Você está na página 1de 13

From Monosilane to

Crystalline Silicon, Part II:


Kinetic Considerations on
Thermal Decomposition of
Pressurized Monosilane
J. O. ODDEN,1 P. K. EGEBERG,1 A. KJEKSHUS2
1
2

Department of Natural Sciences, Agder University College, P. O. Box 422, N-4604 Kristiansand, Norway
Department of Chemistry, University of Oslo, P. O. Box 1033 Blindern, N-0315 Oslo, Norway

Received 10 March 2004; revised 10 January 2005; accepted 14 November 2005


DOI 10.1002/kin.20164
Published online in Wiley InterScience (www.interscience.wiley.com).

ABSTRACT: Kinetic aspects of the thermal decomposition of monosilane at 690830 K and initial pressures of 0.13.7 MPa in a free-space reactor are considered. Neglecting the preparatory initiation period for the reaction (which is difficult to evaluate under the present dynamic
conditions), the onset temperature for the decomposition is stipulated to some 700710 K,
independent of the initial monosilane pressure. The overall reaction appears to be of first order throughout the progressing decomposition process. We observe considerably lower reaction rates under the high-pressure conditions than existing models in the literature suggest.
A modified model is proposed that simulates the observed reaction rates within 1% and moreover predicts credible concentrations of the involved gaseous species. A key feature of the
modified model is incorporation of two third-body assisted surface reactions, which generate
C 2006 Wiley Periodicals, Inc. Int J Chem
monosilane from disilane and disilane from trisilane. 
Kinet 38: 309321, 2006

INTRODUCTION
Silicon is an important constituent of microelectronics and photovoltaic cells (PVCs). The main source of
silicon for PVCs is electronic-grade silicon waste from
the industry. At present, there is no large-scale process
Correspondence to: e-mail: A. Kjekshus; arne.kjekshus@kjemi.
uio.no.
Contract grant sponsor: Research Council of Norway.
Contract grant sponsor: Elkem Research.
Contract grant sponsor: Agder Energy.
c 2006 Wiley Periodicals, Inc.


dedicated to production of solar-grade silicon, and the


rapid growth of the photovoltaic industry is about to
exhaust the sources of electronic-grade silicon scrap.
The expected deficiency in the supply of solar-grade
silicon has led to a renewed interest in the pyrolysis of
monosilane. One process that is considered potentially
interesting is pyrolysis of monosilane in free-space reactors. It seems probable that the economy in such a
process could be improved by turning to elevated reactor pressures and higher inlet concentrations of monosilane, and these aspects have been the main target of the
present project.

310

ODDEN, EGEBERG, AND KJEKSHUS

Pyrolysis of monosilane has been studied in staticreactor experiments [1,2], flow-through reactors [35],
and shock-tube experiments [6,7]. In addition, considerable attention has also been focused on the use of
fluidized-bed reactors, in which silicon is being deposited on seed particles introduced before the onset
of the decomposition starts [811]. An interesting variant of the reaction conditions is to perform the thermal
decomposition on pressurized pure monosilane. The
present series of papers [1214] report on such experiments, and this communication focuses on kinetic
aspects.
The pioneering work on the thermal degradation of
monosilane in general, and the kinetics in particular,
is a paper by Purnell and Walsh in 1966 [1]. These
authors divide the decomposition reaction into four
stages with different reaction-rate regimes. The first
(initiation) stage is reported to cover the range with
01% conversion and a 3/2-order rate-law dependence
with regard to monosilane degradation. In the second
transitional stage, up to about 10% conversion the rate
is accelerated. Then follows a third, first-order stage
with 1030% conversion, and finally a stage where hydrogen inhibition becomes significant. Several papers
on kinetics and reaction mechanisms of the pyrolysis
of monosilane have used the model concepts of Purnell
and Walsh as the starting point. There appears to be
general consent [2,1517] that the reaction responsible
for the initial 3/2-order stage is a hydrogen elimination
reaction to silylene:
SiH4 (g) SiH2 (g) + H2 (g)

(1)

with a first-order pressure dependence. Owing to the


increasing amount of silylene obtained according to
reaction (1), disilane and trisilane (probably also higher
silanes) are generated by the reactions
SiH2 (g) + SiH4 (g) Si2 H6 (g)

(2)

SiH2 (g) + Si2 H6 (g) Si3 H8 (g)

(3)

The steps which now follows are more uncertain, but


the change from one-and-a-half to first-order kinetics
is believed to result from the appearance of SiH3 SiH
(silylsilylene), which eliminates monosilane through a
circle of reactions [2,15]:
Si3 H8 (g) Si2 H6 (g) + SiH2 (g)

(4)

Si2 H6 (g) SiH3 SiH(g) + H2 (g)

(5)

SiH3 SiH(g) + SiH4 (g) Si3 H8 (g)

(6)

Others (see [16]) have pointed at the importance of


heterogeneous surface reactions promoted by the de-

position of (SiH2 )x to explain the increased reaction


order in the monosilane conversion. Silylsilylene and
its isomer as well as higher silylsilylenes have also been
suggested [18] as intermediates in processes leading to
product polymerization and wall deposits [2]. Many
suggestions in the literature about key species in the
formation of the solid deposits illustrate the difficulty
in accounting for all homogeneous and heterogeneous
reactions that take place at the later stages in the decomposition of monosilane. In addition to the mentioned silylsilylenes, nucleated Si and Si2 have been
suggested as important species in the solid deposition
(see [19] and references therein) while short residence
times and lower total pressures [3] as well as lower
temperatures [20] seem to facilitate solid deposition
from SiH4 . At longer residence times and higher total
pressures, however, Si2 H6 [3] and even higher silanes
like Si3 H8 [15,16] become increasingly important in
the solid deposition.
There is more consent that the inhibiting effect of
hydrogen on the rate of the monosilane decomposition is due to regeneration of monosilane from silylene and hydrogen (the reverse of reaction (1)) [2,16].
The inhibiting effect of hydrogen on the pyrolysis of
monosilane was in fact established already in 1936 by
Hogness et al. [21], who also concluded that the inhibition is largest when the decomposition is performed
at low temperatures and high partial pressures of hydrogen. However, the hydrogen is not just responsible
for reduced rates in the homogeneous reactions. The
growth of the silicon particles is also inhibited by high
partial pressures of hydrogen during the decomposition
[12]. At a constant partial pressure of monosilane, and
partial pressure of hydrogen in the range 1301300 Pa,
the overall silicon surface is reported to grow at a rate
proportional to [PH2 ]1/2 at the lower pressures and
proportional to [PH2 ]1 at the higher pressures [22].
In general, reduced total pressure is found to increase
the diffusion coefficient of hydrogen and thereby enhances the growth rate [23]. However, as mentioned
before, studies on the behavior of monosilane under
high-pressure conditions are rare.

