scaleup of fluidized bed reactors

Solar Energy Materials & Solar Cells 107 (2012) 188–200

Contents lists available at SciVerse ScienceDirect

Solar Energy Materials & Solar Cells journal homepage: www.elsevier.com/locate/solmat

Review

Chemical vapor deposition of silicon from silane: Review of growth mechanisms and modeling/scaleup of fluidized bed reactors W.O. Filtvedt a,n, A. Holt a, P.A. Ramachandran b, M.C. Melaaen c a

IFE Kjeller, Norway Washington University, MO, USA c Telemark University College, Tel-Tek, Porsgrunn, Norway b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 12 April 2012 Received in revised form 10 August 2012 Accepted 21 August 2012 Available online 24 September 2012

For an installed silicon based solar cell panel, about 40% of the energy costs involved in the production of the panels can be attributed to the production of the silicon feedstock itself (poly production and crystal growth). Hence reducing the energy consumption in these steps is crucial in order to minimize the energy payback time of installed capacity. For the first step, viz., the poly production, the most promising cost reduction alternative is the fluidized bed reactors (FBR) using silane as a precursor rather than trichlorosilane (TCS) since for TCS the reverse reactions makes the theoretical trichlorosilane conversion substantially lower. Use of silane has, however, many challenges and scaleup to larger capacity can be achieved if associated risks are properly dealt with. This paper outlines some of these challenges and provides a detailed survey of the current status on the growth mechanism and kinetics of silane pyrolysis. The paper also provides a summary of modeling of fluidized bed reactors (FBR) in some depth and give some empirical insight to key aspects in FBR scaleup design. & 2012 Elsevier B.V. All rights reserved.

Keywords: Polysilicon Fluidized bed FBR Silicon

Contents 1.

2. 3. 4.

5. 6. 7.

8. 9.

n

Introduction and scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 1.1. Silicon CVD precursors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 1.2. Fluidized bed reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 1.3. Reviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 Reaction pathways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 Deposition mechanisms and fine formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 3.1. General SiH4 CVD deposition mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 Fluidized bed deposition: microscale effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 4.1. Micro-scale effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 4.2. Microeffects in reactor models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 Product quality: crystallinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 Product quality: porosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 FBR modeling and scaleup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 7.1. Geldart classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 7.2. Reactor models: two phase model and variants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 7.3. Reactor models: detailed CFD models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 7.4. Reactor experimental studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 7.5. Scaleup aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 7.6. Spouted bed reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 Design concepts for FBR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

Corresponding author. Tel.: þ47 63 80 64 01; fax: þ47 63 81 63 56. E-mail address: [email protected] (W.O. Filtvedt).

0927-0248/$ - see front matter & 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.solmat.2012.08.014

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

189

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

1. Introduction and scope Solar cells is an increasing consumer of polysilicon in the world market. Although the rapid expansion in world production capacity have presently made the silicon prices drop, the long term expectations are still growing in PV energy demand and thereby significant growth in polysilicon consumption can be expected in the future [1–3]. The most commonly used method for production of polysilicon is decomposition of a silicon containing reactant gas through heating. Understanding of the fundamental reactions and how they influence product quality is important in order to aid further development. Substantial research has been done by various groups on both decomposition, deposition, product characterization and ultimately reactor design. Monosilane based fluidized bed reactors appear to give the lowest possible cost of operation over the established methods. There are however many challenges with porosity, and impurity encapsulations that one need to overcome if the method is to be become dominant. A second problem is fine formation due to homogeneous reaction in preference to deposition on feed particles. Finally the fluid bed reactor (which currently has the best potential for the silane process) involves complex interaction of hydrodynamics and hence is difficult to scaleup. Hence the use of monosilane in a fluid bed reactor has additional challenges and complexities. In order to advance the technology further and to move to larger scales, a critical review and evaluation of the current knowledge in this field is needed and this is the main scope of this paper. This paper aims to provide an insight to what topics have been examined earlier, where the research stands today and what elements that needs further research to continue the development. A brief background information and references to some earlier review papers is first provided. 1.1. Silicon CVD precursors

The CVD reactors utilizing TCS are however often operated at higher temperatures than 468 1C because some intermediate silicon chlorides have higher decomposition temperatures and since low temperature amorphous structures might cause chlorine encapsulation. These chlorine encapsulations will cause defects and reduce the quality of the material. Typical temperature range for TCS reactors is 850–1200 1C. The lower temperature range for silane and the absence of reverse reactions are the primary reason for switching to silane as the precursor. A detailed understanding of the kinetics and the growth mechanism and the quality of the product is needed for further advances in this technology. 1.2. Fluidized bed reactor The workhorse of today’s industry is the Siemens reactor where either TCS or silane is deposited on silicon rods. Several improvements have been made to the design resulting in a substantial reduction in energy consumption [7]. However, the need for maintaining large temperature differences between the deposition surfaces and the reactor wall, as well as the limited deposition surface area will cause a limit to theoretical reactant yield and lowest energy consumption. TCS has proven successful for these reactors, among several reasons because some reverse reactions will remove loosely bound solid silicon and thereby continuously assure high quality of the deposited material. The drawback is that the product of some of these reactions may form gaseous species that are not decomposed by the reactor operating temperature and this causes a lower reactant gas yield. The yield is defined as the fraction of the introduced reactant gas that becomes solid silicon at the desired deposition surface. Unreacted silicon containing gas in the exhaust as well as undesired depositions and fines formation will cause a reduction in the yield. An alternative to the Siemens reactor is the fluidized bed reactor (FBR). In a FBR, the reactor vessel is filled with silicon particles Fig. 1. A fluidizing gas like hydrogen or helium is injected

In the early days of silicon CVD several gases were explored for the purpose, but in today’s reactors monosilane SiH4 (silane) and trichlorosilane SiHCl3 (TCS) are the ones frequently used. Upon being heated to the decomposition temperature, the gases undergo the following reactions: SiH4 !Si þ2H2

ð1Þ

SiHCl3 þ H2 "Si þ3HCl

ð2Þ

The decomposition temperature for silane is about 420 1C [4]. However, at this temperature the kinetic energy of the individual silicon atoms is too low to form crystalline structures. Low temperature atmospheric pressure decomposition will therefore cause amorphous structures [5]. Silane based CVD reactors are therefore often operated at temperatures in the range 650–800 1C. Some designs utilize after treatment of the beads at even higher temperatures to assure a crystalline structure and complete release of bonded hydrogen. The decomposition temperature of TCS is about 468 1C [4]. Also the reaction is equilibrium limited and thermodynamic conversions are in the range of 25% or so depending on the TCS to hydrogen ratio which necessities TCS recovery and recycle [6].

Fig. 1. The fluidized bed reactor.

190

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

at the bottom of the reactor to fluidize the particles. The definition of fluidizing the particles means the drag force on the individual particles balances the weight of the particle. In this state, the bed of particles behaves like a liquid and the continuous flow of gas keeps the bed in motion [8]. The particles are heated by some means to a temperature above the decomposition temperature and the reactant gas is inserted to the bed. Upon decomposition the silicon is deposited on the particles thus making them grow. After some dwelling time the particles have grown to a size suitable for extraction. The finished beads are then extracted and new small seed particles are either inserted to or produced within the bed. Fluidized bed reactors hold the potential to become the dominating CVD reactor for production of solar grade silicon. Large players in today’s market like Renewable Energy Corporation (REC) and MEMC are producing increasing amounts of their material through FBR instead of Siemens reactors. The two have over 20 years history going back to prior Union Carbide and Ethyl. MEMC has currently entered a joint venture with Samsung Fine Chemicals forming the company SMP [9–16]. Other companies like Wacker have also had substantial research programs on FBR based silicon feedstock production although Wacker has solely published TCS based results. There have been two main challenges with silicon FBRs. For TCS based reactors the reverse reactions have been too aggressive making the overall yield too low. The desired deposition surface in a FBR is the beads. Since the temperature and surface species of the beads varies in time, the silicon atoms will have varying bondage to the surface and may thus be more easily removed than in a Siemens reactor where the surface is more stabile [17]. For silane based reactors the main challenge have been purity and to some degree porosity. There are also several intermediate and reverse mechanisms in silane based reactors. A main difference from TCS is however that the intermediate species to a larger degree are solid and hence some may be captured. The solid intermediate species are in the form of clusters of hydrogen containing silicon. The clusters have dangling bonds as well as new dangling bonds are continuously produced as hydrogen is released. They may thus be scavenged and harvested by forming bonds to the surface of the silicon beads. The expense of this increase in yield is larger inhomogeneity [18]. Even though the material is more inhomogeneous and contains more impurities than what is achievable in a Siemens reactor, the material has proven to be competitive on the world market [2]. The common FBR layout is to have injection of the fluidizing gas through a distribution plate at the bottom of the reactor. The reactant gas is either premixed with the fluidizing gas or injected diluted or undiluted through separate nozzles. The heating can be done in a number of ways at a number of locations through the bed, but the simplest way is to heat the wall of the bed directly by means of heating elements. The volume above the bed which is higher than where the particles reach when getting ejected from bursting bubbles is called the freeboard. The freeboard often suffers from fines clogging. If reactant gas reaches the freeboard without being decomposed it is likely to produce fines because the temperature often is too low to produce crystalline structures although it is above the decomposition temperature. Obvious FBR problems are clogging of reactor internals because of undesired depositions, production of fine dust (fines) because of homogeneous decomposition and agglomeration due to insufficient agitation of the bed. Homogeneous decomposition is decomposition in gas phase while heterogeneous decomposition is decomposition at a depositing surface such as the silicon

