Origin of ω-phase formation in metastable β-type Ti-Mo alloys: cluster structure and stacking fault

The ω-phase formation and its collapsed structures in metastable β-type Ti-Mo alloys were illustrated by first-principles calculations and experimental evidence of a partially collapsed ω-phase in the nano-scale Mo-depleted region under a rapid cooling via high-angle annular dark-field scanning transmission electron microscopy. The ease of ω-phase formation within -Mo-Ti-Mo- poor cluster structure was not only due to the low energy barrier in the collapse pathway, which was caused by the reduced lattice distortion, but also due to the softening of the shear modulus (G111) as a result of the small charge density difference. The most stable collapsed structure of the ω-phase strongly depended on the minimum stacking fault energy among different collapse degrees in accordance to the smallest charge density difference. Therefore, the concurrent compositional and structural instabilities of the ω-phase was attributed to the coupling effect of the cluster structure with stacking fault from the atomic and electronic basis.


Results
(a-f) shows the constructed supercells of the ( 110) β plane with cluster structure and stacking fault. The {111} β planes in the bcc lattice were denoted by "ABCABCA" (z = 0, i.e., β-phase) in Fig. 1(a,d). The adjacent ( 112) β planes at positions −1 and −2 (P −1 and P −2 ) sheared toward each other, thereby changing "AB′C′AB′C′A" (z < 1/6, i.e., partially collapsed or trigonal structure) in Fig. 1(b,e) to "AB′AB′A" (z = 1/6, i.e., fully collapsed or hexagonal structure) in Fig. 1(c,f). Fig. 1(a-c) shows the -Mo-Ti-Mo-linear unit along the [ 111] β direction in the 2Mo supercell. The presence of seven positions from P −3 to P 3 indicated that the cluster moved away from the stacking fault. These cluster positions were labeled by the Ti atom in the middle of the linear unit (e.g., the two cluster positions at P −3 and P 2 in the figure). Fig. 1(d-f) shows the trigonal cluster comprising -Mo-Ti-Mo-linear units along the [ 111] β , [ 111] β, and [001] β directions (in side view) in the 3Mo supercell. Four cluster positions from P −3 to P 3 , such as the two cluster positions at P −3 and P 1 that were labeled by the Mo atom in the middle of the cluster (the left Mo atom in side view), described the position correlation between cluster structure and stacking fault. Fig. 2(a,b) shows stacking fault energy as a function of collapse degree in the constructed 2Mo and 3Mo supercells with different cluster positions. In the initial bcc lattice, no stacking fault energy was observed for all linear and trigonal cluster positions. Fig. 2(a) shows that the stacking fault energy decreases along with collapse degree for each cluster position in the 2Mo supercell. A minimum value (γ SF, min z ) was observed in each curve as denoted by the circle, thereby suggesting that the corresponding collapsed structure of the ω-phase was the most stable among all collapse degrees. With the linear cluster moving away from the stacking fault from P −3 to P 3 , the stacking fault energy decreased, whereas the collapse degree at minimum stacking fault energy increased from 6.2/48 at P −3 to 8/48 at P 2 and P 3 . In the 3Mo supercell with the cluster at P −3 as shown in Fig. 2(b), the stacking fault energy increased along with the collapse degree and its minimum value was zero, thereby suggesting that the bcc lattice was the most stable structure without collapse. Similarly, the stacking fault energy decreased along with www.nature.com/scientificreports www.nature.com/scientificreports/ collapse degree when the trigonal cluster was located far from the stacking fault, and the collapse degree at minimum value increased from 6/48 at P −1 to 8/48 at P 3 . To further evaluate the deviation of collapse degree from the trigonal to hexagonal structure of the ω-phase, the difference of stacking fault energy ( γ ∆ z SF ) between the stable (d-f) 3Mo supercell with -Mo-Ti-Mo-trigonal cluster at two positions (P −3 and P 1 ) for example; (a,d) initial β-phase (z = 0) with the ( 111) β planes labeled by "ABCA;" (b,e) (111) β planes with partial stacking fault (z < 1/6) labeled by "AB′C′A;" and (c,f) (111) β planes with full stacking fault (z = 1/6) labeled by "AB′A". The blue and purple circles represent the Ti and Mo atoms, respectively. The open and solid circles of the Mo atoms in the 3Mo supercell were located at two adjacent (110) β planes.
