactor.

Going from outside of the
reactor toward the inner
reactor chamber which holds
the granular silicon one
encounters the following:
(1) outer reactor pressure wall,
(2) a layer of insulation, (3) a
heating element, (4) the
reactor liner, (5) the inner
reactor chamber

between the many distributor and jet designs are in the placement of holes for the
feed gases and channels for the cooling fluids.
The plugging of the distributor and jets is a significant problem for the long-term
operation of a fluidized bed. Dust can easily plug the narrow gas orifices and disrupt
the flow of feed gas and the fluidization of the bed. Large chunks of silicon from
agglomeration of silicon particles or from the walls of the reactor can drop to the
bottom of the reactor and plug the gas inlets. The only way to clear such blockages is
to empty the reactor and remove the silicon responsible for stopping the gas flow.

Different Designs for Heating Silane-Based Fluidized-Bed Reactors

The main source of heat that brings the silicon particles and feed gas up to reaction
temperature comes from the heated reactor walls, which conduct thermal energy to
the silicon particles that come in contact with the wall. The heat is distributed to the
gas throughout the reactor by the movement of the fluidized particles.
30 L. Jiang et al.

Fig. 7 Reactor with external

induction heating (Clary and
Wikman 1993). Going from
outside to the heating unit
includes the following:
induction coils on the outside
followed by a layer of
insulation. Next comes a
susceptor that is next to the
pressure wall of the reactor.
The heated susceptor conducts
heat to the reactor wall

There are several different types of reactor designs that have been proposed. The
simplest design is a common hot wall metal reactor structure (Type 1), as shown in
Fig. 5a. The Type 1 reactor is typical of many fluidized-bed reactors where the
heated metal walls of the reactor come in contact with the feed gases and fluidized
silicon particles (Gautreaux and Allen 1988).
At the reaction temperature of over 600
C the diffusion rate of metals such as
iron throughout the silicon deposited on the wall is sufficiently rapid that metals are
spread to the fluidized silicon particles by contact with the silicon deposited on the
reactor wall. More recent designs to improve product purity included the use of an
inert liner (Type 2) for the inside of the reactor (see Fig. 5b). The liner is typically
constructed with a hard-ceramic material to reduce the wall erosion by the fluidized
silicon particles.
A more recent approach has the placement of the heating elements inside the
metal pressure containment vessel as shown in Fig. 6 (Kim et al. 2010). An inert liner
is placed next to the heating elements as shown below:
Heat to bring the fluidized bed up to reactor temperature comes from two sources.
As mentioned above a preheater brings the reactants from low temperature to above
300 C. The final heat source is the heated reactor walls which supply the bulk of the
thermal energy and which brings the fluidized-bed reactor temperature to above
600 C.
In addition to electrical resistance heating, induction heating has proven to be a
useful method for introducing the necessary amount of heat energy into a silane-
based fluidized-bed reactor (Clary and Wikman 1993). One of the advantages of
induction heating is that it can bring a large amount of thermal energy uniformly into
a large area of the reactor wall. Below is a diagram showing one of many patented
reactor designs that utilize induction heating (Fig. 7):
This unit consists of a reaction chamber that is surrounded by the induction unit.
Next to the chamber is a susceptor that is heated by the induction coils. After the
susceptor is heated by the induction coils the susceptor brings that reactor wall up to
reaction temperature. Between the induction coils and the susceptor is insulation that
holds most of the heat within the reactor.
Fluidized Bed Process with Silane 31

Another approach to reactor design for silane FB reactors involves separating the
FB reactor into zones (Iya 1989). It has been mentioned that the distributor can be
cooled to prevent silicon deposition while the reactor is heated to create two
temperature zones. Another approach is to try to heat different parts of the reactor
to different temperatures. The idea is to limit the locations where silane decompo-
sition and the deposition can occur to specific areas in a reactor. Given the violent
nature of mixing in a fluidized-bed reactor, the ability to control the temperature at
specific zones in a reactor is limited.
Reactor design remains one of the areas that holds the most promise for improv-
ing the efficiency of a silane fluidized-bed reactor’s operation. Companies continue
to build commercial scale reactors with innovative designs. Modeling of the FB
reactors remains an important tool for evaluating different thermal designs before
they are built and for operating the reactors after they have been constructed.

