1. Introduction
Environmental barrier coatings (EBCs) can play a key role in enabling SiC fiber-reinforced SiC matrix (SiC/SiC) composites to be employed as advanced hot-section components in airplane engines, as a means of realizing exceptional fuel efficiency. EBCs must exhibit excellent oxygen/water vapor shielding characteristics and thermomechanical durability in severe combustion environments [
1,
2,
3,
4,
5,
6,
7]. Thus, a multilayer structure is generally adopted, with each layer having its own unique properties. Naturally, EBCs are exposed to a large oxygen potential gradient (dµ
O) at elevated temperatures. This gradient results in the inward diffusion of oxide ions and outward diffusion of cations, according to the Gibbs-Duhem relationship. Cation transport through an oxide layer typically induces decomposition of the oxide, and can thus cause the EBC structure to collapse. Therefore, the development of a robust EBC with excellent gas shielding characteristics requires both a good understanding of, and control over, mass transfer within the EBC. However, even though the performance of an EBC is greatly affected by both its constituent materials and its microstructure depending on process conditions, there is a crucial lack of knowledge regarding the environmental shielding properties of these materials. Consequently, studies intended to improve EBC performance tend to be based on trial-and-error.
In prior work, our group elucidated the mass transfer processes within the constituent oxides of EBCs by determining the oxygen permeability values of wafers cut from sintered bodies serving as model EBCs [
8,
9,
10,
11,
12,
13]. This technique is believed to accurately evaluate the environmental shielding characteristics of EBC materials under steady state conditions. This is possible because both the dµ
O applied to the wafers and the diffusion length are kept constant. In this technique, the upper and lower surfaces of the wafer are exposed to gases with different oxygen partial pressures (
PO₂) such that the dissociative adsorption of oxygen molecules will proceed at the surface having the higher
PO₂ (the
PO₂(hi) surface), while the reverse reaction progresses on the opposite surface having a lower
PO₂ (the
PO₂(lo) surface). Therefore, the mass transfer mechanisms through the oxides comprising the EBC can be monitored in a facile manner.
Mullite, which is often given the formula Al
4+2xSi
2−2xO
10−x, is regarded as a candidate material for multilayered EBCs because its thermal expansion coefficient is close to that of SiC and it is chemically compatible with Si-based ceramics [
5]. The oxygen shielding properties of polycrystalline mullite, in which
x = 0.25 (corresponding to 3Al
2O
3·2SiO
2 and referred to as so-called “stoichiometric” or stable mullite [
14]) are evidently superior to those of Yb
2Si
2O
7 and Yb
2SiO
5 [
10,
11,
12], although the resistance of mullite to volatilization by water vapor is far inferior to that of the Yb-silicates [
15]. Oxygen permeation through these complex oxides is controlled by the inward diffusion of oxide ions and the outward diffusion of cations (
M), such as Al ions in the case of mullite and Yb ions in the case of Yb-silicates, through grain boundaries (GBs) [
10,
11,
12]. The mobility of Si ions, which comprise the majority of the complex oxides, is evidently too low to contribute to oxygen permeation. Thus, going from the SiC/SiC substrate to the top coat, an EBC typically consists of a Si-based non-oxide bond coat, a mullite layer and Yb-silicate layers. This configuration improves both the oxygen shielding properties and water vapor volatilization resistance of the overall EBC system.
Using the oxygen permeation data resulting from prior work, the oxygen permeability constants of each oxide layer constituting a multi-layered EBC, as well as the contributions of oxide ions and cations to oxygen permeation through the oxide, mass transfer parameters such as chemical potentials, GB diffusion coefficients and fluxes of both ions, can all be calculated at arbitrary temperatures and dµ
O values [
8,
9,
10,
11,
12]. The ability to determine mass transfer through oxides provides useful structural information regarding layer arrangement, layer thickness and grain size that can assist in designing EBCs with excellent oxygen shielding characteristics together with phase stability.
When a mullite layer with superior oxygen shielding characteristics is applied to a conventional bond coat made of Si, the mullite is exposed to an extremely large dµ
O as part of the overall EBC system. The subsequent mass transfer of species diffusing through the mullite is expected to greatly affect the structural stability of the EBC. The present study assessed strategies to improve the oxygen shielding properties and phase stability of a mullite film applied to a SiC/SiC substrate through a Si-based non-oxide bond coat, using previously acquired oxygen permeation data [
10] for mullite at high temperatures. The validity of these strategies was verified via experimental work at 1673 K using bilayer specimens consisting of a mullite film and a bond coat substrate, serving as model EBCs.
