Ind. Eng. Chem. Res.

2001, 40, 5191-5198 5191

Reaction-Bonded Silicon Carbide by Reactive Infiltration

Prasert Sangsuwan, Joaquı́n A. Orejas,† Jorge E. Gatica,* Surendra N. Tewari, and
Mrityunjay Singh‡
Chemical Engineering Department, Cleveland State University, 1960 E 24th Street-SH455,
Cleveland, Ohio 44115-2425

Exothermic reactions between a porous matrix and an infiltrating melt provide a more economic
alternative for synthesizing many ceramics, intermetallics, and composites. It has recently been
demonstrated that infiltration of cast microporous carbon preforms by silicon melt can be used
to fabricate high-density, nearly net-shaped silicon carbide components at significantly reduced
cost. This paper describes the synthesis of reaction-bonded silicon carbide by reactive infiltration
of microporous carbon preforms. The kinetics of unidirectional infiltration of silicon melt into
microporous carbon preforms as a function of pore morphology and melt temperature is
investigated in this paper. Qualitative agreement between experimental data and a mathematical
model for capillarity-driven fluid flow through cylindrical pores is demonstrated. Experimental
evidence of high parametric sensitivity is also presented. A simplified model relating fluid flow,
transport, and reaction phenomena is formulated to interpret experimental evidence of pore
closing and free silicon entrapment. A robust numerical formulation is also described.

Introduction by the silicon melt, and (iii) the chemical reaction to

convert carbon into silicon carbide.
Chemical reactions between a porous matrix and an The process for fabricating the microporous carbon
infiltrating phase provide a more economic alternative preforms is reasonably well understood. Porous carbon
for synthesizing many ceramics, intermetallics, and preforms with varying microstructures [pore size, pore
composites. In reaction-bonded silicon carbide (RBSC), volume fraction, carbon particle (strut) size] can be
for instance, capillary infiltration of a carbon-containing produced by controlling the chemistry of the organic
body is carried out by molten silicon.1 The occurrence mixture and the heat treatment.4
of a chemical reaction between silicon and carbon results
in the formation of silicon carbide particles that bond The process of silicon infiltration, however, is less
existing SiC grains, which are usually added as inert understood. Unlike in other infiltration processes, the
fillers during the preform preparation step. In another presence of a chemical reaction between the porous
approach, carbonaceous materials, such as carbon fiber preforms and the infiltrating fluid adds complexity to
tows, fiber cloths, or felts, are infiltrated by molten the process. Indeed, nonlinear interactions between flow
silicon in a vacuum to form Si/SiC (silicon reinforced and transport phenomena complicate the scenario where
by silicon carbide) composites.2 It has recently been thermomechanical and chemical phenomena determine
demonstrated that infiltration of cast microporous the characteristics of the resulting composite. For
carbon preforms by a silicon melt can be used to example, the reaction between the infiltrating silicon
fabricate high-density, nearly net-shaped silicon carbide melt and carbon results in reduced pore size (there is
components at significantly reduced cost.3,4 Because an approximately 58% volume increase when one mole
molten silicon wets carbon, it wicks up into the mi- of amorphous carbon reacts to form one mole of SiC).
croporous carbon network when brought into contact This affects the permeability of the preform reducing
with the preform, thus converting carbon into silicon the infiltration velocity. Thus, for a preform with a small
carbide. Components fabricated by this technique are initial pore size, the silicon infiltration will stop pre-
expected to have a significant impact in gas turbine maturely, resulting in incomplete infiltration, a phe-
engine components, commercial combustion nozzles, and nomenon frequently referred to as choking. Alterna-
other advanced applications where their refractoriness tively, if the pore constrictions in the preform are large,
(high strength at elevated temperatures), good oxidation the preform will be fully infiltrated, but residual un-
resistance, high thermal conductivity, low density, and converted silicon will remain in the component, with
adequate toughness can be exploited. However, success- deleterious effects for high-temperature applications.
ful commercial exploitation of this technique requires Therefore, it is important to understand (i) the kinetics
a quantitative understanding of the various steps of the chemical reaction, (ii) the kinetics of the melt
involved in the process, namely, (i) the fabrication of infiltration, (iii) the coupling between these two phe-
the microporous carbon preforms, (ii) their infiltration nomena, and (iv) their dependence on the preform
morphology (volume fraction of carbon, pore size dis-
tribution, bed tortuosity, etc.) and the processing condi-
* Author to whom correspondence should be addressed. tions (preform and melt temperatures, pressure, melt
Phone: (216) 523-7274. Fax: (216) 687-9220. E-mail: j.gatica@ characteristics, etc.).
csuohio.edu.
† Facultad de Ingenierı́a, Universidad Nacional de Rı́o A complete analysis of the process would be very
Cuarto, Ruta 36 Km 601, Córdoba, Argentina. complex because of the nonlinear interactions between
‡ NYMA, Inc., NASA John H. Glenn Research Center, 21000 microscale and macroscale phenomena. Microscale phe-
Brookpart Road, Cleveland, OH 44135-3191, USA. nomena, such as the carbon-silicon reaction and mass
10.1021/ie001029e CCC: $20.00 © 2001 American Chemical Society
Published on Web 07/31/2001
5192 Ind. Eng. Chem. Res., Vol. 40, No. 23, 2001