www.nature.com/scientificreports www.nature.com/scientificreports/ structure and the fully collapsed structure was obtained as a function of cluster position as shown in Fig. 2(c). The lower γ ∆ SF z indicated that the stable structure of the ω-phase exhibited a smaller collapse degree. In the 2Mo supercell, γ ∆ SF z had the minimum value of −24.7 mJ/m 2 with a liner cluster at P −3 , thereby suggesting that the stable trigonal structure had the smallest collapse degree (6.2/48). With the cluster moving away from the stacking fault, the γ ∆ SF z increased to zero at P 2 and P 3 corresponding to the stable hexagonal structure. The γ ∆ SF z in the 3Mo supercell exhibited a similarly increasing trend with the trigonal cluster moving away from the stacking fault but was lower than that in the 2Mo supercell at the same position, thereby reducing the collapse degree of the ω-phase, especially at P −3 (bcc lattice) with the lowest value of −310 mJ/m 2 . Fig. 2(d) shows the formation energy (E F ) of the stable structure with minimum stacking fault energy as a function of cluster position in 2Mo and 3Mo supercells. A lower E F indicated that the cluster structure was more stable in the position of the supercell. The E F decreased from 2.5 eV to 2.1 eV in the 2Mo supercell from P −3 to P 3 and from 2.5 eV to 1.4 eV in the 3Mo supercell from P −1 to P 3 . However, the E F in the 3Mo supercell at P −3 exhibited a low value of 1.6 eV due to the stable bcc lattice. Consequently, the -Mo-Ti-Mo-cluster structures should occupy positions far from the stacking fault, and the stable collapsed structure of the ω-phase changed from trigonal to hexagonal due to the low formation and stacking fault energies.
The TEM dark-field image and its selected area electron diffraction (SAED) pattern were obtained in a water-quenched sample of Ti-15Mo alloy. As shown by the arrows in Fig. 3(a), the black area is mainly β-matrix which may contain a small amount of ω-particles, while the white particles are mainly ω-phase with a diameter of less than 10 nm. The SAED pattern in Fig. 3(b) exhibited three sets of spots along the [110] β zone axis that corresponded to two ω-phase variants of ω 1 and ω 2 (denoted by the red and green squares, respectively) as well as the β-phase (denoted by the blue square). The diffraction spots selected to produce the dark-field image was marked by a white circle in Fig. 3(b). The intensity profiles of two ω-variants along the <111> β direction were denoted by the dotted lines and were analyzed on the basis of the SAED pattern in Fig. 3(c). The peak positions of ω-phase spots slightly deviated from relative displacements of 1/3 and 2/3, suggesting the formation of the ω-phase without an ideal hexagonal structure. To quantitatively evaluate its collapse degree, the ratio of ⁎ d 0002ω /d 222 ⁎ β in reciprocal space was obtained by establishing a formula with collapse degree as derived from the Appendix of our previous work 41 . As a result, the ω 1 -variant and ω 2 -variant had collapse degrees of 0.164 ± 0.001 and 0.163 ± 0.001, respectively. In terms of the orientation relationship between the β-phase and ω-phase and the structural evolution of the ω-phase as shown in Fig. 3(d), the ω-phase with a partially collapsed structure was formed in the alloy.  Fig. 3(e)). The two ω-variants (ω 1 and ω 2 ) spots in Fig. 3(g,h) were present in the Mo-depleted regions (denoted by the red and green squares in Fig. 3(e), respectively). The overlay of RGB images was filtered by using reflection spots from the ω 1 -variant (R: red), ω 2 -variant (G: green), and β-phase (B: blue), which corresponded to the Mo-depleted and Mo-enriched regions in Fig. 3(e). The green and red regions indicated that the formed ω-phase had a size of several nanometers. These experimental results indicated that the ω-phase with a trigonal structure was prone to form in the nano-scale Mo-depleted region due to compositional partitioning.  Fig. 4(b) exhibited no collapse in the Mo-enriched region, whereas the ω-phase with partial collapse in Fig. 10(c,d) was close to the boundary of the β-matrix and in the Mo-depleted region. To quantitatively evaluate the collapse degree, a plot of atomic column intensity as a function of distance was obtained along the <111> β direction, and the relative www.nature.com/scientificreports www.nature.com/scientificreports/ positions of peaks were denoted by the dashed lines. The value of z was computed as the ratio of distance that the ( 112) β plane moves along the [111] β direction to 1/2 < 111>a β . In the Mo-enriched region, the peaks of row 2 and row 3 were present at 1/3 and 2/3 positions between two peaks of row 1 in Fig. 4(e), and the value of z was −0.013 ± 0.043. Near the boundary of the β-matrix, the peaks of row 2 and row 3 were closed due to the <111> β atomic collapse as shown in Fig. 4(f), and the value of z was 0.096 ± 0.033. In the Mo-depleted region, the value of z further increased to 0.137 ± 0.015 as shown in Fig. 4(g). Therefore, the collapse degree of the ω-phase continuously increased from the Mo-enriched region to the Mo-depleted region.

