Compositional variations for small-scale gamma prime (c0) precipitates formed at different cooling rates in an advanced Ni-based superalloy

Size-dependent compositional variations under different cooling regimes have been investigated for ordered L12-structured gamma prime (c0) precipitates in the commercial powder metallurgy Ni-based superalloy RR1000. Using scanning transmission electron microscope imaging combined with absorption-corrected energy-dispersive X-ray spectroscopy, we have discovered large differences in the Al, Ti and Co compositions for c0 precipitates in the size range 10–300 nm. Our experimental results, coupled with complementary thermodynamic calculations, demonstrate the importance of kinetic factors on precipitate composition in Ni-based superalloys. In particular, these results provide new evidence for the role of elemental diffusion kinetics and aluminium antisite atoms on the low-temperature growth kinetics of fine-scale c0 precipitates. Our findings have important implications for understanding the microstructure and precipitation behaviour of Ni-based superalloys, suggesting a transition in the mechanism of vacancy-mediated diffusion of Al from intrasublattice exchange at high temperatures to intersublattice antisite-assisted exchange at low temperatures. 2014 Acta Materialia Inc. Published by Elsevier Ltd. This is an open access article under the CCBY license (http://creativecommons.org/licenses/by/


Introduction
Polycrystalline nickel-based superalloys for turbine disc applications typically employ complex alloy chemistry in order to produce the required properties. Excellent fatigue performance and damage tolerance, and good creep resistance at operating temperatures close to 1023 K [1][2][3] originate principally from the presence of a high volume fraction (close to 50%) of gamma-prime (c 0 ) precipitates coherently embedded within the gamma (c) matrix [4,5]. The precipitate size distribution (PSD) for the c 0 phase can be varied in a controlled manner by applying different heat treatments, allowing the mechanical properties of the material to be optimised for certain applications [6]. Changes in the PSD and/or phase chemistry have a direct impact on material performance, so controlling the mechanical properties requires an accurate understanding of precipitate phase chemistry and its evolution during cooling.
Primary c 0 precipitates are intergranular and only form when the superalloy is heat treated below the c-solvus temperature. Supersolvus annealing above the c-solvus dissolves the primary c 0 and subsequent solution cooling, at rates of around 10 K min À1 or less, causes the formation of intragranular c precipitates during the early stage of cooling, commonly termed secondary c. Further cooling results in the growth of secondary c 0 precipitates until the low elemental diffusivities of the c stabilizing elements make it difficult for these elements to reach the comparatively coarse (hundreds of nanometres) secondary c. This results in supersaturation of these elements within the c matrix and consequently drives the nucleation of additional intragranular c. In the remaining cooling range these new precipitates only grow to sizes in the range of tens of nanometres, and are often termed tertiary c 0 [5,7].
In contrast, fast solution cooling results in higher nucleation rates and limits the time available for diffusion to occur. This produces microstructures with smaller c precipitates and a narrower size distribution. As summarised in Fig. 1, the c stabilizing elements have different interdiffusion rates in the matrix and in the c precipitates [8][9][10][11][12]. Considering the complexity of c formation, which occurs over a wide range of temperatures under different cooling regimes, the phase chemistry of chas the potential to be highly size dependent, as demonstrated using three-dimensional atom probe microscopy [13] and in our recent work [14]. What is not yet well understood are details of the diffusion kinetics and how different cooling rates affect the local phase chemistry within different sizes of c 0 precipitate.
