Corrosion of Silicon-Based Ceramics in Combustion Environments
`
`Nathan S. Jacobson*
`National Aeronautics and Space Administration, Lewis Research Center, Cleveland, Ohio
`44 135
`
`Silicon-based ceramics and composites are prime candi-
`dates for heat engine and heat exchanger structural compo-
`nents. In such applications these materials are exposed to
`combustion gases and deposit-forming corrodents. In this
`paper combustion environments are defined for various
`applications. These environments lead to five main types of
`corrosive degradation: passive oxidation, deposit-induced
`corrosion, active oxidation, scale/substrate interactions,
`and scale volatility. Each of these is discussed in detail. The
`key issues in oxidation mechanisms of high-purity silicon
`carbide (Sic) and silicon nitride (Si,N4) in pure oxygen are
`discussed. The complicating factors due to the actual com-
`bustion environment and commercial materials are dis-
`cussed. These discussions include secondary elements in the
`ceramics; additional oxidants, such as water and carbon
`dioxide (COJ; combustion environment impurities; long-
`term oxidation effects; and thermal cycling. Active oxida-
`tion is expected in a limited number of combustion situa-
`tions, and the active-to-passive transition is discussed. At
`high temperatures the limiting factors are scale melting,
`scale volatility, and scale/substrate interactions. Deposit-
`induced corrosion is discussed, primarily for sodium sulfate
`(Na,SO,), but also for vanadate and oxide-slag deposits as
`well. In applying ceramics in combustion environments it is
`essential to be aware of these corrosion routes and how they
`affect the performance of a component.
`
`Introduction
`I.
`ANY potential uses of silicon-based ceramics and com-
`
`M posites involve exposure to combustion gases. These
`
`applications range from hot-section structural components of
`gas turbines and piston engines to heat exchanger tubes for
`industrial furnaces. Gas turbines are used in aircraft and electric
`power generation and have been tested for automobiles. Poten-
`tial ceramic components of these engines include combustor
`liners and, perhaps some day, turbine
`The location of
`these components in a gas turbine engine is shown in Fig. I(a).
`In piston engines the potential ceramic components include
`
`S. M. Wiedcrhorn~ontrihutingeditor
`
`Manuscript No. 1955915. Received June 15. 1992; approved Septembcr 28,
`19?2.
`Member, American Ccramic Society.
`
`valves and piston heads,4 as shown in Fig. I(b). The use of
`ceramic tubes as heat exchangers has been proposed for indus-
`trial furnaces, such as glass remelt furnaces, steel soaking pits,
`and aluminum reclamation furnaces.' An example is illustrated
`in Fig. I(c). In addition, ceramics are under consideration for
`heat exchangers in coal-fired combustors.' A more developed
`application is the use of ceramic tubes for indirect heating,
`where hot combustion gases pass through the tube, heating a
`process on the o u t ~ i d e . ~ The structural components of the hot
`sections are subject to a range of chemical attack processes,
`depending on
`the
`temperature, pressure, and chemical
`environment.
`The focus of this paper is on the silicon-based ceramics sili-
`con carbide (Sic) and silicon nitride (SilN,,). These materials
`are inherently unstable in air and form a thin layer of silicon
`dioxide (SO,) in an oxidizing environment. SiO? has the lowest
`permeability to oxygen of any of the common oxides and forms
`an effective reaction barrier.' Therefore, silicon-based ceramics
`have the potential of substantially better high-temperature oxi-
`dation behavior than metals. The protective oxide scales on sili-
`con, Sic, and Si,N, are shown schematically in Fig. 2 . Note
`that Si,N, forms a silicon oxynitride (Si2N20) layer below the
`SiOz layer.'SiC may also form an oxycarbide layer, but there is
`only limited evidence for this." The analogous protective layers
`on superalloys for high-temperature applications are alumina
`(A120J and chromia (Cr203)." Extensive work has been done
`on the performance of AI,O, and Cr,O, scales in combustion
`environments. In general, our knowledge of the behavior of
`SiO, scales lags behind that of A120, and Cr20,.