and energy transport phenomena, will determine the

flow characteristics within an individual pore. At the
macroscale, the interactions among neighboring pores
will determine the stability of the infiltration front and,
thus, the composite microstructure and overall degrees
of infiltration and silicon carbide conversion. Stresses
due to the severe thermal gradients resulting from the
exothermic reaction could cause cracking, and residual
stresses due to volume change/cooling mechanisms could
have deleterious effects on the component brittleness.
Therefore, when faced with the task of designing the
process leading to a final component, a number of
questions are clearly without a definite answer. In other
words, how can we predict the maximum extent of
infiltration for a given initial melt or preform temper-
ature? What will be the stability characteristics of the Figure 1. Permeability determinations for preforms of type A
(inset shows results from mercury porosimetry).
infiltration front, and what will be its effects on the
composite microstructure (morphology)? How does the
kinetics of infiltration depend on the structural proper- was kept under a vacuum-controlled atmosphere in a
ties of the preform, i.e., preform porosity and tortuosity, bell jar. Details on the experimental setup and protocol
pore size distribution, and pore orientation and inter- can be found elsewhere.5 Infiltrated samples were cooled
connection? Can the resulting temperature and stress to room temperature, sectioned, and examined by optical
distributions in a component be predicted? Can the and scanning electron metallography and electron mi-
overall extent of conversion be predicted by combining croprobe analysis.
knowledge about the chemical reaction and infiltration Microstructural Characterization of Preforms.
kinetics? A combined experimental and numerical mod- The inset in Figure 1 presents the results of mercury
eling approach is followed in this paper to answer these porosimetry determinations carried out on type A
and other related questions. preforms. Both preforms, types A and B, have narrow
The main purpose of this paper is to develop a pore size distributions. Their median pore diameters,
theoretical understanding of the reactive infiltration based either on the pore volume or on the pore surface
process by investigating the infiltration process, the two area, are nearly identical. A comparison between their
phases (kinetics- and diffusion-limited stages) of the skeletal densities, 1.48 and 1.51 g cm-3, and the
chemical reaction between the infiltrating melt and the literature-reported density of glassy carbon, 1.5 g cm-3,
carbonaceous preform, and the associated thermal suggests that the pore networks in these preforms are
phenomena. The present work addresses the following interconnected. In the presence of isolated pores, the
components: (i) Investigation of the kinetics in simple skeletal density calculated from mercury porosimetry
geometries, (ii) microstructural characterization of would be smaller than that corresponding to glassy
porous preforms, (iii) experimental analysis of reactive carbon. However, the B type preform is more porous
infiltration using two different configurations (mi- than the A type, as shown by their fractional porosities,
croporous preforms infiltrated by molten silicon and 0.53 versus 0.48.
silicon/silicon carbide/carbon preforms prepared from Permeability Analysis of Uninfiltrated Speci-
particulate materials), and (iv) phenomenological mod- mens. The porous medium permeability, κ, is an overall
eling and numerical analysis of the infiltration process. macroscopic parameter that is widely used to describe
Preliminary proof-of-concept experimental results are fluid flow through porous structures in many engineer-
presented to lend support to the model formulated. ing applications. This parameter can easily be extracted
from experimental data.
Experimental Characterization For steady-state unidirectional infiltration of a Bouss-
inesq’s fluid, Darcy’s relationship reduces to the well-
The microporous amorphous carbon preforms used in known permeability expression
this study were made from a mixture of furfuryl alcohol
resin, diethylene and triethylene glycols, and p-toluene
κ dp
sulfonic acid. This mixture was polymerized to form a u)- (1)
porous solid polymer. The solid polymer was then heated µf dz
slowly in a flowing argon atmosphere, which resulted
in the production of microporous carbon preforms. where u is the fluid velocity, p is the pressure, µ denotes
Preforms with a range of pore sizes were obtained by the dynamic viscosity of the infiltrating fluid, and κ
varying the composition of the organic mixture. Further represents the permeability of the porous medium. This
details on preform fabrication can be found elsewhere.4 relationship can be used to extract the permeability of
The microstructure of the preforms was examined by the medium from a linear correlation between fluid flow
scanning electron microscopy and characterized by velocity, u, and the applied pressure gradient. Figure 1
mercury porosimetry and permeability determinations. shows the correlation between the water flow rates and
The experimental setup used to analyze this process the applied pressure in a standard permeameter6 for
consisted of an alumina assembly heated via a graphite type A performs. The goodness of fit of these determina-
susceptor using induction coils. The carbon specimens tions supports the applicability of Darcy’s law for the
were inserted in an alumina tube, and the silicon melt preforms used in this study. For a room-temperature
was contained in a quartz crucible. The preform and viscosity of water of 9.6 × 10-10 MPa s, the permeabili-
melt were fitted with tungsten-rhenium thermocouples ties of the two preforms, A and B, can be estimated as
interfaced with an HP data logger. The whole assembly 2.29 × 10-14 and 2.70 × 10-14 m2, respectively.
 Ind. Eng. Chem. Res., Vol. 40, No. 23, 2001 5193

