Introduction

Unlike traditional bulk metals with elastic strain normally less than 1%, shape memory alloys (SMAs) can restore large strains exceeding several percent due to superelasticity, resulting from a reversible stress-induced martensitic transformation (SIMT) between B2 austenite phase and B19’ martensite phase1,2. As the most popular superelastic metal used in practical applications, NiTi SMA typically requires cold working followed by alternative heat treatments to produce ultra-fine grains and nano-precipitates3,4. The resultant grain boundary strengthening and precipitation hardening can increase the critical stress for dislocation slip, thereby suppressing the generation and motion of dislocation defects and improving the reversibility and cyclic stability of superelastic deformation5,6,7. However, the cold working procedure restricts NiTi products to simple forms such as thin wires, plates or tubes, significantly restricting their broader application8,9,10,11.

Additive manufacturing (AM) offers a promising method for fabricating components near net shape. The layer-by-layer construction extends design freedom and manufacturing flexibility, allowing customization of complex geometries12,13,14,15. However, dislocations can slip easily in AM superelastic NiTi due to its coarse grain size of several to tens of microns16,17,18. As a result of exacerbated dislocation slip propensity, large plastic irreversibility and poor cyclic stability are two common drawbacks of AM superelastic NiTi19,20,21. Additionally, the recorded total tensile elongation and the ultimate tensile strength of AM superelastic NiTi are consistently less than 8% and 850 MPa, notably lower than commercially available superelastic NiTi22,23. This puzzle is principally because the high local stress concentration around micro-pore defects generated during AM will intensify material damage and lead to premature fracture17,20,22,23.

A variety of strategies have been proposed to improve the tensile superelasticity of AM NiTi22,23,24,25,26,27. In order to improve the yield strength and curb dislocation slip, attempts were made to refine grain size of AM NiTi by increasing the laser scanning speed and adding nucleating agents16,20. However, high scanning speeds often result in irregular and unfused pore defects, significantly compromising tensile ductility17,20,21. The grain size obtained by adding nucleating agents remains in the micron range, originating in limited improvement in yield strength20,21. For the purpose of balancing strength, ductility, and superelasticity, nano-precipitates with size of several to tens of nanometers were introduced in AM superelastic NiTi by adjusting atomic content and applying an aging treatment22,26. After the introduction of nanoscale Ni4Ti3 precipitates, the superelastic strain of AM NiTi was elevated moderately from 4% to 6%, but its total tensile elongation and ultimate tensile strength were far from satisfactory, because stress concentrations around micro-pore defects and hard precipitates were not alleviated effectively22. To date, achieving AM NiTi with balanced strength-ductility-superelasticity synergy is an elusive yet pressing goal22,23,24,25,26,27.

In this study, we use laser powder bed fusion (L-PBF) followed by a tailored two-step heat treatment to produce a single B2 phase NiTi with high-density Ni-rich local chemical inhomogeneity (LCI) entities with excellent strength-ductility-superelasticity synergy compared to reported AM superelastic NiTi. Combining electron diffraction, atomic-resolution imaging and atomistic simulation, we clarify the nature and formation process of Ni-rich LCI entities, reveal critical insights into the role of LCI on dislocation slip, and uncover the interaction between LCI and SIMT. Our work demonstrates that leveraging LCI opens up new horizons to tackle the long-standing challenge of balancing strength, ductility and superelasticity in AM SMAs.

Results

Ni51Ti49 (at.%) powders, which provided an excess of 2 at.% Ni atoms to facilitate the formation of Ni-rich entities, were selected for L-PBF. The L-PBF specimen prior to heat treatment (as-fabricated specimen) was free from unfused holes, keyholes and cracks, while contained a small amount of spherical micro-pores (Supplementary Fig. 1). The as-fabricated specimen was solutionized at 973 K for 4 h (as-solutionized specimen), followed by aging at 523 K for 24 h (as-aged specimen). The solid solution treatment aimed to dissolve Ni4Ti3 precipitates formed during L-PBF (Supplementary Fig. 2), thus increasing the concentration of dissolved Ni atoms in the matrix. The proposed solid solution temperature (973 K) is lower than the conventional range of 1073–1273 K28,29,30 to avoid undesirable grain growth (Supplementary Fig. 3). The aging temperature (523 K), also lower than the conventional range of 723–873 K28,29,30, was chosen intendedly to slow down the diffusion of Ni atoms, thus preventing the formation of long-range ordered Ni-rich precipitates.