beads. Hence due to these challenges, a detailed understanding of FBR is needed for further progress in this area. 1.3. Reviews There are several reviews on silicon CVD with particular reference to silane pyrolysis. These include Jansinski et al. [19], Jasinski and Gates [20], Giunta et al. [21] and Onischuk and Panfilov [22]. The first two focus on homogeneous intermediate reaction stages of silicon CVD whereas the latter two focus on elementary reactions, the processes of silicon precipitation on the reactor walls, the formation of aerosols and elementary reactions occurring in the solid product in the pyrolysis. The reviews are based on laboratory experiments and modeling, one aim of the investigations is to build a theory around silicon CVD. A general conclusion is that the process is very complex and difficult to describe with a simple model. Cardona provided a review on how hydrogen bonds in silicon and how to identify the different modes of hydrogen bondage [23]. A process not sufficiently investigated is the capture of fines. The process of fines formation has been lively debated, but full understanding is still missing. The material on the industrial development is scarce. The reviews on the polysilicon production development include Filtvedt et al. [24] which investigates the development of silicon CVD with special emphasis on fluidized bed technology and Jianlong et al. which review some silicon FBR technology challenges [25]. In view of these limited studies, the goal of the paper is to provide a comprehensive review on this topic as already indicated in Section 1.1. The paper is structured as follows: Section 2 discusses the proposed kinetic models for silane pyrolysis and the reaction pathways. This is needed to understand the deposition mechanism and the fine formation which is discussed in Section 3. Section 4 reviews the mechanisms in a fluid bed reactor where we indicate the importance of both microscale and macroscale phenomena. Sections 5 and 6 discuss how these mechanisms interplay and affect the product quality (crystallinity, Section 5 and porosity Section 6). Section 7 provides a critical review on reactor models and issues affecting the scaleup of the fluid bed. Section 8 indicates various design concepts proposed in the literature with some evaluation of these ideas. Final Section 9 provides a brief summary and the path forward to further progress in this area.

2. Reaction pathways The earliest studies of silane pyrolysis [26] assumed a straight forward decomposition of the silane molecule into silicon and hydrogen according to Eq. (1). However, later studies [27] introduced an intermediate stage in accordance with Eq. (3). 2SiH4 !Si2 H6 þH2

ð3Þ

The actual chain of events seems however to be more complex and more in depth work has been done on the matter by Purnell and Walsh [28]. The general idea is that there is a complex set of reactions whereof some reversible that ends up with hydrogenated silicon. Defined as silicon that contains some hydrogen, but to refer to it as a pure hydride would be ambiguous. Purnell and Walsh discusses several possibilities but concludes that the reaction described in Eq. (4) would be the most likely first step and that homogeneous initiation is likely. SiH4 !SiH2 þ H2

ð4Þ

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

Ring et al. [29] did several isotope labeling studies and indicated that the reaction described in Eq. (5) would be more likely. The problem of large disilane production was however acknowledged and accounted for with additional reactions. The group still supported the view of a homogeneous initiation of the decomposition. SiH4 !SiH3 þ H

ð5Þ

Robertson, Hils and Gallagher did several low pressure experiments and proposed that the decomposition process is heterogeneous and purely surface initiated [30]. The group agreed to several of the earlier proposed reaction chains but only when initiated from a surface in the CVD chamber. The apparent difference in opinion was later resolved through a number of different experiments and papers [31–33] and the chain reaction initiated homogeneously by the reaction described in Eq. (4) is now considered to be the most probable. Although surface initiated chain reactions are important for low pressure CVD, they cannot be dominating for higher pressures and especially when going to higher concentrations. Further modeling was performed by Coltrin and Breiland et al. [34,35]. They proposed the initial reaction Eq. (4) followed by the reactions described in Eqs. (6)–(8) and later the model grew even further incorporating thermal diffusion [36,37]. SiH2 þSiH4 !Si2 H4 þ H2

ð6Þ

Si2 H4 !Si2 H2 þ H2

ð7Þ

Si2 H2 !Si2 þ H2

ð8Þ

Ho et al. supports the view of Coltrin et al., that Si2 in gas phase serves as a precursor for both gas phase nucleation of silicon and surface deposition. This was based on the identification of Si2 in gas phase above a heated surface by laser excited florescence [32]. The production of SiH2 during monosilane pyrolysis was found by O’Brien and Atkinson [38] and they concluded this to be an important intermediate step. Later, Ho et al. [39] identified the Si atom density profile to be nearly linearly linked to the monosilane profile over a heated disk suggesting that SiH2 is not the dominating source for gas phase Si atoms. The decomposition of monosilane is complex and since especially hydrogen is involved in some of the intermediate reactions, the dilution may play an important role in the overall pyrolysis. A clear indication for this phenomenon is the work by Slootman and Parent [40] where changing carrier gas shifted the critical concentration to temperature ratio. Critical concentration was defined as the concentration at which gas-phase nucleation was initiated. The critical temperature at which homogeneous nucleation occurred was much higher in H2 than in inert atmosphere. This may be driven by H2 accessibility for reverse reactions. By now it was established that Eq. (4) is the likely first step. But the continued reaction sequence was not fully established. A generalized expression for the chain reaction of silane decomposition was proposed by Giunta et al. Eqs. (9) and (10) [21]. The main idea is that the aerosol growth is caused by silylenes. anen 2sylm þ anenm syln 2enen

ð9Þ ð10Þ

In Eqs. (9) and (10) anen representing (poly)silane(s) Sin H2n þ 2 and anenm represents H2. Syln silylene(s) Sin H2n and enen disilene(s) Sin H2n . In this scenario, all unsaturated species (disilenes and silylenes) are taken to be deposition precursors (i.e. they are allowed to stick to either a surface or another nucleation site). From this point on the larger species are considered to be powder that are allowed to grow.

191

Another proposal was given by Becerra and Walsh [41] still focusing on the first intermediate solid species. The initiation of solid formation was considered independent of surface reactions. Five routes where considered. H2 Si ¼ SiH2 , SiðHÞ2 Si and c-Si3 H6 where the c in the latter indicates crystalline. The research team could not find final evidence for either of the proposals and concluded that the process seems to be more complex and also questioned the established gaseous reactions and species. Onischuk et al. [42] proposed that the aerosol growth was caused by the precursors Sin Han ða r 2Þ. The further suggestion was that the hydrogen in the initial stage of particle growth was contained as polysilane chains formed in turn by pyrolysis products with a stoichiometry Sin H2n . At later stages of particle growth the hydrogen was maintained mainly as mono-hydride groups which was considered to be linked to hydrogen-depleted regions. The group also had limited conviction that heterogeneous deposition on the particles had any substantial contribution to the particle growth. The experiments where performed in a 1.1 cm tube and there was substantial wall growth during these experiments. There may therefore be limitations to how representative the findings are for full scale systems, this is also acknowledged by the group. Odden et al. performed a series of decomposition studies under various pressures in a closed box and concluded that the particle size to a great extent was independent of the decomposition pressure [43]. However, there seemed to be higher reaction rates when going to higher pressures than what has been observed earlier at lower pressures. Odden et al. modified the proposed model by Ho et al. [39] and argued that the presence of two dangling bonds per active site could be plausible when going to higher pressures. The work of Odden may also indicate a shift to higher decomposition temperature when going to higher pressure, but more work is needed on this topic before a final conclusion may be drawn. In closing, it appears that extensive work has been carried out on the topic, but a complete model accounting for all effects is still missing. However, a two step pathway, SiH2 formation and decomposition of SiH2, is considered as a simplified reaction scheme and used in many reaction modeling studies.

3. Deposition mechanisms and fine formation 3.1. General SiH4 CVD deposition mechanisms When the monosilane molecule decomposes, a chain reaction is initiated. The chain reaction is complicated as indicated in the previous section and will depend on silane concentration, temperature, pressure, temperature of confining surfaces and the gradient of these variables. At lower pressures there is a distinct tendency to heterogeneous deposition and it may be indicators that the decomposition is surface initiated [30,44–47]. A widely accepted theory is the production of gas phase precursors which have preferential temperature dependent growth on surfaces at low pressures. Type of surface, eventual crystal quality and orientation may also be contributing factors. Gas phase nucleation and formation of loose solid species is identified as a phenomenon when going to higher pressures and seem dependent on reactant concentration. A note must be made that there are at least two sources of solid particles. The first source is the release of loosely bound clusters from deposition surfaces and the other is the formation of solid species in gas phase. Several investigations have been performed on the properties of a deposited film under different conditions [48–50]. For atmospheric