Formulation of a Mathematical Model Φ1 DpγLV cos θ

z2 ) Φt or u ) with Φ ) (4)
Infiltration of Microporous Structures by Reac- 2 z 4µf
tive Fluids. The siliconizing process has been ad-
dressed by a large number of authors, with most of the The parameter Φ, often considered11 an intrinsic
investigations via experiments performed between mol- infiltration rate parameter, can be obtained from the
ten silicon and isolated carbon fibers or plates. However, experimental data by plotting infiltration length or
there has been very little effort to study the most infiltration velocity versus infiltration time. However,
important step in the conversion process: the process the above equation has been derived for single-pore flow
of silicon infiltration through the porous carbon pre- and is strictly valid for a porous structure of parallel
forms. Because the infiltration of the silicon melt pores of diameter Dp. In contrast, the experimental data
precedes the chemical reaction itself, it is a crucial link yield an overall Φ value that is representative of the
in determining the shape, size, and characteristics of pore network (pore size, shape, distribution and micro-
the carbon preforms that can be successfully converted structure) being infiltrated. The Dp in the equation for
into silicon carbide. Φ thus represents an effective pore diameter, i.e., Dpe.
The infiltration of porous structures by a fluid is It is interesting to note here that, assuming a linear
usually described by Darcy’s model, which for a Bouss- dependence for the pressure in the permeability defini-
inesq’s fluid can be written as tion, a relationship between the intrinsic infiltration
parameter and the medium permeability can easily be
∂u µf derived.
F°f ) -∆p - u - F°f[âT(T - T°) + âC(C - C°)]g Most of the attempts to analyze reactive infiltration,
∂t κ
(2) therefore, have naturally followed one of these two
routes. Thus, for instance, Messner and Chiang12 at-
where u is the fluid velocity, p is the pressure, µ is the tempted to develop an analytical model to predict the
fluid dynamic viscosity, F is the fluid density, â is the infiltration phenomenon: based on a varying perme-
expansion coefficient (with C and T standing for con- ability correlated to the pore size reduction with time,
centration and temperature, respectively), g is the the infiltration behavior of silicon through graphite
constant of gravitational acceleration, and κ is the preforms was predicted via Darcy’s law. However, no
(porous) medium permeability. This latter parameter, measurement of the infiltration kinetics was carried out.
which allows one to treat the fluid-solid system as a An early study reported in the literature13 attempted
quasicontinuum medium, is intimately related to the to measure the infiltration front velocity as a function
structure properties (void fraction, pore size distribu- of temperature. To examine the infiltration kinetics,
tion, etc.). A number of empirical relationships have Einset exposed thin tapes of carbon preforms containing
been derived for packed7 and fibrous beds and bundles silicon carbide particles to silicon melt. Thermocouples
of aligned fibers,8 including correction factors for non- were inserted at different points along these tapes to
spherical shapes and misalignment of the fibers, as well. record the onset of the temperature rise due to the
There is, however, no reliable equation9 that can predict reaction at the infiltrating front. These experiments,
the permeability for a given microporous structure. The however, are not an accurate representation of infiltra-
approach, therefore, has been to extract this value from tion phenomena through a porous preform. Indeed, the
experimental measurements as outlined above. silicon melt would preferentially wet and traverse along
A different approach can alternatively be followed. the outer surface of the tapes where there is no
Instead of characterizing the medium by a single resistance to the melt flow and not through the inner
parameter, κ, a description of the capillarity-driven flow tortuous channels of the intricate porous network.
in a single pore can be made using Newton’s law.10 A Preliminary experiments are described next. Hillig’s
force balance for the isothermal infiltration of a vertical simple expression is used to assess the rate-controlling
cylindrical pore (with a constant cross-sectional area) mechanism for the reactive infiltration process.
yields Temperature Dynamics of Reactive Infiltration
Experiments. Typical temperature-time plots re-
corded by the thermocouple assembly described earlier
8µf
∑Fz ) dt(z dt ) ) - 4( dt )
1 d dz 1 dz 2 dz are shown in Figure 2. The thermal response of the
- z - gz +
πrp2Ff rp2Ff dt thermocouples located at 0.3, 0.5, 0.8, and 1.1 cm from
the (bottom) surface of a type A preform exposed to the
2
γ cos(θ) (3) silicon melt are indicated as TC1, TC2, TC3, and TC4,
rpFf LV respectively. The thermal response corresponding to the
silicon melt is indicated as Si. The uncertainty in the
where r is the pore radius; µ and F are the fluid dynamic temperature values is approximately (6 K. Figure 3a
viscosity and density, respectively; γ is the surface shows the different stages involved in the experiment.
tension; θ stands for the contact angle; and z and t The arrows at the bottom of the figure indicate the
represent the infiltration height and time, respectively. different stages of the infiltration experiments. The
The subscripts f and LV stand for fluid and liquid- melt-carbon preform contact occurred at 1853 s, with
vapor interface, respectively. For highly viscous fluids 1.7 cm of the preform being submerged into the melt.
and for short infiltration lengths (neglecting end effects The exothermic reaction between the infiltrating silicon
and body forces) this equation can be solved analytically. melt and the carbon preform resulted in a steep tem-
For most cases, where the flow occurs as a result of perature rise, recorded by all thermocouples. It is worth
capillary pressure, the fluid weight is negligible when noticing, however, that, despite the highly exothermic
compared to the surface tension during the early stages reaction, the temperatures recorded by all preform
of the infiltration (i.e., short infiltration lengths) and thermocouples return to the silicon melt temperature
the above equation yields within 60 s after the initial temperature rise. The
5194 Ind. Eng. Chem. Res., Vol. 40, No. 23, 2001

