Electronic structure of the Si(1 1 1):GaSe van der Waals-like surface termination

. The electronic structure of the Si(1 1 1):GaSe van der Waals-like surface termination has been determined by angle-resolved photoelectron spectroscopy using photons in the energy range hν = 12–170 eV supplied by the BESSY and ALS synchrotron light sources. The Si(1 1 1):GaSe surface is isoelectronic to the passivated Si(1 1 1):H and Si(1 1 1):As surfaces, and also reﬂects the principal building block of layered chalcogenide GaSe single crystals. The electronic structure is discussed in relation to these systems. The chemical bond between the Si and Ga surface atoms is non-polar and therefore similar to the Ga–Ga bond in GaSe single


Introduction
Silicon forms the backbone of modern microelectronics. However, its indirect band gap precludes widespread use in optoelectronic applications. Integration of commonly used optoelectronic materials into silicon technology is hindered by the large lattice and chemical mismatch between Si and III-V or II-VI semiconductors such as GaAs, InAs or ZnSe. Direct incorporation of these materials into silicon-based devices through heteroepitaxy will require development of suitable buffer layers to prevent interdiffusion and to relieve interface strain without creating optically active defects.
van der Waals-epitaxy may provide ultrathin buffer layers for epitaxy of lattice mismatched materials [1,2]. The first step is to establish a stable interface between Si and layered materials. The Si(1 1 1)/GaSe interface is the most intensively studied interface in van der Waals-epitaxy because the close matching between the Si(1 1 1) surface and bilayer building block of the layered GaSe crystal leads to a surface termination with outstanding properties. The van der Waals-surface termination formed by a Ga-Se bilayer, which corresponds to a half-sheet of the layered semiconductor GaSe, on Si(1 1 1) is extensively described in the literature [3]- [15]. It provides a chemically and electronically passivated Si surface, which is stable up to 500 • C [11,12,14].
The bilayer termination of the Si(1 1 1) surface is isoelectronic to both H-and As-terminated Si, and the electronic structure resembles these, though with different interface dipoles between the bulk Si and the outermost surface layer. Alternatively, the electronic valence band structure of the surface may be described by an almost undisturbed GaSe band structure, where the Ga-Ga bond in the centre of the layer is replaced by the interface Si-Ga bond, resulting in an almost one-to-one replacement of the corresponding energy levels. This establishes a non-polar Si-Ga bond in the Si(1 1 1):GaSe half-sheet van der Waals surface termination, which is also suggested by the absence of a surface core-level shift in the Si 2p spectra of the GaSe terminated surface [7,12]. In combination with the chemical inertness and temperature stability of the surface, which 3 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT is superior to the commonly used Si(1 1 1):H, the GaSe termination generates an exceptional surface to be used for Si-based heterostructure devices.
In this paper, we report on determination of the electronic structure of the Si(1 1 1):GaSe surface termination using angle-resolved and energy-dependent valence band photoelectron spectroscopy and scanning tunnelling spectroscopy. After a brief description of the experimental procedure (section 2), the surface structure is reviewed (section 3). The integrated surface electronic structure determined with scanning tunnelling spectroscopy is presented in section 4, followed by detailed band structure measurements with photoelectron spectroscopy in section 5, where results are also presented for single crystal layered GaSe. Finally, in section 6, we present a phenomenological model of the electronic structure of the surface based on molecular orbital considerations and the bulk GaSe band structure, and compare our results with the isoelectronic Si(1 1 1):H and Si(1 1 1):As surfaces.

Experimental
The Si(1 1 1):GaSe termination layer can be prepared in different ways. Either hydrogen terminated Si(1 1 1)-1 × 1:H surfaces or clean Si(1 1 1)-7 × 7 surfaces may be used as a substrate. GaSe deposition may be either performed by separate evaporation of Ga and Se [12,13], or by direct evaporation of the GaSe compound [6,7,15]. The GaSe surface termination layer is stable up to substrate temperatures of ∼550 • C, whereas GaSe multilayer growth occurs only for T 500 • C. Therefore high quality, nearly defect-free termination layers can be formed for intermediate temperatures. The quality of the obtained surfaces was checked by low-energy electron diffraction (LEED), which exhibits clear three-fold diffraction patterns with a low background level.
