S ignificant advances in the synthesis and processing of 2D materials have improved control over optoelectronic properties, enabling a wide-range of device applications. [1][2][3][4] Furthermore, recent device integration of 2D materials with both 1D nanowires and 3D bulk materials has highlighted the need to better understand effects arising from the dimensionality of their interfaces. [5][6][7] Different optoelectronic properties at these interfaces present opportunities for the creation of novel devices. 8 PtSe 2 is a noble-metal chalcogenide that can be either semiconducting (∼1.1 eV, single-layer), semimetallic (if few-layer), or metallic (if thick). 9,10 For materials with a bandgap in this range, the ability to tune band structure is highly desirable for the realization of optoelectronics in the near-infrared range for telecommunication applications. 2,11 PtSe 2 is also a promising material in microelectronics for enabling multifunctional devices that are 3D integrated on Silicon microprocessors due to the relatively low growth and processing temperatures (∼400 C) via vapor phase techniques. 9,10,12 These temperatures are compatible with back end of line (BEOL) integration in Si CMOS processing. In this study, we observe the formation of different Pt-Se phases in situ during high-temperature processing of PtSe 2 and focus on the role interfaces play in the mechanism of their formation. Our observations of the Kirkendall effect in a 2D chalcogenide system opens new opportunities of phase engineering as well as 1D interface engineering in a 2D layer for discovery of new physical phenomena and device applications such as contact engineering.

Diffusion-mediated phase transitions and kinetics of twodimensional chalcogenides such as PtSe 2 are broadly under-stood, where the presence of stable intermediates with a higher metal to chalcogen ratio such as PtSe play a role in the dynamics of the transformation process. 13 The interplay between atomic-scale diffusivity and phase stabilization control the phases that are formed and the underlying microscopic mechanisms. PtSe formation from PtSe 2 occurs through lateral digestion and vertical growth, whereas Pd 2 Se 3 formation from PdSe 2 occurs through vertical fusion. 14,15 The phase transitions in the Pt-Se system are temperature driven, and the crystal structure changes rapidly as the temperature approaches the phase change temperature of the 2D layers. 13,16 This is the primary hurdle in terms of commercial scaling for more exotic 2D materials such as noble metal chalcogenides.

Here, we study the kinetics of temperature-dependent phase transformations of PtSe 2 in situ in the transmission electron microscope (TEM) and ex situ using atomic force microscopy (AFM) and Raman spectroscopy. The in situ experiments involved resistive nonequilibrium heating of 2D PtSe 2 materials ranging from few layers thick to approximately 10 nm in thickness, and an ultrafast detection camera is used to capture the interfacial reaction kinetics of the PtSe 2 to PtSe 1-x transformation. Our experiments are unique not just because of the in situ nature of the study but also because the rate of heating in our study is extremely high (>40 C/sec). This results in vapor pressures, strains, and defect concentrations in samples that are far away from those achieved via slow equilibrium heating in a furnace leading to formation of substoichiometric phases such as PtSe 1-x and other phases such as PtSe that are not observed in equilibrium phase diagram of Pt-Se system. 17 A key observation was that the phase transformation process occurs rapidly at the edge of the 2D layer, which is a 1D boundary, and that vacancies appeared to be involved. Because the interface of a 2D material is 1D, this situation schematically resembles a combination of an interfacial reaction and the Kirkendall effect, except with unique pathways due to the dimensional confinement. The Kirkendall effect involves imbalanced diffusion arising from different chemical diffusivities of two components in a diffusion couple and can be accompanied by interfacial reactions that further contribute to the driving force through a chemical potential gradient. Applying this effect to 3D nanostructures has enabled the synthesis of various hollow core-shell structures. 18,19 The thermolysis of 2D PtSe 2 flakes results in the formation of Se-poor phases with lower Se chemical potential starting at the outer edges. This chemical potential gradient drives faster diffusion of Se from the flake to vacuum and results in the formation of Kirkendall voids in the flake. Heating the sample increases both the vacancy concentration and the diffusion coefficient, resulting in the observed phase transformations (Scheme 1). The interfacial dynamics are completely captured by in situ TEM because the entire material system is electron transparent. By combining in situ and ex situ structural characterization, we gain important insight regarding the temperature-driven PtSe 2 phase transformations. These include the activation energy of the process, the stoichiometry of the chemical intermediates and products (PtSe, PtSe 1-x , and Pt), the dependence of the process on crystallographic orientation. We also characterize the spatial distribution of strain in the layers, which in turn provides indirect information regarding the vacancy concentration.