the presence of free silicon in the preform, as was

confirmed by microstructural analysis.
It is interesting to note that the peak temperatures
show a decrease with increasing distance from the
exposed surface of the preform. This is due to the
continuous decrease in the infiltration front velocity as
it travels into the preform, as predicted by the capillary
flow analysis. Such thermal profiles can be used to
estimate the infiltration front velocity for a quantitative
comparison with theoretical predictions. Detailed met-
allographic analyses and additional runs for different
preforms and conditions can be found elsewhere.5
Dynamics of Reactive Infiltration. The front
velocity was estimated by following the displacement
of the inflection point in the temperature profile. The
time responses of the different thermocouples were
differentiated, and estimates of the front velocity were
thus obtained. Figure 3 shows infiltration data and the
least-squares results (linear correlation and the corre-
sponding 95% confidence intervals) corresponding to
three infiltration experiments. As predicted by the
capillary flow analysis, the infiltration velocity de-
creased with increasing infiltration distance.
An analysis of capillarity-driven infiltration suggests
that the infiltration behavior can be predicted consider-
ing a single-pore representation of the preform structure
and using eq 4. Using the pore geometry characteriza-
tion from permeability measurements and literature
values for physical properties, eq 4 was used to calculate
theoretical infiltration dynamics (Hillig’s parameter or
slope in Figure 3). However, the discrepancies between
experimental (slope in Figure 3) and theoretical (eq 4)
slopes can reach up to 2 orders of magnitude. This
difference, despite the many simplifying assumptions,
suggests that the occurrence of the exothermic chemical
reaction has an accelerating effect on the infiltration
process.
A summary of the possible kinetic mechanisms for
this reaction and their possible integration into a
mathematical model, which would explain the disagree-
ment between experiments and theory, is presented
Figure 2. Thermal responses in a preform of type A. next.
Reaction Mechanism and Kinetics. The siliconiz-
ing process has been addressed by a large number of
authors, with most of the investigations via experiments
performed between molten silicon and isolated carbon
fibers or plates. However, there has been very little
effort to study this process in conjunction with the most
important step in the conversion process: silicon infil-
tration through the porous carbon preforms. The infil-
tration of the silicon melt precedes the chemical reaction
itself and is, therefore, a crucial link in determining the
shape, size, and characteristics of the carbon preforms
that can be successfully converted into silicon carbide.
In a reactive infiltration process, the chemical reaction
Figure 3. Correlation between infiltration velocity and infiltration
height for reactive infiltration.
competes with the infiltration process in becoming the
rate-controlling step. Understanding and characterizing
preform was raised out of the melt at about 2090 s, at the infiltration dynamics and the chemical kinetics will
which time the induction power was switched off. This therefore be crucial for the design and fabrication of
resulted in a rapid cooling, at a rate of approximately actual components by this process.
50 K min-1, recorded by all five thermocouples. The Significant controversy regarding the exact mecha-
isothermal plateau, starting at 2240 s, registered by the nism for the formation of reaction-bonded silicon carbide
thermocouple in the silicon melt corresponds to the (RBSC) has been raised in the literature. Here, the two
solidification of silicon (1687 K). It is interesting to note most feasible basic mechanisms for the RBSC process
that the isothermal trend was also indicated by the are examined. It was initially believed that, after a thin
thermocouples located in the preform, especially TC1, film of silicon carbide formed on the carbon surface,
TC2, and to some extent TC3 (Figure 3c). This indicates diffusion of silicon or carbon through solid SiC (the
 Ind. Eng. Chem. Res., Vol. 40, No. 23, 2001 5195