2. Design to Improve the Oxygen Shielding and Phase Stability of Mullite
In this study, the oxygen shielding obtained from a mullite film formed on a bond coat substrate was examined, employing a simplified EBC system. As described in the Introduction, the oxidation of a Si-based bond coat proceeds via the inward GB diffusion of oxide ions and the outward GB diffusion of Al ions through the exposed upper mullite film following the application of a large dµO at high temperatures. The transport of Al ions reduces the phase stability of the mullite in the vicinity of the junction interface between the mullite film and the underlying bond coat. Therefore, it would be beneficial to design a bond coat capable of supplying Al ions to the adjacent region in the upper mullite film that is otherwise depleted of Al.
β’-SiAlON, having the formula Si
6−zAl
zO
zN
8−z, is a particularly promising candidate for bond coat applications because it provides superior heat resistance compared to Si and its thermal expansion coefficient is close to that of SiC/SiC [
16]. Moreover, β’-SiAlON can dissolve significant quantities of Al, and so could serve to supply Al to the mullite film. In addition, even though the concentration of Al in the β’-SiAlON would be decreased during this process, the β’-SiAlON is able to transition to other oxynitrides during this process, such as O’-SiAlON and the X phase. This continuous variation in phase structure means that the β’-SiAlON is less likely to delay formation of SiO
2 and subsequent crystallization such as cristobalite [
17,
18]. It is well known that phase transition of cristobalite from β-type to α-type at approximately 540 K during cooling leads to significant shrinkage (3.9 vol.%) [
19], resulting in initiation of microcracks at the interfaces and/or in the cristobalite layer. Furthermore, in the case of Si
2N
2O that is a unit structure of O’-SiAlON having the formula Si
2−xAl
xO
1+xN
2, where the
x-value varies from zero to ~0.2, the thermodynamic equilibrium
PO₂ values, in which Si
2N
2O and SiO
2 coexist, are larger than those for Si and SiC [
20,
21]. Therefore, it is expected that the level of oxygen shielding required for oxide films to protect oxidation of underlying SiAlON-based ceramics can be lower than that for Si and SiC. Thus, incorporating a SiAlON bond coat is advantageous for maintaining long-term adhesion between the mullite layer and the bond coat. In the work reported herein, the effects of the bond coat material on the oxygen shielding and phase stability of the mullite film are discussed, based on comparing β’-SiAlON and Si bond coats.
When a mullite film is exposed to a dµ
O at a high temperature, such that each surface of the film is subjected to a different
PO₂, and
PO₂(hi) >>
PO₂(lo), the oxygen permeability constant,
PL, per unit GB length can be expressed as Equation (1) [
10]:
where
P is the oxygen permeability and
L is the wafer thickness. The oxygen permeability term,
PL, normalized by the GB density,
Sgb, is dependent only on the GB characteristics and is unaffected by the grain size of the polycrystalline mullite. The experimental constants
AAl and
AO are related to the adsorption reactions of oxygen molecules that occur on the
PO₂(hi) surface and subsequent movement of Al ions and oxide ions during the oxygen permeation.
AAl and
AO normalized by
Sgb are given by Equations (2) and (3), respectively [
10]:
and
where
R is the gas constant, and
T is the absolute temperature. Thus, the former and latter terms enclosed in parentheses in Equation (1) are related to the movement of Al ions and oxide ions, respectively.In addition, the
PO₂(lo) term in Equation (1) is assumed to equal the minimum equilibrium partial pressure (
PO₂,eq) required to form SiO
2 via the oxidation of Si or β’-Si
3Al
3O
3N
5. The
PO₂,eq values associated with this work were calculated on a thermodynamic basis using the Fact-Sage free energy minimization computer code in conjunction with various databases, including FactPS, FToxide and FTOxCN. In the case of β’-Si
3Al
3O
3N
5, α-Al
2O
3 is formed by the partial oxidation of β’-SiAlON at an equilibrium
PO₂ value below
PO₂,eq. The ongoing oxidation process eventually degrades the various oxynitride products such as O’-SiAlON and the X phase until only SiO
2 and mullite remain. In this study, the
PO₂,eq required to form SiO
2, which would be expected to seriously damage the adhesion of the mullite film, was examined on this basis.