DATA COLLECTION
A fairly detailed outline of the reactor system is given
elsewhere [12]. Here only a more simplified sketch of
the experimental setup is shown in Fig. 1. The desired SiH4 inlet pressure in the reactor is generated
by a high-pressure pump driven by an air compressor
(not shown). During the decomposition both the temperature and pressure inside the reactor are measured
and logged. After the decomposition is completed, the
residual gas mixture is led into a residual gas collector.

FROM MONOSILANE TO CRYSTALLINE SILICON

311

Figure 1 Simplified sketch of the experimental setup used for thermal decomposition of pressurized monosilane.

The function of the pressure-loaded piston indicated


on this container is to provide the pressure needed to
lead the collected residual gas mixtures into the gas
chromatograph (GC), containing a Porapak T column,
in filling amounts. The present considerations on the
kinetics of the thermal decomposition of monosilane
are based on the calculated peak areas for the residual gases as extracted by GC analyses and the logged
data on the pressure and temperature in the reactor as a
function of reaction time. All primary data were logged
on an HP 34070 data acquisition/switch unit. At least
three parallel GC analyses were recorded of the residual gas mixtures after each decomposition experiment
and prior to each analysis the pipeline from the storage
cylinder for the residual gases to the GC was repeatedly flushed with nitrogen until the GC detector, which
was a thermal conductivity detector (TCD), showed
no sign of silanes. Temperatures were measured on the
outside of the reactor, but temperature corrections were
established by calibration runs with termocouples located on the outside and inside of an empty reactor, and
also exposed to the same dynamic conditions (heating
rate 2.5 K min1 ) as during the decomposition experiments. The recordings of the temperature inside the
reactor showed a fully reproducible time lag in relation
to the programmed temperature for the furnace.
The primary pressure data in mV were converted
into pressure units (MPa) through a quality-control
procedure, where monosilane at different pressures, as
read on a calibrated manometer attached to the monosilane source bottle, was fed into the evacuated reactor.
After pressure equalization, the reactor was closed and
the output signal was recorded on the data acquisition

unit. A similar calibration process was performed for


mixtures of hydrogen and monosilane, at different total
pressures, covering the entire intervals 0% hydrogen to
100% monosilane and 100% hydrogen to 0% monosilane, and the actual hydrogen-to-monosilane ratio was
confirmed by the GC analyses. These tests indicate that
these gases form ideal mixtures. The used procedure
gave a simple table for conversion of the output signal from the data acquisition unit to pressure inside the
reactor.
The experimental reproducibility was checked by
performing two or three parallel runs for equal initial
pressures of monosilane. The estimated standard deviation (sd) in the rest concentration of monosilane for
experiments performed at 690820 K varied between
0.5 and 5.0%; typical values (pressure in MPa, number
of parallels, estimated sd in %): 0.13, 2, 2.5; 0.28, 2,
0.5; 0.77, 3, 1.5; 1.13, 3, 0.9; 3.34, 2, 2.4.
In order to extract data for the variation in the
monosilane concentration during the decomposition
process, one has to make use of an equation of state
which connects pressure, temperature, and concentration of the participating gaseous species. The present
study used the van der Waals equation of state (EOS),
and some aspects had to be considered prior to the computational treatment.
The GC analyses showed that the gaseous reactionproduct mixtures contained almost exclusively unreacted monosilane and hydrogen, trace amounts of disilane being detected only in a few experiments (3% of
the residual monosilane concentration at most). Hence,
quantitative information was extracted by modeling the
observed pressure-temperature profiles with the EOS

312

ODDEN, EGEBERG, AND KJEKSHUS

(assuming simple mixtures of monosilane and hydrogen during the decomposition and in the as-analyzed
proportions at the end of the experiments). In other
words, it is assumed that the overall decomposition reaction follows:
SiH4 (g) Si(s) + 2H2 (g)

(7)

The effect of the reduced reactor volume caused


by the formation of the solid silicon was corrected for
by weighing the products and assuming the density
(2.33 g cm3 [24]) listed for crystalline silicon. The
use of this density value simplifies comparison with
other works, without making a major difference in the
corrected reactor volume compared with the use of the
rough estimate for the density [12] of the present amorphous (and porous) silicon (a-Si:H). The neglect to implement the hydrogen situated in the deposited solid
material, approximately 4 atom percent [12], does not
affect the calculations based on the GC data significantly. The so-called Newton procedure was used to
solve the mass-balance equations. The adjusted temperature for the inside of the reactor (as derived from
the measured temperature on the outside thermometer with appropriate corrections from the calibration
procedure; see above) was used as inputs in the EOS.
The short-interval-repeated (every 5 s) recordings of
temperature and pressure were converted to a common
temperature grid by cubic-spline interpolation prior to
introduction in the EOS.