melting point of silicon carbide, 2100 K, is much higher

than that of silicon, 1687 K) is responsible for further
conversion.14,15 Diffusion through a grain boundary was
invoked to explain the fast reaction generally observed.
It was later proposed that the initial silicon carbide
layer spalls off because of the volume increase caused
by the reaction, exposing a fresh unreacted carbon
surface to liquid silicon. Further conversion was then
postulated to occur by dissolution of carbon into molten
silicon at higher temperatures, followed by the precipi-
tation of silicon carbide in regions of lower tempera-
tures.16-18
Diffusion-Controlled Mechanisms. The “shrinking-
core” model assumes that silicon diffuses through the Figure 4. Experimental siliconizing data adapted from Fitzer and
solid SiC and reacts with carbon at the C-SiC interface. Gadow.19
This mechanism has been associated with some of the
and â can readily be associated with well-known kinetic
early work on RBSC in Germany14,15,19 where SiC was
and transport criteria. Namely, â is the inverse of the
observed to grow as a continuous layer on graphite
Damkohler number, and R is associated with the dif-
plates. In this model, the silicon carbide layer is as-
fusion characteristic time. A critical examination of the
sumed to form via a mechanism similar to that involved
data from Fitzer and Gadow19 and Zhou and Singh24
in the oxidation of p-type metals.20,21 This model is
can then be provided in light of this model. Figure 4
simple, and a closed form solution can be obtained for
shows the data and the linear correlation corresponding
the first-order kinetics.22 The “bulging-core” model, on
to the bulging- and shrinking-core models.
the other hand, assumes that carbon diffuses through
Messner and Chiang12 proposed the following model
SiC and reacts with Si at the SiC-melt interface.
for pore shrinkage
Liquid-Phase Reaction Mechanism. This mechanism
was initially proposed by Pampuch et al.18 and later
r(t) ) Ro - ∆(t)
supported by Chiang et al.23 It assumes that carbon first
dissolves in the molten silicon by a self-accelerated
(autocatalytic) mechanism. The dissolved carbon then with ∆(t) ) k1t or ∆(t) ) k2xt (7)
diffuses to the cooler-temperature region of the melt,
where, because of reduced solubility, it precipitates as where the constants k1 ) R/â and k2 ) (2R)1/2 can be
SiC particles. This process can also lead to the growth obtained from experimental data as shown above. These
of larger SiC particles at the expense of smaller ones. experimental data do not support the assumption of
In an attempt to verify this mechanism, Chiang et al.23 kinetically controlled growth (linear with time), as was
carried out experiments similar to those reported by originally assumed by Messner and Chiang:12 the
Fitzer and Gadow19 but focusing on the onset of the Damkohler number, 1/â, cannot be statistically distin-
process (short times, i.e., first-order-dynamics phase). guished from infinity, which suggests a diffusion-
A correlation between the temperature dependence of controlled mechanism. However, it might be that this
carbon solubility in silicon melt and that of the initial absence of a kinetically controlled stage is due to the
SiC growth rate was used to support this mechanism. lack of data at short times or the existence of a
Analysis of Literature Data. Fitzer and Gadow19 condensed-phase reaction during the preheating/melting
measured silicon carbide growth rates in carbon plates stages. Experiments at different temperatures can be
in contact with molten silicon at 1873 and 2073 K. More used to correlate the effective diffusion coefficient of
recently, similar experiments have been reported24 at carbon through silicon carbide via an Arrhenius depen-
lower temperatures, 1703-1783 K. These data can be dence on temperature.
analyzed through the bulging- and/or shrinking-core Support for the Si-diffusion-based mechanism came
models. Indeed, these models yield the following non- from the observation that the siliconized carbon ap-
linear model for siliconizing growth rates proximately retained its original dimensions as it was
converted to SiC. This model seems to correlate well
∆ ) ∆o - â + xâ2 + 2Rt with results reported in Russian literature.25 However,
it is in conflict with more recent experimental observa-
with tions. For example, it is inconsistent with the phenom-
enon of pore closure (choking) observed experimentally
â ) DSi/SiCek̃ or â ) DC/SiCek̃ (6) during siliconization of porous carbon preforms. It is also
contradictory to the observation of carbon dissolution
and during siliconization of carbon fibers.
The mechanism based on the diffusion of C, on the
other hand, would be consistent with (a) the phenom-
R ) FSi/FC or R ) FC/FSi enon of preform choking during silicon infiltration,12 (b)
the growth of a continuous silicon carbide layer observed
where ∆ denotes the SiC layer thickness, t represents on graphite plates in contact with the silicon melt,14,15
time, kr is the reaction rate constant, km is the mass and (c) the observation of carbon fiber dissolution.18,23
transfer coefficient, and De stands for the effective The molar volume increase associated with conversion
diffusion coefficient. of a carbon “particle” into silicon carbide (about 58% in
Although the model breaks down for ∆o ) 0, it volume) reduces the size of the “interparticle” region,
provides a useful tool for examining the controlling containing the silicon melt. The resulting decrease in
mechanism(s) in the RBSC process. The parameters R the permeability will ultimately choke further infiltra-
5196 Ind. Eng. Chem. Res., Vol. 40, No. 23, 2001