`The purpose of this paper is to review our current state of
`knowledge of the interaction of SiO, with combustion environ-
`ments. Because the focus is on the interaction of SiO? with the
`environment, many of the conclusions apply to all Si0,-pro-
`tected materials. This includes composites of S i c and Si,N,,
`such as Sic-fiber-reinforced S i c matrices and SiC-fiber-rein-
`forced Si3N, matrices, and SO,-forming alloys, such as molyb-
`denum disilicide (MoSi,). In this paper the major chemical
`degradation routes are discussed. Particular emphasis is on the
`mechanisms of corrosion and the key questions involving them.
`
`11. Environments
`Combustion environments vary widely, depending on fuel,
`temperature, pressure, and oxidizer. Fuels are complex mix-
`tures of hydrocarbons, and they are classified by boiling points.
`Lower-boiling-point fuels include automotive gasoline, and
`higher-boiling-point fuels include aviation fuel and fuel oils.12
`
`GE-1019.001
`
`

`
`4
`
`,- Fan blades
`
`I
`I
`
`Journul ofthe American Ceramic Society-Jacobson
`
`Vol. 76, No. 1
`
`Stator vanes
`
`\
`
`r Turbine
`blades
`
`I
`
`holder --, Flue gas
`
`Sample
`
`(b)
`
`(4
`
`Schematics of proposed applications lor ceramics where ceramics will be subjected to combustion environments: (a) gas turbine engine, (b)
`Fig. 1.
`piston engine, and (c) aluininum reclamation furnace (see Ref. 5).
`
`Table I lists some common fuels and their approximate impurity
`contents. " -
`l h Note that all fuels have some amount of sulfur,
`which may lead to corrosion. The last category of fuel oils con-
`tains more sodium, potassium, and vanadium than the other
`fuels. These lower-purity fuels are likely to be utilized more in
`the future as petroleum resources decrease. Impurities have a
`major influence on corrosion. Table I does not include coal-
`derived fuels, which generally have even higher impurity
`levels.
`Combustion is the oxidation of these fuels to the stable prod-
`ucts carbon dioxide (C02) and water (H20). Equilibrium com-
`bustion products can be calculated by mixing the fuel and the
`oxidant together in a free-energy-minimization computer
`code. Figure 3(a) shows the equilibrium combustion products
`for a standard aviation fuel (Jet A-CH,
`91xs) as a function of
`equivalence ratio. Equivalence ratio is defined as the fuel-to-air
`ratio at a particular point divided by the stoichiometric fuel-to-
`air ratio for complete combustion to CO, and H20. Thus, an
`
`equivalence ratio of 1 is stoichiometric, an equivalence ratio of
`less than 1 denotes a fuel-lean region, and an equivalence ratio
`greater than I denotes a fuel-rich region. Most gas turbines
`operate in the fuel-lean or stoichiometric regions. These
`regions contain large amounts of oxygen with the CO, and H,O.
`Most corrosion studies have been performed in the fuel-lean
`region. However, some novel combustor designs may involve
`the fuel-rich region, which produces larger amounts of carbon
`monoxide (CO) and hydrogen gas (HJ. However, this region
`also contains the combustion products C02 and H,O. Note that
`these are equilibrium calculations; if equilibrium is not
`attained, other species, such as elemental carbon, may form.
`The goal of a corrosion study is to understand the chemical
`reactions that occur between these combustion products and the
`proposed hot-gas-path structural materials.
`Figure 3(b) shows the adiabatic flame temperature as a func-
`tion of equivalence ratio. This temperature is calculated by
`using a free-energy-minimization computer code and setting
`
`GE-1019.002
`
`

`
`January 1993
`
`Corrosion of Silicon-Based Ceramics in Combustion Environments
`
`5 L
`L
`
`0
`
`.5
`
`1 .o
`1.5
`Equivalence ratio
`
`2.0
`
`2500 r
`
`Si02
`
`SiOp
`
`2000
`
`1500
`
`1000
`
`500
`
`~~~
`
`0
`
`(b)
`
`I
`
`I
`
`I
`2.0
`
`I
`1 .o
`1.5
`.5
`Equivalence ratio
`Fig. 3. Calculated gas composition and Rarne temperature as a func-
`tion of equivalence ratio: (a) equilibrium gas composition and (b) adia-
`batic flame temperature.
`
`Schematic of protective oxide scales: (a) on Si, (b) on Sic,
`Fig. 2.
`and (c) on Si,N,.