DATA TREATMENT

Owing to the produced solid silicon, the available


space for the gaseous species has become reduced between the beginning and end of the experiment, the correction term amounting to x Vm , where Vm is the molar
volume of the solid silicon. From the corrected volume
(V x Vm ) and the number of moles of monosilane
and hydrogen at the end of the experiment, one can stipulate the total pressure inside the reactor and compare
this with the calibrated as-measured pressure. Figure 2
shows that there is a satisfactory agreement between the
thus stipulated (Pstip ) and the measured (Pmeas ) pressure
in the reactor. Pstip is indeed systematically some 5%
lower than Pmeas , but there is a virtually linear relationship between the two pressure assessments. This in turn
implies that the assumed conditions should be credible:
(i) The amount of all gaseous species other then SiH4
and H2 is insignificant at the end of the decomposition. (Modeling (see below) suggests that this is also
the situation during the decomposition.) (ii) The stoichiometry of the overall reaction is satisfactory approximated by reaction (7). (iii) The amount of adsorbed
gases is negligible compared with the amount in the gas
phase. (iv) The linear relationship shown in Fig. 2 supports the above conclusion that SiH4 and H2 form ideal
mixtures.
In order to extract data for the variation in the number of moles of monosilane and hydrogen in the course
of a given decomposition experiment, it is assumed
that the PV T (pressurevolumetemperature) properties of monosilane and hydrogen mixtures are satisfactorily described by the EOS. Expressions for the
stipulated partial pressures of monosilane and hydrogen according to the EOS are accordingly equated

As pointed out above, the overall decomposition reaction appears to follow reaction (7) and the residual gas
mixture contained virtually only hydrogen and unreacted monosilane. Assuming an ideal mixture of these
gases in the as-analyzed amounts, one can estimate the
total pressure at the end of the decomposition which
in turn can be compared with the directly measured
pressure.
Defining n as the number of moles of monosilane at
the start of a given decomposition experiment and 2x
as the number of moles of hydrogen that is present at
the end of the experiment, one obtains an expression
for the mole percentage (m) of unreacted monosilane
at the end of the experiment (determined by the GC
analyses):
m = 100[(n x)/(n + x)]

(8)

The number of moles of monosilane and hydrogen


at the end of the experiment is n[2m/(100 + m)] and
2n[(100 m)/(100 + m)], respectively.

Figure 2 Stipulated (Pstip ) vs. measured (Pmeas ) pressure


of monosilane in the reactor at the end of the decomposition experiments. The dotted line represents the relation
Pstip = Pmeas .

FROM MONOSILANE TO CRYSTALLINE SILICON

to Pmeas :
Pmeas =

RT
V x Vm
nx

bSiH4

RT
V x Vm
2x

b H2

aSiH4
V x Vm
nx

a H2
V x Vm
2x

2

2

(9)

where the van der Waals constants (aSiH4 = 0.437 m6


Pa mol2 , bSiH4 = 0.0000578 m3 mol1 , aH2 = 0.0248
m6 Pa mol2 , and bH2 = 0.0000266 m3 mol1 ) were
taken from [24]. This nonlinear equation was solved
for x by Newtons method with the aid of the computer
program Mathcad [25]. The close correlation in Fig. 2
justifies the present use of the van der Waals EOS to
account for nonideality (which represents a simplification with regard to the number of nonideality parameters compared to other EOSs like the RedlickKwong
or PengRobinson relationship).

RESULTS AND DISCUSSION


The calculated value of x according to this procedure provided the concentration of monosilane (cSiH4 ,
here in mol m3 ) at any time/temperature during a
given decomposition experiment, and selected stipulated monosilane concentration vs. temperature relationships are shown in Fig. 3. The error margins (expressed as relative sd) vary with concentration and are
found to be about 0.5% for concentrations corresponding to initial pressures of monosilane 2.05 MPa. The
relative sd for these experiments rises to about 1.0%
for measurements made at temperatures above 790 K

Figure 3 Variation in monosilane concentration with temperature during various decomposition experiments with different initial pressure of monosilane (given in MPa on the
illustration). Calculated concentrations are derived by the
van der Waals equation of state (EOS).

313

during the decompositions. The relative sd for intermediate (1.11 MPa) and low (0.13 MPa) pressures at
temperatures lower/higher than 790 K is estimated to
be (percentage values) 1.5/2.0 and 4.0/15, respectively.
A normalized version of this illustration in terms of
cSiH4 /c0 vs. temperature (not presented) disclosed only
small variations with pressure, which confirms that
there is no appreciable pressure dependence of the process in the interval studied.
Since there is an approximately linear relationship
between the reaction coordinates temperature and time,
Fig. 3 is easily converted to monosilane concentration
as function of time (needed to derive reaction rates
(dcSiH4 /dt)). Figure 4 shows normalized reaction rates
(1/cSiH4 )(dcSiH4 /dt) as a function of temperature during
decomposition experiments with different initial pressures of monosilane. Calculations based on the above
error margins for the concentrations give relative sd of
about (percentage values) 0.9/1.7, 2.6/3.5, and 6.9/26
for the normalized reaction rates measured at temperatures lower/higher than 790 K at pressures 2.05 (and
above), 1.11, and 0.13 MPa, respectively. The results
presented in Fig. 4 suggest that the temperature variation of the reaction rate is largely independent of the
initial monosilane pressure above about 760 K, and
shows a systematic behavior at lower temperatures with
maxima in reaction rate near 750 K. The pressure dependence of the reaction rate is also relatively small
between ca. 720 and 760 K and, in fact, may be more
or less insignificant when the possible sources of inaccuracies of the present experiments are taken into
account.
Figure 4 also shows an expected exponential increase in the reaction rate in the initial part of the
decomposition region, which in turn is followed by a
decrease in the reaction rate due to hydrogen inhibition.

Figure 4 Stipulated reaction rate vs. reactor temperature


for decomposition experiments on pressurized monosilane.
Initial monosilane pressures (in MPa) are given on the illustration. Curve shape is emphasized with the shading; limited
significance should be attached to the order of the curves.

314

ODDEN, EGEBERG, AND KJEKSHUS

Figure 5 A loglog representation of reaction rate vs. monosilane concentration. Selected examples of isoterms (temperatures
in Kelvin on the illustration). Estimated from the data in Figs. 2 and 3.

At the highest temperatures, the hydrogen inhibition is


again overtaken by the general effect of temperature on
the decomposition reaction.
A derivation of the reaction order requires data on
reaction rates and concentrations at the same temperature. These data were provided by the cubic-spline
interpolation of all data (generated as described above)
onto a common temperature grid. Figure 5 shows selected examples of loglog plots of reaction rate vs.
monosilane concentration from which the reaction order at a given temperature is obtained from the slope of
the lines. Figure 5 shows good linear correlations, but
a certain scatter in the points indicates that the reaction
order may vary a little with temperature (see below).