Reactor Dimensions, Pressure, and Silicon Production Rate

Factors such as reactor height and diameter are critical to the operation of a fluidized-
bed reactor (Filtvedt et al. 2012). The height of the fluidized-bed reactor can affect
the residence time of the feed gas, the silane conversion and the residence time for
scavenging of powders.
Pressure is an important variable when optimizing a silane FB reactor. With
greater pressure one is able to increase the production rate due to increased mass
feed rate. Building a reactor to operate at greater pressure at high temperatures
involves greater design complexity and a reduced safety margin. Working with a
feed gas at higher pressure means that silane will be at a higher concentration and the
fluidization properties of the gas phase will be different than at lower pressures.
The diameter of the reactor is a major factor in determining the FB reactor bed
capacity. Since the heating of the silicon particles to the reaction temperature
typically occurs through the walls of the reactor, the amount of wall heating surface
area is important. The surface to volume ratio of the reactor decreases with increas-
ing diameter. A larger diameter reactor may allow the processing of more silane, but
it becomes more difficult to heat the contents uniformly.

Agglomeration of Silicon Particles

One area of practical concern in operating a FB reactor is the agglomeration of

silicon particles. Chunks of agglomerated silicon particles can form in the reactor
where fluidization may be reduced due to particle segregation or dead flow zones. In
underfluidized areas within the FB reactor chemical sintering of particles can form
agglomerates which settle to the bottom of the reactor and block the distributor or the
jet nozzles. Agglomeration can become a problem that slowly develops for a FB
campaign lasting several months. Such blockages necessitate the shutdown of the
reaction.
32 L. Jiang et al.

Filtvedt et al. (2012) discuss the relationship between the Tamman temperature
and the agglomeration of silicon particles. The Tamman temperature is the temper-
ature at which mobility of molecules in a solid becomes appreciable. Above the
Tamman temperature the surface reactivity and wetting properties of a material
become strong enough to allow powder to stick to the surface. For silicon, the
Tamman temperature is 610 C. In a FB reactor run at 650–850
C the sticking of
silicon powder onto the surface of a silicon particle as shown in the earlier scaveng-
ing mechanism is favorable for the growth of silicon particles. Unfortunately, the
adhering together of several larger silicon particles presents a problem if chunks of
silicon form and fall to the bottom of the reactor. At the bottom of the reactor the
heavier chunks of silicon have limited motion but continue to grow as they interact
with the silane feed gas. Similarly, when pieces of silicon detach from the wall, they
drop to the bottom of the reactor and continue to increase in size. It is necessary to
find flow conditions such that scavenging of silicon dust on granular silicon particles
can occur, but agglomeration of bigger particles is minimized.

Control of the Silicon Particle Size Distribution in the Reactor

In order to operate the fluidized-bed reactor in a continuous mode it is necessary to

maintain the size distribution of silicon particles nearly constant while the particles
grow. This can be accomplished by periodically feeding smaller silicon seed parti-
cles and removing particles that have grown larger. To keep the FB operating
continuously, a portion of the product needs to be reduced in size and returned to
the reactor as seed particles so that the particle size distribution is maintained in a
favorable range for fluidization and silicon growth.
The content of the fluidized bed has a distribution of particle sizes because of the
constant mixing and motion from the normal fluidizing action within the bed. The
favored method for monitoring the granular particle size distribution is by commer-
cial instruments which use digital photography to measure a stream of silicon
particles passing in front of a camera. The digital image can be analyzed by computer
to determine particle size and shape. A plot of particle size distribution of a
statistically significant number of particles can be used to monitor the particle size
distribution in the FB reactor. The digital image method for to monitoring the particle
size distribution is used because it is both rapid and accurate compared to scapening
the particle using a sieve.
The size reduction of the silicon particles needs to be accomplished in such a way
as to minimize contamination of the silicon with metals. Contamination must be
reduced by using equipment that is coated with hard, nonmetallic materials. The
grinding of silicon is especially challenging because silicon is such a hard substance
with a Mohr harness of 7 (Lide 2008).
Fluidized Bed Process with Silane 33