The presence of LCI in our as-aged specimen was verified experimentally by various analytical techniques. The X-ray diffraction (XRD) profile in Fig. 1a shows that our as-aged specimen is single B2 without detectable secondary phase, accordingly all the excess Ni atoms are incorporated into the material as solid solution. Fig. 1b displays the radial distribution function (RDF) curve of the Ni-Ni pair obtained from the atom probe tomography (APT) of our as-aged specimen. A clear positive interaction for the Ni-Ni pair at a distance around 1 nm indicates the existence of Ni-rich LCI31. The upper panel of Fig. 1c presents the high-angle annular dark-field (HAADF) image obtained using aberration-corrected STEM. In the lower panel of Fig. 1c, the atomic concentration ratio between Ni and Ti along the arrow in the HAADF image was calculated by the data from energy-dispersive X-ray spectroscopy (EDS). The pronounced fluctuation of atomic concentration clearly reveals that Ni-rich LCI entities are distributed dispersedly in B2 matrix. Additionally since the transformation temperature depends heavily on Ni content, the alternating nanoscale Ni-rich and Ni-poor regions in the as-aged sample will give rise to a sluggish thermally-induced martensitic transformation, manifested by its rather smooth DSC curves (Supplementary Fig. 4). The lattice distortion nearby LCI entities stabilizes the intermediate martensite, namely R phase, and thus the transformation pathway transits to B2 ↔ R ↔ B19’ in the as-aged sample.

Fig. 1: Evidence of high-density Ni-rich LCI entities in as-aged L-PBF NiTi.
figure 1

a Synchrotron X-ray diffraction profile. b Radial distribution function (RDF) calculated from atom probe tomography (APT) results showing the relationship between bulk normalized concentration (Bulk norm. conc.) and the distance of Ni-Ni pairs. Inset is a 3D reconstruction map of Ni atoms from APT measurement. c HAADF-STEM image and the atomic concentration ratio between Ni and Ti along the red arrow. d Selected-area electron diffraction (SAED) pattern under the [113]B2 zone axis. e Dark-field transmission electron microscopy (TEM) image taken by the aperture positioned in the center of the streaked region in (d) (yellow circle). f HAADF-STEM lattice image. g FFT pattern of (f). h Inverse FFT image taken by the diffuse scatterings of (g) (yellow circles). i, j Simulated Ni concentration maps before and after 108 Monte Carlo/Molecular Dynamics (MC/MD) iterations at 523 K, respectively. k Average diameter (d, red line) and number density (ρ, blue line) of Ni-rich LCI entities as well as potential energy per atom (PE, black line) with respect to MC/MD iteration number.

Full size image

The chemical ordering of Ni-rich LCI entities in our as-aged specimen was further characterized through in-depth aberration-corrected STEM experiments. Under [113]B2 zone axis, the selected-area electron diffraction (SAED) pattern (Fig. 1d) of the as-aged L-PBF NiTi exhibits extra diffuse streaks along {112}B2. The marked contrast of the dark-field image (Fig. 1e) taken with the objective aperture positioned in the center of the streaked region (yellow circle in Fig. 1d) demonstrates that Ni-rich LCI entities have distinct ordering from B2 of matrix. Fast Fourier Transform (FFT) pattern (Fig. 1g) of the HAADF lattice image (Fig. 1f) displays diffuse scatterings at 1/7{321}B2 with the loss of long-range periodicity, implying these Ni-rich LCI entities are most likely precursors of Ni4Ti3 precipitates32,33,34. Inverse FFT analysis (Fig. 1h) along with the statistical analysis (Supplementary Fig. 5) showcase that Ni-rich LCI entities in the as-aged L-PBF NiTi have a size ranging from 1.5 nm to 3.8 nm (2.6 nm in average) and a volume fraction of 22.6%. In this sense, these populous entities are medium-range ordered. It is also worth noting that aberration-corrected STEM observation on the as-solutionized specimen does not evidence a general presence of LCI (no sign of superlattice reflection in Supplementary Fig. 2f), thus doubtlessly verifying high-density Ni-rich LCI entities emerge during aging at 523 K.

