Thermally driven surface phase separation in intermetallic alloys

Abstract

Intermetallic compounds are widely recognized for their high-temperature phase stability and resistance to composition and structural changes. However, we reveal a thermally activated bulk-to-surface mass exchange mechanism that drives surface phase separation, resulting in the formation of surface precipitates with distinct composition and structure from the bulk matrix. Using the archetypal β-NiAl system, we show that asymmetries in vacancy formation energies between Ni and Al atoms induce preferential Ni segregation to the surface, forming Ni-rich γ‘-Ni₃Al precipitates. By integrating in-situ electron microscopy, synchrotron X-ray absorption spectroscopy and first-principles computational modeling, we establish a direct mechanistic connection between bulk thermal defect dynamics, surface compositional evolution, and phase segregation behavior. This bulk-surface coupling mechanism can be a driver of surface phase separation in multicomponent alloys under thermal stress. These results refine the thermodynamic boundaries of intermetallic stability and provide insights into managing the performance and durability of intermetallic alloys for demanding high-temperature applications.

Introduction

Intermetallic compounds exhibit robust phase stability at high temperatures, a property rooted in their highly ordered atomic structures and strong bonding between dissimilar atoms^1,2,3. Unlike conventional alloys, where atomic disorder facilitates compositional fluctuations and phase separation, intermetallics possess rigid stoichiometry and crystallographic order^4,5,6, creating energetic barriers to atomic diffusion and phase decomposition^7,8,9. This inherent stability has positioned them as cornerstone materials for extreme-temperature applications, from turbine blades to aerospace components.

Contrary to the conventional understanding, herein we reveal a thermally driven bulk-surface coupling mechanism that destabilizes intermetallic surfaces through phase separation. While bulk thermal vacancies are recognized as intrinsic defects at elevated temperatures^10,11, their role in driving surface compositional and structural evolution remains unexplored. We show that vacancies in the bulk induce selective elemental migration to the surface, creating phase-segregated domains with distinct compositions and structures.

To investigate this phenomenon, we focus on β-NiAl, a model intermetallic compound for its exceptional mechanical stability under extreme conditions¹². Despite its ability to maintain an ordered phase in the bulk at temperatures as high as 1638 °C¹³, the exponential increase in thermal vacancies with temperature^11,14,15 leads to significant atomic migration from the bulk to the surface. Due to the differing vacancy formation energies of Ni and Al, this migration results in the formation of a high density of Ni-rich surface islands that exhibit distinct phases and compositions in comparison to the bulk.

This bulk-surface coupling mechanism is relevant across a broad range of alloy systems, where inherent differences in equilibrium vacancy behavior can drive atomic redistribution between the bulk and surface (including internal interfaces and grain boundaries) under high-temperature conditions. Such surface compositional and structural variations can significantly influence catalytic activity, corrosion resistance, and mechanical behavior by alternating defect nucleation, migration, and annihilation pathways at surfaces and interfaces.

Results

Ex-situ scanning transmission electron microscopy (STEM) imaging of surface precipitation

Figure 1a presents a representative high-angle annular dark-field (HAADF)-STEM image of the as-prepared NiAl foil (prepared by FIB and unannealed), revealing a homogeneous morphology with uniform intensity, indicative of negligible thickness or composition variations. Figure 1b provides STEM energy-dispersive X-ray spectroscopy (EDS) elemental mapping of the area shown in Fig. 1a, confirming the uniform distribution of Ni and Al throughout the large area. Figure 1c displays a selected area electron diffraction (SAED) pattern obtained from the foil, showing sharp (101) and (110) reflections of the β-NiAl phase. The observed 60° angular separation between these spots aligns with the $\left[\bar{1}11\right]$ zone axis.

Fig. 1: Microstructural and compositional evolution of NiAl during annealing.

a HAADF-STEM image of FIB-prepared NiAl pre-annealing. b EDS-STEM elemental mapping; HAADF image and EDS maps of Ni+Al, Ni (yellow) and Al (red) overlayed across the same field of view. c SAED pattern (zone axis $\left[\bar{1}11\right]$) of β-NiAl. d HAADF image of the NiAl after annealing at 800 °C for 1 h, resulting in Ni-rich precipitates. e EDS mapping: HAADF image and EDS maps of Ni+Al, Ni (yellow) and Al (red) overlayed across the same field of view, with the white arrows highlighting areas containing higher concentrations of Ni in comparison to the surrounding. f SAED pattern of annealed NiAl.

Following annealing at 800 °C for an hour in 0.001 Torr H₂, significant surface restructuring occurs. The HAADF image in Fig. 1d reveals a high density of precipitated particles, with pronounced bright contrast, indicating localized composition changes post–annealing. EDS mapping (Fig. 1e) confirms the Ni enrichment within these precipitates (white arrows in Fig. 1e), while Al distribution remains relatively uniform. Therefore, the brighter HAADF contrast in Fig. 1d is primarily attributed to Z-contrast difference arising from Ni-rich regions (higher atomic number), with minimal contribution from thickness variations. Electron diffraction analysis (Fig. 1f) of the post-annealed NiAl (Fig. 1e) align with the β-NiAl phase observed in Fig. 1c. Additionally, weak superlattice spots (marked by dashed green circles in Fig. 1f) suggest atomic vacancy ordering in the β-NiAl bulk, due to the surface ejection of Ni atoms from the bulk—a phenomenon discussed further in subsequent sections. Notably, the unannealed sample does not exhibit precipitates, streaking of the diffraction spots, and superlattice reflections, as evidenced by uniform HAADF intensity, sharp electron diffraction, and homogeneous EDS mapping. These results demonstrate that thermal annealing drives greater Ni segregation than that of Al, leading to the formation of Ni-rich precipitates on the surface. Furthermore, this surface segregation results in atomic vacancies in the bulk, which undergo self-ordering within the crystal lattice, as indicated by the superlattice reflections in Fig. 1f. This interplay between surface segregation and bulk restructuring highlights how thermal processing influences atomic structure evolution in the alloy.