192

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

pressure, the deposited silicon will contain hydrogen and be amorphous up to a temperature of about 605–620 1C [50,22]. The amount of hydrogen, type of bondage and the temperature at which it is possible to release it will depend on operating conditions [22,51]. The type and amount of hydrogen present is complex and may be altered by elevated pressures. The interested reader is referred to the work of Odden et al. [52,53]. At elevated pressures, the hydrogen content of the silicon appears to be higher, reaching a maximum at about 2 MPa, 3.5 at% [52,53]. For shock tube experiments the maxima appears to be much higher [54]. An advantage of shock tube experiments is that some of the history dependent phenomena may be suppressed and thus one may to a certain extent isolate parts of the reaction sequence. A disadvantage could be that these history dependent species might be important for the overall reaction. An important feature about the experiment is the closed control volume which limits the possibility of having different zones of different purpose. Production of particles in certain zones and production of condensible vapors in other parts whereby the condensible vapors may be scavenged by the particles is not an option. The lack of pressure influence is nevertheless interesting. Hydrogen is distributed throughout the silicon as either different clusters or diluted protons. There are indications that the proton distribution increases in homogeneity when the decomposition happens at elevated pressures [55]. There are however no sign of any pressure dependent alternation of the cluster distribution. If the temperature is sufficiently high, the depositions will be crystalline [49,48]. The growth will depend on the crystal orientation of the deposition surface to some extent. At temperatures above 850 1C TEM analysis indicates that the grain size follows the deposition temperature and that the {110} texture may be dominating [49]. In terms of the crystal orientation, others have reached alternative conclusions [50]. Herrick and Woodruff studied the difference between SiH4 and SiHCl3 and found that reverse mechanisms suppressed fines formation for trichlorosilane at the operating temperatures of a silicon CVD reactor whereas fines formation for a monosilane based reactor would be substantial [17] . The results yielded that trichlorosilane would produce a more shiny and pure material although the yield would be lower. The reverse mechanisms would produce higher order chlorosilanes with a decomposition temperature above the reactor process temperature. Silicon would therefore be lost in the exhaust. This will increase the need for additional recycling steps and decrease the yield as fraction of the inserted reactant gas. The reverse mechanisms will however be more aggressive towards loosely bound formations and therefore selectively remove impurities and defects. This theory is in good agreement with experimental data published by others [6]. The number of produced particles will mostly depend on the temperature, but also concentration, pressure and type of diluent will play a role. It may seem that the hydrogen concentration directly influences the chain reaction. The use of hydrogen as diluent will therefore yield different reaction sequences than other systems, especially, with inert alternatives. Frenklach et al. studied the production of particles produced homogeneously by laser scattering analysis [54]. Their findings was that the production of new particles as a function of temperature reached an apparent maximum at 877 1C. The authors acknowledge the difficult task of explaining the decrease in particle formation as the temperature is increased further. Their explanation was rather that the size of the particles, as the temperature is further increased, invokes different optical properties and will thus be read differently by the laser scattering equipment. In their

modeling they made use of 117 elementary reactions and 42 chemical species. Introducing such a large number of components may serve to pin down what reaction chains are dominating. However, care must be taken since one may end up with a nonphysical model if the results are not matched in detail with the experimental data. For reactor modeling a simplified kinetic model is needed. In this context, Wilke et al. [56] supported the two step mechanism from silane to silylene to silicon indicated in Section 2. Their aim was however to describe silicon deposition on wafers in a hot wall reactor rather than in a fluid bed. Alam and Flagan of Caltech developed Free Space and Fluidized Bed technology over several years and came a long way in defining the particle growth mechanisms [57,58]. A central topic in their research is the production and annihilation of intermediate species and how mean free path to a particle surface in combination with these species determined the balance between production of fines, annihilation of fines and heterogeneous growth on fines. Their theory indicates two major fundamentals: (i) The initial number of formed particles will influence the end size distribution since a large number of particles will all compete for the same remaining vapor and thereby limit subsequent particle growth. (ii) To grow large silicon particles, it is necessary to limit the amount of condensible vapor in the reactor. Their solution is to let the same particles pass through a chamber at conditions that permits scavenging of condensible species but suppresses formation of new particles. Alam’s theory continues and aims at describing how to maximize particle growth and suppress formation of new particles. The work continued later with Wu and Flagan [59] who concentrated on finding the critical concentration at which particle formation was suppressed. According to their work, this development was abrupt and changed drastically at specified concentration/temperature relationships. This has important implications in the control of the micro-environment in large scale reactors and the choice of operating conditions. Gates et al. [60] performed experiments on mono crystalline silicon surfaces and demonstrated that silylene and higher order silanes had several orders higher reactivity than monosilane on this surface; this view was also supported by Meyerson et al. [61]. The aim of Gates work was to underline the already established presumption, that silane CVD undergoes a series of intermediate stages from monosilane to silicon deposition. Buss et al. [62] tried to establish a relative sticking coefficient for silane and disilane on a hot silicon surface. The procedure was to subject the surface to a monitored molecular beam and then record the deposition. What was found was that the reaction was influenced by complex surface chemistry that seem to get even more complex as the reaction temperature and pressure were increased. The roughness of the surface got increasingly more rough as the pressure increased. This could indicate deposition of intermediate species and be the same mechanism as seen in porous areas of a silicon bead produced in a monosilane based FBR [18]. Hashimoto et al. [63] investigated the deposition of monosilane on a silicon rod. The publication reports no fines production and 100% density of the silicon product. There are SEM images of the cross sections in the publication and for the 900 1C run the deposited material seems to be quite compact. However, for the 600 1C run the deposited layer appears porous and actually seems to be surface initiated. A reason for being inclined to this conclusion is the formation of silicon pins that comes radially out from the rod. The wall of the rod reactor is kept at about 73 1C according to the publication. If the earlier contributions of Caltech [57,59] is taken into consideration, this will result in quite a low

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

mean chamber temperature and thus suppress the formation of fines but produce a lot of intermediate highly surface-active species. The lack of fines in the filter after these runs might support this view. The experiment procedure has also recently been revisited by Ramos et al. [64], but so far only TCS based experiments have been published. Onischuk et al. [65] investigated agglomeration of fines downstream of a monosilane CVD chamber. Their theory was that the chain of reactions ending up with a hydrogen containing solid would form dipole agglomerates. If investigating the behavior of these agglomerates just after the CVD chamber there should be evidence of attractive interaction forces. The investigation claimed to have found this evidence and the contribution fits well with the established view of how these mechanisms interact. A strictly fluid mechanical approach was suggested by Korec [66], where boundary layer theory was applied to model wall deposition in a hot wall reactor. The boundary layer approach is more useful if the mass transfer is the rate limiting step rather than the surface kinetics.

193

Fig. 2. Silane CVD mechanisms as described by Hsu et al. (i) Heterogeneous deposition on seed particles or the beads, (ii) homogeneous decomposition, (iii) coalescence of fines, (iv) coagulation, (v) scavenging and (vi) heterogeneous deposition on fines.

4. Fluidized bed deposition: microscale effects The phenomena in fluid beds can be divided into that occurring on the macro scale and that on the micro-scale. The macroscale effects are due to the hydrodynamics in the reactor scale and are discussed in more detail in Section 7.1. In this section, we base the discussion on a widely used phenomenological model. This assumes that the bed can be divided into two regions (a) a bubble phase with a large gas voidage containing very little solids and (b) an emulsion phase which contains most of the seed particles. Hence the general terminology of two phase models are used to denote such models. 4.1. Micro-scale effects A number of micro-scale effects occur in these phases (bubble and emulsion) and several research groups have looked into these effects. The major contributions started out with Iya et al. [67] who did several runs in a fixed bed and empty tube reactor. The next major contribution was done by Hsu et al. [68–71] who proposed six different pathways to the formation of silicon in a FBR. The publication is important since several of the later publications on silicon FBR are based on the basis of these ideas. Another substantial fundamental contribution is the Ph.D. thesis of Marra [72] who investigated laser heated pyrolysis of monosilane. The mechanistic model of Hsu et al. is widely used and this is now described. Hsu et al. proposed a set of governing growth mechanisms present in a silicon FBR, see Fig. 2. (1) Heterogeneous deposition on the crystalline silicon beads is direct decomposition of the silane molecule and deposition of a crystalline layer on the silicon beads. Hsu et al. claims this mechanism to be dominant in emulsion alternatively called dense phase; there is however not any discussion on the surface reactions or intermediate reaction stages between silane and silicon in this paper. Another discussion not given much emphasis is the temperature, temperature gradients and concentrations effect on these intermediate reaction stages. (2) The alternative decomposition path is homogeneous decomposition. This process forms silicon nuclei in gas phase. This is presumably more intense in the bubble phase. (3) Coalescence is the formation of a nucleus from several nuclei. There is a discussion on the growth of these nucleus given with reference to Levenspiel [73]. The general idea is that the

nuclei moves with Brownian motion and the probability of meeting another nuclei is the density or the mean free path between these nuclei as well as the temperature. However, the expression might be short on a term accounting for the production and annihilation of silicon radicals and other intermediate species that may effect the probability of the nuclei to nuclei bondage. (4) Coagulation is the formation of silicon clusters. The formation is by solid to solid collisions and the product is therefore not densely packed. For the given process parameters the loss to silicon clusters or fines is stated to be 10%. (5) Scavenging is the recapture of silicon fines by the silicon beads. Since this process will require silicon beads to be present, the process is happening in dense and not bubble phase. A very interesting and frequently cited concept is the use of the Tamman temperature as explanation of the surface adhesive forces. What is referred to as the Tamman temperature is cited in the Hsu paper from Baker [74] who investigated movement of silver particles on a graphite surface. The concept is that for a certain temperature being about 52% of the absolute melting point of the metal, the wetting ability is greatly effected. For silicon this temperature is 610 1C. Above this temperature the fines would stick to the beads whereas for lower temperatures, the fines would be merely loosely bound. To aid the argument chain, the paper refers to their first investigation [68], where SEM images indicate loosely bound particles at lower temperatures and better bound particles at temperatures from 650 1C and up. Based on these conclusions, the ability of the beads to capture fines would be the same throughout the bed as long as the temperature is above the critical temperature of 610 1C. By taking reference in the Tamman idea, the general equations and concepts of fluidized bed particulate filtration may also be used to model scavenging. These concepts include the work Doganoglu et al. [75] and Peters et al. [76]. The latter proposed a set of different capture mechanisms that are active in a fluidized bed filter. Particle (bead) fine collision as affected by reactor level macrohydrodynamics will affect the scavenging rate based on the impact velocity. The different types of capturing includes interception, impaction, Brownian motion and induced electrostatic Forces. Interception is capturing by the drag behind the particle. Impaction is

194

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

collision where the fines path is independent of bead influence and where the fines has an inertia upon impact. In Brownian Motion the fines path is also uninfluenced by the bead, but it is also random. For the induced electrostatic concept the fines is captured by the beads by electrostatic field. There may however be room for improvement in this model. Certain intermediate surface and gaseous species may contribute to the probability of fines to bead bondage upon impact. Temperature, concentration and pressure is known to produce, condense and remove some of these species and an improvement in the model could perhaps take this consideration into account. More empirical insight to what intermediate gaseous and solid species are important in terms of fine scavenging needs to be provided before an improvement in the model may be proposed. (6) Heterogeneous deposition on fines, nucleus. The mechanism is by Hsu considered to be small based on an argumentation that the surface area of these fines will be limited compared to the beads. The mechanism is considered to happen in emulsion phase. This may however not be the case since the findings of Marra [72] suggest heterogeneous growth on nucleus to be favorable over the formation of new nuclei. The paper is important because the several independent investigations have indicated this mechanism to be important.