`
`the net heat loss equal to zero." This is an idealized case that
`assumes no heat loss through the walls of the combustion cham-
`ber. It does give the maximum temperature of the flame, how-
`ever. In general, the wall materials and the components
`downstream from the flame are at lower temperatures.
`Corrosion occurs not only by gaseous coinbustion products,
`but also by deposits. Perhaps the most common deposit is
`sodium sulfate (Na,SO,), which forms when sodium reacts with
`sulfur fuel impurities.'X The sodium may originate from a
`marine environment, from a salted roadway, or as a fuel impu-
`rity. Corrosion by Na,SO, is termed "hot corrosion" and is dis-
`cussed in detail herein. Other types of deposit-induced
`corrosion originate from vanadium fuel impurities and from
`oxide slags.
`In some ways the piston chamber of an internal combustion
`engine is a more complex environment than the hot section of a
`gas turbine. Pressures and temperatures vary through the
`stroke of the piston. Adiabatic flame temperatures and combus-
`tion gas product compositions can be calculated by the same
`free-energy-minimization program described previously with
`propane (C,H,) as a model fuel. The results of this calculation
`are similar to those for Jet A aviation fuel. Current automotive
`engines run slightly fuel rich to allow proper operation of pollu-
`tion control devices.
`In addition to heat engines, silicon-based ceramics are also
`prime candidates for heat exchanger tubes in industrial fur-
`naces. The environment encountered by a heat exchanger var-
`ies widely. Again, one would expect large amounts of COz and
`
`H 2 0 , but now the impurities play a key role. An aluminum rec-
`lamation furnace and a glass remelt furnace may involve alkali-
`metal salts that can deposit on the heat exchanger.' In a coal-
`tired furnace the atmosphere may be more reducing." In addi-
`tion, a mixture of oxides may form a slag deposit, which can be
`quite corrosive, on the tubes.
`Table I1 summarizes the composition of these combustion
`environments. This list is by no means exhaustive. As noted,
`specific applications generate specific corrodents. For example,
`combustion of municipal wastes may generate hydrogen chlo-
`ride (HCI).'" In summary, combustion environments are com-
`plex, involving not only the combustion products COz and
`H,O, but a variety of other gases and possible deposits as well.
`Temperatures and pressures are again quite dependent on the
`I I . An
`particular system. The ranges are listed in Table
`important issue to consider is thermal cycling. Some applica-
`tions, such as a utility turbine or industrial furnace, involve
`essentially isothermal exposures for long periods. Other appli-
`cations, such as an aircraft gas turbine, involve thermal
`cycling.
`
`Ill. Qpes of Corrosive Attack and
`Experimental Techniques
`Figure 4 shows the major types of corrosive attack as a func-
`tion of pressure and reciprocal temperature. The major types of
`corrosive degradation are passive oxidation, deposit-induced
`corrosion, active oxidation, oxideisubstrate interactions, and
`scale vaporization. The temperature boundaries between these
`types are only approximate and are dependent on the specific
`system. Furthermore, rarely is one mechanism operative. In
`practice, several mechanisms operate simultaneously.
`Laboratory studies of corrosive degradation fall into two
`main categories: burner rig studies and laboratory furnace stud-
`ies. Burners more accurately model the actual combustion situ-
`ation. Figure 5 shows a typical high-velocity burner rig.
`
`Table I. Properties of Some Common Hydrocarbon Fuels
`H C molar
`S content
`Boding range
`(K)*
`Fuel
`rdtio
`(wt%)'
`0.15
`2.02
`300-375
`Unleaded automotive gasoline
`1.92
`450-560
`0.05
`Jet A (commercial aviation fuel)
`1.61
`450-6 15
`Fuel oils
`0.1-1 .0
`~ *Reference 12. 'Refcrcnce 13. 'References 14and IS. 'Reference 15. 'References 15 and 16
`
`N d + K content
`(ppm)'
`3.6
`-10
`10-20
`
`V content
`( p p d
`<0.003
`0.06
`<300
`
`GE-1019.003
`
`

`
`6
`
`Journal of the American Cerumic Sociep-Jucohson
`
`Vol. 76, No. 1
`
`Application
`Heat engines
`Gas turbine
`
`Piston
`Heat exchanger
`
`Coal combustion
`’ I bar = 1o‘Pa. ‘Cycling.