Hybrid Monte Carlo/Molecular Dynamics (MC/MD) simulation was conducted to rationalize the formation process of Ni-rich LCI entities during aging at 523 K. Due to the existence of a large amount of vacancies in L-PBF NiTi (Table S1), a Ni51Ti49 (at.%) model with 0.1% vacancies was used as the initial input for MC/MD simulation (detailed in Methods). Fig. 1i–j and Supplementary Fig. 6 present the predicted Ni concentration map with respect to the number of MC/MD iterations. As MC/MD goes on, excess Ni atoms are driven by a thermodynamic tendency to aggregate, thus creating high-density Ni-rich LCI entities. The average diameter (d) of these entities increases monotonically (red line in Fig. 1k), and their number density (ρ) tends to saturate at a high level of 1 × 1025 m−3 (blue line in Fig. 1k). Additionally, the potential energy per atom (P.E.) of the model (black line in Fig. 1k) gradually decreases, signifying the stabilization of atomic configuration. In a nutshell, the nucleation and growth of high-density Ni-rich LCI entities are energetically favored by Ni51Ti49 during aging at 523 K.

Figure 2a shows the room-temperature tensile stress-strain curves of the as-fabricated and as-aged L-PBF superelastic NiTi. Our as-aged specimen exhibited an average ultimate tensile strength (UTS) of 958.7 ± 27.1 MPa (mean ± SD, n = 3) with a total elongation of 11.2 ± 0.3%, representing a 72.6% improvement in strength and 11% enhancement in ductility compared to the as-fabricated condition (UTS: 565.2 ± 15.1 MPa, elongation: 10.1 ± 0.3%). Fig. 2b compares the tensile fracture stress and total elongation of our specimens with literatured AM superelastic NiTi. The result indicates that our as-aged L-PBF superelastic NiTi exhibits the highest tensile fracture stress and the largest total elongation. Fig. 2c–e display the tensile superelastic curves and the evolution of superelastic strain over ten cycles for as-aged and as-fabricated specimens. In the initial cycle, our as-aged specimen exhibits a superelastic strain of 7.3%, slightly larger than that of the as-fabricated specimen (6.5%). After ten cycles, the drop of superelastic strain in our as-aged specimen is 0.26%, which is only 1/9 of that in the as-fabricated specimen. From Fig. 2f we can see that our as-aged L-PBF superelastic NiTi exhibits the best combination of total tensile elongation and superelastic strain ever reported22,23,24,25,26,27. Surprisingly the tensile superelasticity of our as-aged L-PBF superelastic NiTi even outperforms commercial superelastic NiTi produced by conventional forming method (Supplementary Fig. 7).

Fig. 2: Tensile strength, ductility, and superelasticity of as-fabricated and as-aged L-PBF NiTi specimens at room temperature.
figure 2

a Tensile stress-strain curves. b Comparison of tensile fracture stress and total elongation between our specimens and reported superelastic L-PBF NiTi in the superelastic state22,23,24,25. Standard deviations derived from three repeated measurements are so small that the corresponding error bars are obscured by the data markers. c, d Tensile superelastic curves over ten cycles for as-aged and as-fabricated specimens, respectively. e Attenuation of superelastic strain as a function of cycling number. f Comparison of tensile superelastic strain and total elongation between our specimens and reported superelastic L-PBF NiTi specimens22,23,24,25,26,27. The superelastic strain in this study is extracted from the 10th superelastic cycle, whereas the literature data are the values from the first cycle. Standard deviations are smaller than the size of the symbols and thus not visible in the plot.

Full size image

Stress-biased thermal cycling experiments were conducted to evaluate the impact of high-density Ni-rich LCI entities on the propensity of dislocation slip. Fig. 3a, b show the strain-temperature curves under constant tensile stresses (biasing stresses) for as-fabricated and as-aged specimens, respectively. Fig. 3c illustrates the irreversible strain generated during stress-biased thermal cycling, whose magnitude scales with the extent of dislocation slip during SIMT. As the applied biasing stress increases from 100 MPa to 350 MPa, the irreversible strain in the as-fabricated specimen increases significantly from 0.2% to 4.4%, whereas little irreversiblity accumulates in our as-aged specimen. A microscopic interpretation on this phenomenon is that dislocation slip is weakened markedly in as-aged L-PBF NiTi.