4.2. Microeffects in reactor models Reactor modeling studies which include fine formation and scavenging are now summarized in the following section. Reactor models at the macro-scale level are discussed in a later Section 7.1. Lai et al. [77] continued the Hsu school and based their FBR and ideal back mixed reactor model on the combined heterogeneous deposition and fines scavenging. The scavenging mechanism is based on the FBR filter models proposed by Doganoglu et al. [75] and Peters et al. [76]. This means that the scavenging probability is the same in the whole bed and only depends on the mean bead diameter and fines concentration. Hsu et al. proposed a new FBR model where local concentration of silane and silicon fines was given a more important role. However, the publication still focuses on the generation of Si nuclei from silane and not a complex multistep chain reaction [71]. The fines scavenging mechanism is based on the proposal of Levenspiel [73]. Furusawa et al. [78] did a series of experiments in a fixed bed and a FSR. The temperature range was from 550 to 700 1C. The assumed reaction was in accordance with Eq. (1). The insignificant production of fines is both in accordance with Hsu et al. [68–71] which proposed that fines first and foremost is produced in bubble and not in dense phase, see Fig. 2. The findings are also in accordance with Alam and with Wu [57,59] which both suggest that the mean distance to a nucleation site will determine the concentration of intermediate reactive species and thus be proportional to the inversely probability of nuclei formation. This theory yields that the fines production will be suppressed in a fixed bed since the distance to a particle and thus a nucleation site will be short at all locations in the reactor. Kunii and Levenspiel [79] suggested a pure FBR model not incorporating fines. The model is central since several of the later publications has used this as reference for model building. Kojima and Kimura [80,81] suggested an improvement to the Hsu model by taking the local silane concentration into consideration. The way this influence is considered is however strictly as a source term to fines production. The contribution is however important since the mechanisms resulting in production of silicon

from silane is complex and since the local concentration of intermediate species is likely to influence the process. More indepth research on the role of intermediate species in regards to fines capture would be of great interest and aid further development. Caussat et al. [82,83] improved the Kunii and Levenspiel model [79] and used some concepts introduced by Kato and Wen [84], they also incorporated some intermediate gas phase species. The model was classically divided in bubble and emulsion phase and the various species were treated differently for these different phases. The kinetic constants were selected from the findings of Fayolle [85]. However, the different species that may or may not constitute the fines were not considered. Mahecha-Botero et al. [86] proposed a general FBR model with selective removal of species. The test case was a hydrochloronation FBR and the results was in good agreement with the experimental dataset. Parker [87] proposed a continued model based on Lai and Hsu. Fines scavenging was as earlier proposed as a product of fines concentration and total bead surface, indirectly referring to the Tamman temperature.

5. Product quality: crystallinity If the amorphous silicon is heated with a controlled temperature rate of 20–40 K min  1 in inert atmosphere, hydrogen releases during the heating. By analyzing the evoluted gas by mass spectroscopy, two peaks of hydrogen are identified, one at 300–500 1C and one at 500–650 1C. It is supposed that the low-temperature peak is associated with elimination of hydrogen from the polyhydride groups, while the high-temperature peak is determined by elimination of hydrogen from the mono-hydride groups [22]. The findings are supported by Scott et al. [88] which investigated heterogeneous deposition of silicon at various temperatures. The conclusion of these experiments was that lower temperature decomposition of monosilane in fact forms a hydride where the hydrogen content will depend on the run parameters. The conclusion might be a bit broad since the individual particles constituting the fines will not have identical history and since small perturbations in the chain reactions will influence the end product greatly. The inhomogeneity and properties of the hydrogenated silicon is discussed by Hansen et al. and Odden et al. [55,52,53] which investigated the properties of silicon powder produced from monosilane through CVD in a Closed Box Reactor under various conditions. A deeper investigation on how the protons are dispersed in the silicon is done by Lucovsky [89,90] for which the authors refer for more details. A general conclusion from his work is that films deposited at low substrate temperatures, o 100 1C,contain the bonded hydrogen predominantly in the polymerized or polyhydride bonding arrangements represented by the structural formula, SiH2n, whereas films deposited at higher temperatures, 4 250 1C, contain the bonded hydrogen predominantly in monohydride groups. Substantial attention is also made to how dopant atoms behave in the system and how the hydrogen incorporation is altered by these dopants. A challenge with fluidized bed material is the possible inclusions of hydrogen containing amorphous silicon. Since the hydrogen in these regions will have to diffuse through crystalline material, the required temperature for hydrogen release is higher. Eventually, long treatment time may provide the possibility of lower temperatures. Experiments on this type of hydrogen release was done by Allen and Boone of ethyl [91,92]. 6. Product quality: porosity Different temperatures and concentrations yield different types of crystal growth. One of the early experiments on the

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

topic was performed by Kamins and Cass [49] which did a series of deposition experiments with reactor temperatures of 650– 1200 1C . The findings was that higher temperatures yielded a rougher deposited film, while lower temperatures tended to produce a more uniform film. This result may be caused be at least two different sources. Homogeneous decomposition is known to be a more substantial problem when going to higher temperatures and the rough film may be a result of deposited intermediate solid species. The other explanation is the tendency of preferred Si CVD growth on crystalline surfaces. If an intermediate gaseous species is produced in the reactor chamber it will have a stronger reactivity on a crystalline surface than on other surfaces. This result has been demonstrated by Beers et al. [50]. The result has also been indicated in the hot rod deposition experiment performed by Hashimoto et al. [63] where a deposition temperature of 700 1C yielded surface initiated growth resulting in pin growth. However, when going to higher temperatures the mobility of the surface species is higher and this mechanism yields a tendency of more uniform films. This mechanism is also demonstrated in the Hasimoto paper since the film uniformity increases when going to higher depositing surface temperatures. The apparent difference in results from Kamins and Hashimoto may be related to the low chamber temperature in the Hashimoto reactor. The layout of the reactor is a single rod coaxially placed in a cooled cylindrical chamber. A feature of the Hashimoto publication is that the wall temperatures and thus the average chamber temperature is kept low in all experiments and this suppresses homogeneous decomposition. The rough depositions in Kamins reactor may be related to depositions of intermediate solid species which are suppressed in the Hashimoto reactor. In the literature these two mechanisms make up the prior art knowledge of porosity enhancing mechanisms; deposition of intermediate solid species and the intermediate gaseous species tendency to grow at selective surfaces. A good understanding of these mechanisms is crucial in order to control them during reactor design. Dahl et al. investigated porosity in silicon beads produced by a silane based FBR [18]. The experimental procedure involved slicing the beads and inspecting them by means of SEM and optical microscopy. What was found was a series of radially dispersed lines of more porous material. The distance between the lines was found to be more or less constant alternatively a combination of two periodical patterns. The discussion in the publication revolves around the origin of these porous lines and the conclusion is that the observed lines is a result of deposition of amorphous silicon and subsequent heterogeneous deposition on the amorphous layer. Deposition on amorphous silicon and subsequent recrystallization is also discussed in the publications of Wu and Flagan [59] and was patented as procedure by Flagan and Alam [58] the same year.