Discussion
The ω-phase that was easily formed in the -Mo-Ti-Mo-cluster poor region and its collapsed structure strongly depended on the present minimum stacking fault energy obtained via theoretic calculations. Moreover, the partially collapsed ω-phase was most likely present in the nano-scale Mo-depleted region according to the results of experimental observations. From the atomic and electronic basis, both lattice distortion and change density difference were calculated to study the reasons behind the ω-phase formation via the cluster structure and its collapsed structural evolution combined with stacking fault. The concurrent compositional and structural instabilities of the ω-phase was also discussed from the view of a metastable "frozen" state of Mo atoms due to the rapid cooling condition. ω-phase formation by cluster structure. On the basis of the aforementioned collapse mechanism of the ω-phase, two factors were identified to influence the collapse process on the {112} β plane along the <111> β direction: one is the extrinsic energy barrier in the pathway, whereas the other is the intrinsic resistance of shear along a specific direction. The energy barrier is affected by the lattice distortion, which suppresses the displacive phase transformation 53 . The lattice distortion in this study was caused by the difference in radius between Mo and Ti atoms, thereby resulting in that the ( 112) β plane was not flat with up-and-down hills as shown in Fig. 5(a,b) to supply the energy barriers in the pathway of [111] β direction. In 2Mo and 3Mo supercells, the up-and-down hills in a certain region as denoted by the red dotted lines (i.e., lattice distortion) decreased along with the position of the cluster structure from P −3 to P 3 . The trigonal cluster structure in the 3Mo supercell caused a more serious lattice distortion than that of the linear cluster structure in the 2Mo supercell. To quantitatively evaluate the effect www.nature.com/scientificreports www.nature.com/scientificreports/ of cluster structures, the lattice distortion of the [ 112] β direction (δ) was obtained as shown in Fig. 5(c). The value of δ decreased from 7.2% to 2.2% in the 2Mo supercell and from 10.3% to 3.8% in the 3Mo supercell along with the cluster position from P −3 to P 3 , respectively. In other words, the extrinsic energy barrier for the collapse process of the ω-phase was reduced within -Mo-Ti-Mo-poor region, thereby promoting the formation of this phase.
Shear resistance is reflected by the shear modulus along a certain direction, and its softening effect can benefit the displacive phase transformation. For example, the softening of tetragonal shear elastic constant (C′) with respect to the shear resistance on the {110} β plane along the <110> β direction enhanced the α′′-martensite formation in titanium alloys 41,54,55 . Meanwhile, shear modulus (G 111 ) (i.e., shear resistance on the {112} β plane along the <111> β direction) was associated with the collapse mechanism of the ω-phase. Our previous work 41 reported that the softening of G 111 could benefit the ω-phase formation and strongly depended on the Mo content in metastable Ti-Mo alloys. Therefore, G 111 was reduced when the cluster structure moved away from the initial position of P −3 in Fig. 1(a,d) and softened much easier with the linear cluster than with the trigonal cluster. To further illustrate the correlation between cluster structure and shear modulus, the supercells containing 54 atoms with -Mo-Ti-Mo-linear and trigonal clusters were used to calculate G 111 , which can be derived by the elastic constants of the bcc structure 6,41 . Given the size limitations of the constructed supercells, only the two typical cluster positions of P −3 and P −1 were considered in this study as shown in Fig. 5(d-g). The value of G 111 decreased from −11.8 GPa to −20.6 GPa in Fig. 5(d,e) and from 29.6 GPa to −6.5 GPa in Fig. 5(f,g) along with the shifting of cluster position from P −3 and P −1 in 2Mo and 3Mo supercells, respectively. Therefore, the intrinsic shear resistance for the collapse process was softened within the -Mo-Ti-Mo-cluster poor region.