Photoelectron spectra were partly recorded at the TGM 7 beamline of the electron storage ring BESSY in Berlin, supplying linearly polarized light with photon energies ranging from 11-120 eV. A VG ADES 500 spectrometer with an acceptance angle of ±2 • was used. The overall energy resolution of the setup was below 150 meV for photon energies below 40 eV and ∼350 meV at hν = 80 eV. Angular scans were performed by rotating the analyser in the plane of light polarization (axis of rotation perpendicular to the plane containing the polarization vector, the photon propagation and the electron emission). The angle between the surface normal and incoming photons was set to 45 • . Samples were prepared by selenization of monolayer Ga terminated Si(1 1 1) surfaces. The termination and surface quality was checked by LEED.
Additional measurements were taken at Beamline 7.0.1 at the Advanced Light Source (ALS) in Berkeley, CA, and at the University of Washington (UW) in Seattle, WA. These samples were prepared using a GaSe compound source; a few samples had supplemental Se from an electrochemical source to reduce Ga-terminated regions, which may be observed with STM [15] and result in a metallic component in Ga 3d spectra [16]. At UW, photoelectrons were excited by an unpolarized He resonance lamp with photon energy hν = 21.2 eV and detected by a Leybold EA-11 hemispherical analyser (angle resolution ±1-2 • at an angle of 55 • from the photons. The sample was rotated on an axis perpendicular to plane defined by the photon and electron beams. At the ALS, horizontally polarized photons of energy 70 170 eV were incident at an angle of 60 • from the horizontally emitted electrons; the electrons were detected by a 137 mm PerkinElmer Omni IV hemispherical analyser with angle resolution better than 1 • . The sample was rotated about a horizontal axis lying in the plane containing the polarization and electron emission direction (i.e., perpendicular to that at BESSY or UW).   [4,5] compared to a single GaSe layer. All chemical bonds are saturated, leading to a passivated surface. The surface is isoelectronic to the more common passivations Si(1 1 1):H and Si(1 1 1):As. The Brillouin zones of the hexagonal and the face-centred cubic lattice are shown at the bottom.

Structure of the Si(1 1 1):GaSe surface
The structure of the Si(1 1 1):GaSe surface termination was determined by the group of Eddrief in 1997 using gracing incidence x-ray diffraction [4] and x-ray standing waves [5], and later also confirmed by photoelectron diffraction [6,7] and atomic resolution scanning tunnelling microscopy [15]. The structure of the surface is shown in figure 1. The surface is basically identical to that of the van der Waals surface of bulk layered GaSe. It is characterized by a layer of Ga between the topmost Si atoms and an outermost Se layer [4]- [7]. Each Si atom binds to one Ga atom, while each Ga has one Si (below) and three Se (above) bonding partners. Each Se atom binds to three Ga atoms and has a doubly occupied lone pair orbital. The three Ga valence electrons are shared with the selenium (2/3 per bond) and Si (1 per bond). Four of six Se valence electrons are shared with Ga (4/3 per bond). The remaining two Se valence electrons occupy a lone pair orbital. The electron counting rule [17] is fulfilled and all chemical bonds are saturated. The Ga-Se surface layer has a structure essentially identical to that obtained by cutting the Se-Ga-Ga-Se unit layer of GaSe single crystals at the Ga-Ga bond. A hypothetical cut half-sheet layer of GaSe has an hexagonal array of singly occupied Ga dangling bond orbitals with a lattice constant of 0.374 nm, which matches closely to an unreconstructed Si(1 1 1) surface with a surface atom distance of ( √ 2/2)0.543 nm = 0.384 nm. On the other hand, the surface may also be viewed as isoelectronic to Si(1 1 1):H and Si(1 1 1):As. The surface bilayer of ideally terminated Si(1 1 1) has eight valence electrons per Si pair; seven are in bonds with other Si atoms, and a single electron occupies the surface dangling bond. Addition of hydrogen, with its single electron, saturates