Qualitative Considerations
The temperature dependence of the reaction rate (and
hence its derivative, viz. the rate constant) for the thermal decomposition of monosilane has a similar shape
for all experiments (Fig. 4). A significant reaction rate
is not detected until about 700 K, above which there
occurs a rapid increase to a maximum at about 750 K.
This maximum is succeeded by a minimum at about
790 K, whereupon the reaction rate increases monotonically toward the end temperature (830 K) of the
measurements. Above 820 K, where 8590% conversion is obtained, it proved difficult to obtain reliable
estimates for the gas composition due to the mixed influence of decomposition and thermal expansion as the
source to pressure increase.
Ever since the early pyrolysis experiments by
Hogness et al. [21], three aspects of the thermal decomposition of monosilane have caused much debate and controversy: (i) the onset temperature for the

decomposition, (ii) the effects of the surface-to-volume


ratio of the reaction system, and (iii) the reaction order.
The following qualitative considerations of the experimental findings will focus on these topics.
Onset Temperature for Decomposition of Monosilane. By direct inspection of the as-observed Ptot vs. T
relationships, we earlier [12] estimated the onset temperature for the thermal decomposition of monosilane
under the present high-temperature conditions to 710
720 K. The processed data in Figs. 2, 3, and 5 show
that the assessment should be shifted some 10 K to
lower temperature, viz. Tdecomp 700710 K, which is
still appreciably higher than the temperatures (640
650 K) reported from static reactor [1,2,15,21] and
fluidized-bed reactor [10] experiments. These discrepancies are explained as consequences of the present
dynamic reactor-temperature system working at much
higher pressures and its implications for the initiation
(inhibition) period of the reaction (for static conditions
defined as the time lag between the filling of the reactor and the onset of the pressure increase consequent
on the decomposition). In fact, the findings reported in
[1,21,26] suggest that the occurrence of the initiation
period may make dynamic experiments carried out at
a temperature increase of 2.5 K min1 somewhat unsuitable for obtaining unbiased kinetic information, at
least for temperatures below 700 K. On the other hand,
Robertson et al. [27] argue that the initiation period
is an artifact caused by poisoning of the reactor walls
by oxygen. The present experiments were conducted
in stainless steel tubes, which were opened and exposed to air between different runs. Hence, according
to Robertson et al. [27] the high onset temperature may
be caused by oxygen on the Si-coated reactor walls.

FROM MONOSILANE TO CRYSTALLINE SILICON

315

However, this possibility was ruled out by test experiments in which the reactor was evacuated and reloaded
with monosilane without exposure to the ambient atmosphere. Only minor differences were observed between
such parallels.
Effects of Surface-to-Volume Ratio of the Reaction System. The solid Si obtained in the present experiments
was recovered as a powder with a narrow size distribution (average diameter 3 m, with less than 7% variation between different experiments [12]). Hence, the
ratio between the total surface area of the particles in
two experiments with different initial monosilane pressures should be proportional to the pressure ratio. For
example, the total surface area of the particles in the
3.34 MPa experiment is approximately 25 times larger
than that in the 0.13 MPa experiment. However, over
the temperature range 730780 K, the normalized reaction rate for the 3.34 MPa experiment is only some
40% higher than that for the 0.13 MPa experiment (Fig.
4) and below 720 K the reaction rate for the 3.34 MPa
experiment is actually lower than that for the 0.13 MPa
experiment. Hence, the eventual net effects of the surface area on the kinetics of the monosilane decomposition are small and dependent on temperature.
The absence of significant surface-to-volume effects
on the decomposition of monosilane is also reported
by Hogness et al. [21], Purnell and Walsh [1], and
Onischuk et al. [28]. All in all, it is a reason to insist that the absence of surface-to-volume effects in the
thermal decomposition of monosilane is not yet satisfactorily explained. However, the available knowledge
may be used to advocate that the primary decomposition proceeds as a homogeneous process at moderate
and higher pressures [29].
Reaction Order. The thermal decomposition of monosilane is a complex process, which involves intermediate steps via a number of indirect routes. The reaction
order (n) per definition is related to a particular reaction, which in the present case conveniently is the
gross degradation reaction (7) with the associated rate
equation (where k is the reaction rate constant):
dcSiH4 /dt = k cSiH4 n

(10)

Owing to the many parallel reactions involved, it would


be most appropriate to refer to k as the apparent rate
constant and n as the apparent rate order, but hopefully
there should not arise any misconceptions for neglecting to emphasize this point below.
The reaction order increases from a low value at
low temperatures, passes a maximum at 750 K, and
stabilizes at values close to 1 above 780 K (Fig. 6).

Figure 6 Reaction order vs. temperature. The shaded region


indicates 1 standard deviation.

The complications of the many parallel reactions are


nicely illustrated by the slight unsystematics in the
locations of the isoterms for T > 710 K in Fig. 5
(which, together with all corresponding relations for
T > 710 K, form the basis of Fig. 6). The low reaction
order at low temperatures is connected with the initiation period of the reaction (see above). Above 710 K,
the regressed curves (see Fig. 5) become indistinguishable from straight lines, as evidenced by the narrow
confidence limits, and above this temperature the reaction order also does not depart significantly from 1
within the 95% confidence interval (viz., 0.96 0.04).
This agrees well with the observation by Hogness et al.
[21] that there appears to be no initiation period at
713 K. The results obtained below 710 K will not be
considered further in this report.
Hogness et al. [21] concluded that the thermal decomposition of monosilane is a first-order reaction,
which is inhibited by hydrogen when the conversion
exceeds a certain level. Purnell and Walsh [1] distinguished, as mentioned above, between the initial stage
of the reaction with 3/2 order for less than 3% conversion, and the latter stage (1020% conversion) where
first-order kinetics with respect to monosilane was observed at maximum rate of hydrogen production. Under the same condition (maximum rate of hydrogen
production), Robertson et al. [27] reported 1.2 order
and White et al. [2] first-order kinetics for 330% conversions. As first suggested by Purnell and Walsh [1],
and now commonly believed, the apparent 3/2-order
dependence in the initial stage is caused by the pressure dependence of reaction (1), which indeed is of first
order, and that this reaction is a leading process also at
more advanced stages of the conversion. At 716 K the
reaction progress in the present experiments varies between 4 and 11% conversion; hence our observation of
a reaction order close to 1 is consistent with previous
results. At 750764 K, where the reaction rate is at the
maximum in Fig. 4, the reaction order (1.081.13; see

316

ODDEN, EGEBERG, AND KJEKSHUS

Fig. 6) falls between 1.0 and 1.2 as reported by Purnell


and Walsh [1] and Robertson et al. [27], respectively.
The pressures employed in the three studies decrease
in the order: present study, Purnell and Walsh, and
Robertson et al. Assuming that the first-order reaction (1) is in its fall-off region in our case, it is a reason
to expect that the reaction order should increase in the
said sequence. The apparently constant reaction order
above 780 K (6095% conversion; Fig. 6) is intriguing.