Shear modulus is also associated with the atomic bonding in adjacent planes, which can be evaluated based on the charge density difference. A larger charge density difference corresponds to a stronger atomic bonding 11 . Gao et al. 56 reported that the charge density difference on the (110) plane along the <111> β direction was high for pure Cr but gradually decreased along with the addition of V, thereby reducing G 111 . Here, the shear modulus [ 111] β directions, thereby suggesting that a strong Ti-Mo bonding was formed along these directions within the ( 110) β plane. In the 3Mo supercell, a (110) β plane with a trigonal structure as indicated by the black dotted line in Fig. 6(d) was used to clearly illustrate the charge density difference. In addition, the Ti-Mo bonding was particularly strong in the <111> β direction. To further evaluate the difference of Δρ along specific directions, the line profiles of Δρ were denoted by the white lines from the Ti atom to the Mo atom in Fig. 6(a,d), and the results are shown in Fig. 6(g,h). The [001] β , [110] β , [ 111] β and [ 111] β directions in the 3Mo supercell with a trigonal cluster were analyzed. The Δρ along the <111> β direction was the largest, whereas that along the <100> β direction was the smallest, thereby indicating that the Ti-Mo bonding was strongest along the <111> direction, which was in agreement with the findings of Wang 57 . From an electronic basis, this result further revealed the rationality of the -Mo-Ti-Mo-linear and trigonal cluster structures, which was only analyzed based on cohesive energy in our previous work 41 . Similarly, the Δρ along specific directions was calculated when the linear cluster in Fig. 6(b,c) and trigonal cluster in Fig. 6(e,f) moved from P −3 to P 3 . For the line profiles such as those shown in Fig. 6(g,h), the integral Δρ was obtained for each cluster position as shown in Fig. 6(i). The quantitative results indicated that the charge density difference along a specific direction in the 2Mo supercell was lower than that in the 3Mo supercell and decreased as the clusters moved from P −3 to P 3 . Therefore, the low charge density difference within the -Mo-Ti-Mo-cluster poor region collapsed structure of the ω-phase by stacking fault. Our previous theoretic calculations and experimental observations revealed that the ω-phase structure changed from hexagonal to trigonal along with an increasing Mo content 41 . The experimental evidence also indicated that the ω-phase with a trigonal structure was formed in the Mo-depleted region (Fig. 3). Although the aforementioned extrinsic and intrinsic factors were reasonable enough to explain the easy formation of the ω-phase within the -Mo-Ti-Mo-cluster poor region, a direct correlation between the collapse degree of the ω-phase and cluster structure was not clearly revealed. From the collapse mechanism of the ω-phase, the stacking sequence of the {111} β plane in the bcc lattice changes from "ABCABCA" to "ABABA" along the <111> β direction, whereas the structural evolution of the ω-phase can be influenced by the stacking fault on the {112} β plane 34,42,43 . These results also indicated that the stacking fault energy decreased along with the collapse degree and the collapsed structure was the most stable at the present minimum stacking fault energy ( Figs. 1 and 2). On the basis of the constructed supercells comprising a cluster structure and stacking fault as shown in Fig. 1(b,c,e,f), both the lattice distortion and change density difference were used to examine the stacking fault energy and collapse degree of the ω-phase. On the one hand, the lattice distortion along the [ 112] β direction was not changed by the shear of atoms along the [111] β direction on the ( 112) β plane. For each cluster position in the 2Mo and 3Mo supercells, the value of δ did not change along with collapse degree from 0 to 1/6 as shown in Fig. 7, thereby suggesting that the changing tendency of stacking fault energy in Fig. 1(g,h) was not reasonably explained by the lattice distortion.