In-situ high resolution transmission electron microscopy (HRTEM) imaging of surface phase separation

To attain an atomic-scale understanding of Ni-rich precipitate formation, NiAl films (prepared by e-beam evaporation) are annealed at 800 °C in environmental TEM under a flowing H₂ gas atmosphere (0.001 Torr). This reducing environment prevents surface oxidation during heat treatment. Figure 2a presents an edge-on HRTEM image of the (110) facet, revealing the formation of precipitate particles at the surface. Within the precipitated phase, the measured interplanar spacings of 2.61 Å and the observed 60° rotational symmetry of the lattice planes are consistent with the $\left[\bar{1}11\right]$ zone axis of ordered γ‘-Ni₃Al phase (Fig. 2a), which adopts a cubic L₁₂ structure (PDF #09-0097). The measured (110) interplanar spacing of 2.61 Å is slightly larger than the theoretical room-temperature value of 2.52 Å. This deviation can be attributed to a combination of factors, including thermal expansion at the experimental temperature, slight compositional deviation from the ideal Ni₃Al stoichiometry, and measurement uncertainty. After accounting for thermal expansion^16,17, the remaining lattice discrepancy corresponds to ~1.95%. The fast Fourier transform (FFT) (top-right inset, Fig. 2a) from the precipitate region can be indexed as γ‘-Ni₃Al along the $\left[\bar{1}11\right]$ zone axis. Notably, additional reflections in the FFT indicate the presence of ordered vacancies within the precipitate lattice, similar to those observed in the parent NiAl lattice, as discussed later. This identification of the Ni₃Al phase is further supported by the simulated HRTEM image overlaid in Fig. 2a (red inset).

Fig. 2: Atomic-scale imaging of γ‘-Ni₃Al precipitates formed during annealing of NiAl at 800 °C in 0.001 Torr H₂ for ~1 h.

a HRTEM image of the β-NiAl(110) facet showing a γ‘-Ni₃Al precipitate outlined by yellow dashed lines. The red inset shows a simulated HRTEM image of γ‘-Ni₃Al along the $\left[\bar{1}11\right]$ zone axis, consistent with the FFT (top-right inset). The Ni₃Al/NiAl(110) interface is marked by a red dashed line, with misfit dislocations indicated by green and pink dashed lines. A step-like disconnection defect is highlighted in the magnified inset (white dashed box). b HRTEM image of a γ‘-Ni₃Al precipitate on the β-NiAl(100) facet, viewed along the [001] zone axis. The precipitate, chemically disordered β-NiAl interfacial layer, and β-NiAl substrate are shaded in yellow, cyan, and purple, respectively. c IFFT of (b), generated using the NiAl(100), NiAl(110) and γ‘-Ni₃Al(110) reflections from the FFT of (b), highlighting the stepped Ni₃Al/NiAl(100). d IFFT of (c), obtained using the NiAl($1\bar{1}0$) and γ‘-Ni₃Al($0\bar{1}1$) reflections, showing a coherent (100) interface between β-NiAl and γ‘-Ni₃Al.

A magnified of the white dashed box in Fig. 2a (middle-right inset) reveals a distinct disconnection—a step-like defect—at the interface between the γ‘-Ni₃Al precipitate and the β-NiAl substrate. Such interfacial steps can partially relax lattice mismatch strain in heteroepitaxial systems¹⁸. Moreover, a misfit dislocation at the interface, highlighted by pink and green dashed lines (Fig. 2a), further accommodates the lattice misfit strain (ε ≈ 0.30) between the γ‘ precipitate (L₁₂, a ≈ 3.57 Å) and the NiAl(110) matrix (B2, a ≈2.89 Å). Similar γ‘-Ni₃Al precipitates are observed along different surface facets (Supplementary Fig. 1), confirming the ubiquity of the surface segregation and precipitate formation under annealing.

Figure 2b, c present an HRTEM image of a γ‘-Ni₃Al precipitate formed on the (100) facet of the β-NiAl matrix and its corresponding inverse FFT (IFFT) image generated using the NiAl(100), NiAl(110), and γ‘-Ni₃Al(101) reflections from the FFT of Fig. 2b (Supplementary Fig. 2). The uncolored full field-of-view (Supplementary Fig. 3) reveals a vacuum region above the precipitate, confirming that the γ‘-Ni₃Al forms at the surface rather than within the bulk. At the γ‘-Ni₃Al/β-NiAl interface, a single atomic step (disconnection) is present. Below this interface, several atomic layers of chemically disordered β-NiAl are observed (highlighted in cyan in Fig. 2b). The similar image intensities of Ni and Al atomic columns indicate randomized site occupancy, a hallmark of disorder, suggesting the formation of an interfacial buffer layer. The IFFT filtered generated from the β-NiAl($1\bar{1}0$) and γ‘-Ni₃Al(0$\bar{1}1$) reflections (Fig. 2d; Supplementary Fig. 2) shows that the (100) interface forms a coherent tilt boundary containing a disconnection defect but no misfit dislocations, contrasting with the dislocated (110) interface seen in Fig. 2a. Instead of forming misfit dislocations, the misfit strain at the (100) interface is accommodated by a slight tilt of the γ‘-Ni₃Al lattice relative to the β-NiAl substrate. The measured misfit strain at the (100) interface is ε ≈ 0.08, substantially smaller than that at the (110) interface (ε ≈ 0.30). This enhanced structural compatibility can be enabled by the chemically disordered β-NiAl interfacial layer, which acts as a compliant buffer that relaxes lattice mismatch without the need for dislocation formation.