Understanding the atomic-scale structural characteristics of 2D materials is important to enable their manipulation in functional device applications. We have studied the atomistic transformation of 2D PtSe 2 layers at the atomic scale utilizing in situ vacuum heating inside the TEM. Figure 1a presents atomic models (perspective view, as well as side view) which highlight the van der Waals interactions in this layered material. We have used mechanically exfoliated, (see Experimental Section) few-layer PtSe 2 that was dry transferred to a microfabricated heating chip (see Figure S1). The corresponding height profile is shown in Figure 1b in the AFM (atomic force microscope) topographical image. Nonequilibrium vacuum heating was performed by resistively heating the embedded heating element in the microfabricated chip (see Experimental Section). Temperature is measured with the onchip thermocouple and compared with IR spectroscopic measurements. Figure 1c presents the characteristic Raman spectra of the PtSe 2 flakes before and after heating to 550 °C (see Figure S2, bright-field TEM image before and after heating). We observed a significant intensity reduction of the characteristic Raman peaks of PtSe 2 after heating. The intensity reduction could be consistent with a change in symmetry of the phonon modes due to differences in the crystal structure of the transformed PtSe or PtSe 1-x 13 (see composition measurements, vide infra) and a decrease in the volume fraction of PtSe 2 . To confirm whether the phonon modes changed due to a change in the crystal symmetry, we used near-field, tipenhanced Raman spectroscopy (TERS). TERS from the same area of the transformed PtSe or PtSe 1-x layer (Figure 1c) shows the exact same characteristic Raman peaks (positions of E 2g and A 1g ) as are observed in bare PtSe 2 . The TERS data suggests that the transformed PtSe or PtSe 1-x layer exhibits similar phonon vibrational modes as in the PtSe 2 layer. Because of the lack of any experimental evidence in the literature about the Raman vibrational modes from PtSe or PtSe 1-x , it is difficult to make conclusive statements regarding the phonon interactions in these two materials; further examination, which is beyond the scope of the current work, would be required. We have also analyzed the work function, which is directly proportional to the contact potential difference (CPD) of the transformed phase using Kelvin probe force microscopy (KPFM). As shown in Figure 1d, the CPD map across the edge region of transformed PtSe 2 layer (also see Figure S3, which indicates the specific locations via TEM, topography, CPD mapped images and line profiles) shows a clear change in the work function across the transformed PtSe 1-x phase as compared with the parent PtSe 2 layer.

Figure 1e presents a low magnification, high-angle annular dark field (HAADF) STEM image collected after heating to 550 °C. The contrast in the image shows brighter, higher atomic number domains at the outer edges, consistent with the presence of Se-poor regions, suggesting that thermolysis and phase transformations begin at the outer edges or cracks in the 2D layer of PtSe 2 (see Figure S4). Individual phases have been identified and are denoted with corresponding fast Fourier transformation (FFT) diffractograms (insets) in Figure 1f (see Figure S5). Conversion of the van der Waals layered structure of PtSe 2 to nonlayered, cubic-like phases (PtSe, PtSe 1-x ), as well as complete transformation to Pt particles have also been observed (see Figures S6-S8).

Figure 2a shows the post-transformation structural characteristics of the material. There are three different stoichiometries observed: PtSe 2 , PtSe, and PtSe 1-x . The Se-poor phases, which form from the loss of Se from PtSe 2 , are observed to nucleate and grow at the edge of the PtSe 2 flake. The flake, which is initially uniform, is observed to develop triangular regions of darker contrast in the HAADF-STEM image. These darker regions in Figure 2a indicate decreased high-angle scattering that could be consistent with less material in the area due to triangular voids or vacancy clusters (see Figure S9). Finally, the symmetry of the Se-poor crystals is observed to differ from that of PtSe 2 , as expected. To rationalize these observations, we consider that the activation energy for sulfur vacancy diffusion 20 in MoS 2 is expected to be lower than that of molybdenum vacancies. 21 This would result in a higher diffusion coefficient for the chalcogen compared to refractory metal atoms, especially at higher temperatures. Assuming the chalcogen (Se) vacancies diffuse faster than those of the noble metal (Pt) and combined with the knowledge that PtSe 2 decomposes into Se-poor phases upon heating, 13 we conjecture that the observed structural changes are due to more rapid Se diffusion to the surface, where the decomposition reaction occurs. As the diffusion of Se and Pt is not in balance, the build-up of vacancies in the PtSe 2 is consistent with the Kirkendall Effect. Conventionally, a hollow metal chalcogenide nanocrystal can be obtained via the Kirkendall Effect reacting a metal nanoparticle with a chalcogen; 22 the resulting diffusion across the 2D interface of the 3D material is imbalanced. Our observation of the PtSe 2 structural changes are analogous, except that the material is 2D and the interface is 1D. To rationalize the formation of Pt, we consider that depletion of Se from the sample due to continual flux of Se atoms from PtSe 2 to the edge surfaces and into vacuum would eventually form Pt. As the vapor pressure of Se is 1 Pa at 227 °C (100 kPa at 685 °C), 23 from the original flake, consistent with it having completely depleted its source of Se.