Figure 1 plots the
PO₂,eq values for the oxidation of the bond coat materials Si and β’-Si
3Al
3O
3N
5 as functions of temperature. The line for Si in
Figure 1 corresponds to the oxygen equilibrium partial pressure at the phase boundary between Si and SiO
2 and is in agreement with that in the Si-O volatility diagram [
20]. In the case of SiAlON, the upper side corresponds to the stable region of SiO
2, while the lower side is equivalent to the region where any SiAlONs such as the β’, O’, and X phases exist. It is evident that the
PO₂,eq values for both materials increase with increasing temperature, and that the values for the β’-Si
3Al
3O
3N
5 are larger than those for Si over the entire temperature range. This indicates that applying β’-Si
3Al
3O
3N
5 can lower the oxygen shielding performance required for the upper oxide film compared to Si. The oxygen permeability constants of mullite exposed to various dµ
O were calculated using Equations (1)–(3) with the
PO₂,eq values in
Figure 1.
In the case of a mullite film on a Si bond coat (
Figure 2a) with the mullite surface exposed to a
PO₂(hi) of 10
5 Pa, the plot showing the oxygen permeability constants related to the diffusions of both oxide ions and Al ions almost coincides with the line solely for the diffusion of oxide ions. Therefore, the oxygen permeation along the GBs in the mullite film is controlled by the inward GB diffusion of oxide ions over the entire temperature range. Using the same bilayer structure but a lower
PO₂(hi) of 1 Pa produced the data shown in
Figure 2b, for which the total oxygen permeability constants and the contribution of the diffusion of oxide ions are almost equal to those in
Figure 2a. However, the line associated with the diffusion of Al ions in
Figure 2b drops to about one-tenth of that in
Figure 2a. When the value of the
PO₂(hi) term that appears in Equation (1) was reduced to less than 10
2–10
3 times the
PO₂,eq, there was an evident decrease in the line for the diffusion of oxide ions. Thus, coating the mullite layer with a film made of an oxide such as an Yb-silicate [
11,
12], which provides inferior oxygen shielding, improves the phase stability of the mullite film but does not increase the overall oxygen shielding characteristics of the EBC. In the case of the mullite film on the β’-Si
3Al
3O
3N
5 bond coat with a
PO₂(hi) of 10
5 Pa, as shown in
Figure 2c, the oxygen permeability constant is clearly decreased as a result of the significantly lower contribution of the diffusion of oxide ions. Specifically, the oxygen permeability constant due to the inward diffusion of oxide ions at 1673 K in
Figure 2c was about one-fifth of the values for
Figure 2a,b. It should be noted also that the contribution of the diffusion of Al ions remains much the same as in
Figure 2a. The increase in the
PO₂,eq values improves the oxygen shielding properties but not the phase stability of the mullite film.
Together, these results demonstrate that the oxygen permeation and the contribution of the diffusion of each ion through the mullite film are both significantly affected by the dµ
O value. This occurs because the GB diffusion coefficients of Al ions and oxide ions change greatly as the oxygen chemical potential (µ
O) in the thickness direction of the mullite film varies. That is, the GB diffusion coefficients of Al ions are much larger than those of oxide ions in regions associated with high µ
O values close to the
PO₂(hi) surface, while the reverse is true in regions with low µ
O values close to the
PO₂(lo) surface [
10]. Therefore, an EBC structure comprising a sandwich-type configuration in which the mullite film is between a SiAlON bond coat and an oxide film providing a moderate degree of oxygen shielding (such as an Yb-silicate [
11,
12]) is expected to be especially effective. This structure should simultaneously improve the oxygen shielding properties and phase stability of the mullite film.