Factors to Consider When Optimizing Silicon Particle Size

Distribution

Particle size distribution determines the flow rates of feed hydrogen and silane gas
that are required to maintain fluidization. Larger particles require a greater flow rate
to suspend particles. At some point as the velocity of feed gas increases the
conversion rate will decrease due to reduced gas residence time. Controlling particle
size is a major factor in optimizing a stable reacting bed.
It is most important to maintain stable reactor bed and fluidization state during
reactor operation. Without a stable bed, a commercial process would not be feasible.
One needs dependable reactors that can run for weeks or months in a predictable and
controllable fashion. Based on experience with fluidized bed operated with many
particle size distributions, it has been possible to determine the range of particles
sizes that offer the best chance for long-term stability. For most silane-based FB
reactors the mean particle size ranges from 500 to 1500 microns in diameter (Filtvedt
et al. 2013).
The relationship between particle size and ease of fluidization is presented earlier
in the discussion on the Geldart classification of particles. Because of their ease of
handling and of fluidization, larger particles in Geldart Groups B and D are favored
for silicon fluidized beds, see Fig. 2.
From this discussion on fluidization, heat transfer, and reactivity of silicon
particles as a function of particle size distribution, it is clear why modeling of the
complex reaction bed is vital. All of the above factors can be calculated and
introduced into a model that can describe the physical state of the fluidized bed.

Silicon Powder Formation and Handling in a Fluidized-Bed Reactor

One of the major deficiencies of silane base fluidized-bed production of silicon is the
production of silicon dust (Hsu et al. 1984). As discussed earlier, powder formation
is the result of homogeneous decomposition of silane trapped in bubbles between
silicon particles and any silane present in the free space above the reactor bed.
Several techniques have been considered to reduce powder formation. One approach
is to increase the residence time in the bed by reducing the particle size and the feed
gas flow rate (Gautreaux and Allen 1988). Under this condition the amount of silicon
dust is significantly reduced because there is more time for the silane gas to react.
The silicon powder is more likely to be captured on the surface of a silicon particle.
The downside to reducing the flow rate of silane is that the output of polysilicon is
reduced. Another approach to reduce powder formation is to inject cool hydrogen
into the top of the reactor above the fluidized bed of silicon to lower the temperature
of silane in the off gas below its reaction temperature (Allen 1988). Hwang (1999)
looked at injecting hydrogen chloride into the FB reactor to eliminate silicon dust.
Hydrogen chloride reacts with silicon particles to form trichlorosilane and to elim-
inate silicon powder as observed in Siemens reactor operation. There has been a
patent issued for equipment to reduce the size of bubbles in the FB reactor in an
34 L. Jiang et al.

attempt to limit the homogeneous reaction that produces dust within the bubbles
(Spangler and Miller 2016). Currently silicon dust formation appears to be inevitable
during the operation of the FB reactor with silane. One approach that has been
patented is to collect the silicon powder and pelletize it for use in other applications
(Hariharan et al. 2007). The formation of silicon dust from silane remains an area of
active research (Wyller et al. 2016). The preferred option is to find reactor conditions
that minimize silicon dust formation.