7. FBR modeling and scaleup 7.1. Geldart classification The fluidization characteristics of the particles play a major role in the overall hydrodynamics. Geldart classification is often used to characterize fluidization characteristic of particles. Silicon particles in the range of 202100 mm are classified as type A. These are aeratable, easily fluidized and give small bubbles. The fines formed in the reaction are in this range or even lower. Bed may expand up to certain point without bed bubble formation. Particles above 500 mm are type D and are spoutable and form larger bubbles. The growing seed particles are in this range. Hence

195

in a typical fluid bed we have a bimodal distribution of particles with type A and type D present together. Unfortunately most of the research is on powders of type A since these are the range used in petroleum industry for large scale catalytic cracking process. This presents straightway a problem in scaleup of these systems where both large and small particles are present together. Some large particles scavenge the fines and the data on scavenging coefficient is also not well studied and the effect of reactor scale on this parameter is not well established. The earlier discussion on scavenging mechanisms is useful in this context. 7.2. Reactor models: two phase model and variants A brief review of reactor models with the objective of examining the scaleup aspects is now presented. A widely used type of model is the two phase phenomenological model which assumes that the bed can be divided into two regions: (a) a bubble phase with a large gas voidage containing very little solids and (b) an emulsion phase which contains most of the seed particles as discussed earlier. The heterogeneous reactions take place mainly at emulsion solids. some fines formed here also get scavenged and deposit on the seed particles. The homogeneous reactions occur mainly in the bubble phase. In addition there is a mass exchange between the bubble and emulsion phase. The reactor with smaller bubble sizes promote larger exchange and hence leads to reduced fines formation. This will be a key element to look for in a successful scale-up to large diameter reactors. The key question is how does the bubble diameter change with scaleup. The two phase phenomenological models can further be classified into of four types (20): (A) Patridge and Rowe model (also known as the Davidson–Orcuit model); (B) Kunii–Levenspiel model; (C) modified Kunii–Levenspiel model and (D) Kato–Wen model. These models essentially differ in the manner by which the flow patterns in the bubble and emulsion phase are treated and whether the presence of a cloud phase between the bubble and emulsion phase is explicitly included or not. All the models assume that mass is exchanged between the bubble and the emulsion phase and the mass exchange depends on the bubble size. The Kato–Wen model is somewhat versatile since it treats the bed into a number of interconnected compartments. The size of each compartment can be adjusted (often proportional to the local bubble diameter at that point) and hence various scenarios of mixing pattern of the gas and solid can be examined. Caussat et al. [82] tested the four phenomenological models listed above to study silane pyrolysis in fluid beds. The Kato and Wen model showed closer agreement with the experimental data. Detailed homogeneous reactions were then added to the Kato– Wen model. The major findings were as follows; silane conversion was essentially complete in the first few cm of the bed. The fines are essentially formed by wall deposition of the various polymeric Si species leaving the bed rather than by homogeneous decomposition of silane followed by nucleation. This is in disagreement with other studies where the homogeneous reaction of silane (mainly in the bubble phase) is considered to be responsible for the fine formation. Huang et al. [93] modeled the behavior of a large diameter fluid bed reactor (80 cm) again on the bubble-emulsion phenomenological model. This model was similar to the one proposed by Lai earlier. The major modification was the emulsion phase was considered to be a number of interconnected well mixed tanks rather than one single well mixed tank. The model was compared to experimental data for two runs obtained from an industrial pilot plant and reasonable agreement was found between the model predictions of silane conversion and selectivity to fines. Although one to one comparison was not done with the previous

196

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

models on small diameter beds, it appears that the fine formation rate is larger in larger diameter fluid beds. This may have important bearing during scaleup. The reason attributed to this effect was the poorer bubble to emulsion mass transfer coefficient in larger diameter beds. Additional factors leading to higher fine formation are the low solid holdup in the bubbles, a larger volume for homogeneous nucleation in the bubbles and the lack of scavenging of fines in the bubble phase. The concentration of silane in the emulsion phase was significantly lower than that in the bubble phase indicating that the bubble-emulsion mass transfer resistance is a controlling resistance. Also the silane conversion was almost complete in the lower sections of the bed in the emulsion phase but complete conversion was not achieved in the bubble phase in contrast to the predictions in smaller diameter beds. Based on the model further parametric effects of various operating parameters were examined. The silane conversion increased slightly with the increase in the inlet silane mole fraction but the quantity of fines produced increased as well. Hence there is an optimum value for the inlet concentration of silane for which the productivity will be a maximum. The effect of wall temperature profiling on the productivity was also examined. These parametric studies are useful for reactor optimization and to minimize wall deposition. The model itself coupled with pilot scale studies provides a solid foundation for scaleup to larger sized plants. 7.3. Reactor models: detailed CFD models A second class of model is the detailed fluid dynamic model which uses a point to point momentum balance to compute the local velocity and void fraction and do not explicitly assume the presence of a separate bubble phase. Detailed models can be of two types.

One example is the use of MFIX code (Euler–Euler model) developed by NETL, Morgantown, WV for silane decomposition in a 5 cm diameter and 63 cm tall reactor. It was found that the reactions were essentially complete in the 4 cm region above the distributor. The model predictions for macroscopic quantities (such as the H2 production rate) agreed with the data reported by Caussat et al. [82] but this is not a very accurate test for a detailed computational model. More elaborate pointwise (microscopic level) data for the comparison of the detailed model are needed rather than comparing the overall performance. An interesting observation of this study was a 2-D axisymmetric model showed significant difference compared a full 3-D simulation. No study of the application of DEM to silane pyrolysis has been reported in the literature and a comparative study of the two approaches (Euler–Euler vs discrete element) would be useful. 7.4. Reactor experimental studies On the experimental side, a number of studies have been published on silane pyrolysis. The data has been used for model validation. A problem with these studies when used for scale-up is that most of the published data are for small diameter reactor and the direct use of such data to design a larger diameter reactor is not possible and experimental studies in a pilot scale are still needed. It should noted that small diameter bed operate in the slugging mode since the bubble size is comparable to the bed diameter. It may be noted that the slugging is somewhat reduced with larger particles. Defluidization of the bed is also a possibility upon scale-up due to changing particle size and more pronounced local disturbances in temperature in large diameter reactors. Control strategies to detect and alleviate the problem should be in place. 7.5. Scaleup aspects

(i) Euler–Euler model which treats the two phases (gas and solid) as an interpenetrating continuum and solves for the momentum equations for both gas and solid phases. The model equations are essentially the gas and the solid phase continuity equations, momentum balances with phase interaction terms and the energy balances. These equations are supplemented by a large number of closures (the constitutive relations) for the phase interaction terms. It may be noted that the model predictions are sensitive to the parameters used for the closure. (ii) Discrete element model (DEM) or discrete particle model which treats the gas as a continuum but treats the solid phase by motion of individual particles modeled by the Newton’s second law of motion. These type of models are also known as Euler– Lagrangian models. Although it is physically more realistic to treat the solids as a discrete phase rather than continuous, this model is computationally more difficult than the Euler–Euler model.

Scaleup is usually based on geometric and dynamic similarity considerations. The key dimensionless analysis groups are the Reynolds number at minimum fluidization condition and the Froude number. Also u=umf is kept constant. In that case, the bed behavior will be similar in the sense that you have similar bubble diameters and gas hold up in the bubble and emulsion phase. A factor often overlooked is matching of the reaction parameters. The key dimensionless parameter in this regards is the ratio of the exchange coefficient to the reaction rate constant. In order to match this to a larger scale, the variation of the exchange parameter on the column size is needed. However, the effect of column diameter on exchange coefficient is not well established. The exchange parameter has both diffusional and convection contribution and the two contributions scale up differently with column size. In general the diffusive contribution to the exchange parameter decreases as the column size is increased. 7.6. Spouted bed reactor

A third emerging type of model is the discrete bubble model where the bubbles are treated as though they are discrete pseudo-particles. It is claimed that this type of model is suitable for large diameter beds. In principle, these detailed model should provide significantly more information than the phenomenological model. More insight on the gas distribution, mode and location of gas injection points, local temperature gradients and other information needed for scale-up can be gathered from such models. However, the use of such detailed models is bogged down by large computational time and the need to provide sub-models for many closure terms such as particle–particle interaction etc. In view of this only limited studies of application of such models to silane pyrolysis are published.

Since type D particles are spoutable, a spouted bed reactor can be used as an alternative to the fluid beds. Spouted bed is a modification of the fluid bed where the inlet gas is introduced by a nozzle somewhat above the bed. The particles are fluidized in the region above this nozzle while in the bottom part, the solids circulate as a spout. Some advantages claimed by this design are as follows: (i) larger solid holdup in the spout region which permits heterogeneous reaction and fine capture; (ii) smaller gas residence time and bubble region is reduced reducing fine formation; and (iii) more intimate contact with larger particles and fines promoting scavenging of fines.

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

Modeling of such a reactor needs a spout-emulsion transport region for the bottom part and a bubble-emulsion region for the fluidized part and such a model developed by Pina et al. [94]. Their model is similar to that proposed by Lai et al. [77] earlier who consider a jet-emulsion and a bubble-emulsion region. The main difference is that a detailed population balance model was used to predict the particle size distribution in the bed for both the fines and the larger seed particles while Lai used the method of moments only for the fines. In either of the two approaches, the model for the particle size distribution requires two additional parameters: scavenging constant and the agglomeration constant as discussed earlier. These parameters were fitted to the experimental data in the work of Pina et al. The comparisons were done for five experiments and these parameters had to be adjusted for each case. A unique set of values could not be assigned. Hence the models for particle size prediction do not appear to be predictive and require further development. The fluid bed model itself predicts that the silane conversion is complete in the first 10% of the bed. The model predictions were sensitive to the exchange coefficients needed in the two phase model. In particular the sensitivity was greater for the spout-emulsion exchange coefficient compared to the bubble-emulsion exchange coefficient. The scaleup of spouted beds is also approached in the same manner. Key dimensionless groups are matched to keep the dynamic similarity. Again the Reynolds number and Froude number are the key hydrodynamic parameters while the exchange coefficient to the rate coefficient is the key reactor level parameter. Also scale up to larger diameter may be problematic due to central introduction of the gas rather than a uniform injection as is the case in traditional fluid beds. Key parameters to study in scaleup are the spout diameter and the exchange rate between the spoutannular regions. Discrete element simulation may be useful tool in this regards since it tracks the motion of individual particles. The published results appear to be sensitive to the particle– particle interaction constants. Also reacting systems with both homogeneous and heterogeneous reactions have not been studied. More work can be expected in this area in view of the importance of the process.