`
`Tcrnperaturc (K)
`
`Summarv of Combustion Environments
`Table 11.
`Gas atmosphere
`Pressure (bar)’
`
`1173-1673
`
`1273-1 873’
`1273-1 873
`
`1473-1673
`
`1-50
`
`1-10
`
`1-10
`
`- I
`
`Fuel lean: NZ, O?, CO,, HZO;
`fuel rich: N2. COz, HzO, CO, H?
`N2, COZ, HZO,O2
`Oxidizing
`
`Reducing
`
`Potential deposit
`
`Na2S0,, NaZV,O,.
`
`Alkali halides, Na2S0,,
`transition-metal oxides
`Acidic or basic coal slags
`
`- NaqSOq-induced
`‘ corrosion
`
`Melting point
`of SiO, 7
`-
`
`Passive oxidation
`
`Reciprocal temperature, 1/T, K -
`
`SiO, has some unique properties that influence its perfor-
`mance as a protective oxide. In most oxidation experiments,
`SiO, forms as an amorphous film and then crystallizes to either
`cristobalite or tridymite. This is significant because these
`phases have very different physical and chemical properties.”
`It is generally accepted that the mobile species is oxygen, not
`silicon. For this reason the chemical reaction occurs at the
`Si02/Si or SiOz/ceramic interface.”-23 Transport through SiO,
`can occur by diffusion of molecular oxygen as interstitials or by
`network exchange of ionic
`The latter is referred to
`as either “network exchange diffusion” or “ionic diffusion” in
`this paper. Molecular oxygen diffusion coefficients are mea-
`sured by examining permeation through a SiO, membrane;”
`network exchange diffusion coefficients are measured by an
`isotope exchange technique."^" Table 111 lists both types of dif-
`fusion coefficients. The molecular oxygen diffusion coefficient
`is roughly 10‘ times greater than the ionic oxygen diffusion
`coefficient. There have been attempts to correlate these two
` coefficient^;'^ however, Cawleyj” has
`types of diffusion
`reported the difficulties with such a correlation due to network/
`interstitial exchange and mobile network ions. These issues and
`the disparity of the diffusion coefficient measurements (Table
`111) indicate there are still some unresolved issues in these fun-
`damental transport quantities.
`Before discussing the oxidation of SIC and Si,N,, it is appro-
`priate to discuss the oxidation of pure silicon. Because this pro-
`cess is critically important to the semiconductor industry, it has
`been extensively studied. The classic paper in this area is by
`Deal and Grove.” They view oxidation as consisting of three
`distinct steps, as shown in Fig. 8. The three steps are transfer of
`the gaseous oxidant to the outer surface of the oxide film, diffu-
`sion through the oxide film, and reaction at the oxideisilkon
`interface. From these steps they derive the following linear-
`parabolic equation:
`x’ + Ax = B(t + T)
`(1)
`where x is the scale thickness, B the parabolic rate constant, t
`the time, and T the time shift corresponding to the presence of
`an initial oxide layer. The quantity BIA is the linear rate con-
`stant. The parabolic rate constant is given by
`B = 2D,,,C*(02)/N,,
`where D,,, is the diffusion coefficient through the film, C*(O,)
`is the equilibrium concentration of oxidant in the scale, N,, the
`number of oxygen molecules incorporated into the SiO, scale
`per unit volume. Note that No must be modified for S i c oxida-
`tion due to the formation of CO. For short oxidation times, oxi-
`dation follows a linear rate law. The physical interpretation of
`this linear region is still controversial; it has been attributed to
`interface control” or to diffusion control which is nonparabolic
`due to strain effects in the oxide.’2 For longer times, oxidation
`follows a parabolic law, and diffusion through the thicker oxide
`is rate controlling. An activation energy can be determined
`from a plot of In B vs l/T. The magnitude of the activation
`energy can reveal useful information about the diffusion pro-
`cess. Deal and Grove” report that care must be taken to inter-
`pret the oxidation process at the appropriate times (i.e.,
`parabolic behavior cannot be assumed when the process is still
`in the linear region). They have shown that Eq. ( I ) describes a
`
`9
`
`8
`
`7
`
`6
`
`5x1 O4
`
`1200
`
`1600 1800 2000
`1400
`Temperature, T, K
`
`Major types of corrosive attack and degradation as an
`Fig. 4.