Removing the Silicon Deposited on the Reactor Wall

Another downside to the use of silane-based FB reactors is silicon deposition on the

inside reactor wall that must be periodically removed. Because the hottest part of the
reactor is its walls, the space near the walls is the most chemically active part of the
reactor. As the thickness of deposited silicon on the reactor wall increases over time
the heating characteristics of the wall change. Eventually the silicon producing
fluidized-bed reactor must be shut down to remove the reactor wall deposits.
Several techniques have been developed for lengthening the time between reactor
shut downs for silicon removal from the walls. One obvious method is to minimize
the temperature on the walls to reduce silicon deposition. Another approach men-
tioned earlier is to flow hydrogen or an inert gas along the wall surface to reduce the
concentration of silane on the reactor wall (Kulkarni et al. 2013). Proposals to install
baffles in the reactor have been made to dilute silane next to the wall and to reduce
the exposure of the reactor wall to silane (Bhusarapu et al. 2015).
Despite all efforts to minimize the deposition of silicon on the walls of the reactor,
a point is reached where it is necessary to remove the silicon layer that has formed on
the reactor walls. The buildup of silicon on the reactor walls leads to several
problems. As the layer of silicon becomes thicker, the transfer of heat from the
wall and through the silicon layer becomes less efficient. As the silicon deposits
become thicker, the probability of chunks of silicon breaking off the wall and falling
to the bottom of the reactor increases.
The primary method for removing silicon deposited on the reactor wall is to
chemically remove the silicon. The reactor can be emptied of granular silicon and
then a gas mixture containing hydrogen chloride (HCl) is passed over the deposited
silicon to convert the deposited silicon to chlorosilanes. The reactor walls have to be
heated to achieve an acceptable silicon etch rate:

Si þ 3HCl ! HSiCl3 þ H2 : (47)

By using dry hydrogen chloride with a very low water content and diluting the
hydrogen chloride with hydrogen, it has proven possible to etch off the silicon from
the reactor wall and to minimize corrosion of the metal parts of the reactor (Chung
and Sturm 2007).
Fluidized Bed Process with Silane 35

Safety Issue with Silane and Hydrogen

Although it is often not stated, safety is one of the major concerns when operating a
fluidized-bed reactor to produce silicon. Both silane and hydrogen can form explo-
sive mixtures when combined with air. In addition, silane is pyrophoric and com-
busts spontaneously in contact with air. One is faced with the possibility that any
leaks or failures in parts of the plant piping containing hydrogen and silane could
result in fires or explosions. The risk factor is increased because the reactor is
operated at such a high temperature. All parts of plant design must take into
consideration the inherent dangers encountered with hydrogen and silane-based
processes.

Atmospheric Contamination of Silicon

Unlike many chemical processes, nitrogen cannot be used as an inert gas for high
temperature silicon processes such as the fluidized bed. The safety issue arrives from
the fact that at elevated temperatures above 900
C silicon reacts with nitrogen to
form silicon nitride (Riley 2000):

3 Si þ 2 N2 ! Si3 N4 : (48)

The reaction of silicon and nitrogen is quite exothermic so that there is a danger of
with thermal runaway. The reaction between silicon and nitrogen may start slowly
but if the heat generated is not dispersed, the temperature of the silicon will rise and
further increase the reaction rate. Often argon is required as a blanket gas when
dealing with high-purity silicon. Since hydrogen is used as a reaction gas and is
available, it is often used as the blanket gas above hot silicon.
In addition to considering the reaction of nitrogen and silicon at elevated tem-
peratures, the interaction between silicon and oxygen is an issue. Any silicon surface
exposed to air or other source of oxygen at room temperature or above will
instantaneously form multiple atomic layers of silicon oxide. This formation of an
oxide outer layer becomes more of an issue when dealing with granular silicon
because of the greater surface area provided by the small particles of silicon. In the
fluidized-bed reactor and the silicon handling lines, it is preferred to eliminate
oxygen and keep the silicon covered with an inert gas or hydrogen.

