S. gi Joumal of the European Ceramic Society 29(2009 )539-550 film widths in the range 0.6-0.9nm. The creep deforma and accumulated in grain boundaries under tension. 3 The mea- process, did, hence, not have any pronounced effect on surements of intergranular film widths presented in this paper distribution of the intergranular glass do not, however, show any clear evidence of glass redistribution during these creep tests. The thickness of the investigated grain 5. Discussion boundary films was in the range 0.6-0.9 nm in both as-sintered and crept specimens, and a possible redistribution might, there- 5.1. The polycrystalline mullite fore, be difficult to detect. The virtually unchanged low dislocation densities in the 5.2. The mullite/SiC nanocomposite crept mullite specimens strongly suggest that dislocation glide was not active to any significant extent under the applied The addition of Sic nanoparticles resulted in a reduced testing conditions. This is in agreement with previous exper- mullite grain size (Table 1). This indicates that the nanopar- imental studies which indicated a very limited (if any at all) ticles suppressed grain growth during sintering through grain dislocation mobility during plastic deformation of both single boundary pinning. The SiC particles located at the grain bound crystal and polycrystalline mullite specimens. 10, 12, 13,28A previ- aries were larger(30-90 nm)than the intragranular particles ous TEM investigation of dislocations in mullite by Gustafsson (10-50 nm). A critical Sic particle size for effective mullite and Falk29,30 revealed comparatively large Burgers vectors of grain boundary pinning would, hence, be in the range 30-50nm the type b=<100>, <00>, <1 lo> and <l l 2>. This may, during hot pressing under a pressure of 40 MPa at 1600C together with the complex mullite crystal structure, result in a Strain contours were generally not observed in the mullite limited dislocation activity. matrix around the Sic particles in the as-sintered specimen. This The changes observed in the microstructures of the crept mul- indicates that the thermal expansion mismatch between mul- lite specimens indicate that lattice diffusion is an active creep lite(a=5.3 x 10-6oC-)6 and SiC (a=4. 7 x 10-6oC-)7is deformation mechanism, both at 1300 and 1400C. The elon- too low for the introduction of compressive residual stresses gation and enlargement of some intragranular cavities, and the of any significant magnitude at the SiC/mullite interface. In development of thin intragranular cavity channels, indicate dif- addition, the analysed Sic/mullite interfaces contained amor fusion activity within the grains, and thus a contribution from phous films or pockets(Fig 9). The SiC/mullite interfaces are, Nabarro-Herring creep. In addition, the presence of a continuous hence, not likely to be significantly more rigid than the glass intergranular glassy phase at the grain boundaries(Figs. 4 and 5) containing mullite/mullite grain boundaries. This is in contrast would provide rapid diffusion paths and thereby promote creep to SiC reinforced alumina nanocomposites where the thermal deformation by grain boundary diffusion. The intergranular expansion mismatch between alumina(a=88x 10-6oC-)7 glass may, hence, contribute to an increased creep rate as com- and Sic puts the particle/matrix interface under compression pared to the model discussed in Section 2, see Fig. la during cooling from the sintering temperature. It has been pro- Creep deformation by diffusion, either through the grains or posed that this leads to a more rigid interface bonding and through the grain boundary glass, would give a stress exponent thereby to a reduced creep rate due to the suppression of of n=l. This is in good agreement with the stress exponent of the nucleation and annihilation of point defects during creep n=1. 2 that was determined from the creep tests performed at deformation. The proposed rigid interface bonding was sup- 1300.C8 Solution-reprecipitation creep, limited by the trans- ported by theoretical calculations and TEM observations of glass port of matter through the glassy grain boundary films, is also free SiC/alumina interfaces. This interface mechanism would, expected to give a stress exponent of n=l. There is, however, hence, not be active to any significant extent in the present no microstructural evidence for such a process mulliteSiC nanocomposite material Diffusion creep only would, however, not result in the The fraction of Sic particles located at mullite grain bound stress exponent n=2 determined from the creep tests carried aries and multi-grain junctions was unchanged(80%)in the out at 1400C.A large number of cavities had formed at specimen that had been creep tested under a stress of 50.0 MPa multi-grain junctions in these specimens(Fig. 11), and the sur- at 1400C. This suggests that moving grain boundaries dragged rounding grains showed strain contrast(Figs. 10 and 11). This the SiC particles during creep deformation. This process requires was most pronounced in the specimen that had been tested at a directional flow of atoms from one side of the particle to the 486MPa. These observations suggest that rigid grain bound- other. 3 The diffusional migration may occur along three dif- ary sliding, facilitated by softening of the intergranular glass, ferent paths: through the SiC particle, in the thin glassy film contributed to the strain during creep testing. Such a process separating the SiC particle from the mullite matrix or around the values of n>1. A stress exponent close to n=2, and evidence interfacial reactions, may be rate controlling 33nprocesses, or would increase the stress exponent, since cavitation alone gives particle through the matrix. One of these diffusion processes,or of cavitation and grain boundary sliding, have been observed A model proposed by Clegg and co-workers 8, suggests that previously in other creep studies of glass containing mullite the creep rate would be limited by self-diffusion through the low diffusivity SiC particles. It was assumed that the particles would It may be expected that the amorphous grain boundary phase move with a velocity equal to that of the grain boundary at which would redistribute during rigid grain sliding because the glass they are situated. It was also assumed that the global defor- may be squeezed out from grain boundaries under compression mation of the body would be caused by self-diffusion within548 S. Gustafsson et al. / Journal of the European Ceramic Society 29 (2009) 539–550 film widths in the range 0.6–0.9 nm. The creep deformation process, did, hence, not have any pronounced effect on the distribution of the intergranular glass. 