The HAADF STEM image shown in Figure 2b shows that the transformation from PtSe 2 to PtSe 1-x occurs more rapidly in some regions of the sample and exhibits anisotropy. As the Se-poor domains that form are faceted, the anisotropic transformation shows that the growth rate of PtSe 1-x depends on crystallographic orientation. Figure 2b inset is a lower magnification STEM image of the same region in Figure 2b. The region in Figure 2b is magnified to atomic resolution in Figure 2c, showing the atomic arrangement and the facets at the interface between PtSe 2 and PtSe 1-x . The white lines shown in Figure 2c correspond to the [211 0], [101 0], and [011 0] orientations of PtSe 2 . The interface between the PtSe 2 and PtSe 1-x is highly stepped, and there is an epitaxial relationship between the PtSe 2 and the PtSe 1-x , with [011 0] PtSe2 //[100] PtSe1-x and [21 1 0] PtSe2 //[010] PtSe1-x . The micrographs show that the facets at the 1D interface between PtSe 2 and PtSe 1-x have different lengths, consistent with the expected anisotropies in their interface energies. The first order diffraction spots of PtSe 2 have also been marked in the respective FFT pattern, shown as insets of Figure 2c (see Figure S10). The Se-poor PtSe 1-x domain that has formed has a rectangular shape (see the bright contrast extending inward to 2D PtSe 2 layer). We have further examined the elemental compositional of the three different phases. Figure 2d presents energy dispersive X-ray spectroscopy (EDS) maps for the selected nanoscale region, which includes the PtSe 2 and PtSe 1-x phases. The Pt and Se elemental maps show regions with three different levels of stoichiometries which further visualized by spectral intensity plot shown in Figure 2e. By normalizing the EDS spectra intensities with respect to the Pt, we find that regions 1 and 3 are consistent with PtSe 1-x and that region 2 is consistent with PtSe 2 . Figure 2f shows a comparison of normalized spectra for these three regions, indicated by 1 (PtSe 1-x L), 2 (PtSe 2 ) and 3 (PtSe 1-x R) phases. We can clearly observe the Pt to Se ratio is significantly reduced in the two regions, consistent with the formation of PtSe 1-x . We attribute these results to a possible dependence of the stoichiometry of the transformed PtSe 1-x phase on the different 1D interfaces and Se vacancy-based reaction kinetics (see Figure S11).

Figure 3 shows 4D-STEM 24 maps of transformed and untransformed regions of PtSe 2 . Convergent beam electron diffraction (CBED) patterns are collected as a function of spatial position, allowing crystallographic orientation and lattice constant to be obtained as a function of spatial position (Figure 3a). Post transformation, we observe that CBED patterns taken from regions with different contrast in the HAADF-STEM image have different lattice parameters and crystallographic symmetries, as expected based on the Pt to Se stoichiometry (Figure 3b). This directly detects the different phases through the diffraction patterns because PtSe 2 is hexagonal whereas the Se-deficient PtSe 1-x and PtSe are observed to be cubic. Schematically, strain mapping in two component directions is obtained by measuring differences in the spacing between two diffracted beams with g-vectors perpendicular to each other and the transmitted beam in reciprocal space as a function of position. Thus, this only makes sense within a region in which the crystal symmetry is the same. Within a single cubic crystallographic domain, we map the strain distribution and find that the lattice parameter in the two in-plane directions varies significantly, on the order of a few percent. Since the sample stays relatively flat (Rq ≤ 1.25 nm, as measured in the AFM), physical undulation alone in the crystal is unlikely to result in this amount of strain variation. Instead, the lattice constant variation could be consistent with spatially varying selenium vacancy concentration. 25 Our post-transformation HR-STEM results also show contrast consistent with varying vacancy concentration, vide infra, Figure 4. Finally, we perform a strain map across a grain boundary between two cubic Se-poor domains imaged using aberration corrected STEM (Figure 3d and see Figure S12). These show an approximately 6% difference in the two in-plane lattice parameters (Figure 3e). The results suggest that the formation of selenium poor phases at the edge of PtSe 2 is correlated with a local variation in the lattice parameter.