Current Status of Commercial Research and Development

of Silane Fluidized-Bed Reactors

As noted above, there are many industrial processes that have successfully used
fluidized-bed technology in chemical processes. Such processes are developed and
optimized over a number of years and reach a stage where the process is predictable
and stable on the commercial scale. The use of a fluidized bed for the deposition of
36 L. Jiang et al.

silane to produce high purity silicon has not followed this pattern. Even though the
theoretical advantages of fluidized bed for silane deposition are well recognized, the
actual development of the technology has proven to be difficult. Over the last few
years REC, GCL, and SunEdison have all operated commercial scale silicon pro-
duction plants and have gained experience with silane-based fluidized-bed silane
units (Bernreuter 2014). There are several FB reactor projects currently under
construction in China (Roselund 2015). Despite decades of experience with the
silane-based FB technology, challenges still remain before the technology can
compete with the Siemens process. The difficulties result from the nature of the
reaction that is being carried out and the need to resolve conflicting factors.
The optimization of the silane-based fluidized-bed technology has proven to be
complex and has not reached its theoretical potential. Most of the issues found when
designing and running a FB silicon plant have been discussed above. The following
is an organized list of five largely unique, complex, and interrelated issues that are
encountered while operating a silane deposition FB reactor:

1. Silicon particle deposition versus powder formation: In a silane fluidized-bed

reactor, there are competing reactions between the desired deposition of silane
onto silicon particles and the undesirable formation of very fine silicon powder. In
practice one cannot eliminate powder formation. It is important to develop
techniques for minimizing the amount of silicon powder produced.
Reactor design and operating regime is critical in minimizing powder formation
in a FB reactor. The goal is to minimize the volume of silane that is heated above
its reaction temperatures when the silane is not in contact with a silicon surface.
For a fluidized-bed reactor, this means limiting slugging and reducing the size of
bubbles.
2. Agglomeration of silicon particles: At the high temperature that a silane-based FB
reactor operates, agglomeration of silicon particles can become a problem. Large
chunks of silicon particles made up of many attached silicon particles can disrupt
the flow of feed gas and silicon particles in the reactor. One must find reactor
conditions such that silicon particles will not interact sufficiently to allow
agglomeration to occur. At the same time, it is important to encourage the
scavenging reaction where silicon dust adheres to silicon particles and adds to
the growth of individual silicon particles.
3. Depositing silicon on the reactor walls: By design, silicon is mostly deposited on
the silicon particles in the reactor, but it is inevitable that some silicon deposition
occurs on the reactor walls. Deposition on the vessel walls becomes a significant
issue if the reactor is heated through the walls and the walls are the hottest points
in the reactor where silicon will preferentially be deposited.
Much effort in reactor design has been expended to reduce wall deposition.
Approaches using the bottom distributor or jets to blanket the walls with hydro-
gen or inert gas have been patented (Kulkarni et al. 2013). Other approaches
using baffles have been suggested (Osborne et al. 2011). The ultimate answer to
minimizing wall deposition may be found in developing novel approaches to
heating the reactor.
Fluidized Bed Process with Silane 37

4. Heating the reactor bed to a high reaction temperature: By its nature, a fluidized-
bed reactor requires extensive heating. There is a large, continuous energy
demand to heat the incoming gas and maintain fluidization of the silicon particles
at the reaction temperature. Several different heating designs have been
presented.
Reactor design has centered around using electrical resistance or induction
heating to supply the necessary thermal energy to the reactor. The problem is to
find a reactor design such that sufficient heat can be introduced while minimizing
silicon wall deposition and dust formation and optimize growth of granular
silicon.
5. Achieving high purity in the product silicon: Operating cost and product purity
are the most important factors in evaluating FB silicon.