Model Considerations
There appears to be consensus that the thermal decomposition of monosilane starts off by a rupturing of the
gaseous molecule into silylene and hydrogen. It is debatable whether the initial stage of this reaction takes
place solely in the gas phase as indicated in reaction (1),
occurs as a gassurface reaction, or involves both gas
gas and gassurface processes.
Later in the decomposition process, SiH2 is formed
(and consumed) along a member of routes and homogeneous as well as heterogeneous mechanisms are clearly
involved (see, e.g., [30,31]). Any way SiH2 is a very
reactive SiH fragment, which definitely has a key role
throughout the entire decomposition process. SiH2 is
inter alia participating in the formation of Si2 H6 (reaction (2)), Si3 H8 (reaction (3)), and even higher silanes.
All these silanes may in turn be decompositionally reformed to silylsilylenes (see, e.g., reactions (4)(6)).
H3 SiSiH is another highly reactive SiH species, which
through reaction with SiH4 produces SiH2 :
H3 SiSiH(g) + SiH4 (g) Si2 H6 (g) + SiH2 (g) (11)
and thus accelerates the overall decomposition of
monosilane. H3 SiSiH can also undergo isomerization
to H2 SiSiH2 , the latter being believed [28] to have some
influence on which SiH terminations that occur in the
deposited silicon. H3 SiSiH may also react with H2 to
regenerate SiH4 and SiH2 or itself decompose to SiH4
and Si. It seems moreover reasonable that one by involving silanes higher than trisilane will generate more
silylsilylenes which in turn will enhance the overall
decomposition reaction.
As should be evident from the preceding outline of
probable reaction paths, the thermal decomposition of
monosilane is far from a simple process. This complexity is also reflected in the different models designed to
describe the process, which indeed span a wide range
of sophistication and detail level. For example, Kleijn
[32] modeled the chemical vapor deposition of silicon
from monosilane in a cold-wall, single-wafer reactor by
the means of only five species, whereas Girshick et al.
[31] included more than 100 SiH species and about

400 reactions in order to adequately predict the particle nucleation process. Ho et al. [30] (see their
Tables 1 and 2) used some 10 species and 31 reactions
to model the monosilane decomposition in the gaseous
phase. The model of Ho et al. should for our purposes
provide a sufficiently sophisticated base for a realistic representation of the pressure change inside the reactor (viz., describes the decomposition process) and
give mathematical expressions that are solvable with
reasonable computational resources. The model of Ho
et al., with their appurtenant mathematical codes, was
therefore used to describe the process under the present
conditions, unless otherwise is stated in the text. This
model, which involves gasgas as well as gassurface
mechanisms, also constitutes a central part of the more
complex model of Girshick et al. [31].
Under the present experimental conditions, gasphase reactions are in the high-pressure regime. However, the pressure dependences of Ho et al. [30] were
included in order to facilitate comparative modeling of
available data obtained at lower pressures. The silylene
insertion reactions (reactions (2) and (3)) are central
elements in this reaction scheme. It should be noted
that the kinetic parameters used by Ho et al. [30] are
in excellent agreement with the more recent data of
Beccera et al. [33,34].
For the gassurface reactions it is also difficult to
identify the key species. In the model of White et al.
[2] deposition of Si takes place through adsorption of
SiH4 and Si2 H4 . Ring and ONeal [16] modeled the experimental data of Purnell and Walsh [1] assuming decomposition of polysilanes (Si3 H8 and higher silanes)
at the surfaces. The same data were modeled by Becerra
and Walsh [15] using only Si3 H8 as a transfer agent.
Hence, none of these studies included disilane in the
deposition to the solid silicon, whereas Gates and coworkers [35,36], Kulkarni et al. [37], and references
therein maintained that disilane has a high surface reactivity. In fact, most of the more recent modeling
attempts consider disilane as an important Si carrier
(see, e.g., [38]). Central elements in the gassurface
mechanisms of Ho et al. [30] are the two-site dissociative adsorption of mono-, di-, and trisilane, followed
by first-order hydrogen elimination to the gas phase.
In the present model computations, the rates of disilane and trisilane decomposition were set (based on
the experimental data for the sticking coefficients [39])
to 10 times the reaction rate for monosilane and assigned the same temperature dependence as monosilane (E a = 157 kJ mol1 ).
Experimental data from Purnell and Walsh [1] have
frequently been used in development and calibration
of models for the decomposition of monosilane [2,15
17] and have also been used to check the present model

FROM MONOSILANE TO CRYSTALLINE SILICON

calculations. The original model of Ho et al. [30] was


developed for decomposition at very low monosilane
pressures; hence it is not surprising that this could
not be directly fitted to the data from Purnell and
Walsh [1]. As already pointed out, the deposition is
the least clarified part of the decomposition process.
Ho et al.s assumption of a two-site dissociative adsorption process requires the participation of two neighboring dangling bonds in the composite decomposition
deposition step. Because of the very dense atmosphere
(viz., many small silicon particles and many different gas species of which monosilane and/or hydrogen
certainly exhibit(s) high pressures), the probability of
finding two neighboring dangling bonds is much lower
under the present experimental conditions than implicit
in the model of Ho et al. This is accordingly a natural
spot to explore for a possible modification of the model.
First, we neglected the effect of surface area. This is
justified by our own and others observations [1,21,28].
Conceptually, we envisage that only one dangling bond
is likely to be involved per deposited SiH fragment
in a dense atmosphere. Along with the suggestions of
Kulkarni et al. [37], this would be reconcilable with
pyrolysis of mono-, di-, and trisilane according to the
reactions:
SiH4 (g) SiH2 (s) + H2 (g)