4. Results and Discussion
The sample composed of a mullite film on a Si disk showed complete bonding of the mullite to the underlying Si disk in the thickness direction prior to heating, as shown in
Figure 4a. No damage was observed in the vicinity of the joining interfaces in bright-field STEM images or in the corresponding EDS elemental maps of the cross-section around the interface. In contrast, after the heat treatment, damage to the mullite film near the interface was evident (
Figure 4b). It is noteworthy that there was no degradation to the thicker 5 mm mullite layer at the opposite interface, which retained the same microstructure as before the heat treatment. Thus, the observed damage to the mullite is attributed to the increased driving force for the GB diffusion of Al and oxide ions resulting from the reduced film thickness. The uneven thickness of the damaged region in
Figure 4b is likely related to variations in the GB density of the mullite film adjacent to the interface, because the GBs in the mullite provide paths for the rapid diffusion of ions.
Figure 5 presents a bright-field STEM image and the corresponding EDS elemental maps of the cross-section around the bonding interface of the annealed sample shown in
Figure 4b. Because Al ions diffuse along the GBs in the mullite film in response to the application of the dµ
O, both SiO
2-rich mullite and SiO
2 are expected to be formed by the decomposition of mullite near the interface. However, the damaged region shown in
Figure 4b instead was found to consist of Al
2O
3 and Si. A Si-O phase was present only in areas enclosed by the mullite grains and located less than one micrometer from the interface. In addition, microstructural changes were not observed in the Si disk near the interface.
Figure 6 presents a plot of the free energy curve for mullite in a SiO
2-Al
2O
3 system at 1673 K. Based on this plot, the concentration of Al
2O
3 in the mullite near the interface should change from an initial value, x
i, to a new value on the SiO
2-rich side of the plot, as a result of the outward GB diffusion of Al ions. When the Al
2O
3 concentration reaches a critical value, x
cr, a portion of the mullite will break down to produce both SiO
2-rich mullite and SiO
2, corresponding to the Si–O phase seen in
Figure 5. In this study, the specimens were fabricated by joining parts made of mullite and Si using hot pressing with a carbon heater, and so the µ
O for oxygen dissolved in the Si disk was likely extremely low. Consequently, the SiO
2 formed at GBs of the mullite layer in the vicinity of the bonding interface due to the outward GB diffusion of Al ions was considered to be easily reduced to produce Si particles and oxygen that immediately dissolved in the Si substrate. Thus, voids under negative pressure were thought to be formed adjacent to the Si particles derived from the mullite layer. When mullite is exposed to a total pressure of 10
−2 Pa or less at 1673 K, the decomposition reaction to form Al
2O
3 and SiO gas proceeds thermodynamically. Therefore, the mullite located around the voids was probably decomposed, resulting in generation of the Al
2O
3 particles. The voids might be also filled with Si that migrated from the Si disk, since the exposure temperature was near the melting point (1687 K) of Si. On the other hand, although the amount of oxygen permeated through the mullite layer calculated from Equations (1)–(3) was much larger than the value corresponding to the solid solubility limit of oxygen in Si [
22], obvious SiO
2 scale formation due to oxidation of the Si disk was not observed at the interface. Oxide ions diffused inward along the GBs in the mullite layer might have been captured by oxygen vacancies in the mullite layer before reaching the interface between the mullite layer and the Si disk, in which a large amount of the vacancies were created around the GBs in the mullite layer during sample preparation in the reducing atmosphere.
In the case of an actual EBC system in which a mullite layer is formed on a Si bond coat using an atmospheric plasma spray process, some ambient oxygen will dissolve in the lower Si bond coat during the high temperature coating, resulting in a sufficiently high µO in both the bond coat and mullite film compared with the specimens employed in this study. Therefore, a large amount of Si–O phase will be formed by decomposition of the mullite film due to the outward diffusion of Al ions and also by the oxidation of Si due to the inward diffusion of oxide ions, and will be present near the interface in the absence of oxide reduction.
Figure 7 provides TEM images of cross-sections around the bonding interface of a sample in which a 0.25 mm thick mullite layer was applied overtop of a SiAlON disk, before and after annealing in oxygen at 1673 K for 10 h.
Figure 8 shows a bright-field STEM image and the corresponding EDS elemental maps of a cross-section around the bonding interface of the annealed sample shown in
Figure 7b. The microstructures around the interfaces before and after the heat treatment are seen to be similar, and there is no damage to the mullite film resulting from the interdiffusion of Al and oxide ions. This result is believed to be related to the reduced inward GB diffusion of oxide ions due to the increased µ
O value in the vicinity of the interface and to the improved phase stability of the mullite film due to the Al supply from the underlying SiAlON.