Three-dimensional atom probe microscopy was the first technique used that was able to quantify the phase chemistry of c 0 precipitates [13]. Energy-dispersive X-ray (EDX) spectroscopy within a scanning transmission electron microscope (STEM) provides an alternative approach to measuring the phase chemistry within thin samples, as well as providing complementary information from electron diffraction [14][15][16][17]. Traditional electropolished transmission electron microscope (TEM) thin film samples facilitate accurate EDX compositional measurements for precipitates with diameters approaching the foil thickness ($100 nm). However, for smaller sizes of precipitate, it is difficult to distinguish the composition of the precipitate from that of the surrounding matrix. To overcome this problem, an alternative technique for TEM sample preparation was demonstrated by Mukherji et al. [18], who employed an electrochemical method to extract individual c 0 precipitates by dissolving the surrounding c matrix. More recently, we have demonstrated the use of this technique coupled with absorption-corrected EDX spectroscopy to semi-quantitatively analyse precipitate compositions [14]. In this paper, we have applied this same approach to study the size-dependent compositional variations of c 0 precipitates formed by employing different cooling regimes. Our results reveal new insights into the role of diffusion kinetics for determining precipitation mechanisms in the commercial powder metallurgy (PM) Ni-based superalloy RR1000.

Materials and heat treatment
The commercial PM Ni-based superalloy RR1000 studied in the present work has the nominal composition shown in Table 1 and is used for disc applications in aero engines. This polycrystalline Ni-based superalloy was manufactured via the powder metallurgical route followed by subsequent forging, and typically has trimodal c 0 PSD [14]. In this work, blanks of size 10 Â 10 Â 10 mm 3 were first heat treated for 2 h at a supersolvus temperature (20 K above the c 0 -solvus) in order to homogenise the microstructure and chemistry of the material. These blanks were then subjected to a super-c-solvus heat treatment at 1453 K followed by cooling to room temperature using a range of controlled cooling rates, both slow (1 and 10 Kmin À1 )and fast (100 and 360 Kmin À1 ), to produce different PSDs with which to study size-dependent precipitation behaviour. In order to control and monitor the heat treatment, equivalent size samples were fabricated and a thermocouple fitted into the centre of the block in order to allow continuous monitoring of the cooling rate.

Analysis methodology of large c precipitates
Samples were mechanically polished using standard metallography and finished with a final polish using colloidal silica solution for 30 min in order to obtain the required surface quality. To reveal the different microstructures, specimens were etched using a two-part etchant to dissolve the c 0 phase but retain the matrix [19]. High-resolution scanning electron microscope (SEM) imaging was  25 and Ni 74 Al 26 , respectively. All data has been extended from 6.7 Â 10 À4 K À1 (1500 K) to 13 Â 10 À4 K À1 (770 K) using Arrhenius curves, assuming a dilute solid solution. The data was obtained by summarising results from Refs. [8][9][10][11][12]. Current diffusivity data for Ta and Hf in c 0 was only available at 1470 K, so the temperature dependence of these elements could not be predicted. performed using an in-lens detector fitted into an FEI Sirion field-emission gun SEM. Appropriate image thresholding procedures were applied within the ImageJ software in order to characterise c 0 precipitates within a large number of SEM images and thereby quantify the precipitate volume fraction and the PSD in the size range 20-300 nm at different cooling rates [14]. The precipitate size is simply defined throughout this work as the square root of the measured visible area. For transmission electron microscope (TEM) imaging, electron transparent foils were prepared by cutting and grinding the material to a thickness of 100-150 lm using standard metallographic methods. Discs of 3 mm diameter were then punched out and electropolished using a Struers Tenupol twin-jet electropolisher with 8% perchloric acid in acetic acid at room temperature and 28V [20]. A small objective aperture was used to select the {1 0 0} superlattice reflection in order to reveal the ordered c 0 precipitates as bright regions in dark-field TEM images.
TEM and STEM imaging were performed using an FEI Tecnai F30 S/TEM at an accelerating voltage of 300 kV. High-angle annular dark field (HAADF) imaging was performed with a convergence semi-angle of 12.5 mrad and a HAADF inner angle of 30 mrad.