this dangling bond and brings the dangling-bond orbital well below the valence band maximum [18,19], passivating the surface. Replacement of the top Si and H by an arsenic atom, with five valence electrons, likewise leaves the surface passivated with a doubly occupied lone-pair and three Si-As bonds. The electronic structure is similar to Si(1 1 1):H, except that the large charge transfer between the Si and As results both in a large (0.75 eV) core-level shift on the interface Si 2p [20] and the lone-pair state lies about 3 eV closer to the Si valence band maximum [21] than does the Si-H bond. Exchanging the Si-As bilayer with a Ga-Se bilayer again results in the same number of valence electrons and a passivated surface with a doubly occupied lone pair surface orbital. As discussed below, the reduced surface dipole of Si(1 1 1):GaSe lowers the energy of the lone-pair orbital relative to Si(1 1 1):As.
Scanning tunnelling microscopy (STM) images are shown in figure 2. On the atomic scale (figure 2(a) and (b)), an hexagonal close-packed 1 × 1 structure is observed with positive and negative sample bias (see also [15]). Low-energy electron diffraction also shows a 1 × 1 pattern with a threefold symmetry [12,13], expected for a single domain surface termination. Wide scan STM images show a terrace structure resulting from the misorientation of the wafer surface with only very few defects.
Surfaces with a low miscut angle (<1 • ) exhibit large terrace widths of more than 100 nm with small holes (marked with 'A' in figure 2) and some small islands (marked with 'B'). As reported in [14,15], the bottom of the holes is of the same height as the adjacent lower terrace. Likewise, the small islands have the same height as the adjacent higher terrace. We attributed these features to the two main reorganization processes, which occur during the GaSe half-sheet formation [13]. First the gallium is moved from the T 4 -site in the √ 3 × √ 3 R30 • -Ga (1/3 ML) to a substitutional site in the 6.3 √ 3 × 6.3 √ 3 R30 • -Ga (1ML) reconstructions. Therefore, one monolayer of silicon must be carried away. In the second stage, the gallium-reconstruction is removed stepwise by the  GaSe half-sheet. The gallium is located at the top position in the GaSe-half-sheet and therefore one monolayer of silicon must be transferred to step-edges. If the diffusion-length is comparable to the size of the terrace-width (100 nm), the number of depressions and small island is very low and of the order of 10 10 cm −2 as shown in figure 2. Larger terraces lead to an increased density of these two features as earlier presented by Ueno et al [14] and Ohta et al [15]. The displaced atoms also attach at the step edges, making them more meandering than the straight edges typical of the initial substrate. Additional line and point defects related to orientational domain boundaries and substitutional defects have also been reported [15]. High-resolution Si 2p, Ga 3d and Se 3d core-level spectra recorded using excitation of hν = 140 eV taken from the U49/2-PGM 2 undulator beamline at BESSY [22] are shown in figure 3. All three core-levels show only a single-chemical component at very high resolution, consistent with previous observations [7]. The core-level spin-orbit splittings of the Si 2p, Se 3d and Ga 3d level were determined from least-squares curve fitting with Voigt profiles as 0.61, 0.86 and 0.45 eV, respectively. All values are in good agreement with reported data for Si, GaAs and WSe 2 [23]- [25]. In contrast to other Si surfaces [18,21,23], the Si 2p core-level shows no surface core-level shift (SCLS) within our experimental resolution of 150 meV, which is determined from the Gaussian broadening of the Fermi edge emission of a clean Au surface. The absence of a surface core-level shift at the Si(1 1 1):GaSe surface termination has been attributed to the non-polarity of the Si-Ga bond [7,12]. While Ga-covered Si(1 1 1) surfaces clearly show a 300 meV SCLS due to a polar Si-Ga bond [12,13,26], the polarity of the bond is most likely removed by the strong attraction of electrons of the outermost Se atoms [12].