On the other hand, the stacking fault energy was affected by the charge density difference or atomic bonding, and the variation of stacking fault energy could be analyzed by the redistribution of change density, where some bonding may become either stronger or weaker 58,59 . Shang et al. 60 reported that the larger charge density difference of Al along the [ 1120] direction led to higher stacking fault energy and shear strength along the [ 1010] direction in Mg-based alloys. Kioussis et al. 61 found that the variation of stacking fault energy in fcc metals resulted from the change of charge density in three directions across to the {111} slip plane. Compared with the charge density difference without a collapse degree as shown in in Fig. 6, the values of integral ∆ρ for each cluster position in the 2Mo and 3Mo supercells were calculated at different collapse degrees along specific directions. Here, the difference of ∆ρ (d∆ρ) with and without collapse degree was calculated to describe the redistribution of charge density as shown in Fig. 8. In the 2Mo supercell, the d∆ρ along the <100> β direction increased at each cluster position along with an increasing collapse degree but did not show any changes at P 3 as shown in Fig. 8(a). As shown in Fig. 8(b), the d∆ρ along the <110> β direction did not obviously change along with collapse degree at all cluster positions. By contrast, the d∆ρ along the <111> β direction exhibited a decreasing trend at each cluster position with an increasing the collapse degree, except at the position of P 3 as shown in Fig. 8(c). Although the trends of d∆ρ (Fig. 8(d-f)) were similar to those recorded in the 2Mo supercell, larger variations in the values of d∆ρ were observed along the <100> β and <111> β directions in the 3Mo supercell. To synthetically examine the effect of charge density difference on stacking fault energy, the total values of ∆ρ and d∆ρ along three directions were calculated at different collapse degrees as shown in Fig. 9. The total ∆ρ decreased for each collapse degree when the cluster structure moved from P −3 to P 3 and decreased along with an increasing collapse degree at each cluster position, except at P 3 (where no obvious changes were observed) in 2Mo and 3Mo supercells and at P −3 (where an increasing trend was reported) in the 3Mo supercell. Moreover, a minimum charge density difference was observed at a collapse degree of less than or equal to 1/6 when the cluster position was near to or far from the stacking fault. The minimum charge density difference corresponded to the zero collapse degree (bcc lattice) in the 3Mo supercell with the cluster at P -3 . The variations in charge density difference along with cluster position and collapse degree agreed with the trends of stacking fault energy and indicated the collapsed structural evolution of ω-phase, even when the overestimated and underestimated values were considered. www.nature.com/scientificreports www.nature.com/scientificreports/ On the basis of the theoretic calculations of stacking fault energy and charge density difference, in the Mo-depleted region, the ω-phase with a hexagonal structure was easily formed, while the experimental evidence confirmed the formation of a trigonal ω-phase due to the compositional instability under the water-quenching condition, that is, a metastable "frozen" state of Mo atoms, thereby leading to the present minimum stacking fault energy (minimum charge density difference) with a partial collapse degree (Figs. 1 and 9). If the Mo atoms sufficiently moved out of the Mo-depleted region, then the collapse degree of the ω-phase would increase up to 1/6, as reported in ref. 10,33 . However, the formation of an isothermal ω-phase was not the focus of this work. The local instability of the composition is known to be present in quenched β-type titanium alloys, which often leads to local structural instability 10,34,36,62,63 . In addition, the nano-scale compositional partitioning resulted in continuous changes in the solute content between the depleted and enriched regions as shown in Figs. 3 and 4. Therefore, the structural instability of the ω-phase changed continuously due to compositional instability. Consequently, the structural evolution of the ω-phase from hexagonal to trigonal in metastable β-type Ti-Mo alloys was schematically drawn along with Mo content and collapse degree in Fig. 10. The concurrent compositional and structural instabilities of the ω-phase was attributed to the coupling effect of the cluster structure with stacking fault.
In conclusion, the -Mo-Ti-Mo-cluster structures should occupy positions far from the stacking fault in the constructed supercells with a linear cluster along the <111> β direction and a trigonal cluster along the <111> β and <001> β directions. The stable structure of the ω-phase with different collapse degrees on the {112} β plane www.nature.com/scientificreports www.nature.com/scientificreports/ along the <111> β direction strongly depended on the present minimum stacking fault energy and the formation energy of cluster structures. The ω-phase with a partially collapsed structure was prone to form in the nano-scale Mo-depleted region due to compositional partitioning under a rapid cooling condition. The structural instability of the ω-phase continuously changed along with compositional instability from the Mo-enriched region to the Mo-depleted region due to the metastable "frozen" state of Mo atoms. Within the cluster poor region, the lattice distortion along the <112> β direction on the {112} β plane decreased, thereby leading to the low energy barrier in the collapse pathway. Moreover, the charge density difference corresponding to the atomic bonding along the <111> β , <011> β and <001> β directions on the (110) β plane was reduced, thereby softening the shear modulus (G 111 ) along the <111> β direction on the {112} β plane, which benefited the ω-phase formation. The redistribution of charge density exhibited an obvious anisotropy among the <111> β , <011> β , and <001> β directions with different collapse degrees, and the changing tendency of total charge density difference on the (110) β plane with cluster structure and collapse degree was in accordance with the stacking fault energy and corresponded to the collapsed structural evolution of the ω-phase from trigonal to hexagonal.