Specifications for Polycrystalline Silicon for PV Applications

Polysilicon with higher purity yields solar cells with higher efficiencies. Even with
the use of high-purity silane and high-purity silicon particles, processing of materials
in the fluidized-bed reactor requires special techniques and materials to prevent
contamination. Materials of construction and environmental elements such as oxy-
gen and carbon in the plant can contaminate the silicon product. It is challenging to
build and operate a silicon plant where contamination in the product silicon is
maintained below the ppm (6 N) level.
Because the main purpose for producing polysilicon using FB technology or the
Siemens process is to obtain silicon sufficiently pure for photovoltaic and semicon-
ductor applications, it is worth discussing how high-purity polysilicon is obtained
and analyzed. Handling of polysilicon has to be done in such a way that the silicon
does not come in contact with any metal or material that could cause contamination.
At low temperatures, the polysilicon particles can be handled in plastic lined
containers or tubes. At high temperatures, inert liners and hard coatings are neces-
sary to protect fluidized silicon from contamination.
The determination of metals in polysilicon down to the 6 N and 9 N levels has
been revolutionized by the use of the analytical technique of inductively coupled
plasma mass spectroscopy (ICP-MS). Silicon samples can be dissolved in a
hydrofluoric acid/nitric acid solution, vaporized in a plasma, and analyzed by mass
spectroscopy after sample workup. Metals and other impurities can be monitored in
silicon down to the ppb level or below using ICP-MS. Analysis for dopants such as
boron and phosphorous can be carried out by low temperature infrared (IR) analysis.
It is necessary to prepare an appropriate silicon sample for IR analysis by growing
single crystal rods of the granular silicon product. The organization SEMI (Semi-
conductor Equipment and Materials International) and other international organiza-
tions provides approved analytical methods and standards for the analysis of
polysilicon. (SEMI 2013).
38 L. Jiang et al.

Improving FB Technology with Modeling

These five problem areas listed above form the basis for improving silane FB reactor
technology. The issues are interrelated in complex ways. With such complicated
interactions occuring simultaneously, optimizing silane-based FB technology can be
a complex task. Reactor modeling offers a way to evaluate several of the issues at the
same time. Using simulation methods differing reactor configurations and operating
conditions can be tested before building a reactor. The effects of changing param-
eters such as temperature, flow rate, concentration, and pressure while operating a
FB reactor can be difficult to anticipate. Reactor modeling can provide insights into
trends and fundamental chemical and fluid mechanical mechanisms.

Summary and Conclusions

There is renewed interest in the use of the fluidized-bed process for commercial
production of polysilicon for solar applications. Fluidized-bed technology and its
continuous processing shows promise as a means to significantly reduce the price of
polysilicon. The process uses considerably less electrical energy than the competing
Siemens process. Silane-based fluidized-bed reactors produce granular polysilicon,
this reduces the cost of downstream crystallization processing. There has been a
gradual improvement in silane technology but there remains a large difference
between the commercial state of the process and the theoretical potential for
improvement. The developing photovoltaic market for silicon has prompted several
companies to further investigate the fluidized-bed route to high-purity polysilicon.
The downside to the use of the silane-based fluidized-bed process is that the
installed capital cost is higher and it is more difficult to operate than a Siemens plant.
As with all fluidized-bed units, the reactor operation is very dynamic and prone to
upsets. Silicon powder formation is a constant problem along with agglomeration of
silicon. Cleaning the walls of deposited silicon is a periodic maintenance task.
The issue of scalability remains. If FB silicon is to overtake silicon produced by
the Siemens process, FB plants will have to be able to produce ever larger amounts
of polysilicon. This means that larger and more productive FB reactors will have to
be designed and operated. Because of the unpredictability of scale up in FB plants, it
is difficult to predict if scaled up FB reactor will operate successfully.
The one overriding issue that ultimately could determine the success of silicon FB
technology is the purity of the silicon can be produced. Roughly, optimized FB
silicon plants of today can produce material which has impurities in the 6 N to 9 N
level (IHS 2014). Siemens product is often in the 9 N to 11 N range. Higher solar cell
efficiencies require higher purity silicon. FB silicon will eventually have to be
superior to Siemens product not only in cost but also nearly equal in purity.
Fluidized Bed Process with Silane 39