On both the NiAl(100) and (110) facets, the precipitates retain their ordered structure, as the experimental temperatures remain below the order-disorder transition temperature (1330 °C) of γ‘-Ni₃Al^4,5. On the (100) facet, γ‘-Ni₃Al initially adopts a coherent, two-dimensional morphology, establishing a low-energy, lattice-matched interface with the NiAl substrate (Fig. 2b–d). As the layer grows and reaches a critical thickness, strain accumulation may drive a transformation into three-dimensional (3D) islands to partially relieve interfacial strain. In contrast, γ‘-Ni₃Al precipitates on the (110) facet exhibit a 3D island morphology from the outset, due to the significantly higher lattice misfit (0.30) that favors incoherent bonding. Interfacial disconnections and steps results in a tilted interface (Fig. 2a), partially accommodating the strain. These observations highlight how lattice mismatch governs precipitate morphology, where lower strain on the (100) facet permits the formation of an initially coherent layer that can later evolve into 3D islands, while high strain on (110) drives immediate 3D island growth to minimize interfacial energy.

In-situ X-ray photoemission electron microscopy (XPEEM) and X-ray photoelectron spectroscopy (XPS) analysis of surface precipitation

The surface formation of precipitate particles is further examined by annealing a bulk β-NiAl(100) single crystal at 800 °C under ultrahigh  vacuum (UHV) conditions in an XPEEM system. During annealing, the growth of a precipitate island is monitored by near edge X-ray absorption fine structure (NEXAFS) mapping (Fig. 3a–c). To investigate local compositional variations, spatially resolved Ni L₃-edge NEXAFS spectra and maps are acquired using XPEEM across the island and the surrounding pristine NiAl(100) surface. The NEXAFS data are collected by acquiring a series of spatial resolved XPEEM images using secondary electrons emitted at different photon energies. The photon energy is incrementally increased by 0.2 eV from 850 to 870 eV, with snapshots captured at each step using secondary electrons ejected from the surface. The XPEEM system enables pixel-wise analysis of the selected field of view, where higher secondary electron counts from a given surface region result in increased signal intensity for corresponding pixels. This intensity variation is tracked across photon energy increments to construct a pixel-wise NEXAFS map, facilitating simultaneous acquisition of spectra from both the newly grown island and the surrounding NiAl surface.

Fig. 3: Spectroscopic characterization of γ‘-Ni₃Al precipitate formation.

a–c Time- and spatially resolved NEXAFS images (extracted from Supplementary Movie 1) obtained using secondary electrons generated under X-ray illumination at photon energies of 850.4 eV (a), 856.6 eV (b), and 868.6 eV (c), capturing the growth of a γ‘-Ni₃Al precipitate island during UHV annealing of a β-NiAl(100) single crystal at 800 °C. In these images, the cyan region indicates the area with higher secondary electron intensity, while the pink region corresponds to lower secondary electron counts. d Temporal evolution of the precipitate island area. Error bars (6%) reflect uncertainties from edge detection procedures used to define the island boundary. e Spatially resolved NEXAFS map acquired using secondary electrons at 853.4 eV, distinguishing the γ‘-Ni₃Al island (cyan) from the surrounding β-NiAl surface (pink). f NEXAFS image snapshot at photon energy of 865 eV showing the island morphology. g Upper panel: experimental Ni L₃-edge NEXAFS spectra from a pure Ni reference (black), the γ‘-Ni₃Al island (cyan), and the surrounding pristine NiAl(100) (pink) at 800 °C, acquired by measuring secondary electron intensity across photon energies from 850 to 865 eV. Peak B at 853.4 eV corresponds to the characteristic lower-energy feature of the γ‘-Ni₃Al phase. Lower panel: Simulated Ni L₃-edge NEXAFS spectra for Ni, γ‘-Ni₃Al, and β-NiAl. The raw structure files used to generate the simulated NEXAFS spectra are provided in the “Source Data Files” folder.

Additionally, because each image slice in the NEXAFS stack is acquired over time, a time-lapse movie can be generated to visualize island growth in real time (Fig. 3a–c), showing the linear growth of the island area (Fig. 3d). The error bars associated with the plotted area vs time arise from the ~6% error due to island edge-detection challenges. While the NEXAFS spectra profiles from the island and island free area appear similar at higher photon energies (Fig. 3g), the island remains distinguishable due to differences in work function between the island and the surrounding area, reflecting compositional variations.

The pristine NiAl(100) region exhibits characteristic Ni L₃-edge peaks at 856.0 eV (peak D) and 858.6 eV (Peak E) (Fig. 3g), corresponding to the β-NiAl phase. In contrast, the precipitate island area displays an additional distinct peak at 853.4 eV (Peak B in Fig. 3g), superimposed on the β-NiAl spectral profile. The simulated spectra closely match the XPEEM-measured NEXAFS spectra in terms of profiles and peak positions (peaks at 855.9 eV (Peak D′) and 857.6 eV (Peak E′) for β-NiAl and 853.5 eV (Peak B′) for the γ‘-Ni₃Al phase as shown in Fig. 3g. In comparison to β-NiAl, the peak at 858.6 eV (Peak E) is shifted to a lower energy at 858.5 eV in the island area. This feature can be explained by the additional peak that is 1 eV below the peak at 857.6 eV (Peak C′) in the simulated γ‘-Ni₃Al spectrum. Furthermore, the peak of metallic Ni at 852.6 eV (Peak A in Fig. 3g) is absent in the experimental spectra of both the precipitate island and surrounding β-NiAl region (Fig. 3g). This absence rules out elemental Ni segregation, confirming that the precipitate island is γ‘-Ni₃Al.