Fig. 3: Effect of high-density Ni-rich LCI entities on dislocation slip.
figure 3

a, b Strain-temperature curves of as-fabricated and as-aged NiTi specimens under constant tensile stresses. c Evolution of irreversible strain (Irr. ε) with respect to tensile stress in as-fabricated and as-aged specimens. d, e MD-derived cyclic shear stress-strain curves of specimens embedded with a (011)[100]B2 edge dislocation before and after MC/MD, respectively. f, g Simulated dislocation movement in specimens before and after MC/MD, respectively. Green lines represent dislocation configurations at various time points: P0, P2, P3 (before MC/MD) and P0, P2, P5 (after MC/MD). LCI entities are visualized by gray atoms. h, i Simulated evolutions of the average dislocation displacement for specimens before and after MC/MD, respectively. j Generalized stacking fault energy (GSFE) profiles along (011)[100]B2 in specimens before and after MC/MD. Unstable stacking fault energy (USFE) yields the peak value of GSFE. ‘u’ represents the displacement and ‘b’ represents the magnitude of Burgers vector.

Full size image

To further elucidate dislocation motion during superelastic cycling, MD simulation was performed on specimens embedded with a (011)[100]B2 edge dislocation, which belongs to the most energetically favorable slipping system of superelastic NiTi35. As shown by MD-derived cyclic shear stress-strain curves (Fig. 3d, e), the irreversible strain of the specimen before MC/MD increases rapidly and reaches 3.8% after five cycles, which is three times larger than that of the specimen after MC/MD (1.2%). Figure 3f, g illustrate the position and morphology of the (011)[100]B2 edge dislocation in both specimens. Before MC/MD the dislocation maintains relatively straight, while after MC/MD the dislocation is found to interact with Ni-rich LCI entities via shearing mechanism due to the small size of LCI entity and its excellent lattice coherency with the matrix32,33,34. Figure 3h, i depict the relationship between average dislocation displacement and shear strain across five cycles. The average dislocation displacement reaches 12.7 nm in the specimen before MC/MD (Fig. 3h), which is pronouncedly larger than 3.5 nm of the specimen with high-density Ni-rich LCI entities (Fig. 3i). This event is attributed to the LCI-induced dislocation pinning effect, evidenced by the wavy trajectory of the dislocation (Fig. 3g) and the substantial increment of unstable stacking fault energy (USFE) from 619.7 mJ/m2 (before MC/MD) to 719.8 mJ/m2 (after MC/MD) (Fig. 3j). Therefore MD simulation cross validates experimental observation that dislocation slip resistance is reinforced pronouncedly by high-density Ni-rich LCI entities.

To scrutinize the influence of high-density Ni-rich LCI entities on SIMT, an in-situ synchrotron XRD experiment was conducted on our as-aged specimen throughout a tensile cycle. Fig. 4a displays a sequence of XRD profiles captured during superelastic deformation. Fig. 4b shows the volume fractions of B2, R and B19’ phases as functions of applied strain, which are determined by the percentage of diffraction peak areas of B2 (100), R (10\(\bar{1}\)) and B19’ (001) in Fig. 4a. These results indicate that our as-aged specimen follows the B2 → R → B19’ pathway upon loading and the B19’ → R → B2 pathway during unloading. Loading to 6% tensile strain causes 90% of B2 phase to convert to B19’ phase and the rest to R phase. Unloading to stress-free state does not leave B19’ or R phase, indicating SIMT and its reverse are thorough in our as-aged specimen.

Fig. 4: Effect of high-density Ni-rich LCI entities on SIMT.
figure 4

a In-situ synchrotron X-ray diffraction profiles captured during a tensile cycle in our as-aged specimen. b Evolution of the volume fractions of B2, R and B19’ phases during a tensile cycle in our as-aged specimen. c, d Simulated atomic configurations of NiTi specimens before and after MC/MD during tensile loading at strains of 0%, 2% and 8%. B2 phase and B19’ phase are colored in white and red, respectively. The other lattice structures are colored in gray. e MD-derived evolution of B19’ phase volume fraction with applied strain of NiTi specimens before and after MC/MD.