`approximate function of reciprocal temperature (P,,,,,, = Pc,x,ddn,).
`
`However, burner rig tests are expensive, and it is difficult to
`precisely control all the operating parameters. Laboratory fur-
`nace tests offer much more accurate control of system parame-
`ters. Gas composition, pressure, and temperature can be
`accurately controlled. The extent of corrosion can be monitored
`by following the weight gain with a microbalance or by measur-
`ing scale thickness with various optical techniques. Figure 6
`shows a standard laboratory microbalance apparatus for iso-
`thermal oxidation studies. Standard electron-optical techniques
`have been useful for characterizing corroded specimens. In gen-
`eral, burner rig tests are helpful for an overall assessment of
`behavior in the actual environment. Laboratory furnace tests
`complement burner rig tests by permitting isolation of individ-
`ual effects so that they can be understood on a fundamental
`basis.
`
`IV.
`
`Isothermal Oxidation
`
`(1) General Considerations and Pure Silicon Oxidation
`A large amount of research has been performed on the iso-
`thermal, passive oxidation of silicon-based ceramics. This
`research is complex because of the various secondary effects,
`such as other oxidants and impurities in the ceramics and the
`environment. These effects are noted in Fig. 7. Fundamental
`studies are at the center of the circle-pure materials and pure
`oxygen environments. Eliminating the complicating factors
`allows detailed studies of the basic oxidation mechanism.
`Actual combustion environments are more complex, involving
`the various factors on the outer circle in Fig. 7.
`The focus of this section is on fundamental studies of S i c and
`Si,N, oxidation. These studies provide an atomistic understand-
`ing of the mechanism of SiO, scale growth. These include
`determining the slow reaction step (or steps) and understanding
`the diffusion mechanism. Despite the large number of papers in
`this area, many questions remain unanswered. This paper is not
`intended to he an exhaustive review but rather a discussion of
`some of the key issues involved.
`
`GE-1019.004
`
`

`
`January 1993
`
`Corrosion o j Silicon-Based Ceramics in Combustion Environments
`
`7
`
`Salt solution
`
`Ignitor
`
`(a)
`
`Fig. 5. Mach 0.3 burner rig. (Courtesy of M. Cuy, NASA Lewis.)
`
`wide range of oxidation data. Calculations of the parabolic rate
`constant that are based on the molecular oxygen diffusion coef-
`ficient of Norton2'show good agreement with the measured val-
`ues, and the activation energy for parabolic oxidation is close to
`that for molecular oxygen diffusion through SiO?. Thus, it is
`generally accepted that molecular oxygen diffusion through the
`SiO, layer is rate ~ o n t r o l l i n g . ~ ~ . ~ ~
`The oxidation of S i c and Si3N4 is more complex. Figure 2
`illustrates some of the processes involved. In both ceramics
`
`there is a countercurrent of gas. The following reactions are
`generally accepted:
`SiC(s) + 1.502(g) = SiO,(s) + CO(g)
`(3)
`Si,N,(s) + 0.7502(g) = I.SSi,N,O(s) + 0.5N2(g)
`(4a)
`SizN20(s) + 1.50z(g) = 2Si02(s) + N,(g)
`(46)
`However, the actual processes may involve other reactions. The
`production of CO in the oxidation of S i c has been observed
`
`GE-1019.005
`
`

`
`8
`
`Journal of the American Ceramic Society-Jacobson
`
`Vol. 76, No. 1
`
`r-
`
`Purified gas in
`
`L-
`
`Thermocouple
`Gas out
`
`Fig. 6 . Schematic of oxidation balance apparatus
`
`Other
`oxidants
`
`Impurities
`(SO,, HCI, Na, K),
`transition metals
`
`additive
`effects
`
`i
`/ \ isothermal
`Ceramic /
`
`oxidation,
`pure materials,
`pure oxygen
`
`Thermal
`
`crystallization
`
`Long-term
`isothermal
`effects
`
`Complications to isothermal oxidation due to additive-con-
`Fig. 7.
`taining ceramics and combustion environments.