Scanning tunnelling spectroscopy
To verify the absence of electronic states at the surface with energies inside the Si band gap, and to determine the surface band gap, the Si(1 1 1):GaSe surface has also been investigated by scanning tunnelling spectroscopy (STS). Scanning probe images clearly confirm the unreconstructed 1 × 1 structure of the surface for positive and negative sample bias (see figure 2). The voltage dependence of the tunnelling current and its derivative are shown in have neither nearby substitutional defects, defects at the measuring points in the next STM-image, nor vibrations visible in the I-V curves. Our electronic noise level was ≈1 pA.
Taking the edges of the surface bands to be the point where the absolute value of the current drops below our 1 pA noise level, we identify the surface band gap for a GaSe bilayer on an n-type Si substrate to be 2.08 eV (figure 4(a)). A similar value is also observed on p-type substrates. This value is within the range of all layered bulk GaSe polytypes (∼2.0 eV; see [27] and references therein). It thus appears that orbital interactions giving rise to the band gap in pure layered GaSe are comparable to those for the single Ga-Se layer on Si(1 1 1):GaSe.
The normalized conductance derivative in figure 4(b) is proportional to the density of states [28]- [30]. The band-gap and two important points (A and B) are marked in the right plot. The value for the band-gap is taken from the left figure. The state 'A' at −1.6 V (negative voltages correspond to occupied states) is most likely due to the frontier lone-pair Se p z -orbitals (see below and [31]). To image the GaSe half-sheet with negative bias this is the minimum voltage that must be applied (see figure 2(a) and [15]). The first maximum at positive bias, labelled 'B', is ∼1.8-2.0 eV above the Fermi level. By using this voltage, the best images at positive bias can be recorded (see figure 2(b)). In accordance with single crystal band structures, the calculation from Camara et al [31] shows that the conduction band of the Si(1 1 1):GaSe surface is also mostly derived from Se s-orbitals. As noted in [15], selenium orbitals are therefore most likely responsible for both tunnelling directions.

Normal emission valence band spectra
Normal emission spectra probe the electronic structure perpendicular to the surface and the surface¯ point. This corresponds to the L direction of the silicon Brillouin zone and to the A direction of the hexagonal GaSe Brillouin zone. Figure 5 presents normal emission valence band spectra from a vacuum-cleaved GaSe single crystal (a) and a Si(1 1 1):GaSe terminated surface (b) for a wide range of photon energies (and hence k ⊥ ). The valence band spectra and the derived band structure of the cleaved single crystal GaSe(0 0 0 1) surface are in good agreement with previous studies [32]- [34]. The bands can be grouped into four different groups A-D, as indicated at the top of figure 5(a). In normal emission, these correspond to emission from bands which are predominantly derived from Se p z (A), Ga and Se p x and p y (B) and from the bonding (D) and anti-bonding (C) orbitals of the Ga-Ga bond (Ga s orbitals) [32]- [37]. The band structure is well modelled by ab initio density functional theory [33], which closely reproduced the semi-empirical tight binding calculation of Doni et al [37], which will be used for a description of the electronic structure later. As can be expected, the bands derived from the p x -and p y -orbitals show minimal dispersion in normal emission, while the other bands show pronounced binding energy shifts with variation of photon energy (or k ⊥ ). assuming direct transitions to free-electron final states with an inner potential of 12.5 eV. This value of the inner potential is consistent with the value of 13 ± 1 eV found when fitting low kinetic energy photoelectron diffraction data [7]. In general, the final state band structure can deviate significantly from the free electron parabola at low energies (see e.g. [38]), resulting in wrong calculations of k ⊥ . A detailed analysis of the band symmetries of GaSe based on spin-resolved normal emission valence-band spectra given recently [34], indicating indeed a rather complex final-state band structure for bulk GaSe for low photon energies. However, a free-electron final state model has also led to a reasonable interpretation of data for electronically decoupled thin films of InSe [39]. In addition, the results shown in figure 6 are taken for photon energies resulting in kinetic energies >10 eV.