Full size image

The interaction between high-density Ni-rich LCI entities and SIMT is examined by MD simulation. Fig. 4c shows that the specimen before MC/MD manifests rather homogeneous nucleation of B19’ phase, while in the specimen after MC/MD (Fig. 4d), SIMT is initiated in a more heterogeneous fashion. Some atoms were neither recognized as B2 phase nor B19’ phase due to the large lattice distortion nearby. They are categorized artificially as “other” phase, which are indeed Ni-rich LCI entities. The regions adjacent to Ni-rich LCI entities (with evident lattice tension/compression and depleted in Ni, Supplementary Fig. 8) are energetically favored sites for the commencement of SIMT, because lower Ni content and larger lattice distortion reduce the nucleation energy barrier of B19’ phase36,37. Fig. 4e illustrates that the volume fractions of B19’ phase for both specimens evolve in similar manner with respect to the applied strain. The saturated volume fraction of B19’ phase in the model after MC/MD is more than 90%, in reasonable agreement with the synchrotron XRD results.

Discussion

The undesirable compromise among strength, ductility, and superelasticity has become the Gordian knot of AM NiTi22,23,24,25,26,27. The introduction of high-density Ni-rich LCI entities can overcome this issue and make our as-aged L-PBF NiTi a good candidate for critical applications such as load-bearing and self-centering devices. Given systematic microstructural characterizations and atomistic simulations in this study, we will elucidate why high-density Ni-rich LCI entities are capable to improve the strength, ductility, and superelasticity of L-PBF NiTi substantially and synergistically.

Recoverability and cyclic stability are two critical metrics of superelasticity9,10. We start with deciphering the origin of recoverability in our as-aged specimen from three aspects. First, compared to the well-documented nano-precipitation (interspacing d ∼ 20 nm, average size D ∼ 50 nm and number density ρ ∼ 3 × 1023 m−3) used to strengthen L-PBF NiTi26, our Ni-rich LCI entities manifest more than an order higher density (ρ ∼ 1 × 1025 m-3) and much smaller size and interspacing (d ∼ 4.8 nm, D ∼ 2.6 nm), which renders more frequent intersection with dislocations. Therefore the dislocation slip is suppressed more effectively and the plastic irreversibility stemming from dislocation slip is eliminated more drastically38,39,40,41,42. Second, the crystallographic compatibility during SIMT is improved by the intermediate R phase (Fig. 4a, b), so possibility of the activation of transformation-induced dislocations at the phase interface is decreased43. Third, unlike secondary phases (such as Ni₄Ti₃) with long-range ordered crystal structures that hinder martensitic transformation during loading44, our Ni-rich LCI entities may undergo SIMT because of their small size and coherency to matrix (Fig. 4d), so the saturated volume fraction of B19’ phase and the resultant peak superelastic strain are maintained at high levels in the as-aged specimen (Fig. 4e). To recapitulate, inhibited plastic irreversibility and high superelastic strain result in extraordinary reversibility of our as-aged sample.

The superior cyclic stability in our as-aged specimen not only shares the reasons for superb recoverability mentioned above, but also benefits from the high mechanical stability of LCI entities38,40,41,42. Figure 1b, h elucidate that the present Ni-rich LCI is medium-range ordered and its compositional fluctuation is far from the statistical value in a random case. This kind of LCI is able to sustain repetitive SIMT (10 cycles in this work) and dislocation slips with a fairly high probability. As a result, the degradation of the superelastic response, such as the decrement of transformation stress, the drop of hysteresis stress and the decline of superelastic strain, can be well controlled upon cycling (Fig. 2c)45,46,47.