NEXAFS maps generated using secondary electrons upon shining X-rays at 853.4 eV (cyan in Fig. 3e) and 865 eV (Fig. 3f) reveal the morphology of the precipitate island as it produces a higher intensity of secondary electrons in comparison to the surrounding. Repeated XPEEM measurements on the β-NiAl(100) single crystal under similar UHV annealing conditions consistently reveal precipitate islands with the same Ni L₃-edge NEXAFS feature at 853.4 eV (Supplementary Fig. 4), demonstrating the reproducibility of surface phase separation and precipitate formation at elevated temperatures. These results align with the HRTEM observations (Fig. 2a, b), further confirming that high-temperature annealing promotes the surface segregation of γ‘-Ni₃Al precipitates.

The formation of γ‘-Ni₃Al precipitates, rather than elemental Ni islands, is further supported by the NEXAFS edge positions, which typically correlate with the charge state of a given metal. As shown in Fig. 3g, the absorption edges of β-NiAl and γ‘-Ni₃Al occur at higher photon energies than those of metallic Ni, indicative of electron transfer away from Ni and towards Al. This Ni-to-Al electron transfer, as inferred from NEXAFS, is further corroborated by X-ray photoelectron spectroscopy (XPS) measurements (Fig. 4a, b) obtained using An Al Kα X-ray source for a larger probing depth. Three standard single crystals (Al(111), Ni(111) and NiAl(100)) were analyzed to obtain reference spectra to investigate the relative core level shifts in binding energy between Ni and Al in pure Ni(111), Al(111) and NiAl(100) crystals. The NiAl(100) single crystal was the same sample used for the NEXAFS studies shown in Fig. 3. The stoichiometry of the NiAl(100) single crystal was determined from peaking fitting of the Al 2p/ Ni 3p region, yielding a composition of Ni ~49% and Al ~51% (Supplementary Fig. 5). The Ni 2p peak in β-NiAl(100) appears at a binding energy of 853.2 eV, which is higher than 852.6 eV observed for metallic Ni. Conversely, the Al 2p peak in β-NiAl(100) is observed at 72.6 eV is slightly lower than the 72.9 eV measured for pure Al. These shifts indicate electron donation from Ni to Al, as an increased binding energy corresponds to a more positive charge state. In particular, the higher binding energy of the Ni 2p peak in β-NiAl compared to pure Ni confirms a net charge transfer away from Ni in β-NiAl.

Fig. 4: XPS and EFL maps showing charge transfer from Ni to Al in NiAl and Ni₃Al.

a Ni 2p XPS spectra, with the Ni 2p3/2 peak at 853.2 eV for Ni in a β-NiAl(100) single crystal and at 852.8 eV for Ni in a Ni(111) reference crystal. b XPS spectra of the Al 2p/Ni 3p region, highlighting the Al 2p peak at 72.9 eV for metallic Al in an Al(111) single crystal and at 72.6 eV for Al in a β-NiAl(100) single crystal. c, d EFL maps of the (110) plane for (c) β-NiAl and (d) γ‘-Ni₃Al, with corresponding line profiles along the indicated arrows, showing a clear shift of electron density toward the Al sites. An Al Kα X-ray source was used to collect the spectra in (a, b). The same β-NiAl(100) single crystal was used for the measurements in both Figs. 3 & 4. The raw structure files used to compute the ELF maps are provided in the “Source Data Files” folder.

This trend in binding energy shifts extends to γ‘-Ni₃Al, where previous studies indicate that the Ni 2p binding energy in γ‘-Ni₃Al (853.0 eV) falls between pure Ni (852.6 eV) and β-NiAl (853.2 eV)^19,20,21, supporting a progressive charge transfer with increasing Al content. However, the subtle binding energy difference ( ~ 0.1 eV) between NiAl and γ‘-Ni₃Al makes this shift challenging to resolve experimentally. Additionally, fully capturing the charge transfer is further complicated by the fact that the γ‘-Ni₃Al phase exists as a surface layer atop a NiAl(100) bulk substrate, resulting in convoluted signals that are difficult to deconvolute. Nevertheless, the Ni-to-Al charge transfer is further corroborated by DFT-calculated electron localization function (ELF) maps of the (110) crystal planes of β-NiAl and γ‘-Ni₃Al (Fig. 4c, d). These maps reveal a shift in electron density away from the Ni atoms and toward the Al atoms, confirming the charge transfer trends observed in the XPS and NEXAFS analysis.