8. Design concepts for FBR Various designs and alternations have been proposed through the years and the interested reader is referred to the review of Filtvedt et al. [24] for a broader insight to the development history. This section will go through some highlights and aims to link the different designs to the various mechanisms active in a silicon CVD FBR. Early designs were closely linked to reactors built for catalytic cracking [91,92]. One of the early topics was generating the seed particles. The seed particles naturally needs to be as pure as the finished product and this might be challenging if conventional crushing techniques are applied. There are generally two established approaches to the solution. Either produce seeds from a part of the produced material or to produce the seeds directly from the reactant gas. A version of the first solution was proposed by Iya of Union Carbide Corporation who suggested the use of a high velocity jet to produce seed particles within the reactor [95]. The problem of the solution was that this high velocity may produce large bubbles which will increase the mean free path to a solid nucleation site when the silane molecule reaches thermal decomposition [68,69]. This will in turn yield higher production of fines as well as the long high temperature dwelling time will increase the probability of producing amorphous silicon complexes which is harder to capture.

197

A version of the alternative design was proposed by Hsu et al. [70] which suggested producing fines within a separate reactor and introducing them to a FBR after some growth. One of the benefits of the solution is that the particles produced in a FSR tends to be spherical or semi spherical and the use of spherical seeds will enhance the probability of ending up with a spherical finished product. One challenge with the method is to grow the seeds in such a way that they consist of crystalline silicon of low porosity. A more recent version of the solution was proposed by MEMC in 2007 [11]. The problem of fines generation is known to be related to a combination of mean free path to a solid nucleation site, the concentration of reactant gas and the temperature. Within a FBR, increasing the mean free path is the same as increasing the virtual velocity and thus the mean bubble diameter. Gautreaux et al. [91] acknowledged this relation and proposed running their FBR in two modes. A high velocity fines generating mode and a low velocity fines capturing mode. The solution was taken further by Lord et al. of ASIMI which proposed a conical reactor in order to have the two modes at the same time [96,97]. REC silicon continued the development of the ASIMI technology and has proposed several later improvements [9,10]. Another classical Si CVD FBR problem is heating. Substantial effort has been made in resolving this issue but this review will briefly present some of the key concepts. The earliest designs were based on basic wall heating and the whole bed was arranged as one zone with heating beginning just above a common inlet for both reactant and fluidization gas. The fundamental problem is to heat the beads without heating the gas in order to suppress fines formation at the same time as the reactant gas reaches decomposition upon contact with the beads. The first inherent problem is to minimize heat loss. The reactant gas needs to enter at a temperature below 350 1C but also needs to be subjected to a temperature above 650 1C within the reactor to assure correct crystal growth. An early attempt to resolve this issue was also proposed by Iya [98]. The solution was to divide the reactor into two zones. The premixed reactant and fluidization gas was injected to the lower part of the bed which was kept cold while the uppermost part of the reactor was heated. The two zone idea was taken further by Kim et al. [99,100] and later Wacker Chemie [101,102] which proposed to have the heating zone below the reaction zone and also to heat the beads by means of microwaves in order to prevent the fluidization gas from being a main heat carrier. Controlling a microwave field in a silicon CVD FBR has so far been challenging. The idea of dividing the reactor into zones was taken further by Kulkarni et al. of MEMC [103,14] which proposed dividing the reactor into radial zones. The bottom distribution plate was divided in a central zone where the reactant gas was introduced and a radially surrounding inert zone where the fluidization gas was introduced.

9. Conclusion Deposition in a silane based fluidized bed is complex and designing a reactor capable of producing a product of high quality and low operation cost is challenging. Interplay of reaction mechanisms, reactor scale microscale effects and macroscale hydrodynamic effects all need to be combined and used judiciously to arrive at an optimum design and to obtain the desired product quality. This present review provides the key summary of the literature information on this field and hopefully aid in future improvements of this technology. The kinetics of silicon deposition from silane is fairly complicated and there seems to be some controversy on the detailed

198

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

pathway leading to the deposition. It is agreed that there exist a series of intermediate reactions and species between monosilane and silicon and the local physical properties within the reactor will influence the production and annihilation of these species. The general process of monosilane CVD is the gradual removal of hydrogen moving into more complex structures. The advantage over trichlorosilane is less reverse mechanisms. The overall yield of the process may thus be higher. The first decomposition is known to start at temperatures as low as 370 1C depending on several conditions, and a series of different stages may be active up to 610 1C where the last bounded hydrogen will be released if the dwelling time is long enough. The fines produced within the FBR may consist of amorphous silicon, silicon hydride, silicon-hydrogen complexes and even crystalline silicon depending on operation conditions and the process history of the individual particles constituting the fines. It should be noted that if the finished beads exhibit amorphous hydrogen containing inclusions and further that the time for hydrogen release through the crystalline parts may be long at temperatures as low as 610 1C. Higher temperatures have proven to reduce this time substantially. Certain intermediate species such as silylene has proven to have a higher surface reactivity than monosilane. In order to enhance heterogeneous or surface deposition, the mean free path between reactive species needs to be kept high compared to the mean free path to a solid surface. There exists at least two strategies to comprehend this problem. Either decreasing the mean free path to the beads or increasing the mean free path between the reactive species. Decreasing the distance between the beads is most efficiently done by reducing the mean bubble size or alternatively by utilizing complex designs where the reactant gas is introduced to parts of the reactor where the bubble size is known to be smaller. Increasing the mean free path between reactive species may be done either by reducing the pressure or the use of a dilution. Both these strategies decreases the capacity and a known solution is injection of reactant gas at several locations along the bed in order to keep a larger active production volume while at the same time limiting the concentration of silanes at all locations. There are several indications that the reactivity of the intermediate species is temperature dependent and keeping the temperature sufficiently low is crucial in order to prevent high fines formation. The drawback of this strategy is that the silicon formation below 610 1C may be amorphous and contain hydrogen. A suggested option is to operate the reactor just above this temperature or to have different zones throughout the reactor and alternate the beads between these zones. Fines capture is complex and includes the production of surface reactive species which interact with other intermediate species. The mechanisms are not sufficiently investigated and more experiments on amorphous silicon encapsulation and intermediate species interaction is important to aid further process development. Hydrogen will release from amorphous silicon at 610 1C upon going from amorphous to crystalline structure. The structure transaction is not clearly visible on SEM, but may be identified by X-ray diffraction or TEM. It is probable that crystalline fines will be harder to encapsulate than amorphous due to the lack of dangling bonds and thus a definite reduction in sticking coefficient. The necessity of not passing this transaction point is therefore crucial if the design aims at enhancing fines scavenging. Fluidized bed design for silicon production is complex and more research is necessary in order to understand the fundamental mechanisms active within the bed. Models based on two

phase description of the bed seems to be in vogue and is fairly useful for scaleup if backed by data on pilot plants and cold flow studies. A number of variations of the two phase models are also available and one promising model is the compartmental model following the lines of Wen and Kato. Scaleup of fluid beds is also a challenging problem especially when coupled with the complex chemistry of silicon deposition. The information provided in this paper will serve as a useful starting guideline.

Acknowledgments We thank The Research Council of Norway project num. 188026 in collaboration with Hydro and Umoe Solar for all funding and support in this project.

References [1] Yijun Liu, Yingqi Liu, The industrial research about China solar PV based on the price of polysilicon, Advanced Materials Research 383–390 (2012) 7043–7048. [2] Rob Steemana, Jeannie Yonga, Yvind Mjsb, Adolphus Songb, Integrating the value chain: the impact of silicon quality on cell performance, Energy Procedia 15 (2012) 20–29. [3] Johannes Bernreuter, The polysilicon market—perspectives on demand, supply and production technologies, in: Silicon for the Chemical and Solar Industry, vol. 11, 2012, pp. 281–284 (vol. 15, 2012, pp. 20–29). [4] S.P. Walch, C.E. Dateo, Thermal decomposition pathways and rates for silane, chlorosilane, dichlorosilane and trichlorosilane, Journal of Physical Chemistry 105 (2001) 2015–2022. [5] W.A.P. Claasen, J. Bloem, The nucleation of CVD silicon on SiO2 and Si3N4 substrates, Journal of the Electrochemical Society 127 (1) (1980) 194–202. ¨ [6] Tim Popken, Raymund Sonnenschein, Process for producing silicon, Assigned to Evonik Degussa GmbH, EP 1 720 800 B1 (2005). ˜ izo, A. Luque, Temperature homogeneity of [7] G. del Coso, I. Tobı´as, C. Can polysilicon rods in a Siemens reactor, Journal of Crystal Growth 299 (2007) 165–170. [8] Wen-Ching Yang, Handbook of Fluidization and Fluid-particle Systems, CRC Press, 2003. [9] Paul E. Ege, Jeffrey Hansen, Levi C. Allen, Silicon spout-fluidized bed, Assigned to REC Silicon, WO 2007.012027 (2007). [10] W. Wayne Osborne, Michael V. Spangler, Levi C. Allen, Robert J. Geertsen, Paul E. Ege, Walter J. Stupin, Gerald Zeininger, Silicon fluid bed reactor, Assigned to REC, US 8.075.692 B2 (2011). [11] Jameel Ibrahim, Melinda G. Ivey, Timothy Dinh Truong, High purity granular silicon and method of manufacturing the same, Assigned to MEMC, EP 1 833 759 B1 (2007). [12] John D. Holder, John D. Hilker, Systems and methods for measuring and reducing dust in granular material, Assigned to MEMC, US 7.291.222 (2007). [13] Melinda G. Ivey, Timothy D. Truong, Method of making dust-free granular silicon, Assigned to MEMC, EP 1 900 686 A1 (2008). [14] Milind S. Kulkarni, Puneet Gupta, Balaji Devulapalli, Fluidized bed reactor systems and methods for reducing the deposition of silicon on reactor walls, Assigned to MEMC, WO 2010/002815 A2 (2010). [15] Henry E. Erk, Fluidized bed reactor systems and distributors for use in same, Assigned to MEMC, US 2011/0158857 (2011). [16] MEMC Inc, Building on the Past, Ready for the Future: A Fiftieth Anniversary Celebration of MEMC Electronic Materials, Inc, vols. 55–56, 2011, pp. 38–40, MEMC electronic materials /www.memc.comS. [17] Carlyle S. Herrick, David W. Woodruff, The homogeneous nucleation of condensed silicon in the Gaseous Si–H–Cl system, The Journal of the Electrochemical Society: Solid-state Science and Technology (1984) 2417–2422. [18] M.M. Dahl, A. Bellou, D.F. Bahr, M.G. Norton, E.W. Osborne, Microstructure and grain growth of polycrystalline silicon in fluidized bed reactors, Journal of Crystal Growth 311 (2009) 1496–1500. [19] J.M. Jasinski, B.S. Meyerson, B.A. Scott, Mechanistic studies of chemical vapor deposition, Annual Review Physical Chemistry 38 (1987) 109–140. [20] Joseph M. Jasinski, Stephen M. Gates, Silicon chemical vapor deposition one step at a time: fundamental studies of silicon hydride chemistry, Accounts of Chemical Research 24 (1991) 9–15. [21] Carmen J. Giunta, Richard J. McCurdy, Jonathan D. Chapple-Sokol, Roy G. Gordon, Gas-phase kinetics in the atmospheric pressure chemical vapor deposition of silicon from silane and disilane, Journal of Applied Physics 67 (2) (1990) 1062–1075. [22] A.A. Onischuk, V.N. Panfilov, Mechanism of thermal decomposition of silanes, Russian Chemical Reviews 70 (4) (2001) 321–332. [23] M. Cardona, Vibrational spectra of hydrogen in silicon and germanium, Physica Status Solidi (b) 118 (1983) 464–481.