`
`with the gas ~hromatograph,.'~ although the possibility of fur-
`ther oxidation to CO? must be considered. There has been no
`clear observation of the production of elemental carbon. The
`oxidation of Si,N, may lead to the production of nitrous oxide
`(NO(g)) in addition to molecular nitrogen gas, as observed with
`
`a mass ~pectrometer.~",~~ Note that, even though the reactions
`lead to a net weight gain, there is also some weight loss due to
`gas evolution. These reactions may be monitored by following
`net weight gain, scale thickness as a function of time, or even
`gas evolution. The following factors allow conversion of para-
`bolic rate constants from scale thickness to net weight gain:"
`for S i c
`( 5 )
`1 (m'is) = I .67 x lo-" (kg'/(m4.s))
`for Si,N,
`I (m'is) = 3.77 X lo-" (kgLi(m4.s))
`(6)
`Note that Eq. ( 5 ) is based on Eq. (3) and that Eq. (6) is based on
`a combination of Eqs. (4a) and (4b), i.e., the oxidation of Si,N,
`to SO,. This is appropriate since SiO, is formed in much larger
`amounts than Si2N,0.'
`The oxidation process for S i c and Si,N, involves five steps:
`( I ) transport of molecular oxygen gas to the oxide surface, (2)
`
`diffusion of oxygen through the oxide film, (3) reaction at the
`oxideiceramic interface, (4) transport of product gases (e.g.,
`CO and nitrogen) through the oxide film, and ( 5 ) transport of
`product gases away from the surface. This process is even more
`complex in the Si,N, case because of the duplex film formation.
`Many questions remain about these steps. The key question
`concerns the rate-controlling, or slowest, step. This has been an
`area of controversy for both S i c and Si,N, and is discussed in
`further detail herein. Another question deals with the transport
`mechanism through the SiOz scale. Is it permeation through the
`network by oxygen molecules or a network exchange mecha-
`nism of 02-? Even less is known about the transport mecha-
`nism of product gases (CO and nitrogen) outward through the
`SiOz scale. As noted, SiOz undergoes a number of phase
`changes. This section deals primarily with amorphous scales.
`Crystallization, which complicates the oxidation process, is
`discussed separately.
`(2) Silicon Carbide
`First consider the oxidation of Sic. The key observations
`have been summarized by several in~estigators."~" Generally,
`only one scale forms, although there is limited evidence that an
`oxycarbide may form between the oxide and the Sic."' Figure 9
`shows a representative kinetic curve. Many investigators have
`observed a brief (<< 1 h) linear region followed by a parabolic
`region."+" The focus of most investigations has been on the
`parabolic region, as shown in Fig. 9. Figure I0 shows the para-
`bolic rate constants for a number of S i c materials. Only the
`higher-purity materials
`(i.e., chemically-vapor-deposited
`(CVD) S i c and single-crystal Sic) are discussed in this section.
`The key question concerns the rate-controlling step. There
`are three possibilities: ( I ) oxygen diffusion inward, (2) CO dif-
`fusion outward, and/or (3) an interfacial reaction. Recently,
`Luthra"' systematically examined these possibilities based on
`available experimental data. His general approach is taken
`here. Table IV summarizes some of the experimental observa-
`tions and their implications.
`( A ) Oxygen Diflusion Inward:
`In general, most of the
`data imply oxygen diffusion inward as rate limiting. Rates are
`parabolic and dependent on the partial pressure of oxygen
`(Po,). Motzfeld4' has reviewed the literature to about 1963 and
`carefully compared the rate constants for silicon and Sic. After
`correcting for the stoichiometry difference (i.e., the additional
`oxygen necessary to oxidize carbon), he finds that silicon and
`Sic have essentially the same rates and the same activation
`energies. Therefore, he concludes that the same process that
`controls silicon oxidation also controls S i c oxidation (i.e.,
`oxygen diffusion inward). Much of the data he uses are for S i c
`powder. The difficulty in a precise determination of the surface
`area and possible impurities suggests that high-purity coupons
`may give more precise data for Sic. Recent measurements on
`single-crystal
`give rates somewhat slower than those for
`silicon, as shown in Fig. 10.
`Table V summarizes the activation energies for oxidation of
`high-purity Sic. Note that there is generally a low-temperature
`(T < 1623 K) and a high-temperature (T > 1623 K) regime.