In-situ high resolution transmission electron microscopy and ex-situ scanning transmission electron microscopy imaging of the bulk lattice

To further investigate the mechanism driving phase separation and the formation of the Ni-rich γ‘-Ni₃Al phase on the surface at elevated temperatures, HRTEM and HAADF-STEM imaging of bulk regions of annealed β-NiAl samples is performed and analyzed. The primary phase present in the bulk is identified as B2-ordered β-NiAl, as confirmed by the HRTEM image (Fig. 5a). The diffractogram of the HRTEM image along the [001] zone axis (Fig. 5b) further validates the β-NiAl phase, revealing the (100)-type reflections. The lattice spacing of the (100) planes is measured at 2.87 Å, consistent with d₍₁₀₀₎ = 2.88 Å for ordered β-NiAl (PDF # 44-1188). Similarly, the (110) lattice spacing, determined from the diffractogram in Fig. 5b, is found to be 2.01 Å, closely matching the expected d₍₁₁₀₎ = 2.04 Å for β-NiAl, within experimental uncertainty. To further support the structural identification, HRTEM image simulations are performed using the B2-ordered β-NiAl structure along the [001] zone axis. The simulated image, overlaid onto the experimental HRTEM image (yellow inset in Fig. 5a), demonstrates strong agreement, with both images exhibiting alternating atom columns of high and low image intensity, characteristic of the B2 structure. The persistence of this ordered phase at experimental temperatures of 800 °C aligns with expectations, as this temperature is well below the order-disorder transition temperature of NiAl, which is close to its melting temperature of 1639 °C²².

Fig. 5: Atomic-scale structural analysis of bulk β-NiAl and vacancy ordering.

a In-situ HRTEM image of β-NiAl bulk during annealing at 800 °C in 0.001 Torr H₂, overlaid with a simulated HRTEM image (yellow outline) based on the B2-ordered β-NiAl structure. (b) Diffractogram (top) of the HRTEM image in (a) and the corresponding atomic structure model of ordered β-NiAl (bottom). c Ex-situ HAADF-STEM image of the annealed β-NiAl bulk. d, e Intensity profiles of atom columns along the yellow and cyan rectangles in (c), corresponding to the <100> and <110> crystallographic directions, respectively. f HRTEM image acquired during annealing β-NiAl at 800 °C in 0.001 Torr H₂. g, h Diffractograms of the regions marked by the blue and orange squares in (f), where blue circles indicate fundamental reflections, and green circles highlight superlattice reflections associated with vacancy ordering. i, j β-NiAl models illustrating different ordering configurations of vacancies (yellow spheres) and their corresponding simulated SAED patterns along the [100] zone axis. k Ex-situ HAADF-STEM image of an annealed β-NiAl bulk region, showing orthogonal dark-contrast bands along <110> directions (yellow arrows). The inset shows the corresponding diffractogram. l Intensity profiles along the purple and cyan rectangles in (k). Yellow arrows mark Ni atom columns with reduced intensities relative to the neighboring Ni columns, indicative of Ni vacancies.

Concurrently, the HAADF-STEM image (Fig. 5c) of the bulk region is analyzed by examining the intensity profiles of atom columns along the <100> and <110> crystallographic directions (Fig. 5d, e, respectively). The resulting line profiles reveal distinct superlattice contrast arising from the atomic mass difference between Ni and Al. The brighter intensity peaks correspond to Ni atom columns, as Ni has a higher atomic number (Z = 28) than Al (Z = 13), leading to increased electron scattering at higher angles and higher image intensity. In the <100> direction (Fig. 5d), the intensity profile exhibits a consistent peak amplitude, indicating that the atom columns along this direction have uniform composition, as expected for the ordered β-NiAl. In contrast, the line profiles taken along the <110> direction (Fig. 5e) show alternating intensity variations, with adjacent atom columns exhibiting lower intensities relative to their neighbors. This intensity oscillation reflects the alternating Ni and Al atomic arrangement in the <110> direction when viewed along the [001] zone axis, further confirming the ordered B2 structure of β-NiAl.

Direct imaging of individual vacancies is inherently difficult; however, a combination of in-situ and post-cooling analyses provides compelling evidence for their presence and preferential nature. In-situ HRTEM imaging of the bulk reveals regions with distinct diffractograms (Fig. 5g, h), corresponding to the areas highlighted by the blue and orange boxes in Fig. 5f. These neighboring regions exhibit differences in their diffractograms, suggesting variations in local vacancy ordering. In addition to the fundamental reflections of β-NiAl, diffractogram analysis of these areas reveals superlattice reflections (highlighted by green circles in Fig. 5g, h), which can be attributed to the ordering of atomic vacancies in β-NiAl and other solids^23,24. Despite the high stability of the ordered phase, the presence of thermal vacancies in the bulk is inevitable at these temperatures. While randomly distributed vacancies are difficult to detect via HRTEM and HAADF, their presence becomes observable when they exhibit long-range ordering. In such cases, the resulting FFT patterns show distinct superlattice reflections related to the periodic arrangement of vacancies²⁵, as shown in Fig. 2a and Fig. 5f–j. Our results reveal two adjacent regions in the bulk with differing vacancy concentrations (Fig. 5f–j).

To attain a better deeper insight into the nature of vacancy ordering, structural models are developed to match the observed diffractograms. The region highlighted by the blue box in Fig. 5f is modeled using a (2 × 2 × 2) supercell containing a single vacancy (left panel, Fig. 5i). The corresponding simulated electron diffraction pattern (right panel, Fig. 5i) closely matches the experimental diffractogram (Fig. 5g), confirming the proposed long-range vacancy ordering configuration. Similarly, a second model with a (4 × 4 × 4) supercell (left panel, Fig. 5j) is constructed to represent the vacancy arrangement observed in the orange-box region of Fig. 5f. The simulated electron diffraction (right panel, Fig. 5j) reproduces the periodic weak superlattice reflections superimposed on the fundamental β-NiAl reflections, consistent with the experimental diffractogram (Fig. 5h).