Analysis methodology of fine c 0 precipitates
The thin foil samples described above were suitable for EDX spectroscopy compositional measurements when precipitate diameters were greater than the foil thickness. However, for smaller precipitates we employed the approach of Mukherji et al. [18], whereby an electrochemical method dissolves the surrounding c matrix in order to extract individual c 0 precipitates. Once extracted, the precipitates are prepared for S/TEM imaging by being drop cast from solution onto holey carbon TEM support grids.
Watanabe et al. [15,21] have shown that, for Ni 3 Al thin films, the difference in absorption of the AlK a X-rays (1.48 keV) compared to NiK a X-rays (7.47 keV) can lead to EDX compositional measurements that vary by $12% for thicknesses in the range of 80 nm. To compensate for these errors in thin foil samples, they introduce an absorption-correction factor (ACF), which is given as [15,21]: where l/q] A Sp is the mass-absorption coefficient of species A in the specimen, q is the density of the specimen, t is the specimen thickness and a is the detector take-off angle. tcoseca in the exponentials of Eq. (1) can be considered as the equivalent X-ray penetration depth with specimen tilt as the gradually attenuated X-rays travel through a specimen. In order to apply the ACF to samples consisting of isolated spherical particles, we have previously shown that this equation can be modified such that the particle radius R, as the equivalent X-ray penetration depth, is substituted for tcoseca [14], with the new ACF given by: Conveniently, the similarity of these equations means that it is possible to apply the standard ACF implemented in commercial software packages to isolate spherical particles by simply calculating the input value for thickness as R/coseca. In this work we employed an Oxford Instruments X-Max N 80T SDD for EDX spectroscopy with a take-off angle of 20°fitted into an FEI Tecnai F30 S/TEM. EDX spectroscopy measurements with counts of greater than 250,000 were performed using a STEM probe with a convergence semi-angle of 12.5 mrad and a spot size of around 1 nm, so that even for the smallest precipitates, with diameters of less than 20 nm, the interaction volume includes tens of thousands of atoms and the measured compositions can be approximated as homogeneous at this length scale.  Fig. 3. The total volume of c 0 precipitates decreases with increasing cooling rate, although the fraction of precipitates with sizes less than 300 nm increases significantly. Slow cooling rates (1 and 10 Kmin À1 ) are found to show a multi-modal precipitate size distribution, whilst the fast cooling rates (100 and 360 Kmin À1 ) exhibit narrower or unimodal precipitate distributions.

Composition-size relationships for small-scale c 0 precipitates
Previously, we have demonstrated the measurement of size-dependent c 0 compositional variations for the asreceived alloy RR1000 [14]. In this work we follow the same analytical approach for calculating size-dependent elemental compositions as a function of cooling rate for the smallscale c 0 precipitates (less than 300 nm). Our experimental measurements of the different size/cooling dependant compositional behaviours are shown in Fig. 4, together with dark-field TEM images showing the microstructure of the precipitates within electron transparent thin films.
For the two slower cooling rates (1 and 10 Kmin À1 ), the Al content of c 0 is found to be increased in the smallest c 0 precipitates whilst the Ti and Co concentrations are decreased. The smallest c 0 precipitates are likely to be generated at the slowest cooling rate and have the highest Al content (16.7 at.% Al in 20 nm tertiary c 0 for a cooling rate of 1 Kmin À1 , compared to 15.0 at.% Al in 20 nm tertiary c 0 for a cooling rate of 10 Kmin À1 ). The 240 nm secondary c 0 precipitates have lower Al contents: 12.5 and 12.0 at.% Al for cooling rates of 1 and 10 Kmin À1 , respectively. Ti has the opposite size-dependent variation for both the slower cooling rates, increasing from $7.5 at.% in 20 nm tertiary c 0 to $9.6 at.% Ti in 250 nm secondary c 0 . Ta is also slightly depleted in the fine-scale (<80 nm) tertiary c 0 produced by a cooling rate of 1 Kmin À1 , but is approximately constant at $1.9 at.% Ta for larger precipitates and for all sizes of precipitate formed at a cooling rate of 10 Kmin À1 . The Co content is virtually constant at approximately 7.3 at.% for all sizes of precipitate formed at a cooling rate of 1 Kmin À1 , but decreases from 10.5 at.% in $250 nm secondary c 0 to 7.6 at.% in $25 nm tertiary c 0 generated at a cooling rate of 10 Kmin À1 . The contents of both Cr and Mo increase slightly in the smallest precipitates for both the 1 and 10 K min À1 cooling rates, although the variability in both elements is larger in the 10 K min À1 data set. In contrast, the compositional data for the fast cooling rates of 100 and 360 K min À1 (Fig. 4c and d) shows much smaller compositional variations, even for the 100 K min À1 cooling rate, which has a significant size distribution.