tion of silicon melt. At first, the observation of carbon in the bulk phase of the silicon melt and act as
dissolution might appear inconsistent with the forma- nucleation sites for further SiC precipitation/growth.
tion of a continuous layer of silicon carbide. However, This mechanism, although “simplistic” in appearance,
as discussed below, this conclusion is misleading be- would add the displacement equations and precipitation
cause these observations were made under two very kinetics to the above diffusion mechanism. However, it
different sets of experimental conditions. The carbon presents a more realistic representation of the reaction
dissolution was observed for carbon fibers in contact during silicon infiltration of large-pore carbon preforms.
with silicon melt, whereas formation of the continuous Preforms with a small pore size will most likely not
SiC layer was observed for carbon plates. A thin layer manifest this mechanism. Indeed, the growth of cylin-
of silicon carbide instantaneously forms as soon as drical silicon carbide shells is accompanied by stresses
carbon is brought in contact with silicon. This SiC acts that can exceed the material tensile strength. The
as a barrier for further reaction, and the process siliconizing of fibers creates silicon carbide layers subject
becomes diffusion-controlled. It is generally believed to tensile stresses against a free surface, which therefore
that this layer is subject to increasing tensile stresses can exhibit spalling. For the silicon carbide growth on
with its further growth. The tensile strength of silicon the inside of carbon pores, on the other hand, silicon
carbide is between 50 and 200 MPa. Experiments in carbide layer breakage is contained by the unconverted
which dissolution of carbon and spalling off of the silicon carbon, and therefore, the layer can grow to larger
carbide layer were reported were on carbon fibers 10- thicknesses before the hollow SiC cylinder collapses.
100 µm in diameter (or porous preforms). It is obvious A Modified Model and its Numerical Solution.
that the radial stresses in a ring-like silicon carbide In summary, diffusion-limited SiC growth appears to
be the most feasible mechanism in the siliconizing of
layer will be far larger than those in a thin (0-50 µm)
microporous carbon preforms. The infiltration analysis
silicon carbide layer growing on a flat graphite plate
presented earlier dealt with a single-pore representation
(0.1 cm thick). The spalling, therefore, is likely to occur
of a structure consisting of a randomly oriented array
during the early stages of siliconizing of fibers or porous
of multisize pores. The pores were considered to remain
structures, whereas the process of siliconizing graphite unaltered during the infiltration process. However,
plates will not exhibit the spalling phenomena. infiltration of silicon through microporous carbon is
Messner and Chiang12 identified a transition in the accompanied by structural changes in the pores, which
growth dynamics. An apparent linear correlation con- calls for a dynamic percolation model. Any predictive
sistent with a dissolution mechanism was confirmed in tool, therefore, cannot be based on static measurements
a later paper.3 A closer inspection, however, of the alone (permeability experiments) but must also include
experimental data reported by Chiang et al.3 clearly the dynamics of the structure changes.
shows nonzero intercepts for the linear correlation. This Messner and Chiang12 proposed a simple model for
is clear evidence of the existence of a protective layer pore shrinkage
and, therefore, the presence of a diffusion mechanism.
For reasons similar to those explained above, this r(t) ) Ro - ∆(t)
protective layer is not likely to spall in the siliconizing
of small-sized pores. Moreover, growth data correspond- with ∆(t) ) k1t or ∆(t) ) k2xt
ing to diffusion-controlled mechanisms will appear
distorted if plotted as degree of carbon conversion as a where the constants k1 ) R/â and k2 ) (2R)1/2 can be
function of time and will then be misleading. The obtained from experimental data as described above.
square-root dynamics typical of diffusion-limited pro- However, in addition to the difficulty of justifying the
cesses will thus be masked. To compound the problem, linear assumption (i.e., a kinetically controlled siliconiz-
the dissolution process is highly exothermic and, there- ing process), this model has a major drawback: the
fore, should accelerate itself until mass transport be- above equations, valid for flat plates, cannot be used
comes the controlling mechanism or heat conduction for cylindrical fibers without a proper reformulation. For
promotes an equilibrium. This calls for a more detailed a cylindrical geometry, they should be
analysis of the correlation between deposition dynamics r(t) ) Ro - ∆(t) ) Ro(1 - δ(t)) (8)
and carbon dissolution. Moreover, the important ques-
tion of how the cycle of dissolution and precipitation is with δ(t) being the solution to
terminated remains unanswered. These observations
suggest a “hybrid” mechanism, which would be consis- 4R
tent with the arguments supporting the above-described (δ - 1)2 [1 - 2 ln(1 - δ)] ) 1 - t (9)
Ro2
theories.
A Hybrid Stress-Controlled Mechanism. The diffusion This simple modification predicts substantially different
mechanism is consistent with many of the experimental dynamics compared with that predicted by assuming a
observations. However, a realistic RBSC model has to simple extension of the plate results to fiber bundles.
include the stress-induced “spalling off” feature dis- Changes in the infiltrated structure as the infiltration
cussed earlier. The diffusive growth of SiC will introduce proceeds have been treated at several levels of complex-
tensile stresses in the SiC layer. As explained above, ity: from infiltration of a porous preform by a pure
these stresses will rapidly exceed the tensile strength metal or an alloy where reactions at the fluid-preform