The primary β-NiAl reflections (marked by blue circles in Fig. 5g, h) remain unchanged, while the additional weak reflections (green circles in Fig. 5g, h) arise from vacancy ordering. In the blue-box region in Fig. 5f, the model suggests that vacancies are evenly distributed along the <$110$> directions, with an inter-vacancy spacing of ~8.14 Å, which corresponds to twice the spacing along the (110) planes in β-NiAl. The spacing between successive parallel rows of ordered vacancies is also ~8.14 Å due to the symmetry of the (2 × 2 × 2) supercell. In the orange box region in Fig. 5f, the model suggests a different periodicity, where vacancies are arranged in distinct rows. Our model (Fig. 5j) shows that the spacing between adjacent vacancies within each row is ~4.07 Å along the <$1\bar{1}0$> direction, while the separation between consecutive vacancy-containing rows is ~16.28 Å. This variation in vacancy concentration suggests vacancy migration, driven by the natural tendency to redistribute from high- to low-concentration regions. Such vacancy ordering also indicates atomic migration within the bulk, as vacancy migration is inherently linked to atomic diffusion.

Upon cooling, the equilibrium concentration of thermal vacancies decreases, leading to their aggregation. A wider field-of-view HAADF image acquired after cooling from 800 °C to room temperature (Fig. 5k) provides further evidence of vacancy-related features. The image reveals orthogonal low-intensity bands extending along the [$1\bar{1}0$] and [110] directions (yellow arrows in Fig. 5k), consistent with preferential aggregation of vacancies in these regions. Notably, the corresponding diffractogram of the HAADF image (inset of Fig. 5k) lacks the superlattice reflections observed at high temperature (Fig. 5g, h). This discrepancy suggests that cooling disrupts the ordered vacancy configurations, leading to their aggregation into <110 > -oriented bands rather than maintaining long-range ordering.

Quantitative intensity line-profile analysis of randomly selected {110} planes (purple and cyan rectangles in Fig. 5k) is shown in Fig. 5l. The profiles reveal alternating high-intensity (Ni) and low-intensity (Al) columns, characteristic of B2 NiAl and consistent with Fig. 5e. However, Ni columns exhibit significant intensity variations (yellow arrows, Fig. 5l), providing direct evidence of preferential Ni vacancies. By contrast, Al columns display far more uniform intensities, suggesting a lower concentration of Al vacancies. Collectively, these results establish that Ni vacancies dominate in the bulk, driving Ni segregation to the surface and facilitating the subsequent precipitation of γ‘-Ni₃Al.

Density functional theory (DFT) modeling

Our DFT calculations on an β-NiAl bulk model indicate that the vacancy formation energies for Ni and Al atoms are 0.98 eV and 1.20 eV, respectively (Fig. 6a), in agreement with other studies^26,27. The lower vacancy formation energy of Ni suggests that Ni vacancies form more readily than Al vacancies, leading to an increased presence of Ni adatoms on the surface that serves as a sink for ejected atoms from the bulk to result in a higher concentration of thermal vacancies at the elevated temperature.

Fig. 6: Energetics of vacancies and atomic diffusion in β-NiAl.

a Calculated vacancy formation energies in bulk β-NiAl: 0.98 eV for a Ni vacancy and 1.20 eV for an Al vacancy. b Surface exchange mechanism on Al-terminated NiAl(100), in which an Al adatom displaces a Ni antisite (yellow) with an energy barrier of 0.55 eV and a net system energy decrease of 0.31 eV. c, d Diffusion pathways for Ni and Al atoms along the <110> directions (top panel) and <100> directions (bottom panel) and their associated energy barriers. The raw structure files used in the NEB models to compute the diffusion energy barriers are provided in the “Source Data Files” folder.

Additionally, our calculations reveal that once Ni atoms diffuse outward and occupy sites in the top-most layer, an exchange process with Al adatoms on the surface can occur, further increasing the concentration of Ni adatoms at the surface. As shown in Fig. 6b, the Ni antisite atom in the topmost layer (highlighted in yellow) moves outward, allowing an Al adatom (highlighted in dark blue) to replace it, forming a perfect Al terminated NiAl(100) surface. This exchange mechanism results in an overall system energy change of −0.31 eV, with an associated energy barrier of 0.55 eV. An Al-terminated NiAl(100) surface is chosen for these calculations, as indicated by its lower surface energy compared to the Ni surface termination²⁸. This Al termination is further supported by our XPS measurements of Al 2p and Ni 3p spectra acquired using different photon energies, showing an increased Al-to-Ni ratio at lower photon energies (Supplementary Fig. 6), consistent with surface Al-enrichment reported in the literature²⁹.

Furthermore, the analysis of diffusion barriers indicates that, in the bulk, Ni and Al have comparable overall mobility, consistent with previous computational studies of stoichiometric β-NiAl^30,31. Along the <110> direction, Ni migration into an adjacent Al vacancy exhibits a relatively low barrier of 0.41 eV (Fig. 6c), while the reverse process, where an antisite Ni atom returns to a Ni vacancy, requires 0.87 eV. For Al, migration to the next-nearest Ni vacancy is unlikely, whereas an Al antisite atom can spontaneously relocate to a neighboring Al vacancy (Fig. 6d). Along the <100> direction, energy barriers are generally higher, with Ni diffusion requiring 2.13 eV and Al migration being improbable if adjacent sites are fully occupied. However, introducing additional vacancies facilitates consecutive jumps, enabling net <100> migration and significantly lowering the energy barriers to 0.74 eV for Ni and 0.92 eV for Al. These results indicate that, in bulk NiAl, the diffusion rates of Ni and Al are broadly comparable, with differences largely arising from the local distribution of vacancies and antisite defects. However, at the surface, these differences in bulk diffusion are amplified by additional mechanisms. Calculations indicate that Al adatoms can exchange with antisite Ni atoms in the topmost lattice layer, promoting Ni accumulation at the surface. This surface-mediated process, coupled with differences in vacancy formation energies of Ni and Al, drives the preferential formation of Ni-rich γ‘-Ni₃Al precipitates.