Equilibrium calculations
In order to gain a deeper understanding of the experimentally measured compositional variations, equilibrium thermodynamic calculations were performed using Pandat software [22] with data from the Ni-based superalloy database [23] for a temperature range between 873 and 1323 K. The c-solvus (1323 K) can be considered as the upper temperature limit for c nucleation. The lower temperature limit for c 0 precipitation and growth can be estimated by considering the elemental diffusivity of the c 0 stabilizers, Al and Ti, in the c matrix phase (Fig. 1a), together with the minimum precipitate size observed experimentally. The smallest c precipitates observed in this work were $10 nm in terms of square root particle area. Considering the slowest cooling rate of 1 K min À1 and a broad range of possible starting temperatures (803-933 K), the corresponding diffusion distance for Al can be calculated as shown in Fig. 5. This data shows that, for c nucleation temperatures below 873 K, it is not possible for Al and Ti to diffuse more than 10 nm in the c matrix phase during continued cooling to room temperature. Therefore, all observed precipitates are expected to have nucleated above 873 K, and this can be considered as the low-temperature limit for equilibrium thermodynamic calculations.
As shown in Fig. 6, for most elements these calculations predict a thermodynamically stable composition that is largely independent of temperature in the range 873-1323 K, with mean values of 12.3 ± 0.3 at.% Al, 8.7 ± 0.1 at.% Ti, 1.5 ± 0.1 at.% Ta, 0.3 ± 0.1 at.% Hf 2.5 ± 0.2 at.% Cr and 0.6 ± 0.1 at.% Mo. Co is the exception, showing a marked decrease with decreasing temperature, from 11 at.% Cr at 1323 K to 6.0 at.% Cr at 873 K.
In a binary Ni-Al alloy the c 0 phase has the composition Ni 3 Al, with the ordered L1 2 structure, in which the facecentred sublattice (a) is occupied by Ni atoms and the corner sublattice (b) is occupied by Al atoms. For the more complex case of the PM Ni-based superalloy RR1000, the chemistry of the c 0 phase is generally given as (Ni, Co, Cr, Mo) 3 (Al, Ti, Ta, Hf). The elements Ni, Co, Cr and Mo are generally assumed to substitute into the a-sublattice, while Al, Ti, Ta and Hf occupy the b-sublattice, as predicted by first principle calculations [24][25][26][27][28][29]. We note that in this work precipitates in the size range of interest have Hf contents below experimentally detectable levels, hence Hf is not considered in further discussions. The ratio C (Ni, Co, Cr, Mo) /C (Al, Ti, Ta) can be used to estimate the stoichiometry of the precipitate phase chemistry for different sizes of precipitate formed at different cooling rates. As shown in Fig. 7, for all cooling rates, the large ($240 nm) precipitates show mean ratios of 3.41 ± 0.08 (3r), but this decreases for smaller precipitates, with a mean ratio of 2.85 ± 0.30 (3r) for precipitates of less than 70 nm. Our thermodynamic calculations predict an almost constant   superalloy RR1000 cooled at a rate of (a) 1 K min À1 , (b) 10 K min À1 , (c) 100 K min À1 and (d) 360 K min À1 . Bottom: the corresponding sizedependent elemental compositions of c 0 precipitates for each microstructure, calculated using absorption-corrected STEM EDX spectroscopy. The error bars were calculated using the standard deviation of four measurements taken from different regions near the centre of the same precipitate. value of 3.4 for the equilibrium C (Ni, Co, Cr, Mo) /C (Al, Ti, Ta) ratio at all temperatures between 873 and 1323 K. The experimentally measured values for the C (Ni, Co, Cr, Mo) / C (Al, Ti, Ta) ratio are close to this equilibrium prediction for the larger particles in this study but decrease quickly for smaller precipitates, with