(12)

Si2 H6 (g) SiH2 (s) + SiH4 (g)

(13)

Si3 H8 (g) SiH2 (s) + Si2 H6 (g)

(14)

Some support for such reaction steps is provided by


Weerts et al. [40] who followed the development in the
concentrations of di- and trisilane during the course of
the reaction. Above 650 K, these authors found that
adsorption of di- and trisilane is accompanied by formation of monosilane. With regard to the present experiments, where the onset of the decomposition takes
place at 700710 K, most disilane formed according
to reaction (14) will immediately decompose further
according to reaction (13). Rate constants for reactions
(12)(14) (5.8 106 , 3.0 102 , and 7.7 102 s1 ,
respectively) are estimated by fitting our thus simplified model to the steady-state data reported by Purnell
and Walsh [1] for 0.025 and 0.005 MPa and 703 K.
The temperature variation of the rate constants was assumed to follow an Arrhenius-type relationship, and
numerical values were deducted by supposing the same
temperature dependence as for the sticking coefficients
[30]. The unit s1 for the rate constants is a consequence of our decision to neglect the effect of surface
area. We have no explanation for the absence of effects
of surface area, but its inclusion in the model would
be contrary to experimental observations [1,21,28].

317

Figure 7 Temperature dependencies of modeled concentrations of SiH species involved during thermal decomposition of monosilane (data for 3.34 MPa initial monosilane
pressure). Scale factors applicable to the different curves in
order to obtain concentrations in mol cm3 are SiH4 , 105 ;
H2 , 105 ; Si2 H6 , 107 ; Si3 H8 , 109 ; SiH2 , 1016 ; H3 SiSiH,
1018 ; H2 SiSiH2 , 1019 ; and Si, 1023 .

The activation energy for mono- and disilane was accordingly set to 148 and 119 kJ mol1 , respectively,
and the latter value was also ascribed to Si3 H8 . The
deposition reactions for other gaseous Si species (notably SiH2 , H3 SiSiH, H2 SiSiH2 , and Si) were assigned
the same rate constant as Si2 H6 . Besides, the modeling showed (Fig. 7) that the contributions of the just
bracketed species to the deposition of silicon are insignificant at the temperatures concerned. In fact, even
increasing their rate constants by several orders of magnitude would not alter this inference, viz., large errors
in the estimates of these constants have little effect
on the output of the model calculations. We consider
most of the deposition to take place through reactions
(12)(14), but do not rule out the possibility that
other reactions may contribute with minor amounts. In
fact, the distribution of SiH groups (dilute monohydrides, clustered monohydrides, and clustered polyhydrides) in the generated solids suggests a more complex
deposition scheme [13].
Modeled and experimental reaction rates are compared in Fig. 8 for three initial pressures of monosilane (0.13, 1.13, and 3.34 MPa). For the lowest initial monosilane pressure and low temperatures, the
modeled data compare favorably with the rate constants
determined by Purnell and Walsh [1] and White et al.
[2]. The increasing deviation between the calculated
and observed data with increasing initial pressure and
temperature is caused by the pressure sensitivity of reaction (1). Owing to the continuous decomposition, the
reaction atmosphere gets more dense with increasing
temperature. The model predicts almost constant and
pressure insensitive rate constants above 740 K. The

318

ODDEN, EGEBERG, AND KJEKSHUS

Figure 8 Experimental and modeled (corresponding line


symbols in both sets of curves; 0.13 MPa: dotted, 1.13 MPa:
stippled, 3.34 MPa: solid) monosilane decomposition reaction rate as function of temperature. Experimental data quoted
from Purnell and Walsh [1] and White et al. [2] are marked
by filled and open symbols, respectively.

inhibiting effect of hydrogen on the monosilane decomposition, first reported by Hogness et al. [21], has
been observed in all subsequent pyrolysis experiments
extending beyond 1030% conversion, and has been attributed to the reaction between silylene and hydrogen
(viz., the reverse of reaction (1)). The model calculations (Fig. 8) show that at temperatures higher than
some 740 K the amount of hydrogen liberated by the
various reactions is sufficient to balance the stimulating effects of pressure and temperature on the reaction
rate. Qualitatively, the experimental rate constants resemble the model prediction in that the initial loading
pressure is seen to have little effect, and that on this
scale (see, however, Fig. 4) the reaction rate becomes
relatively constant at higher temperatures. The obvious
difference is that the observed reaction rates are more
than an order of magnitude smaller than those predicted
by the model.
Unfortunately there are no other data available
on the decomposition of monosilane at the present
experimental conditions (temperatures: 710820 K,
hydrogen pressures: 0.0922 MPa, monosilane pressures: 0.292.9 MPa) that can be used for an independent evaluation of the model. However, a comparison of
the observed rates with the extrapolated rate of Purnell
and Walsh [1] suggests that the model largely integrates
the present knowledge of the kinetics of the monosilane
decomposition, but predicts higher rates than observed.
The inadequacy of the model may then either be attributed to errors in the experimental rate constants, or
reveal that some important reaction mechanisms (significant only at higher pressures) have not been taken
into account.
To a large extent, the gas-phase model is driven
by three reversible reactions: the monosilane dehydro-

genation reaction (reaction (1)) and the two silylene


insertion reactions (reactions (2) and (3)). These reactions have been studied in detail, and their rate constants and pressure sensitivities are apparently well established [41,42]. The remaining gas-phase reactions
do not affect the net rate of monosilane decomposition to any appreciable extent (see Fig. 7). This leaves
uncertainties in gassurface transport reactions as candidates for a possible explanation of the high-pressure
inadequacy of the Ho et al. [30] model. The modelpredicted rate of monosilane decomposition will to a
certain extent be lowered by increasing the activation
energies and still reproduce the results of Purnell and
Walsh [1] at low temperatures. However, this leads to
intolerably high concentrations of Si2 H6 and Si3 H8 , unless the production rates of these species are reduced
correspondingly. However, only traces of disilane were
observed in a few experiments. A lowering of the rate
of deposition by blocking of the dangling bonds with,
for example, hydrogen would have the same effect as
increasing the activation energies. All in all, it seems
impossible to account for the high-pressure inadequacy
of the Ho et al. [30] model by adjusting kinetic parameters of the involved reactions within reasonable limits.
The inference is accordingly that there must be some
components missing from the model.
In order to reduce the modeled rate without increasing the concentrations of Si2 H6 and Si3 H8 , the missing
components must regenerate monosilane from the latter species. The present experiments are characterized
by the exceptionally dense atmosphere, high particle
density, and high partial pressure of hydrogen. Hence,
the search for additional reactions was limited to those
sensitive to these parameters. In line with this, it seems
natural to postulate that the missing components are
the two reactions:
Si2 H6 (g) + H2 (g) 2SiH4 (g)