Discussion

Intermetallic compounds such as NiAl are widely used in high-temperature applications due to their ordered structures and robustness. However, at elevated temperatures, dynamic processes such as vacancy generation, atomic diffusion, and surface segregation can compromise their stability. These effects are difficult to probe directly due to rapid atomic dynamics and the exponential increase in thermal vacancies governed by Arrhenius behavior. While computational studies have predicted comparable bulk diffusion rates for Ni and Al in stoichiometric NiAl^30,31, our experimental results reveal a different outcome at the surface: Ni preferentially segregate outward, driving the precipitation of γ‘-Ni₃Al. This surface-driven process provides an important experimental benchmark, underscoring the need to account for surface and near-surface kinetics when evaluating the long-term stability of intermetallic alloys, where degradation often initiates.

Unlike oxidation-driven Ni₃Al formation—caused by preferential Al consumption during Al₂O₃ scale growth³²—or bulk decomposition in Ni-rich alloys that require quenching and aging^22,33,34,35, the γ‘-Ni₃Al precipitates observed here form near-stoichiometric NiAl through simple annealing under inert conditions. This indicates that surface-segregation-induced phase separation is an intrinsic behavior, occurring even in the absence of an oxidizing environment. Observed under ultra-high vacuum, this phenomenon reflects a fundamental material response that arises from surface-bulk coupling rather than bulk diffusion disparities or external chemical driving forces. The morphology of the precipitates is highly facet-dependent. On NiAl(100), γ‘-Ni₃Al initially forms a coherent wetting layer that may transform into 3D islands after a critical thickness due to accumulated strain, whereas on NiAl(110), the larger lattice mismatch promotes 3D island formation from the outset. These observations highlight how interfacial strain and crystallographic orientation govern nucleation pathways and growth modes in intermetallic systems.

Our multimodal analysis provides mechanistic insight into this process. EDS measurements confirm localized Ni enrichment after annealing, while in-situ HRTEM and NEXAFS identify the Ni-rich precipitates as γ‘-Ni₃Al. Complementary TEM imaging of bulk regions reveals that Ni out-diffusion from the NiAl matrix generates a substantial population of thermal vacancies, which self-organize into ordered arrangement at high temperature and subsequently aggregate upon cooling. Similarly, FFT patterns of the γ‘-Ni₃Al phase segregated at the NiAl surface also exhibit superlattice reflections, indicative of vacancy ordering. Together, these results establish a direct link between bulk vacancy dynamics and surface segregation, revealing how thermally activated processes in the bulk can propagate to the surface and drive phase separation.

This segregation-induced phase separation has important implications for material performance. Ni-rich surface precipitates reduce the availability of Al required to form a uniform, protective Al₂O₃ scale, thereby promoting non-uniform oxide growth and localized oxidation that compromise the overall oxidation resistance. Beyond oxidation, the altered surface composition modifies catalytic activity by changing the balance of Ni-Al active sites and shifting the electronic structure^36,37, with potential consequences for adsorption, reaction pathways, and selectivity. Moreover, the surface separation introduces microstructural heterogeneity and local strain, which may drive microcracking during thermal cycling and degrade mechanical stability. Over time, continuous Ni depletion also perturbs bulk composition and phase stability, raising concerns for applications requiring precise structural integrity. Addressing these challenges requires the development of targeted strategies, such as alloying, doping, surface coatings, or tailored processing routes—designed to mitigate excessive surface or interface segregation while preserving surface and bulk functionality.

Overall, this work experimentally corroborates that high-temperature environments drive surface phase separation in NiAl alloys. At elevated temperature, thermal vacancies in the bulk facilitate atomic migration, leading to the ejection of Ni and Al atoms to the surface. Due to the lower vacancy formation energy of Ni compared to Al, a greater number of Ni atoms migrate to the surface, where they agglomerate into Ni-rich γ‘ -Ni₃Al precipitates. The morphology of these precipitates is dictated by interfacial misfit strain—low-strain coherent interfaces favor planar wetting layers, while higher strain promotes island-like morphology to minimize interfacial energy. These findings reveal how elevated temperatures alter the surface composition and crystal phase, while simultaneously depleting Ni in the bulk. This dual effect—surface enrichment coupled with bulk depletion—highlights the delicate balance between bulk thermal vacancy dynamics and surface restructuring in the NiAl system, with critical implications for material stability and performance in high-temperature applications.

Methods

In-situ transmission electron microscopy (TEM) analysis

NiAl thin films ( ~ 50 nm thickness) were prepared using electron-beam co-evaporation of Ni and Al onto a NaCl (100) substrate, with the alloy composition controlled by adjusting the evaporation rates of the individual Ni and Al sources. After deposition, the NiAl film was transferred from the NaCl substrate by floating it in deionized water, washed, and mounted on a through-hole MEMS chip, and was then loaded into the TEM column. In-situ TEM experiments were performed using a DENS solution heating holder inside an image aberration-corrected environmental TEM (FEI Titan 80-300) equipped with a differential pumping system. The thin film was annealed at 900 °C and in 0.001 Torr H₂ to maintain a reducing atmosphere and prevent oxidation. This annealing process led to film tearing, exposing atomically clean crystal facets—resembling the mechanical cleavage of crystals in vacuum during surface science experiments. These pristine facets enabled direct observation of surface phase separation dynamics via edge-on TEM imaging.

