Formation and Healing of Defects in Atomically Thin GaSe and InSe

: GaSe and InSe are important members of a class of 2D materials, the III-VI metal monochalcogenides, which are attracting considerable attention due to their promising electronic and optoelectronic properties. Here an investigation of point and extended atomic defects formed in mono-, bi-, and few-layer GaSe and InSe crystals is presented. Using state-of-the-art scanning transmission electron microscopy (STEM), it is observed that these materials can form both metal and selenium vacancies under the action of the electron beam. Selenium vacancies are observed to be healable; recovering the perfect lattice structure in the presence of selenium or enabling incorporation of dopant atoms in the presence of impurities. Under prolonged imaging, multiple point defects are observed to coalesce to form extended defect structures, with GaSe generally developing trigonal defects and InSe primarily forming line defects. These insights into atomic behavior could be harnessed to synthesize and tune the properties of 2D post transition metal monochalcogenide materials for optoelectronic applications.

3 Layered III-VI post transition metal monochalcogenides, such as GaS, GaSe, InS, and InSe, have recently generated much interest within the expanding field of 2D materials beyond graphene. 1 In particular, GaSe and InSe have shown excellent optoelectronic properties, with strong second harmonic generation reported in monolayers of both materials, 2,3 and potential photonic applications. 4,5 These almost lattice-matched films possess high electron mobility (values over 1000 cm 2 V -1 s -1 reported in few layer InSe devices) and thickness-dependent, tunable band-gaps, 6-10 a desirable combination lacking in graphene and most transition metal dichalcogenides (TMDs). In contrast to most TMDs (for example MoS2), [11][12][13] both GaSe and InSe possess a direct bandgap in bulk form, which transitions to an quasi-direct bandgap when reduced to the monolayer limit. [14][15][16] These properties have been successfully applied in few-layer photodetection and field effect devices, 6,9,[17][18][19][20][21] and when used together in GaSe/InSe heterostructures, have shown tunable photoemission from interlayer excitons. 10 The performance of devices based on GaSe and InSe can be affected by the presence of impurities or by exposure to air, light and/or moisture. 7,22,23 As an example, it is generally considered that the presence of selenium vacancies in InSe and their interaction with ambient oxygen species underpins the loss of ambipolarity in InSe FET devices, 23 and is also a cause of the lower ambient stability of InSe relative to expectation. 24 However, in order to fully understand the role of point defects in determining the properties of optoelectronic devices based on III-VI post transition metal monochalcogenides, it is necessary to characterize their structure and dynamic behavior. Previous transmission electron microscopy (TEM) investigations of these few layer monochalcogenides have been limited to structural verification of monolayer GaSe 25 and InSe crystals. 3 Here the formation and stability of atomic vacancy and extended defects in monolayer and few-layer GaSe and InSe is investigated using low voltage, aberration-corrected 4 scanning transmission electron microscope (STEM) imaging and image simulation, combined with density functional theory (DFT) calculations and electron energy loss spectroscopy (EELS).
The crystal structure of GaSe and InSe consists of van der Waals bonded layers, with each layer comprising four parallel planes of atoms in the order VI-III-III-VI (e.g. Se-Ga-Ga-Se, see Figure 1). The individual layers exhibit three-fold honeycomb symmetry and are covalently bonded in-plane. As with most 2D materials, the layer stacking sequence is essential in determining their band structure, with 2Hb and 3R stacking (commonly referred to as ε and γ phases) being the most thermodynamically stable polytypes for bulk GaSe and InSe, respectively. 26 These polytypes are illustrated in Figure 1a, 1b, 1e and 1f, and have unit cells 5 containing n layers, where n = 2 for 2Hb and n = 3 for 3R, giving rise to hexagonal (P-6m2) and rhombohedral (R3m) stacking symmetries, respectively.
To enable atomic resolution characterization of these highly beam sensitive materials, few layer GaSe and InSe crystals were encapsulated between two graphene sheets. Graphene encapsulation in an argon glovebox has been previously employed to improve the stability of 2D materials when exposed to high resolution imaging with a focused STEM electron probe. [27][28][29] The graphene also serves to isolate the material from the air but trace oxygen, carbon, hydrogen and silicon impurities can become trapped beneath the graphene sheets, even when dry peel encapsulation is performed in an inert atmosphere. 30 The trapped contamination provides a source of impurities and therefore enables the study of lattice doping with atomic resolution.