these smaller precipitates having non-stoichiometric compositions significantly away from equilibrium. Fig. 8 compares mean compositions at different cooling rates for three elements in different size ranges: Al, Ti and Co content was compared for particles of size (a) 0-70 nm, (b) 70-140 nm and (c) 140-300 nm. Note that, for fast cooling rates, the c 0 precipitates only have a limited size range, and consequently it is only possible to show compo-sitions for certain sizes of precipitate (only 70-140 nm at 100 K min À1 and only 0-70 nm for 360 K min À1 ). We note that, for the smallest precipitates, the size range observed in the TEM measurements (as illustrated in Fig. 4) is slightly different to that determined from the SEM images (Fig. 2). In particular, at 100 K min À1 , TEM analysis shows fewer very small precipitates than the SEM measurements (Fig. 3a). However, as the size analysis of the extracted precipitates (Fig. 4) agrees with the unimodal distribution observable by dark-field TEM, this PSD difference is likely to be the result of an artefact in the SEM analysis where chemical inhomogeneity results in unwanted pitting of the matrix being identified as small-scale precipitates.

Discussion
The fast-cooled microstructures show small size distributions and negligible compositional variations, consistent with all precipitates being formed over a small temperature range with a limited time for diffusion. In contrast, clear size-dependent compositional variations are observed at slow cooling rates. Secondary c 0 generated at all cooling rates display compositions that are closer to thermodynamic predictions than tertiary c 0 . For example, at our slowest cooling rate of 1 K min À1 , 240 nm secondary c 0 precipitates have effectively near-field compositions, displaying mean values of 12.5 at.% Al and 8.9 at.% Ti, compared to the equilibrium means of 12.3 ± 0.3 at.% Al and 8.7 ± 0.1 at.% Ti, respectively (Fig. 4a). Tertiary c 0 are farther away from the near-field compositions predicted by equilibrium thermodynamics, having compositions of 16.7 at.% Al and 8.3 at.% Ti for a 1 K min À1 cooling rate. This implies that our experimentally observed compositional variations are unlikely to be caused by differences in the thermodynamically stable composition at different temperatures.
An alternative possible cause of non-equilibrium elemental distributions in fine-scale particles is the Gibbs-Thomson effect [30], where a particle 0 s composition in the   7. The ratio of c-forming elements (Ni, Co, Cr, Mo) to c 0 stabilizers (Al, Ti, Ta) in the PM Ni-based superalloy RR1000 (C (Ni, Co, Cr, Mo) /C (Al, Ti, Ta) ) for different sizes of precipitate in microstructures produced at different cooling rates. Note that Hf is traditionally a c 0 stabilizer but is not included in these calculations as it is below the experimental detection limits for these precipitates. For comparison, the value of C (Ni, Co, Cr, Mo) /C (Al, Ti, Ta) predicted by thermodynamic calculations is 3.4. presence of a curved interface under tension is altered so as to lower the free energy of the system. The free energy resulting from interfacial tension, DG C , is given as [31]: where C is the interfacial tension and r is the precipitate radius. However, in Ni-based superalloys the interfacial tension is exceptionally low (C = 0.01 J mol À1 m À2 , compared to typical values of 1 J mol À1 m À2 in Al-Cu alloys [32]). For the smallest precipitates considered here, r = 5 nm, an interfacial energy, DG C , of 4 Â 10 6 J mol À1 m À3 is calculated when using Eq. (3). Experimentally, we observe an increase of 4.3 at.% Al in the slow-cooled samples, from 12.4 at.% Al in the 250 nm secondary precipitates to 16.7 at.% Al in the 20 nm tertiary ones (Fig. 4a).