`Consider the low-temperature regime. Here the activation
`energy is low, about 120 to 140 kJ/mol, and is similar to that for
`oxidation of pure silicon and molecular oxygen diffusion
`through amorphous SiO,. This supports molecular oxygen dif-
`fusion inward as a rate-controlling step for S i c oxidation below
`I623 K.
`Consider further the type of transport through the growing
`SiO, layer. As mentioned, the activation energy for oxidation
`becomes much greater after 1623 to 1673 K for controlled-
`nucleation thermally deposited (CNTD) S i c and the carbon
`face of single-crystal Sic. Zheng et ~ 1 . ~ ~ . ~ ~
`have explained this
`phenomenon as a change in the oxygen transport mechanism.
`Below about 1623 K , the activation energy for the parabolic
`rate constant is close to that for silicon oxidation, and diffusion
`of molecular oxygen through the scale is rate controlling. They
`show this by forming a scale in '"0 and then reoxidizing in ''0.
`
`GE-1019.006
`
`

`
`January 1993
`
`Corrosion of Silicon -Rri.srd Ccrntnics in Cornbus tion Eli virotinwnrs
`
`9
`
`Table 111. Diffusion Coefficients for Oxygen in Amorphous SiO,
`Typc 01 d i f t u w n coctlicicnt.
`Icmpcralurc rlrnpc ( K )
`MolecularO,, 973-1 373
`Network exchange. I 123- IS23
`Network exchange, 1198-1498
`Network exchange, 1423-1 703
`
`Twhniquc
`Permeation
`Isotope exchange
`isotope exchange
`Isotope exchange
`
`Ditfu\ion cocflicicnt
`valuc ( I l l ' / \ )
`2.9 x 10 'exp( - 112900iRTj
`2.0 X 10 '.'exp(- 121336/RT)
`1.5 X 10 "exp(-29790l/RT)
`4.4 X 10- "exp( - 82425iRT)
`
`lnvestigator
`
`Norton"
`
`Sucov2'
`Muehlenbachs and Schaeffer'"
`'E* in Jimol.
`
`Gas
`
`C'
`
`= 3K"D::P,;:,'"
`where the subscripts i and R refer to the SiC/SiOl interface and
`SiOJgas interface, respectively; D,, is the diffusivity of oxygen;
`and the K' and K" are constants. Equation (9) simplifies since
`P,,ol >> P$,(), and there is a negative vacancy gradient as
`described by Eq. (76). Thus, k, shows very little dependence on
`external Po,. This is in accordance with the oxidation literature
`for scale growth on materials with this type of oxide defect
`structure.'" The change in Po, dependence of k,, with tempera-
`ture4' is consistent with a transition from molecular diffusion to
`a larger contribution of ionic diffusion. Note also that, with a
`defect structure different from that in Eq. (7a), the Po, depen-
`dence of k, changes.
`(5) CO Diflusion Ourwurd: Some investigators have
`attributed the higher activation energies at higher temperatures
`to a transition to CO-diffusion-outward rate control. K'' Two
`factors tend to discount this. Zheng rt a/.'' have oxidized Sil'C
`and examined the resultant scale by using SIMS. They found no
`carbon gradient, as one would expect if CO diffuses outward
`slowly. Thermodynamic arguments also support the rapid trans-
`port of CO outward. Suppose the reverse is true and CO dif-
`fuses outward slowly. Then Po, at the SiC/SiOz interface must
`be close to I , and, as required by diffusion control. Eq. ( 3 )
`equilibrium is maintained; the Pco has a value of about loz7 bar
`( I bar = 10' Pa) at 1600 K . Extremely high pressure would be
`expected to blow the scale off, but this is not observed. Some
`type of "CO pressure gauge" is needed at the SiCiSiO- , inter-
`face. The formation of bubbles in the scale may provide some
`indication of pressure buildup." However, there are two difti-
`culties with this. First, the precise pressure to form a bubble is
`not clear-it
`depends on local surface tension and a variety of
`other factors. Second, the source of bubbles at high tempera-
`tures may also be the S i c + SiOL reaction,'' as discussed later
`in this paper.