In contrast to the bulk GaSe data, the binding energies of most of the transitions observed for the Si(1 1 1):GaSe surface do not depend on k. Such dispersionless states are expected for surface state emissions, and are highlighted with lines in the figures. Also visible, especially in the fourth Brillouin zone, are two dispersing Si bulk states, which are indicated by dashed thick lines in figure 6. The domination of the spectra by surface state emissions is not surprising as the GaSe half-sheet is almost as thick as the minimum of the photoelectron escape depth, which is attained at ∼50 eV electron kinetic energy. Away from the sample normal, however, bulk states are somewhat more visible, as seen in Si(1 1 1):As [21] and Si(1 1 1):AlSe [40].
Of the five surface states observed at¯ , only the highest (associated with the lone pair) and lowest (the Si-Ga bonding state) are strong in both the high and low photon energy data. This may be due to different orientations of the sample relative to the light polarization in the two experiments.

Angle-dependent valence band spectra
Angle-dependent photoelectron spectra taken with hν = 21 eV excitation energy along the M direction of the hexagonal Brillouin zone for a vacuum-cleaved single crystal GaSe(0 0 0 1) surface and along the corresponding¯ M direction of the surface Brillouin zone of the Si(1 1 1):GaSe terminated surface are presented in figure 7. Spectra taken along the K direction of the same samples are given in figure 8.
The single crystal spectra show a large number of intense emissions corresponding to the number of atoms in the unit cell (8). Large binding energy shifts along both crystallographic directions are evident. The energy dispersion agrees with those given in the literature by Thiry et al [32] and Plucinski et al [33]. The emissions with lowest binding energy are observed at 0 • emission angle, indicating a valence band maximum at in agreement with the literature [32,33,35,37,41]. The spectra taken from the Si(1 1 1):GaSe surface termination also show pronounced structures with strongly dispersing transitions for both symmetry directions. As expected from the crystallographic structure of an ideal surface termination, they are noticeably different for the two directions, indicating a well-prepared surface termination layer.
On first glance at figures 7 and 8, spectra taken from the GaSe-terminated Si(1 1 1) surface show little resemblance to the single crystal data. However, the dissimilarity is mostly caused by different intensities of the various emissions. This is seen clearly in figure 9, which shows the energy dispersions along M and K (determined from the spectra given in   contributions to the electronic structure of bulk GaSe. An analogous approach has recently been used to discuss inter-layer interactions in layered metal dichalcogenides [42,43]. In the following section, we discuss the bands in terms of the isoelectronic Si(1 1 1):H, Si(1 1 1):As and Si(1 1 1):AlSe and the connections between passivated Si(1 1 1) and layered GaSe. A detailed description of the electronic structure of III-VI-compounds is given in the literature [33,35,37,41]. GaSe is a direct gap semiconductor with a valence band maximum and a conduction band minimum located at for the β (2H) polytype, or at Z for the γ (3R) polytype, respectively. The difference in electronic structures of the various polytypes is mainly explained by the different extension of the unit cell along c. Since we are only dealing with a half-sheet of a single layer, the distinction between different polytypes is not important here (see figure 1). Energy bands for bulk β-GaSe (2H polytype), as calculated by Doni et al using a semi-empirical tight-binding approach [37], are reproduced in figure 10(a). These band structures quite well reproduce the experimental band structures of GaSe and InSe (see also [39]) and are very close to state-of-the-art density functional theory calculations [33].