of SiC, and the film will crack and/or spall. This will interface are neglected26,27 to the mechanisms where
then be followed by a rapid reaction stage due to the nucleation, thermophoresis, and kinetics interact to
exposure of fresh carbon surface to molten silicon. A new deposit films on the pore walls.28,29 However, reactive
diffusion layer will be formed and the diffusion- infiltration has received limited attention. The infiltra-
controlled stage will follow until the stresses again tion of a channel with a time-dependent cross-sectional
expose fresh carbon. The SiC fragments will precipitate area is governed by equations far more complex than
 Ind. Eng. Chem. Res., Vol. 40, No. 23, 2001 5197

Figure 6. Infiltration dynamics vs infiltration height. Numerical

Figure 5. Schematic of an infiltrated pore.
results for infiltration of type B preforms.
eq 3, and the simplifications leading to simple perme-
ability-structure relationships can no longer be made formulations. The time integration is carried over using
(i.e., the simple geometric model for permeability in a standard adaptive fourth-order Runge-Kutta method,
which the square of the infiltration length is propor- with the domain being updated at each integration step.
tional to time can not be recovered). The dynamics of Typical results are presented in Figure 6. These
the infiltration process is then described by a modified results intend to mimic the infiltration of type B
formulation of Newton’s law preforms. The dynamics corresponding to the infiltration
height (h), pore radius (r), and reaction-diffusion
∑F(z,h) ) F ∂t(πr2h ∂t )
∂ ∂h interfaces (C/SiC/Si) are presented in a format depicting
(10)
the predicted structure at the time that pore closure
occurs. To avoid dealing with the singularity of the
or, in detail reaction at time t ) 0, the initial infiltration stage was
correlated to experimental data extracted from Fitzer
d2h and Gadow19 and the analytical solution to the momen-
FV(t) ) A(h)γLV cos(θ) - FV(t)g -
dt2 tum equation. Despite these assumptions, the numerical
1 solution shows a striking resemblance to the experi-
A(0,t)F v(0,t)2 - FD(t) (11) mental results. Indeed, the infiltration lengths and
4
times are on the order of magnitude of experimental
where z is the infiltration direction, r is the radius of observations (i.e., infiltration lengths of 1-2 cm are
the pore, and h is the position of the silicon melt predicted, with pore closure occurring in the first 10 s
meniscus. A and V represent the cross-sectional area of infiltration, cf. Figure 3). Moreover, the pore-closing
and volume of the silicon melt, respectively; the cross- and free silicon entrapment predictions (consistent with
sectional area will depend on both the position along the evidence of free silicon found for both types of
the infiltration path and the time. FD represents the preforms) are also clearly demonstrated by the numer-
viscous drag effects. Here, surface tension, gravitational, ical solution. Issues related to the initial reaction-
viscous-drag, and end-effect forces will be time-depend- infiltration stages and thermal interaction between the
ent as they are interrelated to structural changes. different regions, physical properties, and diffusion-
Moreover, the infiltration of a single pore will then reaction parameters remain unresolved by the present
necessitate two additional governing equations, a mass model. These phenomena might be responsible for the
balance and an energy balance, which would govern the instabilities and high parametric sensitivity observed
dynamics of the infiltration geometry. To compound the experimentally but not captured by the formulated
difficulty introduced by these two equations, the de- mathematical model.
scription of the infiltration domain requires the redefi- The above analysis, however, is based on isothermal
nition of a dynamic three-region domain. Figure 5 observations. For the small fluid (melt) volume con-
presents an schematic of the different regions identifi- tained inside the pores, a strong interaction with flow
able in the reactive infiltration of a single pore, i.e., a and transport phenomena will occur. The analysis of
carbon layer, a silicon carbide wedge and flowing liquid such an infiltration process would most likely require
silicon. A detailed formulation of the corresponding additional dynamic measurements in order to further
mathematical model and boundary conditions is beyond understand the siliconizing mechanism, and it will be
the scope and extent allotted for this paper, and it will the subject of future contributions.
be published elsewhere. The macrodynamics of the infiltration-reaction pro-
The numerical solution of the governing momentum, cess has received very limited attention. Some efforts
mass, and energy transport governing equations is in this direction have recently been reported in the
better suited by a method that treats the different literature.31 These studies involved the use of com-
regions illustrated in Figure 5 separately. The authors mercial CFD software (FIDAP) to model mass and
have adopted orthogonal collocation on finite elements energy transport phenomena in a regular arrangement
(OCFEM),30 a method that has been reported to be of parallel graphite fibers in contact with a silicon melt.
successful in many diffusion-reaction problems. This Flow was assumed to occur parallel to the fiber axes,
method is particularly suited to represent many of the and a two-dimensional concentration and temperature
qualitative features of the problem, it results in a robust distribution was formulated. The reaction was assumed
and simple formulation, and it is generally more ac- to be under kinetic control and the initial (only the
curate than finite difference (FD) approaches while dynamic analysis for the first 0.1 µs was reported)
being simpler than more traditional finite element stages of the Si/C counterdiffusion process were simu-
5198 Ind. Eng. Chem. Res., Vol. 40, No. 23, 2001