We finally shift our focus to the excellent tensile ductility and fracture strength in our as-aged specimen. At the atomic scale, SIMT involves collective uniform lattice shear9,10, which differs significantly from dislocation slip featured with highly localized lattice distortion44,48. Since the deformation of our as-aged L-PBF NiTi is mostly accommodated by SIMT (Fig. 3b), the uniform lattice shear of SIMT greatly releases the local stress around micro-pore defects and postpones catastrophic failure thereafter49. In addition, traditional strengthening of L-PBF NiTi was based on precipitation hardening arising from dislocation bypassing of long-range ordered precipitates, such as Ni4Ti347,50. However, the strong microstructural inhomogeneity between precipitates and matrix introduces significant local stress concentration, thereby easing crack initiation and limiting tensile ductility48,51. In contrast, Ni-rich LCI entities share better lattice compatibility with the matrix due to their medium-range ordered essence, and they interact with dislocations through shearing mechanism (Fig. 3g)40,41. The resultant LCI-induced strain hardening becomes pronounced in the as-aged sample when dislocations are massively activated at a strain higher than 6% (Supplementary Fig. 9). In this way, the as-aged sample is ductilized by reduced local stress concentration and strengthened by LCI-induced strain hardening. All these aspects jointly settle the strength-ductility tradeoff in L-PBF NiTi.

To conclude, the study presents an strategy to promote the tensile superelasticity of L-PBF NiTi. A specialized microstructure, i.e., high-density Ni-rich LCI entities (with a measured volume fraction of 22.6% and an average size of 2.6 nm) distributed uniformly in the B2 matrix, was formed after a tailored two-step heat treatment, realized by the well-controlled diffusion of excess Ni atoms. In lieu of other approaches to strengthening L-PBF NiTi such as nano-precipitation, our medium-range ordered Ni-rich LCI entities are unique to increase the resistance against dislocation slip, promote SIMT, and relieve local stress concentration around micro-pore defects and entity interface, thereby enhancing the tensile strength, ductility and superelasticity of L-PBF NiTi synergistically. Among literatured AM superelastic NiTi, our L-PBF superelastic NiTi with high-density Ni-rich LCI entities exhibits excellent tensile superelasticity, i.e., tensile ultimate strength of 958.7 MPa, total tensile elongation of 11.2%, superelastic strain exceeding 7%, and superior cyclic stability. The findings from this work are expected to inspire the development high-performance AM SMAs with LCI engineering.

Methods

Material processing and heat treatment

NiTi spherical powders with size ranging from 15 to 53 μm were gas-atomized from a Ni51Ti49 (at.%) ingot. The assembled powder was additively manufactured via L-PBF using an Eplus M100-T 200 W system. Optimized processing parameters (laser power of 120 W, scanning speed of 800 mm/s, hatch spacing of 110 μm, and powder layer thickness of 30 μm) were used to fabricate rectangular plates with dimensions of 70 × 8 × 3 mm3. Subsequently these plates were machined to specimens with required sizes for various tests. The conditions for solid solution and aging are 973 K 4 h and 523 K 24 h, respectively.

Room temperature tensile tests and stress-biased thermal cycling experiments

Specimens with dimensions of 70 × 2.6 × 0.9 mm3 were sliced from L-PBF rectangular plates via electrical discharge machining. Tensile tests were conducted using a KQL universal testing machine at a strain rate of 5 × 10−4 s−1. Experiments were repeated three times to ensure the reproducibility and reliability of the data (Supplementary Fig. 10). Stress-biased thermal cycling experiments were performed by a KQL universal testing machine with a temperature controller, with a heating and cooling rate of 10 K/min. An Epsilon Technology Model 3442 extensometer was utilized to ensure precise strain measurement.

Differential scanning calorimetry (DSC) analysis

The transformation temperature of the specimens was analyzed through DSC measurements, conducted with a NETZSCH DSC 214 instrument at a cooling/heating rate of 5 K/min.

TEM characterization

FEI Tecnai G2 F20 was used for bright-field and dark-field imaging, as well as SAED. HAADF and super four-probe EDS equipped on the FEI Titan Themis were used to analyze the atomic-resolution structure and elemental distribution.