5. Discussion 5.1. The polycrystalline mullite The virtually unchanged low dislocation densities in the crept mullite specimens strongly suggest that dislocation glide was not active to any significant extent under the applied testing conditions. This is in agreement with previous exper￾imental studies which indicated a very limited (if any at all) dislocation mobility during plastic deformation of both single crystal and polycrystalline mullite specimens.10,12,13,28 A previ￾ous TEM investigation of dislocations in mullite by Gustafsson and Falk29,30 revealed comparatively large Burgers vectors of the type b = <1 0 0>, <0 1 0>, <1 1 0> and <1 1 2>. This may, together with the complex mullite crystal structure, result in a limited dislocation activity.12 The changes observed in the microstructures of the crept mul￾lite specimens indicate that lattice diffusion is an active creep deformation mechanism, both at 1300 and 1400 ◦C. The elon￾gation and enlargement of some intragranular cavities, and the development of thin intragranular cavity channels, indicate dif￾fusion activity within the grains, and thus a contribution from Nabarro-Herring creep. In addition, the presence of a continuous intergranular glassy phase at the grain boundaries (Figs. 4 and 5) would provide rapid diffusion paths and thereby promote creep deformation by grain boundary diffusion. The intergranular glass may, hence, contribute to an increased creep rate as com￾pared to the model discussed in Section 2, see Fig. 1a. Creep deformation by diffusion, either through the grains or through the grain boundary glass, would give a stress exponent of n = 1. This is in good agreement with the stress exponent of n = 1.2 that was determined from the creep tests performed at 1300 ◦C.18 Solution-reprecipitation creep, limited by the trans￾port of matter through the glassy grain boundary films, is also expected to give a stress exponent of n = 1. There is, however, no microstructural evidence for such a process. Diffusion creep only would, however, not result in the stress exponent n = 2 determined from the creep tests carried out at 1400 ◦C.18 A large number of cavities had formed at multi-grain junctions in these specimens (Fig. 11), and the sur￾rounding grains showed strain contrast (Figs. 10 and 11). This was most pronounced in the specimen that had been tested at 48.6 MPa. These observations suggest that rigid grain bound￾ary sliding, facilitated by softening of the intergranular glass, contributed to the strain during creep testing. Such a process would increase the stress exponent, since cavitation alone gives values of n > 1.31 A stress exponent close to n = 2, and evidence of cavitation and grain boundary sliding, have been observed previously in other creep studies of glass containing mullite ceramics.10,12,13 It may be expected that the amorphous grain boundary phase would redistribute during rigid grain sliding because the glass may be squeezed out from grain boundaries under compression and accumulated in grain boundaries under tension.32 The mea￾surements of intergranular film widths presented in this paper do not, however, show any clear evidence of glass redistribution during these creep tests. The thickness of the investigated grain boundary films was in the range 0.6–0.9 nm in both as-sintered and crept specimens, and a possible redistribution might, there￾fore, be difficult to detect. 5.2. The mullite/SiC nanocomposite The addition of SiC nanoparticles resulted in a reduced mullite grain size (Table 1). This indicates that the nanopar￾ticles suppressed grain growth during sintering through grain boundary pinning. The SiC particles located at the grain bound￾aries were larger (30–90 nm) than the intragranular particles (10–50 nm). A critical SiC particle size for effective mullite grain boundary pinning would, hence, be in the range 30–50 nm during hot pressing under a pressure of 40 MPa at 1600 ◦C. Strain contours were generally not observed in the mullite matrix around the SiC particles in the as-sintered specimen. This indicates that the thermal expansion mismatch between mul￾lite (α = 5.3 × 10−6 ◦C−1) 16 and SiC (α = 4.7 × 10−6 ◦C−1) 17 is too low for the introduction of compressive residual stresses of any significant magnitude at the SiC/mullite interface. In addition, the analysed SiC/mullite interfaces contained amor￾phous films or pockets (Fig. 9). The SiC/mullite interfaces are, hence, not likely to be significantly more rigid than the glass containing mullite/mullite grain boundaries. This is in contrast to SiC reinforced alumina nanocomposites where the thermal expansion mismatch between alumina (α = 8.8 × 10−6 ◦C−1) 17 and SiC puts the particle/matrix interface under compression during cooling from the sintering temperature. It has been pro￾posed that this leads to a more rigid interface bonding and thereby to a reduced creep rate due to the suppression of the nucleation and annihilation of point defects during creep deformation.5 The proposed rigid interface bonding was sup￾ported by theoretical calculations and TEM observations of glass free SiC/alumina interfaces.5 This interface mechanism would, hence, not be active to any significant extent in the present mullite/SiC nanocomposite material. The fraction of SiC particles located at mullite grain bound￾aries and multi-grain junctions was unchanged (80%) in the specimen that had been creep tested under a stress of 50.0 MPa at 1400 ◦C. This suggests that moving grain boundaries dragged the SiC particles during creep deformation. This process requires a directional flow of atoms from one side of the particle to the other.33 The diffusional migration may occur along three dif￾ferent paths: through the SiC particle, in the thin glassy film separating the SiC particle from the mullite matrix or around the particle through the matrix. One of these diffusion processes, or interfacial reactions, may be rate controlling.33 A model proposed by Clegg and co-workers18,19 suggests that the creep rate would be limited by self-diffusion through the low diffusivity SiC particles. It was assumed that the particles would move with a velocity equal to that of the grain boundary at which they are situated. It was also assumed that the global defor￾mation of the body would be caused by self-diffusion within