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

[24] W.O. Filtvedt, M. Javidi, A. Holt, M.C. Melaaen, E. Marstein, H. Tathgar, P.A. Ramachandran, Development of fluidized bed reactors for silicon production, Solar Energy Materials & Solar Cells 94 (2010) 1980–1995. [25] Li Jianlong, Chen Guanghui, Zhang Pan, Wang Weiwen, Duan Jihai, Technical challenges and progress in fluidized bed chemical vapor deposition of polysilicon, Chinese Journal of Chemical Engineering 19 (5) (2011) 747–753. [26] T.R. Hogness, Thomas L. Wilson, Warren C. Johnson, The thermal decomposition of silane, Journal of the American Chemical Society 58 (1936) 108–112. [27] K. Stokland, Investigations on silanes with light and heavy hydrogen, Det Kongelige Norske Videnskapelige Selskaps skrifter 3 (1948). [28] J.H. Purnell, R. Walsh, The pyrolysis of monosilane, Proceedings of the Royal Society Series A 293 (1966) 543–561. [29] M.A. Ring, M.J. Punentes, H.E. O’Neal, Pyrolysis of monosilane, Journal of the American Chemical Society 92 (16) (1970) 4845–4848. [30] R. Robertson, D. Hils, A. Gallagher, Silane pyrolysis, Chemical Physics Letters 103 (5) (1984) 397–404. [31] M.A. Ring, H.E. O’Neal, Kinetics and mechanism of silane decomposition, in: Proceedings of the Flat Plate Solar Array Project Workshop on the Science of Silicon, Material Preparation, DOE/JPL, 1012-81, 1983, pp. 63–75. [32] Pauline Ho, William G. Breiland, Observation of Si2 in a chemical vapor deposition reactor by laser excited fluorescence, Applied Physics Letters 44 (1) (1984) 51–53. [33] J. Howard Purnell, Robin Walsh, Some comments on kinetics and mechanisms in the pyrolysis of monosilane, Chemical Physics Letters 110 (3) (1984) 330–334. [34] Michael E. Coltrin, Robert J. Kee, James A. Miller, A mathematical model of the coupled fluid mechanics and chemical kinetics in a chemical vapor deposition reactor, Journal of the Electrochemical Society 131 (2) (1984) 425–434. [35] William G. Breiland, Michael E. Coltrin, Pauline Ho, Comparisons between a gas-phase model of silane chemical vapor deposition and laser-diagnostic measurements, Journal of Applied Physics 59 (9) (1986) 3267–3273. [36] Michael E. Coltrin, Robert J. Kee, James A. Miller, A mathematical model of silicon chemical vapor deposition, Annual Review of Physical Chemistry 38 (1987) 109–140. [37] Michael E. Coltrin, Robert J. Kee, Greg H. Evans, A mathematical model of the fluid mechanics and gas-phase chemistry in a rotating disk chemical vapor deposition reactor, Journal of the Electrochemical Society 136 (3) (1989) 819–829. [38] J.J. O’Brien, G.H. Atkinson, Role of silylene in the pyrolysis of silane and organosilanes, The Journal of Physical Chemistry 92 (1988) 5782–5787. [39] Pauline Ho, Michael E. Coltrin, William G. Breiland, Laser-induced fluorescence measurements and kinetic analysis of Si atom formation in a rotating disk chemical vapor deposition reactor, The Journal of Physical Chemistry 98 (40) (1994) 10138–10147. [40] Frank Slootman, Jean-Claude Parent, Homogeneous gas-phase nucleation in silane pyrolysis, Journal of Aerosol Science 25 (1) (1993) 15–21. [41] Rosa Becerra, Robin Walsh, Some mechanistic problems in the kinetic modelling of monosilane pyrolysis, The Journal of Physical Chemistry 96 (1992) 10856–10862. [42] A.A. Onischuk, V.P. Strunin, M.A. Ushakova, V.N. Panfilov, On the pathways of aerosol formation by thermal decomposition of silane, Journal of Aerosol Science 28 (2) (1997) 207–222. [43] J.O. Odden, P.K. Egeberg, A. Kjekshus, Thermal decomposition of pressurized monosilane, International Journal of Chemical Kinetics 38 (2006) 309–321. [44] R. Cadoret, Mechanisms of Silicon monocrystalline growth from SiH4/H2 at reduced pressures, Journal of Crystal Growth 61 (1983) 259–274. [45] J. Bloem, Y.S. Oei, H.H.C. de Moor, J.H.L. Hanssen, L.J. Giling, Epitaxial growth of silicon by CVD in a hot-wall furnace, Journal of the Electrochemical Society 132 (8) (1983) 1973–1980. [46] Hong H. Lee, Silicon growth at low temperatures SiH4–HCl–H2 system, Journal of Crystal Growth 69 (1984) 82–90. [47] M.H. Brodsky, Manuel Cardona, J.J. Cuomo, Infrared and Raman spectra of the silicon-hydrogen bonds in amorphous silicon prepared by glow discharge and sputtering, Physical Review B 16 (8) (1977) 3556–3571. [48] C.C. Mai, T.S. Whitehouse, R.C. Thomas, D.R. Goldstein, Properties of polycrystalline silicon deposited on silicon nitride layers, Journal of the Electrochemical Society 118 (2) (1971) 331–334. [49] T.I. Kamins, T.R. Cass, Structure of chemically deposited polycrystallinesilicon films, Thin Solid Films 16 (1973) 147–165. [50] A.M. Beers, H.T.J.M. Hintzen, H.G. Schaeken, J. Bloem, CVD silicon structures formed by amorphous and crystalline growth, Journal of Crystal Growth 64 (1983) 563–571. [51] W.B. Pollard, G. Lucovsky, Phonons in polysilane alloys, Physical Review B 26 (6) (1982) 3172–3180. [52] J.O. Odden, P.K. Egeberg, A. Kjekshus, From monosilane to crystalline silicon, Part I: decomposition of monosilane at 690–830 K and initial pressures 0.1–6.6 MPa in a free-space reactor, Solar Energy Materials & Solar Cells 86 (2005) 165–176. [53] J.O. Odden, P.K. Egeberg, A. Kjekshus, From monosilane to crystalline silicon. Part III. Characterization of amorphous, hydrogen-containing silicon products, Journal of Non-Crystalline Solids 351 (2005) 1317–1327. [54] Michael Frenklach, Larry Ting, Hai Wang, Martin J. Rabinowitz, Silicon particle formation in pyrolysis of silane and disilane, Israel Journal of Chemistry 36 (1996) 293–303.