Results/Discussion
InSe and GaSe TEM specimens were prepared by exfoliation of the bulk crystals in an inert argon gas environment to produce flakes that ranged in thickness from 1-5 layers, as determined by the optical contrast (Figure 1c and g). The materials were then encapsulated between two graphene sheets to protect them from atmospheric degradation as well as to reduce electron beam induced radiation damage. 27,28 Selected area electron diffraction (SAED) patterns agree with those expected for few layer GaSe and InSe and also show the reflections associated with the two encapsulating graphene sheets (Figure 1d and h). Monolayer regions are identified using optical contrast prior to encapsulation and by atomic resolution annular dark field (ADF) STEM imaging, with the latter demonstrating an excellent qualitative match to ADF multislice image simulations as shown in Figure 2. The encapsulating graphene layers have been neglected from the ADF STEM defect simulations as their relative lateral position is unknown and they were found to have a negligible effect on the relative intensity of different columns (as shown in 6 Figure S1). In GaSe, the relative positions of the gallium and selenium lattice sites could not be distinguished from the ADF intensity alone. However, as the selenium-selenium spacing projected along the z-axis is larger than the gallium-gallium spacing, a given tilt away from the [001] zone axis causes a greater lateral shift for the selenium atoms than the gallium, revealing In GaSe, vacancies are observed on both gallium and selenium lattice sites with the defect in Figure 3b showing a good match to the expected contrast of a gallium vacancy. Selenium site defects were also observed but comparison with simulations revealed that these defects are likely to be substituted with light atom impurities (Figure 3f). The most likely available light dopant species is considered to be oxygen, but other light impurities, such as nitrogen or carbon, are also possible, as discussed later. Point vacancy defects are also observed in monolayer InSe but here no evidence was found to support light atom substitution. For example, in Figure 4b, the defect at the indium site is a good match to a pristine indium vacancy and the defect at the selenium site in Figure 4f is a better match to a selenium vacancy, rather than a light atom substitution (such as oxygen). The effect on the ADF intensity of a selenium site defect is primarily observed not at the selenium site itself but at the adjacent metal sites. This is due to the missing chalcogenide ion  from frame to frame. This, together with a small sample tilt, prevents a fully quantitative analysis of the dopant species from these images.
In order to better understand the stability and mobility of point defects in this class of 2D materials, beam-induced defect migration in both monolayer GaSe and InSe is investigated ( Figure 5). In monolayer regions of both materials with low contamination, vacancies are observed to form, travel to adjacent crystal positions and then recombine with a relevant adatom to completely recover the perfect lattice (healing). In relatively clean areas of the sample, most single point defects heal within approximately 10s of continuous imaging (cumulative electron fluence of ~1.5 x 10 6 e -Å -2 ). In this time vacancy hopping behavior is observed to occur in to nearest neighbor or next nearest neighbor positions and defect diffusion is generally limited to <2 nm from the initial formation site before healing.
The adatoms liberated from vacancy formation are found to be far more mobile than the vacancies themselves, as they are generally not observed in the vicinity of their vacancy during imaging. This fast surface diffusivity could explain why selenium vacancies are sometimes filled by impurity species rather than the original selenium atom, even in apparently clean regions of the crystal. In contrast, in crystal regions with high levels of trapped contamination, the point defects did not re-heal but rapidly multiplied, resulting in the local destruction of the crystal at an electron fluence of <1 x 10 6 e -Å -2 . This difference in behavior for clean versus contaminated crystal regions suggests than when impurity species are present in the vicinity of a selenium vacancy, dopant substitution is favored over recovery of the original lattice. This conclusion is supported by the DFT calculations which found that oxygen substitution is energetically more favorable than healing of the original lattice (see Figure 5l).
In order to understand the mechanism of defect formation, the elastic Mott scattering cross section the maximum energy transfer and the displacement threshold energies for the experiments have been calculated. These calculations demonstrate that the intrinsic knock-on damage threshold for selenium (or gallium/indium) in GaSe or InSe is considerably greater than the energy available in an 80 keV electron beam, even for atoms adjacent to pre-existing defects (see Supplementary Figure S4). It is therefore unlikely that the irradiation induced defects observed are produced via direct knock-on damage (sputtering) of the material. In contrast, light atom molecules and hydrocarbons are well known to be highly susceptible to knock-on damage at 80 kV. 31,32 Consequently, the unavoidable presence of an oxygen-containing hydrocarbon residue trapped between the encapsulating graphene sheets and the III-VI crystal is likely to serve as a source of oxidative radical species generated through interaction with the electron beam. It is therefore suggested that selenium vacancies can be formed by interaction with oxidative species produced through radiolytic, interatomic Auger decay processes. 33,34 Chalcogenide vacancies have been predicted as energetically favorable sites for O2 and H2O adsorption and dissociation in both GaSe and InSe, 23 Table   S1).