References
R.H. Allen, U.S. Patent 4,748,052, 1988
J. Baeyens, D. Geldart, Chem. Eng. Sci. 29, 255–265 (1974)
S. Balaji, J. Du, C.M. White, E. Ydstie, Powder Technol. 199, 23 (2010)
J. Bernreuter, FBR Polysilicon Technology – Promise or Hype? (PV-Tech, 2014.) www.pv-tech.org/
guest-blog/fbr_polysilicon_technology_promise_or_hype. Accessed 2 March 2017
J. Bernreuter, Cut-throat Competition on the Polysiicon Market (World News, 2016.) https://
eresearch.fidelity.com/eresearch/markets_sectors/news/story.jhtml?storyid=201611240400PR
_NEWS_USPR_____enUK201611224068&provider=PR_NEWS_&product=USPR____
&category=sectorNews&gic=453010. Accessed 3 March 2017
L. Bertrand, C.M Olsen, U.S. Patent 3,012,862, 1961
S. Bhusarapu, P. Gupta, Y. Huang, U.S. Patents 9,114,996, 2015 and 9,114,997, 2015
W.C. Breneman, U.S. Patent 4,676,967, 1987
W.C. Breneman, U.S. Patent 9,352,971, 2016
L. Cadoret, N. Reuge, S. Pannala, M. Syamlal, C. Coufort, B. Caussat, Surf. Coat. Technol. 201,
8919 (2007)
B. Caussat, M. Hemati, J.P. Couderc, Chem. Eng. Sci. 50, 3625 (1995)
K.H. Chung and J.C. Sturm, https://pdfs.Semanticscholar.org/2ff4/3f7680437746363 fap
33038725145 eee607 pdf, 2007
D.W. Clary, A.O. Wikman, U.S. Patent 5,260,538, 1993
W.O. Filtvedt, Production of Polysilicon from Silane Pyrolysis in a Fluidized bed, thesis, Telemark
University College, 2013. Available from https://teora.hit.no/handle/2282/1453
W.O. Filtvedt, M. Javidi, A. Holt, M.C. Melaaen, E. Marstein, H. Tathgar, P.A. Ramachandran, Sol.
Energy Mater. Sol. Cells 94, 1890 (2010)
W.O. Filtvedt, A. Holt, P.A. Ramachandran, M.C. Melaaen, Sol. Energy Mater. Sol. Cells 107,
188 (2012)
W.O. Filtvedt, T. Mongstad, A. Holt, M. Melaaen, H. Klette, Int. J. Chem. React. Eng. 11(1) (2013)
R. Fu, T.L. James, M. Woodhouse, IEEE J Photovoltaics 5, 515 (2015)
M.F Gautreaux, R.H. Allen, U.S. Patent 4,784,840, 1988
D. Geldart, Characterization of fluidized powders, in Gas Fluidization Technology, ed. by
D. Geldart (Wiley, Chichester, 1986)
H. Gutschel, U.S. Patent 3,042,494, 1962
A.V. Hariharan, M. Chandra, K.P. Gupta, U.S. Patent 7,175,685, 2007
J.R. Howard, Fluidized Bed Technology: Principles and Applications (Adam Hilger,
Bristol/New York, 1989), p. 30, p. 202
G. Hsu, R. Hogle, N. Rohtagi, A. Morrison, J. Electrochem. Soc. 131, 660 (1984)
G. Hsu, N. Rohatgi, J. Houseman, AICHE J. 33(5), 784 (1987)
N.W. Hwang, J. Cryst. Growth 205, 59 (1999)
IHS, Technology: Fluidized Bed Reactor Technology Stakes its claim in solar Polysilicon
Manufacturing. http://news.ihsmarkit.com/press-release/design-supply-chain-media/fluidized-
bed-reactor-technology-stakes-its-claim-solar-poly, 2014
S.K. Iya, U.S. Patent 4,684,513, 1987
S.K. Iya, U.S. Patent 4,818,496, 1989
L. Jianlong, C. Guanghui, Z. Pan, W. Weiwen, D. Jihai, Chin. J. Chem. Eng. 19, 747 (2011)
D.K. Kerner, R.W. Davis, E.F. Moore, S.H. Ehrman, J. Cryst. Growth 247, 333 (2003)
H.Y. Kim, K.K. Yoon, Y.K. Park, W.C. Choi, U.S. Patent 7,771,687, 2010
T. Kimura, T. Kojima, Journal de Physique IV Colloque, 1991, 02 (C2), pp. C2-103–C2-110
M.S. Kulkarni, P. Gupta, B. Devulapalli, J. Ibrahim, V. Revankar, K. Foli, U.S. Patent 8,404,206,
2013
D. Kunii, O. Levenspiel, Fluidization Engineering, 2nd edn. (Butterworth-Heinemann, Newton,
1991), p. 35, p. 52, and p. 78
S. Lai, M.P. Dudukovic, P.A. Ramachandran, Chem. Eng. Sci. 41, 633 (1986)
40 L. Jiang et al.