In the simplest description of the electronic structure of layered III-VI compounds, the valence band maximum is derived from chalcogen p z -orbitals and the conduction band minimum from metal p z -orbitals (symmetrized combinations of molecular orbitals are given by Doni et al [37] and are shown in figure 10(d). In the real electronic structure, hybridization between chalcogen and metal states is important. A more detailed analysis of the contributions to the band edge states for the isostructural and isoelectronic compound InSe is given by Gomes da Costa et al [41], revealing a valence band maximum composed of 70% Se p z and 30% In p z -orbitals, while the conduction band minimum is composed of Se s (37.5%), Se p z (25%) and In s-orbitals (37.5%).
As a result of hybridization, the bands derived from the given symmetrized orbital combinations do not directly correspond to calculated energy bands. In terms of energy band structure, bands may only cross without interaction (hybridization) if they have different symmetry, while hybridization of energy bands leads to an energy gap. Using symmetrized combinations of atomic orbitals [37], hypothetical energy bands without hybridization can be outlined for bulk GaSe. The result is shown in figure 10(b). The construction of nonhybridized energy bands is easily accomplished for the M direction, which corresponds to the crystallographic y-axis as shown in figure 10(d). 7 The given hypothetical energy bands for the different symmetrized orbital combinations, shown in figure 10(d), exhibit the dispersions expected from simple tight-binding considerations [44,45]. The bands derived from s-and p z -orbitals (labelled I, II, and V in figure 10(b)) disperse upwards with increasing k , while those derived from p x -and p y -orbitals (labelled III and IV in figure 10) disperse downwards. Binding energies for the various orbital combinations at and M are taken from Doni et al [37].
Of interest for the Si(1 1 1):GaSe surface are those electronic states expected to persist for an isolated half-sheet layer. These can, at least partly, be identified a priori. First of all, inter-layer interactions, which are indicated in figure 10 by the shaded grouping of bands (perpendicular dispersion), have to be omitted (see also the description of electronic states in [42]). As a consequence, the pairs of hypothetical, non-hybridized bands indicated by roman numerals I, II, and V in figure  10(b) have to be replaced by a single band located approximately in the centre of the shaded regions at the boundaries of the Brillouin zone. The bands derived from p x -and p y -states (groups III and IV) have a negligible inter-layer splitting and are therefore not significantly affected.
In addition, splitting the layer in half between the two Ga atoms will strongly modify some intra-layer interactions. Cutting the layer between the Ga planes has the strongest impact on states derived from p z and s orbitals, as these have a large overlap along the c-direction, and a weaker impact on the p x and p y (Ga-Se bonding) states. Since the Se atomic layers are well separated (0.478 nm), the main consequence of splitting the layer in half is the removal of the Ga-Ga interaction. The bonding and anti-bonding combinations with predominant Ga 4s character are given by bands I and II (see figure 10). An isolated half-sheet of GaSe should have a single band, instead of the two bands I and II, as the Ga-Ga bond is removed. Hybridization of the chalcogen p x and p y -states with the Ga p x and p y -states (bands labelled III and IV in figure 10) is not strongly modified since the Ga-Se bond length and geometry is nearly unchanged. The (small) splitting 16 Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT between the intralayer bonding (labelled as x + in [42], figure 2) and anti-bonding (x − )-states is removed, while maintaining a similar dispersion along M.
Using the above arguments, tentative energy bands for a hypothetical, isolated GaSe halfsheet layer are drawn in figure 10(c). The p x and p y -bands (bands III and IV) are modified only by the missing inter-layer interactions. The former Ga-Ga bonding and anti-bonding states (groups I and II) are represented by dashed lines in figure 10(c). The splitting between them should be absent for an isolated half-sheet (with no Ga-Ga bonds); the two bands are replaced by a single band approximately half-way between the dashed bands, indicated in figure 10(c) by a solid line dispersing upwards along M. However, at the Si(1 1 1):GaSe surface, the Ga-Ga bond is replaced by a Si-Ga bond. Although the dispersion of the corresponding bonding and anti-bonding states will in principle be similar to those of the Ga-Ga bond, the energies and band widths of the states belonging to the Si-Ga bond are not known. One might expect noticeable differences between the asymmetric Si-Ga bond and the non-polar Ga-Ga bond in GaSe.