Ex-situ scanning TEM (STEM) analysis

NiAl foils ( ~ 50 nm thickness) were extracted from bulk NiAl single crystal using focused-ion beam (FIB) lift-out and mounted onto a Mo Omniprobe lift-out grid. The FIB-prepared NiAl foils were further polished using a NanoMill TEM specimen preparation system with Ar+ ions to minimize surface damage and contamination. The NiAl foils were then annealed at 800 °C in 0.001 Torr H₂ for 1 h in a vacuum chamber. Elemental mapping and electron diffraction of the annealed NiAl samples were conducted using a FEI Talos F200X microscope, operated at 200 kV and equipped with a four-quadrant EDS detector. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) imaging, at both low and high magnifications, was performed with a probe-corrected Hitachi 2700 STEM.

Gatan Digital Micrograph Software (V3.50.3584.0) was used for processing HRTEM and HAADF-STEM images.

X-ray photoemission electron microscopy (XPEEM)

Spatially resolved Ni L₃-edge near edge X-ray absorption fine structure (NEXAFS) spectroscopy data were collected using XPEEM at the Electron Spectro-Microscopy (ESM) beamline (21-ID) at the National Synchrotron Light Source II (NSLS-II), Brookhaven National Laboratory (BNL)³⁸. A top-hat-shaped NiAl(100) single crystal (99.9999% purity, 1 mm thick, 8 mm in diameter) was obtained from Princeton Scientific Corp. for this study. The surface of the crystal was cleaned by repeated cycles of sputtering (1.0 × 10⁻⁵Torr Ar at 1 keV for 10 min) followed by vacuum annealing at 750 °C for ~10 min. This process was repeated until sharp c(√2×3√2) R45° reconstruction spots, characteristic of a clean NiAl(100) surface, were observed using low energy electron diffraction (LEED), confirming an atomically clean NiAl(100) surface. Spatially resolved near-edge X-ray absorption fine structure (NEXAFS) measurements were performed in partial electron yield mode, by collecting the secondary electrons with a 2 eV energy analyzer slit centered at the maximum of the secondary electron emission peak. Pixel-wise absorption spectra were obtained by recording a series of XPEEM images at each energy across the absorption edge range in sequential 0.2 eV increments, with the incident X-ray beam at ~17^o grazing incidence relative to the sample surface.

X-ray photoelectron spectroscopy (XPS)

pure Ni-foils (99.999% purity) and the Al(111) top-hat shaped single crystal were measured using with an XPS spectrometer (SPECS Phoibos 150 MCD analyzer) with a delay-line detector. Al-Kα X-ray radiation is used for the XPS measurements after repeated cycles of sputter (1.0 × 10⁻⁵Torr Ar at 1 keV for 10 min) and annealing at 400 °C and 550 °C for the Al(111) and Ni-foil samples respectively. CasaXPS V2.3.26 was used for peak fitting of the XPS spectra³⁹. XPS peak fits of the Al 2p and Ni 3p region were obtained using a Shirley background with 3 arbitrary peaks (30% Lorentzian and 70% gaussian product) being used to define the Ni 3p region, with the remaining counts being attributed to the presence of Al (Al 2p region). An asymmetric line shape was used to define the area associated with Al⁰. The Al³⁺ region was defined using a single GL(30) peak. A pass energy of 50 eV was used to acquire the spectra presented in Supplementary Fig. 5.

Density functional theory (DFT)

DFT calculations were carried out using the Vienna Ab initio Simulation Package (VASP). Energy barriers for atomic diffusion processes were determined using the nudge elastic band (NEB) method. To analyze electron transfer dynamics, the electron localization function (ELF)⁴⁰ was evaluated using the converged electron density from self-consistent field (SCF) calculations. ELF visualization was conducted with the VESTA software⁴¹. For β-NiAl, ELF maps were generated by cleaving the (110) crystallographic plane, where Ni and Al coexist. The ELF intensity profiles along the Ni-Al co-existence direction were extracted and plotted to quantify electronic interactions. Ni L-edge NEXAFS spectra were calculated using the excited-electron core-hole method implemented in VASP 6.2.1^42,43,44. Input files were generated with the Lightshow package^45,46, and many-body final state effects were incorporated through a Kohn-Sham self-consistent solution of the excited state with the Perdew–Burke–Ernzerhof exchange-correlation functional⁴⁷. A full core hole was introduced, where the core electron is placed at the bottom of the conduction band. GW-type projector augmented wave (PAW) pseudopotentials were employed to enhance the accuracy of high-energy scattering state descriptions⁴⁸. Supercells were constructed with sizes larger than 9 Å along each lattice vector to minimize spurious inter-cell interactions. The k-mesh resolution was set with an effective crystal size of 43 Bohr to ensure convergence⁴⁶. Simulated spectra were post-processed with an energy-dependent Voigt broadening⁴⁹ (with a 0.53 eV core-hole life time broadening and a 0.3 eV Gaussian broadening⁵⁰) and aligned using the ΔSCF procedure⁴⁶ for site- and material-specific edge positions. This approach has demonstrated strong agreement with XPS experiments in both molecules and solids⁵¹. Finally, a global energy shift was applied to align the calculated main peak of Ni foil with experiment at 852.6 eV.