lated. Although the permeability was related to the (8) Jackson, G. W.; James, D. F. The Permeability of Fibrous
cross-sectional area change due to the reaction, the Porous Media. Can. J. Chem. Eng. 1986, 64, 364.
dynamics of the reactive infiltration process was not (9) Dullien, F. A. L. Porous Media. Fluid Transport and Pore
Structure; Academic Press: New York, 1979.
analyzed. The authors concluded that a phenomenologi- (10) Martins, G. P.; Olson, D. L.; Edwards, G. R. Modeling of
cal model (similar to the one proposed above) where Infiltration Kinetics for Liquid Metal Processing of Composites.
carbon dissolution and breakage of the SiC layer Metall. Trans. 1988, 19B, 95.
combine to complete the siliconizing process was needed. (11) Hillig, W. B. Melt Infiltration Process for Making Ceramic
Nevertheless, the unidirectional flow assumption and Composites. In Fiber Reinforced Ceramic Composites. Materials,
the ordered geometry reduce the applicability of these Processing and Technology; Mazdiyasni, K. S., Ed.; Noyes Publica-
tions: Park Ridge, NJ, 1990; pp 260-277.
results to the problem analyzed in this paper. (12) Messner, R. P.; Chiang, Y. M. Liquid-Phase Reaction-
Bonding of Silicon Carbide Using Alloyed Silicon-Molybdenum
Conclusions Melts. J. Am. Ceram. Soc. 1990, 73 (5), 1193.
(13) Einset, E. O. Transport and Reaction in Silicon Melt
The results indicate the need for a dynamic percola- Infiltration of Porous Carbon Preforms. J. Am. Ceram. Soc. 1996,
tion model that includes structural changes in the pores. 79 (2), 333.
The fluid contained within the microchannels (1-5 µm) (14) Fitzer, E.; Gadow, R. Fiber Reinforced Silicon Carbide.
interacts thermally with the solid skeleton, and the local Ceram. Bull. 1986, 65 (2), 326.
fluid properties change drastically with the sharp tem- (15) Minnear, W. P. Interfacial Energies in the Si/SiC System
and the Si + C Reaction. Am. Ceram. Soc. Commun. 1982, Jan,
perature rise. The species distribution is also influenced. C-10-11.
Structural changes are local phenomena, strongly in- (16) Ness, J. N.; Page, T. F. Microstructural Evolution in
fluenced by the temperature and species distribution. Reaction Bonded Silicon Carbide. J. Mater. Sci. 1986, 21, 1372.
Therefore, any predictive tool must include the dynam- (17) Hase, T.; Suzuki, H.; Isekei, T. Formation Process of â-SiC
ics of the flow environment, thermal interactions be- During Reaction Sintering. J. Nucl. Mater. 1976, 59, 42.
tween the melt and the solid, temperature effects on the (18) Pampuch, P.; Walasek. E.; Bialoskorski, J. Reaction Mech-
siliconizing process, and residual reaction/thermal ef- anism in Carbon-Liquid Silicon Systems at Elevated Tempera-
tures. Ceram. Int. 1986, 12, 99.
fects resulting from of the free silicon entrapped ahead (19) Fitzer, E.; Gadow, R. Investigation of the Reactivity of
of the choking front. In other words, a model based Different Carbons with Liquid Silicon. In Proceedings of the
purely on flow analysis considerations, even when the International Symposium on Ceramics for Engines; Technoplaza
pore structure is adapted dynamically to account for the Co. Ltd.: Tokyo, Japan, 1983; pp 561-572.
occurrence of a chemical reaction, might not be adequate (20) Rode, H.; Hlavacek, V.; Viljoen, H. J.; Gatica, J. E.
for predicting all features of a reactive infiltration Combustion of Metallic Powders: A Phenomenological Model for
the Initiation of Combustion. Combust. Sci. Technol. 1993, 88, 153.
dynamics. (21) Zhou, H.; Singh, R. N. Reaction Kinetics and Mechanism
in Si-C and Si-C-Mo Systems. In Advances in Ceramic-Matrix
Acknowledgment Composites; American Ceramic Society: Westerville, OH, 1993;
pp 91-102.
Alppreciation is expressed to Pat Dickerson, Dave (22) Levenspiel, O. Chemical Reaction Engineering; John Wiley
Epperly, Jim Barker, and Christopher Palda for their & Sons: New York, 1972; Chapter 12.
technical assistance and to Thomas K. Glasgow, Chief, (23) Chiang, Y. M.; Haggerty, J. S.; Messner, R. P.; Demetry,
Processing Science and Technology Branch, at the C. Reaction-Based Processing Methods for Ceramic-Matrix Com-
NASA Glenn Research Center for partial support of this posites. Ceram. Bull. 1989, 68 (2), 420.
research. One of the authors (J.E.G.) gratefully acknowl- (24) Zhou, H.; Singh, R. N. Kinetics Model for the Growth of
Silicon Carbide by the Reaction of Liquid Silicon with Carbon. J.
edges support from the Established Full-time Faculty Am. Ceram. Soc. 1995, 78 (9), 2456.
Research Development (EFFRD) program at Cleveland (25) Lukin, B. V. Determination of the Linear Velocity of the
State University. Dr. Orejas acknowledges the support formation of â-SiC. Izv. Akad. Nauk SSSR 1971, 7 (7), 1169.
given by the Secretarı́a de Ciencia y Técnica Univer- (26) Mortensen, A.; Masur, L. J.; Cornie, J. A.; Flemmings, M.
sidad Nacional de Rı́o Cuarto (UNRC) and by the C. Metall. Trans. A. 1989, 23, 2535.
Consejo de Investigaciones Cientı́ficas y Técnicas de la (27) Biswas, D. K.; Gatica, J. E.; Tewari, S. N. Dynamic
Analysis of Pressure Infiltration Processes. Ceram. Trans. 1995,
provincia de Córdoba (CONICOR), Argentina.
58, 105-110.
(28) Jensen, K. F. Micro-Reaction Engineering: Applications
Literature Cited of Reaction Engineering to Processing of Electronic and Photonic
Materials. Chem. Eng. Sci. 1987, 42, 923.
(1) Popper, P. The Reaction of Dense Self-bonded Silicon
(29) Gomez, A.; Rosner, D. E. Combust. Sci. Technol. 1993, 89,
Carbide. In Special Ceramics; Heywood: London, 1960; p 209.
335 (see also Koylu, O. U.; Tandon, P.; Xing, Y.; Rosner, D. E.
(2) Mehan, R. L. Effect of Silicon Carbide Content and Orienta- Experimental and Theoretical Studies of the Structure of Inorganic
tion on the Properties of Si/SiC Ceramic Composite. J. Am. Ceram. Particle-Producing Seeded Laminar Counterflow Diffusion Flames.
Soc. 1977, 60 (3-4), 177. Presented at the Annual AIChE Meeting, Miami, FL, 1995; Paper
(3) Chiang, Y. M.; Messner, R. P.; Terwilliger, C. D.; Behrendt, 59p.
D. R. Reaction Formed Silicon Carbide. Mater. Sci. Eng. 1991, (30) Finlayson, B. A. Nonlinear Analysis in Chemical Engineer-
A144, 63. ing; McGraw-Hill, New York, 1980.
(4) Singh, M.; Behrendt, D. R. Studies on the Reactive Melt (31) Rajesh, G.; Bhagat, R. B.; Nelson, E. Modeling the flow of
Infiltration of Silicon and Silicon-Molybdenum Alloys in Porous molten silicon in porous carbon preforms and the subsequent
Carbon; NASA Report TM-105860; National Aeronautics and formation of silicon carbide. In Ceramic Matrix Compositess
Space Administration: Washington, D.C., 1992. Advanced High-Temperature Structural Materials; Lowden, R. A.,
(5) Sangsuwan, P.; Tewari, S. N.; Gatica, J. E.; Singh, M.; Ferbere, M. K., Hollmann, J. R.,Chawla, K. K.,DiPietroand, S. G.,
Dickerson, R. Reactive Infiltration of Silicon Melt Through Mi- Eds.; MRS Symposium Proceedings 365; Materials Research
croporous Carbon Preforms. Metall. Trans. 1999, 30B, 933. Society: Pittsburgh, PA, 1995; pp 147-152.
(6) Simhai, N.; Datta, S. K.; Singh, M.; Dickerson, R.; Smith,
J.; Gatica, J. E.; Tewari, S. N. Characterization of Microporous Received for review December 5, 2000
Preforms: Analysis of Permeability. Metall. Mater. Trans. A 1996, Revised manuscript received June 6, 2001
27, 3669. Accepted June 6, 2001
(7) Ergun, S. Fluid Flow through Packed Columns. Chem. Eng.
Prog. 1952, 48, 89. IE001029E