In-situ synchrotron high-energy X-ray diffraction measurement

In-situ synchrotron HE-XRD was performed on the 11-ID-C beamline of the Advanced Photon Source at Argonne National Laboratory, USA. High-energy X-rays with an energy of 115 keV, a wavelength of 0.1173 Å and a beam size of 0.5 × 0.5 mm2 were diffracted in transmission geometry towards a PerkinElmer large area detector to obtain two-dimensional diffraction patterns. One-dimensional XRD line spectra were generated from the two-dimensional diffraction patterns using the FIT2D software package. Gaussian fitting was used to determine the areas of diffraction peaks. Errors in the measurement of relative peak intensities were estimated to be less than 0.1%.

Atom probe tomography measurement

Needle-shaped specimens were prepared by a standard electropolishing technique. APT were run on a LEAP 4000X Si (CAMECA Scientific Instruments) using a laser wavelength of 355 nm. APT acquisitions were carried out at a temperature of 20 K, a pulse frequency of 200 kHz, a pulse energy of 40 pJ, and an evaporation rate of 0.6%. APT data reconstruction was accomplished using the shank angle algorithm. APT data analysis was performed using the IVAS 3.8.4 software.

Atomistic simulation

In this study, all atomic models were built by open-source program ATOMSK52. The initial NiTi configuration is provided as Supplementary Data 1, and the aged NiTi configuration with Ni-rich LCI entities is provided as Supplementary Data 2. Molecular dynamics (MD) simulations were performed using the software package LAMMPS53 within NPT ensemble. Interatomic interactions were characterized using a second nearest-neighbor modified embedded-atom-method (2NN MEAM) potential54. The Ni51Ti49 (at.%) model was established by replacing 1 at.% Ti with Ni randomly in the Ni50Ti50 (at.%) prototype with ideal B2 configuration. Equal amounts of Ni and Ti atoms were then removed to create vacancies.

Hybrid Monte Carlo MD (MC/MD) simulation was carried out at the aging temperature of 523 K. The dimensions of the simulation cell are 9.4 × 9.6 × 9.4 nm3 with x // [011]B2, y // [21\(\bar{1}\)]B2, and z // [\(\bar{1}\)1\(\bar{1}\)]B2, respectively. Periodic boundary conditions (PBCs) were applied in all three orthogonal directions. At each iteration, an attempted swap of one randomly selected Ni atom with one randomly selected Ti atom was performed using the Metropolis algorithm. 100 MC swaps were conducted between each MD step.

Based on the same simulation cell, the tensile response was simulated via MD method at a temperature above Af (prescribed as 500 K). PBCs were set for all three orthogonal directions. After relaxation for 100 ps, a displacement was applied along x direction such that the specimen was deformed uniaxially with a strain rate of 1 × 109 s−1.

The mobility of dislocation was investigated under cyclic shear-unloading via MD method. The size of the simulation cell was changed to 12.2 × 12.8 × 8.5 nm3 with x // [100]B2, y // [01\(\bar{1}\)]B2, and z // [011]B2. A (011)[100]B2 edge dislocation was inserted near the mid-plane of the cell. PCBs were imposed along x and y directions. Following relaxation for 100 ps, a displacement was applied to the upper z surface along x direction, while the lower z surface was fixed. The specimen was sheared repeatedly for 5 times at a shear strain rate of 1 × 109 s⁻¹.

To evaluate the energy barrier during the course of dislocation slip, GSFE landscape was constructed using the rigid shear method. PBCs were set for x // [100]B2 and y // [01\(\bar{1}\)]B2. Divided by the slip plane of (011)[100]B2 edge dislocation, the upper half of the cell was rigidly displaced along x, while the lower half of the cell was fixed. GSFE was equal to the free energy difference per unit area between the dislocated state and the initial state. USFE was equal to the peak value of GSFE.

Statistics and Reproducibility

Tensile tests (n = 3 per condition) were conducted three times per condition to ensure consistency. Superelastic tests, stress-biased thermal cycling tests and DSC measurements (n = 1 per condition) tests were conducted on a single representative specimen for each condition. Microstructural analyzes (SEM, TEM) covered multiple regions (5 for as-aged, 3 for as-fabricated) from a single sample, while APT data were collected from one representative tip. In-situ synchrotron XRD tracked phase evolution dynamically in one specimen. Molecular dynamics simulations used a single model but validated trends through five loading cycles. No data or samples were excluded from analysis.