13 appears due to stacking faults that develop in the crystal; differences in crystal stacking are considered in detail later). MoS2, 38 and parallel to the edges of the trigonal defects observed in GaSe. At high vacancy concentrations these linear defects coalesce to form trigonal gaps or islands (Figure 6g). It is possible that the presence of intrinsic trigonal defects in GaSe may contribute to its lower ambient stability relative to InSe, due to the introduction of exposed edges that are susceptible to chemical attack.
14 The difference in the observed morphologies of the extended defects can be assigned to the different behavior of the post transition metals (gallium and indium). Trigonal lattice defects like those seen in GaSe tend to form when the energy needed to remove atoms is similar for both species (as for electron beam induced knock-on damage of hBN). [42][43][44] Linear defects form in TMDs where the probability of removing the chalcogenide is much greater than the transition  the crystal, forming a thicker amorphous oxide layer at the edge. In addition, the hydrocarbon bubble has an outer rim (~5 nm thick) that is depleted in oxygen relative to the center (Figure   7g), supporting the hypothesis that hydrocarbon contamination is the source of oxygen.
Finally, the presence of multiple stacking faults in both GaSe and InSe few layer crystals is identified. The ability of van der Waals crystals to exist in multiple different low energy stacking configurations, each with different optical and electronic properties, can provide a potential route for tuning the material's behavior for different applications. [49][50][51][52] Both GaSe and InSe crystals are known to occur in several different polytypes that are structurally equivalent except for their stacking sequence. 26,53 However, the optoelectronic properties, such as the bandgap, are predicted to vary for different polytypes. 14 It is therefore important to understand the presence and nature of stacking defects in order to accurately interpret optoelectronic device behavior.  Figure S9), similar to the 60° shear partial dislocations previously observed in graphene bilayers, which generate AB ↔ AC transitions. 54 The boundary between the two pristine stacking regions extends over a distance of greater than 10 nm, similar to faults reported in other 2D chalcogenides. 55,56 Previous work by Zhou et al has identified a γ ↔ β stacking transition in bilayer InSe, 3 but their characterization found the boundary to be atomically abrupt (discussed in the Supporting Information). The difference between this and the data presented in this work is easily attributed to the different synthesis and processing for their material: Zhou et al studied InSe synthesized by a bottom-up PVD method, whilst this work utilizes top-down mechanically exfoliated materials. It is suggested that in bilayers, abrupt stacking boundaries result from different stacking regions impinging during crystal growth, whereas the extended boundaries observed here are not intrinsic to the bulk crystal but form as a result of mechanical stress during mechanical exfoliation and encapsulation. 57 The extended stacking faults are of interest as they introduce sizable regions of different crystal symmetry. This is particularly relevant to InSe, where honeycomb-like hexagonal polytypes generally do not form spontaneously, as opposed to GaSe which exists in four stable stacking phases (β, γ, δ, and ε), of which the β polytype (also called 2Hc, with P63/mmc symmetry) possesses honeycomb symmetry. 26,58 The authors are not aware of any previous experimental or theoretical works that report AA-stacked (1H) InSe or GaSe materials, hence the known effects of hexagonally symmetric β-stacking as well as the possible effects of AAstacking are considered briefly here. In both GaSe and InSe, when ε-or γ-stacking transitions into β-stacking, it will have an effect on the rules governing spectroscopic selection; since βstacking is the only stacking order to exhibit inversion symmetry. In vibrational spectroscopy, IR and Raman activity are mutually exclusive in the presence of inversion symmetry, thus a different response is expected in β-compared to ε-/γ-stacked bilayers. In the bulk limit, ε-GaSe is known to exhibit two Raman active modes around 210-215 cm -1 , 59 while only one of those is Raman active in β-GaSe. 58 The splitting between the two modes is small which would make it challenging to identify ε-or β-stacking from the Raman spectrum alone, demonstrating the importance of gaining a full microstructural understanding of the material when interpreting optical data.