D. R. Lide (ed.), CRC Handbook of Chemistry and Physics, 89th edn. (CRC Press, Boca Raton,
2008), pp. 12–217
B. Mazumder, Silicon and Its Compounds (Science Publishers, Plymouth, 2000), p. 7
NETL, https://mfix.netl.doe.gov/
R.I Newman, W.E. Watson, U.S. Patent 3,370,938, 1968
E.W. Osborne, M.V Spangler, L.C. Allen, R.J Geertsen, P.E. Ege, W.J. Stupin, G. Zeininger,
U.S. Patent 8,075,692, 2011
S. Pannala, M. Syamlal, T. O’brien, Computational Gas-Solid Flows and Reacting Systems
(Engineering Science Reference, Hersey, New York, 2010)
J.M. Parker, Int. J. Chem. React. Eng. 9(A40) (2011)
J. Pina, V. Bacala, N.S. Schbib, P. Ege, H. Ignacio, I. Lasa, Int. J. Chem. React. Eng. 4, A9 (2006)
REC, https://www.recsilicon.com/products/solar-grade-polysilicon. (2017)
N. Reuge, L. Cadoret, B. Caussat, Chem. Eng. J. 148, 509 (2009)
V. Revankar, S. Lohoti, U.S. Patent 8,657,958, 2014
F.L. Riley, J. Am. Ceram. Soc. 83, 245 (2000)
C. Roselund, The slow grind of FBR polysilicon, (pv magazine, 2015). www.http:/www.bernreuter.
com/fileadmin/user_upload/library/Slow-grind-of-FBR-polysilicon-pv-magazine-9-2015.pdf.
Accessed 2 Mar 2017
SEMI, http://ams.semi.org/ebusiness/standards/semistandard.aspx?volumeid=12, (see methods
SEMI PV49-0613 2013 and SEMI PV 17–1012, 2012)
M.V. Spangler, M.J. Miller, U.S. Patent 9,404,177, 2016
M.V Spangler, Glen Stucki, U.S. Patent 8,926,929, 2015
R. Steemann, J. Yong, O. Mjos, A. Song, Energy Procedia 15, 20 (2012)
R.D. Toomy, Chem. Eng. Process. 48, 220 (1952)
Wacker., https://www.wacker.com/cms/media/documents/investor-relations/cmd16/cmd16_polysilicon.
pdf. 2016
S.P. Walch, C.E. Dateo, J. Phys. Chem. 105, 2015 (2001)
D. Weidhaus, I. Crossmann, F. Schreieder, U.S. Patent Application 2005/0135986, 2005
G.M. Wyller, T.J. Preston, H. Klette, O. Nordseth, T.T. Mongstad, W.O. Filtvedt, E.S. Marstein,
Energy Procedia 92, 904 (2016)
Yang (ed.), Handbook of Fluidization and Fluid-Particle Systems (CRC Press/Taylor & Francis
Group, Boca Raton, 2003), p. 261
G. Zhang, F. Quyang, Ind. Eng. Chem. Process. Des. Dev. 24, 430 (1985)