199

[55] Eddy Walther Hansen, Arne Kjekshus, Jan Ove Odden, On the hydrogen in hydrogen-containing amorphous silicon, Journal of Non-Crystalline Solids 353 (2007) 2734–2743. [56] T.E. Wilke, K.A. Turner, C.G. Takoudis, Chemical vapor deposition under reduced pressure in hot-wall reactors, Chemical Engineering Science 41 (4) (1986) 643–650. [57] M.K. Alam, R.C. Flagan, Controlled nucleation aerosol reactors: production of bulk silicon, Aerosol Science and Technology 5 (1986) 237–248. [58] Richard C. Flagan, Mohammed K. Alam, Reactor for producing large particles of materials from gases, Assigned to California Institute of Technology, US Pat 4.642.227 (1987). [59] Jin Jwang Wu, Richard C. Flagan, Onset of runaway nucleation in aerosol reactors, Journal of Applied Physics 61 (4) (1987) 1365–1371. [60] S.M. Gates, B.A. Scott, D.B. Beach, R. Imbihl, J.E. Demuth, Chemisorption of silanes on the Si(111)-(7  7) surface, Journal of Vacuum Science and Technology A 5 (4) (1987) 628–629. [61] B.S. Meyerson, B.A. Scott, R. Tsui, Experimental and chemical kinetic modelling study of silicon CVD from monosilane and disilane, Chemtronics 1 (1986) 150–155. [62] Richard J. Buss, Pauline Ho, William G. Breiland, Michael E. Coltrin, Reactive sticking coefficients for silane and disilane on polycrystalline silicon, Journal of Applied Physics 63 (April (8)) (1988) 2808–2819. [63] Kenji Hashimoto, Kouichi Miura, Takao Masudo, Masaaki Toma, Hiroyuki Sawai, Motoaki Kawase, Growth kinetics of polycrystalline silicon from silane by thermal chemical vapor deposition method, Journal of the Electrochemical Society 137 (3) (1990) 1000–1007. [64] A. Ramos, C. del Canizo, J. Valdehita, J.C. Zamorano, A. Rodriguez, A. Luque, On track for solar grade silicon through a Siemens process-type laboratory reactor: operating conditions and energy savings, Silicon for the Chemical and Solar Industry 11 (2012) 231–242. [65] A.A. Onischuk, S. di Stasio, A.I. Levykin, V.P. Strunin, V.V. Karasev, K.K. Sabelfeld, V.N. Panfilov, The role of dipole interactions in coagulation of silicon agglomerates, Journal of Aerosol Science 31 (1) (2000) S950–S951. [66] Jacek Korec, Deposition of polysilicon as an example of a pyrolytic CVD process, Journal of Crystal Growth 49 (1980) 547–558. [67] S.K. Iya, R.N. Flagella, F.S. DiPaolo, Heterogeneous decomposition of silane in a fixed bed reactor, Journal of the Electrochemical Society 129 (7) (1982) 1531–1535. [68] G. Hsu, R. Hogle, N. Rohatgi, A. Morrison, A fines in fluidized bed silane pyrolysis, Journal of the Electrochemical Society: Solid-State Science and Technology (1984) 660–663. [69] G. Hsu, A. Morrison, N. Rohatgi, R. Lutwack, T. MacConnell, Fluidized bed silicon deposition, in: IEEE Photovoltaic Specialist Conference (7 Kissimee), 1984. [70] George C. Hsu, Harry Levin, Richard A. Hogle, Ananda Praturi, Ralph Lutwack, Fluidized bed silicon deposition from silane, Assigned to California Institute of Technology, US Pat 4.444.811 (1984). [71] George Hsu, Naresh Rohatgi, John Houseman, Silicon particle growth in a fluidized-bed reactor, AIChE Journal 33 (5) (1987) 785–791. [72] Robert A. Marra, Homogeneous Nucleation and Growth of Silicon Powder from Laser Heated Gas Phase Reactants, Ph.D. Thesis, Massachusetts Institute of Technology, 1982. [73] O. Levenspiel, The Chemical Reactor Omnibook, Oregon State University Bookstore Inc, 1979. [74] R.T.K. Baker, The relationship between particle motion on a graphite surface and Tammann temperature, Journal of Catalysis 78 (1982) 473–476. [75] Y. Doganoglu, V. Jog, K.V. Thambimuthu, R. Clift, Removal of fine particulates from gases in fluidized beds, Transactions on IChemE 56 (1978) 239–248. [76] M.H. Peters, Liang-Shih Fan, T.L. Sweeney, Simulatin of particulate removal in solid fluidized beds, AlChE Journal 28 (1) (1982) 39–49. [77] S. Lai, M.P. Dudukovic, P.A. Ramachandran, Chemical vapor deposition and homogeneous nucleation in Fluidized bed reactors: silicon from silane, Chemical Engineering Science 41 (4) (1986) 633–641. [78] Takehiko Furusawa, Toshinori Kojima, Hiroyuki Hiroha, Chemical vapor deposition and homogeneous nucleation in monosilane pyrolysis within interparticle spaces—application of fines formation analysis to fluidized bed CVD, Chemical Engineering Science 43 (8) (1988) 2037–2042. [79] Daizo Kunii, Octave Levenspiel, Fluidized reactor models 1. For bubbling beds of fine, intermediate, and large particles 2 for the lean phase: freeboard and fast fluidization, Industrial and Engineering Chemistry 29 (1990) 1226–1234. [80] T. Kojima, T. Kimura, M. Matsukata, Development of numerical model for reactions in fluidized bed grid zone—application to chemical vapor deposition of polycrystalline silicon by monosilane pyrolysis, Chemical Engineering Science 45 (8) (1990) 2527–2534. [81] T. Kimura, T. Kojima, Numerical model of a fluidized bed reactor for polycrystalline silicon production-estimation of CVD and Fines formation, Journal de Physique IV Colloque cZ, suppl. au Journal de Physique 11 1 (September) (1991) 103–110. [82] B. Caussat, M. Hemati, J.P. Couderc, Silicon deposition from silane or disilane in a fluidized bed-Part II: theoretical analysis and modeling, Chemical Engineering Science 50 (22) (1995) 3625–3635. [83] N. Reuge, L. Cadoret, B. Caussat, Multifluid Eulerian modelling of a silicon fluidized bed chemical vapor deposition process: analysis of various kinetic models, Chemical Engineering Journal 148 (2009) 506–516. [84] K. Kato, C.Y. Wen, Bubble assemblage model for fluidized bed catalytic reactors, Chemical Engineering Science 24 (1969) 1351–1369.

200

W.O. Filtvedt et al. / Solar Energy Materials & Solar Cells 107 (2012) 188–200

[85] F. Fayolle, Silicon Deposition from Silane or Disilane in a Fluidized Bed—Part II: Theoretical Analysis and Modeling, Ph.D. Thesis, I.N.P. Toulouse, 1993. [86] Andres Mahecha-Botero, John Grace, Said S.E.H. Elnashaie, C. Jim Lim, Comprehensive modeling of gas fluidized-bed reactors allowing for transient, International Journal of Chemical Reactor Engineering 4 (2006). (Article A11). [87] James M. Parker, Validation of CFD model for polysilicon deposition and production of silicon fines in a silane deposition FBR, International Journal of Chemical Reactor Engineering 9 (2011), A40. [88] B.A. Scott, R.D. Estes, J.M. Jasinski, The role of surface reactions in monosilane pyrolysis, Journal of Chemical Physics 89 (4) (1988) 2544–2549. [89] Gerald Lucovsky, Local bonding of hydrogen in a-Si:H, a-Ge:H and a-Si, Ge:H alloy films, Journal of Non-Crystalline Solids 76 (1985) 173–186. [90] Gerald Lucovsky, Hydrogen in amorphous silicon: local bonding and vibrational properties, Journal of Non-Crystalline Solids 141 (1992) 241–256. [91] Marcelian F. Gautreaux, Robert H. Allen, Polysilicon fluid bed process and product, Assigned to Ethyl Corporation, US Pat 4.784.840 (1988). [92] Robert H. Allen, James E. Boone, Process for preparing polysilicon with diminished hydrogen content by using a fluidized bed with a two-step heating process, Assigned to Ethyl Corporation, US Pat 5.242.671 (1993). [93] Yue Huang, P.A. Ramachandran, M.P. Dudukovic, Fluidized bubbling bed reactor model for silane pyrolysis in solar grade silicon production, in: Paper 463c, AIChE Annual Meeting, Salt Lake City, November, 2007. ˜ a, Vero´nica Bucala´, Noemı´ Susana Schbib, Paul Ege, Hugo Ignacio [94] Juliana Pin de Lasa, Modeling a silicon CVD spouted bed pilot plant reactor, International Journal of Chemical Reactor Engineering 4 (2006) (Article A9).

[95] Sridhar K. Iya, Fluid jet seed particle generator for silane pyrolysis, Assigned to Union Carbide Corporation, US Pat 4,424,199 (1984). [96] Stephen M. Lord, Robert J. Milligan, Method for silicon deposition, Assigned to Advanced Silicon Materials Inc, US Pat 5,798,137 (1995). [97] Stephen M. Lord, Robert J. Milligan, Silicon deposition apparatus, Assigned to Advanced Silicon Materials Inc, US Pat 5,810,934 (1998). [98] Sirdhar K. Iya, Zone heating for fluidized bed silane pyrolysis, Assigned to Union Carbide Corporation, US Pat 4,684,513 (1987). [99] Hee Y. Kim, Young M. Song, Jong Y. Jeon, Dae H. Kwon, Kang M. Lee, Jae S. Lee, Dong S. Park, Heating of fluidized bed reactor by microwaves, Assigned to Korea Research Institute of Chemical Technology, US Pat 5,374,413 (1994). [100] Hee Y. Kim, Young M. Song, Jong Y. Jeon, Dae H. Kwon, Kang M. Lee, Jae S. Lee, Dong S. Park, Fluidized bed reactor heated by microwaves, Assigned to Korea Research Institute of Chemical Technology, US Pat 5,382,412 (1995). [101] Franz Schreieder, Young Kim, Process for preparing highly pure silicon granules, Assigned to Wacker-Chemie GmbH, US Pat 6,007,869 (1999). [102] Harald Hertlein, Rainer Hauswirth, Method and device for producing granulated polycrystalline silicon in a fluidized bed reactor, Assigned to Wacker-Chemie AG (GmbH), US Pat 2008.0.241.046 (2008). [103] Milind S. Kulkarni, Puneet Gupta, Balaji Devulapalli, Jameel Ibrahim, Vithal Revankar, Kwasi Foli, Fluidized bed reactor systems and methods for reducing the deposition of silicon on reactor walls, Assigned to MEMC Electronic Materials Inc, US Pat 2009.0.324479 A1 (2009).