(15)

Si3 H8 (g) + H2 (g) Si2 H6 (g) + SiH4 (g) (16)


These reactions are envisaged to occur at the surfaces
when the unstable complexes Si2 H6 or Si3 H8 together
with hydrogen (from the gas phase) attach to a particle
site with a dangling bond. Both reactions are exergonic
at our experimental conditions.
Some experimental evidence in support of this
hypothesis comes from a study of Gates [43] on the
kinetics and mechanisms of surface reactions in epitaxial growth of Si from SiH4 and Si2 H6 . Gates points
out that all studies of the chemisorption of Si2 H6 on Si
surfaces agree that Si2 H6 adsorbs through a molecular precursor state with many characteristics of a twodimensional gas, weakly interacting with the surface.

FROM MONOSILANE TO CRYSTALLINE SILICON

The plausibility of the postulated adjustments of the


model of Ho et al. [30] depends on how well and
logically they account for the experimental findings.
Kinetic parameters for reactions (15) and (16) were
determined by forcing the thus modified model to reproduce the observed reaction rate for the decomposition of monosilane at the experimental conditions
(temperature, pressure, and concentrations of monosilane and hydrogen). In order not to introduce unmotivated constraints on the kinetics, the effect of reactions
(15) and (16) was entered into the model simply as
first-order terms, with respect to di- and trisilane. Numerically stable solutions could only be obtained by
assuming that the rate constants for the two reactions
were correlated, and identical activation energies were
accordingly assumed.
The observed reaction rates between 710 and 820 K
could be modeled to within 1% by the modified
model. In none of the simulations did the concentration
of di- and trisilane exceed 0.6 and 0.04% of the total
gas concentration, respectively. The relative concentration distribution of the various gaseous species (Fig. 7)
showed only minor variation between simulations for
different experiments. According to the model, the local reaction maximum at about 750 K (Fig. 4) is caused
by maxima in the concentrations of di- and trisilane at
730 K (Fig. 7; note that the temperature dependence
of the formation rates will displace the corresponding maxima in the deposition to somewhat higher temperatures). Observe that the concentration of the third
most abundant siliconhydrogen species, silylene, is
78 orders of magnitude lower than that of trisilane.
Hence, the contribution to the gas-to-surface transport
of Si from species other than mono-, di-, and trisilane
is negligible.
The reaction rate of, say, reaction (15) (and analogously for reaction (16)) was entered into the model as
k15 cSi2 H6 cH2 , assuming a simple first-order dependence also on the hydrogen concentration. At constant
temperature, k15 should be a real constant or approach
a constant value only at higher concentration of ctot if
the reaction is third body assisted. In line with expectations, Fig. 9 suggests that the latter possibility applies
to the present case.
Examination of the results revealed that the rate constants k15 and k16 are strongly dependent on the concentration of hydrogen at low H2 concentrations, but
virtually independent of the concentration of hydrogen
above a certain level. In an attempt to isolate the effect
of hydrogen, the rate constant data were interpolated on
a common hydrogenconcentration grid so that results
obtained at different temperatures (but at the same hydrogen concentration) could be extracted. This also has
the effect of eliminating differences in particle density

319

Figure 9 Dependence of modeled rate constants k15 and


k16 on the total concentration of gaseous species. Curves are
guides for the eye.

(surface area) between different experiments, because


the amount of silicon particles formed is directly proportional to the concentration of hydrogen. The regression procedure gave rate constants which could be fitted
to the Arrhenius plots (Fig. 10) with good correlation
coefficients (0.989, n = 90.999, n = 6). The estimated
sd for the slope of the curves varies between 0.8 and
1.1. This is certainly not sufficient to prove that reactions according to reactions (15) and (16) are actually
in operation, but is a minimum requirement for such
an event. The slopes on such Arrhenius plots are dependent on the hydrogen concentration (partial pressure). From the thus established modeled relationship
between activation energy and total concentration of
gases (Fig. 11), a systematic increase in E a is seen
(about 250 kJ mol1 at zero H2 concentration (extrapolated) and ca. 350 kJ mol1 at the end of the experiment when hydrogen dominates the pressure, the
corresponding estimated sd being 10 and 7 kJ mol1 ,
respectively).

Figure 10 Arrhenius plots for the modeled rate constants


k15 and k16 . Note that the lines in pairs refer to different
hydrogen concentrations (partial pressures).

320

ODDEN, EGEBERG, AND KJEKSHUS

somewhat third-body assisted, and their rate constants


and activation energies depend on the concentration of
hydrogen.

BIBLIOGRAPHY

Figure 11 Modeled activation energy vs. ctot (cH2 ) for reactions (15) and (16). The curve is a guide for the eye.

CONCLUSIONS
A study of the thermal decomposition of pure monosilane at elevated pressures has been performed with
focus on extracting reaction-kinetic information. The
onset temperature of the monosilane decomposition
under these conditions is appreciably higher (700
710 K) than indicated in the literature for static-reactor
conditions. This discrepancy reflects the initiation period of the reaction, which also makes data obtained at
temperatures below 710 K under our conditions unsuitable for extraction of unbiased kinetic information.
At monosilane conversions beyond 1030%, hydrogen strongly inhibits the decomposition (as also earlier
recorded at lower initial pressures of monosilane) due
to back-reaction with silylene. Surface area/volume effects seem to have a small, but temperature-dependent
influence on the reaction kinetics of the thermal decomposition of monosilane, again in agreement with earlier
findings at lower pressures. The overall reaction order
above 710 K is close to one, also consistent with results published for monosilane conversion in the range
330% at lower pressures. The simulation model of
Ho et al. [30] (with the gassurface mechanism modified to only one dangling bond per active site) was first
used to fit the experimental results. This processing disclosed a conspicuous discrepancy between the model
predictions with regard to reaction rates (much too low
observed rates). The most likely explanation for this
controversy appears to be that additional mechanisms
are in operation under the present reaction conditions.
As a remedy, the Ho et al. model was modified to include reactions between hydrogen and disilane or trisilane, both regenerating monosilane. Thus the modified
model was found to produce trustworthy distributions
of gaseous species throughout the studied temperature, pressure, and composition ranges. The additional
reactions thus introduced in the model appear to be