Our thermodynamic calculations predict that for RR1000 this approximately equates to a chemical free energy increase of between 4 Â 10 8 and 8 Â 10 8 J mol À1 m À3 . Thus the interfacial free energy is only able to account for $0.5% of the observed increase in Al content for the 10 nm diameter precipitates and therefore the Gibb-Thomson effect is not believed to be responsible for the Al enrichment we observe experimentally for the fine-scale tertiary c 0 precipitates formed on slow cooling. Having determined that neither the interfacial energy nor the chemical free energy of Ni-based superalloys could produce the changes in Al content we observe experimentally in the fine-scale tertiary c 0 at slow cooling rates, it is necessary to next consider the influence of diffusion kinetics. The smallest precipitates (sizes less than 70 nm) are likely to have formed last and at the lowest temperature. These show clear cooling rate dependent compositional variations (Fig. 8a), while precipitates larger than 140 nm (Fig. 8c) were generated at high temperatures and show fairly flat mean compositions for all cooling rates. The temperature for fine-scale c 0 formation and growth decreases with decreasing cooling rate, so we can predict that diffusion effects will have the largest influence on the fine-scale precipitates and the slowest cooling rates. In fact, this is what we observe experimentally, with larger precipitates and fast cooling rates showing equilibrium compositions while small-scale precipitates at slow cooling rates have enhanced Al contents and reduced Ti and Co contents. Compositional variations are smaller for fast compared to slow cooling rates: comparing Fig. 8a and b shows that the increase in Al content for precipitates of 0-70 nm compared to 140-250 nm is 7.3 at.% at 1 K min À1 but 5.3 at.% at 10 K min À1 . With regular intrasublattice diffusion limited growth, tertiary precipitates will have been generated from a supersaturated matrix in which c-forming elements are concentrated. Consequently, tertiary c would be expected to be enriched in c formers such as Co and Cr, while c stabilizers such as Al and Ti are likely to be depleted or have constant compositions compared to larger precipitates. For this intrasublattice diffusion mechanism, slowing the cooling rate can homogenise materials and drive precipitates to closer to the equilibrium composition. This is clearly not consistent with the experimentally observed differences of Al content compared with Ti and this suggests an anomalous diffusion mechanism of Al. The interdiffusion coefficients for different alloying elements in c and c 0 phases shown in Fig. 1 are summarised from Refs. [8][9][10][11][12]. Most alloying elements considered here diffuse significantly more slowly in the c 0 precipitates than in the c matrix. The only exception to this behaviour is aluminium, which has a comparable diffusion coefficient in both c 0 and c across the whole temperature range considered. In particular, at low temperatures, the interdiffusivity of Al in the c 0 phase is comparable to that in the c phase (3.0 Â 10 À19 in c 0 compared to 1.1 Â 10 À18 in c) and is significantly larger than the diffusivities of other elements. The diffusion behaviour of most alloying elements in the ordered L1 2 c 0 phase is explained well by intrasublattice vacancy-mediated diffusion occurring only within the relevant sublattice [10]. For example, Ti and Ta are confined to b-sublattice, while the c formers Cr, Co and Mo are confined to the a-sublattice. All these elements diffuse slowly, especially at low temperatures, as there is a high activation barrier for diffusion via vacancy-atom exchange. Aluminium 0 s anomalously fast diffusion behaviour at low temperatures has recently been explained by intersublattice diffusion, i.e. vacancy-mediated antisite-bridging diffusion [9,10,33]. This allows aluminium to diffuse preferentially within the b-sublattice via the asublattice due to the lower energy barrier compared to the direct vacancy-atom exchange diffusion. This unusual diffusion mechanism is particularly important at low temperature and will enhance the diffusivity of Al in the c 0 precipitates. Gopal and Srinivasan [34] have recently demonstrated that this mechanism provides a good fit to experimental data measured by diffusion couples [35] for the binary Ni 3 Al system and validates the argument of antisite-assisted diffusion by Mishin [36]. Al antisite diffusion results in more complex non-equilibrium kinetic effects as Al is no longer confined to the b-sublattice.