Similarly, the photoluminescence (PL) excitation energies of InSe and GaSe films are also modulated by stacking order, which matters for both β-and AA-stacking, although this effect is small due to spin-orbit coupling. In the absence of spin-orbit coupling, symmetry forbids the lowest energy optical transition, and while this restriction is lifted by band mixing promoted by spin-orbit interaction, 60 it remains weak, hence the absence of the lowest energy PL line in monolayer InSe. 7 In contrast, whilst ε-or γ-stacking activates this transition, in AA-stacking the retained z/-z reflection symmetry of the crystal forbids the transition which is only enabled by spin-orbit interaction, while in β-stacking the same occurs due to the presence of inversion symmetry. Stacking faults could, in principle, be observed by mapping the PL intensity, as a drop in intensity is likely to occur at AA-and β-stacked regions. However, the relatively poor

Methods/Experimental
γ-InSe and ε-GaSe single crystals were grown by a modified Bridgman method 62,63 from high purity(>99.999%) gallium, indium, and selenium source materials. From these bulk crystals, few-layer specimens were mechanically exfoliated by peeling with Nitto Denko BT-150E-CM tape before being pressed onto a Si/SiO2-90nm wafer (heated to 60°C) which had a thin (~200nm) spin-coated film of poly(propylenecarbonate) on the surface (to improve adhesion).
Suitable flakes were then identified using optical microscopy and picked up with monolayer graphene using the poly(methyl-methacrylate) (PMMA) dry peel transfer technique using a bespoke micromanipulation setup. Following this, a second flake of graphene was picked up from Si/SiO2-90nm fully encapsulating the specimen while on the PMMA membrane.
Exfoliation and transfers took place within an argon glovebox to prevent sample degradation. 64,65 After full encapsulation with graphene on each side, the specimen was removed from the glovebox and transferred onto a custom SiN TEM grid, using the PMMA dry transfer technique, cutting around the PMMA membrane. To promote adhesion between the specimen and the grid, annealing was performed at 130°C for 5 minutes. To remove polymer on the surface of the specimen, the sample was cleaned by full immersion in acetone, IPA and hexane for 5 minutes each. The specimen was then removed and flash dried with nitrogen to remove any surface solvent residue. Multislice ADF STEM image simulation was performed using QSTEM, 66 using the above experimental parameters, source size of 1.1 Å, at Scherzer defocus and averaged for 25 randomized phonon configurations. Reference bulk crystal structures were obtained from the Inorganic Crystal Structure Database 67 and processed using the Atomic Simulation Environment. 68 Structural models were rendered using VESTA. 69 EELS analysis was performed using HyperSpy v1.3, 70 with principle component analysis dimensionality reduction to reduce the noise level followed by model fitting of the elemental edges. 71 To aid visualization, all ADF STEM images were Gaussian filtered.
Density functional theory (DFT) as implemented in the VASP code 72 was used to calculate the formation energies of point defects in 5x5 supercells of monolayer InSe and GaSe, and the interlayer binding energies of pristine bilayer InSe and GaSe. All calculations used a plane-wave basis with a cutoff energy of 600 eV and neglected spin-orbit coupling. A 20 Å separation in the vertical direction was included to model the freestanding 2D crystals in a 3D cell, and in the point defect calculations a 6x6x1 Monkhorst-Pack k-point grid was used. The presence of graphene was neglected in the calculations as no significant charge transfer is expected from graphene to InSe or GaSe, based on recent ARPES measurements of monolayer GaSe with graphene contacts where the Dirac point of graphene is found to be located over 2 eV above the valence band edge of GaSe and over 1 eV below the conduction band edge. 73 For InSe, while charge transfer is possible for n-doped layers, it is small as the layers are atomically thin; therefore, the integral value of the available charge is small.
In the point defect calculations, a full structural relaxation was performed on both InSe and GaSe monolayers in a 5x5 supercell, containing one of the following point defects: a single selenium vacancy, single metal vacancy, metal-selenium divacancy, selenium-selenium divacancy, and selenium to oxygen substitution. The total energies were compared to that of the relaxed pristine 5x5 supercell and defect formation energies were computed as the difference between the total energy of the system after and before the introduction of the defect.

Supporting Information.
The Supporting Information is available online. All authors commented on the manuscript.