1. Purnell, J. H.; Walsh, R. Proc R Soc London A 1966,


543.
2. White, R. T.; Espino-Rios, R. L.; Rogers, D. S.; Ring,
M. A.; ONeal, H. E. Int J Chem Kinet 1985, 17, 1029.
3. Scott, B. A.; Estes, R. D.; Jasinski, J. M. J Chem Phys
1998, 89, 2544.
4. Slootman, F.; Parent, J.-C. J Aerosol Sci 1994, 25, 15.
5. Wiggers, H.; Starke, R.; Roth, P. Chem Eng Technol
2001, 24, 261.
6. Frenklach, M.; Ting, L.; Wang, H.; Rabinowitz, J. Isr J
Chem 1996, 36, 293.
7. Newman, C. G.; ONeal, H. E.; Ring, M. A.; Leska, F.;
Shipley, N. Int J Chem Kinet 1979, 11, 1167.
8. Matsukata, M.; Odagiri, T.; Kojima, T. J Phys IV C2
1991, 1, 483.
9. Hsu, G.; Hogle, R.; Rohatgi, N.; Morrison, A. J Electrochem Soc Solid-State Sci Technol 1984, 131, 660.
10. Caussat, B.; Hemati, M; Couderc, J. P. Chem Eng Sci
1995, 50, 3615.
11. Caussat, B.; Hemati, M.; Couderc, J. P. Chem Eng Sci
1995, 50, 3625.
12. Odden, J. O.; Egeberg, P. K.; Kjekshus, A. Sol Energy
Mater Sol Cells 2005, 86, 165.
13. Odden, J. O.; Egeberg, P. K.; Kjekshus, A. J Non-Cryst
Solids 2005, 351, 1317.
14. Hansen, E. W.; Kjekshus, A.; Odden, J. O. J Phys Chem
B (submitted).
15. Becerra, R.; Walsh, R. J Phys Chem 1992, 96, 10856.
16. Ring, M. A.; ONeal, H. E. J Phys Chem 1992, 96, 10848.
17. Roenigk, K. F.; Jensen, K. F.; Carr, R. W. J Phys Chem
1987, 91, 5726.
18. Giunta, C. J.; McCurdy, R. J.; Chapple-Sokol, J. D. J
Appl Phys 1990, 67, 1062.
19. Onischuk, A. A.; Strunin, V. P.; Ushakova, M. A.;
Panfilov, V. N. J Aerosol Sci 1997, 28, 207.
20. Tao, M.; Hunt, L. P. J Electrochem Soc 1992, 139, 806.
21. Hogness, T. R.; Wilson, T. L.; Johnson, W. C. J Am Chem
Soc 1936, 58, 108.
22. Duchemin, M. J.-P.; Bonnet, M. M.; Koelsch, M. F. J
Electrochem Soc 1978, 125, 637.
23. Kuiper, E. T.; Brekel, J. H.; Groot, J.; Veltkamp, G. W.
J Electrochem Soc: Solid-State Sci Technol 1982, 129,
2288.
24. Weast, R. C.; Astle, M. J. (Eds.). CRC Handbook of
Chemistry and Physics, 60th ed.; CRC Press: Boca
Raton, FL, 19791980.
25. Mathcad 2001i Profesional, MathSoft Inc, Cambridge,
MA. http://www.mathsoft.com/.
26. Stokland, K. Trans Faraday Soc 1948, 44, 545.
27. Robertson, R.; Hils, D.; Gallagher, A. Chem Phys Lett
1984, 103, 397.

FROM MONOSILANE TO CRYSTALLINE SILICON

28. Onischuk, A. A.; Levykin, A. I.; Strunin, V. P.; Ushakova,


M. A.; Samoilova, R. I.; Sabelfeld, K. K.; Panfilov,
V. N. J Aerosol Sci 2000, 31, 879.
29. Onischuk, A. A.; Panfilov, V. N. Russ Chem Rev 2001,
70, 321.
30. Ho, P.; Coltrin, M. E.; Breiland, W. G. J Phys Chem
1994, 98, 10138.
31. Girshick, S. L.; Swihart, M. T.; Suh, S.-M.; Mahajan,
M. R.; Nijhawan, S. J Electrochem Soc 2000, 147, 2303.
32. Kleijn, C. R. J Electrochem Soc 1991, 138, 2190.
33. Becerra, R.; Frey, H. M.; Mason, B. P.; Walsh, R. J Chem
Soc, Faraday Trans 1996, 91, 2723.
34. Becerra, R.; Frey, H. M.; Mason, B. P. J Organomet
Chem 1996, 521, 343.
35. Gates, S. M. Surf Sci 1988, 195, 307.

321

36. Gates, S. M.; Greenlief, C. M.; Beach, D. B.; Kunz,


R. R. Chem Phys Lett 1989, 154, 505.
37. Kulkarni, S. K.; Gates, S. M.; Greenlief, C. M.; Sawin,
H. H. Surf Sci 1990, 239, 26.
38. Weerts, W. L. M.; Croon, M. H. J. M.; Marin, G. B. J
Electrochem Soc 1998, 145, 1318.
39. Buss, R. J.; Ho, P.; Breiland, W. G.; Coltrin, M. E. J Appl
Phys 1988, 63, 2808.
40. Weerts, W. L. M.; Croon, M. H. J. M.; Marin, G. B. Surf
Sci 1996, 367, 321.
41. Meyerson, B. S.; Jasinski, J. M. J Appl Phys 1987, 61,
785.
42. Jasinski, J. M.; Chu, J. O. J Chem Phys 1988, 88,
1678.
43. Gates, S.M. J Cryst Growth 1992, 120, 269.

Você também pode gostar