`(C)
`Interfircia1 Rraction: The other possibility for rate
`control is interfacial reaction. Some investigators have reported
`linear reaction rates for the silicon face of single-crystal Sic,
`which is attributed to interfacial reaction
`However,
`most of the data for pure S i c indicate parabolic kinetics for the
`majority of the reaction period. Table IV summarizes the exper-
`imental observations and the rate-controlling step supported.
`Most of the evidence implies oxygen diffusion inward as being
`rate controlling.
`However, the situation is not so clear. The most recent data
`indicate that single-crystal Sic has rates slower than that of
`pure silicon, even with the stoichiometry correction. The
`(0001) silicon face of single-crystal Sic shows rates slower
`than that of the (000T) carbon face." There is no explanation
`for this. Furthermore, the assumption of oxygen diffusion
`inward and CO diffusion outward assumes that one is fast and
`the other is slow. Although CO is polar and molecular oxygen is
`not, it is difficult to imagine that they would have dramatically
`different permeation properties. These three facts make it dif-
`ficult to accept oxygen diffusion inward as completely rate con-
`trolling. Physical phenomena do not always fall into distinct
`categories. As Luthra'X reports, it may be that a mixed control
`mechanism is operative and both diffusion and the interface
`
`'
`
`Gas-to-oxide
`transfer,
`Jg = h (C' - Co)
`
`Diffusion
`in oxide,
`
`Interface
`reaction,
`J3=kCj
`
`J2 = Deff xg
`c0-ci
`Processes involved in silicon oxidation. (After Deal and
`
`Fig. 8.
`Grove. " j
`
`An accumulation of "0 at the SiO,/SiC and SiOjgas interface,
`as determined by secondary ion mass spectrometry (SJMS),
`supports this. Above 1673 K the activation energy becomes
`higher, and this is attributed to a mixture of network exchange
`diffusion and molecular oxygen diffusion. Zheng et ~ 1 . ~ ' ~ ' ~
`have
`observed a constant distribution of "0 through the scale in their
`reoxidation experiment. They attribute this to the exchange
`process. However, Cawley and Boyce4' have reported that,
`without some type of concentration gradient, it is difficult to
`draw conclusions about the diffusion mechanism.
`A transition from molecular oxygen to network oxygen dif-
`fusion should be reflected in the dependence of the parabolic
`rate constant k, on Po,. For molecular oxygen diffusion, the rate
`constant should be pioportional to Po-, assuming that the solu-
`bility of oxygen in SiO, obeys Henry's law. For network
`exchange diffusion, one expects k, to show a weak dependence
`on P0,.4x.s0 Zheng et
`have found k, to be dependent on
`(P,2y, where n varies from 0.6 at 1473 K to 0.3 at 1773 K. This
`variation is generally consistent with the proposed transition
`between the two types of diffusion. At lower temperatures one
`might expect a value of n closer to I for molecular oxygen dif-
`fusion. Zheng et al. attribute the difference to deviations from
`Henry's law for the solubility of oxygen in SO2. Narushima
`et al." have found k, to be proportional to (Po2)''2 at 1823 to
`1948 K. From the high activation energy, Narushima et al. con-
`clude that network exchange diffusion of oxygen inward is rate
`controlling.
`In the case of network exchange diffusion, the expected Po,
`dependence is important to explore further. Using standard
`Kroger-Vink notation and assuming that oxygen vacancies are
`the primary defect,
`0, = V;; + 2e' + 0.502
`[V;] = K' Pi,'"
`Do = DZPG2""
`According to the Wagner oxidation theory'" for ionic diffusion,
`
`(70)
`
`GE-1019.007
`
`

`
`JournuI of the American Ceramic Society-Jucobson
`
`Vol. 76, No. I
`
`20
`
`40
`
`60
`Time, hr
`
`80
`
`100
`
`1 20
`
`0
`
`.I6 -
`
`Kp = 1 .3x104 mg2/cm4h
`
`0
`
`20
`
`40
`
`60
`Time, h
`Sample oxidation curves for CVD Sic in dry oxygen (7 = I673 K)?' (a) weight gain versus time ( I mgicm' = 10' kgim') and (b) (weight
`Fig. 9.
`gain)2 versus time ( 1 mg'/(cm'.h) = 3.6 X lo7 kg'/(m''.s)).
`
`80
`
`100
`
`120
`
`reaction are rate controlling. Because of t