Methods calculation methodology.
The calculations were carried out based on density functional theory and by using the Vienna Ab-initio Simulation Package code 64 and the projector augmented wave method for investigating the core-valence electrons interaction 65 . The exchange correlation potential was measured via generalized gradient approximation and expressed in Perdew-Burke-Ernzerhof form 66 . The plane wave cut-off energy was set at 400 eV. Relaxation was stopped when the energy reached 10 -3 eV convergence. The Monkhorst-Pack mesh for this system was 9 × 11 × 1. Supercells comprising 24 atoms were constructed with 12 ( 112) β planes and 2 and 3 Mo atoms (Ti-15Mo or Ti-22Mo alloys, respectively). According to the collapse mechanism of the ω-phase, the two adjacent Stacking fault energy (γ SF z ) was calculated as follows in terms of the constructed supercells with cluster structure and stacking fault 48 : Figure 9. Total charge density difference (total ∆ρ) in (a) the 2Mo supercell and (b) the 3Mo supercell, and total difference of charge density difference (total d∆ρ) in (c) the 2Mo supercell and (d) the 3Mo containing <001> β , <011> β and <111> β directions on the {110} β plane as a function of collapse degree (z).
where E z and E 0 are the total energy of a supercell with a collapse degree and the total energy of a perfect bcc supercell, respectively, and A is the area of the fault plane that is a cross product of supercell lattice vectors a 0[ 110] β and a 0 [111] β . To further evaluate the stability of an Mo atom occupying a certain position in the supercell, the formation energy (E F ) was calculated as where E Ti hcp and E Mo bcc are the energies per atom of bulk Ti in the hexagonal close-packed structure and that of Mo in the bcc structure, respectively, n is the number of Mo atoms in each supercell, and E SF Ti is the stacking fault energy without Mo atoms. Following this definition, a lower formation energy corresponds to a more stable Mo atom at a certain position. The formation energy was calculated only in the structure with the minimum stacking fault energy.
The lattice distortion (δ) of each supercell was calculated due to the different radiuses of the Mo and Ti atoms, and the quantitative value of the ( 110) β plane along the [ 112] β direction was computed as x cos cos y y cos z z cos where x′, y′, z′ and x, y, z are the fractional coordinates of atoms with distortion and non-distortion, respectively, and α 1 , α 2 , and α 3

Ti Mo Ti
where ρ Ti-Mo is the charge density of the supercell containing Ti and Mo atoms, and ρ Ti is the charge density of the supercell with Mo replaced by Ti atoms. The charge density difference on the ( 110) β plane along the <001> β , <011> β and <111> β directions was quantitatively evaluated.
Material preparation and microstructural observation. Ti-15Mo alloy was prepared via cold crucible levitation melting. The ingot weighed approximately 1 kg and was homogenized at 1273 K for 3.6 ks, hot forged into a block at 1273 K, hot rolled into a plate with dimensions of 290 mm (l) × 50 mm (w) × 10 mm (t) at 1173 K, and eventually air cooled. After cutting into 10 mm (l) × 10 mm (w) × 5 mm (t), the plate was solution treated for 3.6 ks at 1173 K and underwent water quenching. Our previous works 67,68 reported that the chemical compositions of Ti-Mo alloys were controlled well by using this melting method and that each alloy had an oxygen content of approximately 0.1 wt.% with small amounts of nitrogen and carbon.
The sample of 10 mm (l) × 10 mm (w) × 1 mm (t) was cut from the solution-treated plate via wire cut electrical discharge machining and then mechanically ground to 0.1 μm in thickness. Subsequently, the foils for microstructural characterization were thinned via dimple grinding (Gatan 656 Dimple Grinder) and ion milling (Gatan PIPS II 695). TEM and atomic resolution HAADF-STEM observations were carried out by using a 300 kV probe corrected electron microscope (FEI Titan G2). HAADF images were captured by using a probe with a convergence semi-angle of 21.4 mrad by setting the camera length at 100 mm. The inner and outer detection angles were 76 mrad and 200 mrad, respectively. To illustrate the structural features at the atomic level, Fourier filtering processing was performed by using a digital micrograph software to remove noisy background from raw images, and the local structural information was analyzed by Fast Fourier Transform (FFT) patterns.