This vacancy-mediated antisite-bridging diffusion of Al will increase Al antisite occupancy at lower temperatures and therefore explains the experimentally observed anomalous enrichment of Al in the smaller precipitates on slow cooling ( Fig. 4a and b). The greater Al enrichment for small precipitates found for the 1 compared to the 10 K min À1 cooling rate further supports the importance of diffusion. The simultaneous depletion of Ti and Ta for the small precipitates formed at slow cooling rates (1 and 10 K min À1 ) can also be explained by the relatively lower diffusivity of these elements at lower temperature. Minish [36] used first-principle calculations to demonstrate that, in c 0 , Alrich off-stoichiometry is characterised by high Al antisite concentrations, while Ni-rich off-stoichiometry has dramatically fewer Al antisites. This supports the conclusion that our experimentally measured decrease in the C (Ni, Co, Cr, Mo) /C (Al, Ti, Ta) ratio for smaller precipitates and slow cooling rates (Fig. 7) is produced by increased concentrations of Al antisites. This again demonstrates that the enhancement of Al in the fine-scale tertiary precipitates results from the surviving antisites within the b-sublattice and the importance of the antisite diffusion mechanism at low temperatures.
It is worth noting that, for the fast cooling rates, all precipitates are likely to have been nucleated at relatively high temperatures, similar to the temperatures at which the secondary c 0 precipitates were generated within the slowcooled samples, in comparison to those of the tertiary formation, i.e. faster cooling rates result in fine-scale precipitates being formed at higher temperatures compared to those of the same sized precipitates generated at 1 or 10 K min À1 . Consequently, at 360 K min À1 , precipitates of 30-50 nm grow by the same diffusion mechanism as observed for 240 nm secondary precipitates formed at cooling rates of 1 K min À1 , i.e. growth occurs via intrasublattice vacancy-atom exchange, and these precipitates do not show anomalous compositions. This implies a transition for aluminium from intrasublattice diffusion at high temperature to antisite at low temperature.

Conclusions
In conclusion, absorption-corrected EDX spectroscopy has been performed in an S/TEM on a large number of extracted c 0 precipitates in order to study the size-dependent compositional variations present for the PM Ni-based superalloy RR1000 at different solution cooling rates. Our results provide new experimental data to support the importance of considering differences in diffusion kinetics for different alloying elements when predicting the precipitation and phase chemistry. In summary: (1) The secondary c 0 precipitates (140-300 nm) are found to have near-field equilibrium compositions, whilst the fine-scale tertiary c 0 precipitates have farfield compositions. (2) Aluminium is observed to be enriched in the finescale tertiary c 0 for slow-cooled alloys and exhibits different compositional variation behaviour from the other c 0 stabilizers (Ti and Ta are depleted in these same precipitates). (3) A slow cooling rate of 1 K min À1 led to greater Al enrichment compared to the same sizes of tertiary c 0 precipitates formed with a cooling rate of 10 K min À1 . This suggests abnormal diffusion kinetics in aluminium, which can be explained by the importance of the antisite diffusion mechanism at low temperature. The experimentally observed differences in the Al enrichment observed for different cooling rates suggests a transition of Al diffusion kinetics from intrasublattice at high temperature to antisite at low temperature. Our results provide new experimental evidence validating Al antisitebridged diffusion at low temperatures and provide valuable structural data towards improving the accuracy of predicting the precipitate phase chemistry and microstructural evolution in Ni-based superalloys.