By performing an in situ heating experiments at different temperatures and observing the dependence of the growth rate of the Se-poor PtSe 1-x phase, we can quantify the kinetics of the PtSe 2 transformation process. A schematic of the process by which Se vacancy diffusion drives the PtSe 2 phase change during heating is shown in Figure 4a,b. Imbalanced diffusion results in a vacancy flux into the PtSe 2 flake while regions of differing stoichiometry (Region I, hexagonal PtSe 2 ; Region II, cubic PtSe 1-x ; Region III, metallic Pt) are formed due to interfacial reactions. To measure the growth rate, we examined the in situ TEM video data collected during heating and measured the velocity of the PtSe 2 /PtSe 1-x interface as shown in Figure 4c (also see Figure S13, S14 and in situ Videos S1, S2, and S3). After performing this experiment multiple times at temperatures ranging from 500 to 600 °C, we observe that the interface velocity exhibits an Arrhenius dependence on temperature (Figure 4d). The activation energy obtained by fitting the data is 2.41 eV. This is consistent with the expected diffusion barrier for Se in PtSe 2 , which has been calculated to be approximately 2.56 eV. 26 Because the activation energy for diffusion of an isolated Se vacancy is higher (4.25 eV), our data is consistent with a material with a typical vacancy concentration where interaction of multiple vacancies lowers the activation energy. 27,28 This mechanism involving multiple vacancies also is consistent with the observation of dark triangular regions, which appear to be vacancy clusters or aggregates. While extrinsic contamination in 2D materials is well-known and could potentially impact the nucleation of substoichiometric Pt-Se species, we do not find significant spatial variation in extrinsic components in the EDS spectra in the fresh sample, and most of the extrinsic X-ray signal in the EDS spectra can be attributed to hole count (Figure S15). 29 Furthermore, the dominant variable affecting the temperatureinduced phase transformations is the distance from the flake edge, consistent with a diffusional transformation (as opposed to nucleation and growth). The activation energy extracted from the in situ data is also consistent with this picture.

In summary, we have studied and analyzed the 1D interfacial kinetics and reactions that occur in 2D PtSe 2 layers at elevated temperatures. Ultrafast in situ heating of PtSe 2 has been performed utilizing a microfabricated heating chip coupled with analysis of the phase transformations and accompanying stoichiometric changes. Multiple Pt-Se phases with Se-poor stoichiometries are formed during PtSe 2 heating. PtSe, offstoichiometric PtSe 1-x , and Pt are observed at the 1D interface of PtSe 2 . The Se-poor regions are highly anisotropic, and within Se-poor PtSe 1-x domains, there is significant variation in lattice strain consistent with large variations in vacancy concentration. Darker, triangular contrast in the dark field STEM images of PtSe 2 is also consistent with vacancies arising from imbalanced diffusion and the Kirkendall effect. We observe the kinetics of the phase transformation in situ in the TEM, and by extracting an activation energy, find that the mechanism is consistent with Se vacancy diffusion. Our observations of the Kirkendall effect in a Pt-Se systems suggests that with availability of multiple phases is a given 2D material system there is more room for tunability of the material for contact engineering in device applications. Another implication of this study from the basic science side is the discovery of new and metastable phases that are unknown and yet to be predicted with potentially novel properties such as PtSe 1-x which has been observed for the first time. Finally, the 1D heterojunctions and interfaces occurring at the phase boundaries as reported in this study could potentially host interesting electronic states that have not been uncovered yet since the density of states and spin-orbit coupling can abruptly vary across the 1D interface.

■ EXPERIMENTAL SECTION Materials Synthesis and Processing. Two-dimensional few-layer PtSe 2 flakes were obtained from the bulk single crystal PtSe 2 (purchased from hQ graphene, Netherland) using the Scotch tape method. Exfoliated few-layer PtSe 2 was transferred to prefabricated microfabricated heating chip (Hummingbird Scientific LLC) using dry transfer technique with help of thin PDMS stamp. After transferring to a chip, we utilized a forming gas-based annealing of the PtSe 2 flake up to 300 °C to remove all the residual PDMS contaminant over 2D layer, which was present after dry transfer process.

Surfaces and Spectroscopic Analysis. The thickness of the PtSe 2 layer was determined using atomic force microscopy height imaging (AFM, OmegaScope-R (AIST-NT) setup). Also, Kelvin-probe force microscopy (KPFM) measurement was accomplished with the above-mentioned AFM setup. A gold tip was used and biased by 3 V while the sample was grounded. Tip calibration was done with the help of a freshly cleaved highly oriented pyrolytic graphite (HOPG) sample. Spectroscopic signatures from the PtSe 2 were analyzed using Raman Spectroscopy (LabRAM HR Evolution -Horiba) utilizing 633 nm LASER. Near-field, tip-enhanced Raman spectroscopy was performed using the above-mentioned AFM system, which is attached and upgraded with the LabRam Horiba Raman spectroscopy system. Au-coated OMNI-TERS probes (APP Nano) were used for the measurement of tipenhanced Raman spectrum.

Structural and Strain Analysis. TEM characterization and 4D-STEM strain mapping were performed using a JEOL JEM-F200, and aberration corrected (probe) STEM characterization was performed using a JEOL NEOARM. The 4D-STEM data were collected using STEMx (Gatan) with a Gatan OneView camera (4k × 4k) on the F200. Diffraction patterns were acquired for 0.04 s, and drift correction was performed for each row.