Finally, we note that the experimentally observed energy bands should follow the dispersion of hybridized bands and not those of the outlined hypothetical non-hybridized bands. Rather, construction of non-hybridized bands enables identification of those parts of the GaSe band structure expected to persist in the half-sheet layer.

Comparison with experiment
We start with the band we expect to be the least perturbed between the bulk and half-layer GaSe, namely the p x -and p y -states of the Ga-Se bilayer. The emission at a binding energy of BE ≈ 3 eV at¯ (see figures 7(b) and 8(b)) can be attributed to these states (band III and IV in bulk GaSe). The p y -states (band III) strongly disperse downwards and give rise to strong emissions at BE ≈ 8 eV at the M point of the Brillouin zone (see figure 7(b) at 40 • emission angle). The downward dispersion of the p x -states is much weaker since they are oriented perpendicular to the¯ M direction. They are expected to show up at BE ≈ 4 eV at theM point. Similar binding energies are expected atM from the upward dispersing Si-Ga bonding and anti-bonding states (bands I and II), from which they cannot be distinguished. However, it can be expected that the p x -states contribute to the number of intense emissions nearM at this energy.
Interestingly, the Si(1 1 1):GaSe surface shows emissions at at binding energies close to those of the single crystal Ga-Ga bonding and anti-bonding states. This is evident from the general agreement between both band structures (figure 9). These emissions are most likely attributed to states from the Si-Ga bond, implying that the binding energies of the bonding and anti-bonding Si-Ga states of Si(1 1 1):GaSe are very close to those of the Ga-Ga states in GaSe single crystals. This evidence for non-polarity in the Si-Ga bond is consistent with the absence of a Si surface core-level shift in high-resolution Si 2p core-level spectra (see [7,12] and figure 3) and the observation of flatband conditions at the surface [12]. As discussed in [12], a Si-Ga bond is generally polar due to the larger electronegativity of Si compared to Ga (Pauling electronegativity = 1.90 and 1.81, respectively), but the interface Si-Ga bond in Si(1 1 1):GaSe becomes non-polar as a result of the Ga-Se bonds. The large electronegativity of Se (Pauling electronegativity = 2.55) leads to charge transfer from Ga to Se and reduces the ability of Ga to donate electrons to the silicon, effectively increasing its electronegativity to essentially equal that of Si.
The top of the GaSe valence band (transitions with lowest binding energies) has − p z states. The corresponding lone-pair emissions are also clearly observed at the Si(1 1 1):GaSe surface, and absent on the splitting associated with inter-layer interactions. The upward dispersion with increasing k , expected for non-hybridized p z -states (figure 10(c)), is not observed, which is a consequence of the hybridization between p z -and p x , p y -states (see section 6.1). It is also not expected that the hybridization of these states is absent at the surface, as it has a lower symmetry than the corresponding GaSe single layer (missing horizontal reflection plane). The similar hybridization of the layer compared to bulk states has the important consequence of producing a surface without electronic states within the Si band gap, leading to an electronically passivated surface [12]. A noticeable difference between the band structures of GaSe and the Si(1 1 1):GaSe surface termination is the bulk band dispersing downwards from BE ≈ 5 eV at¯ to BE ≈ 8 eV atM orK. The corresponding states of the Si(1 1 1):GaSe give rise to weak emissions close to¯ , which are not observed at higher emission angle (see figures [7][8][9]. A downward dispersing band is detected in the 72 eV spectra, which, however, is more likely due to emission from a Si bulk band. For the GaSe single crystal, this band is mainly composed of Ga s and Se p z orbitals at and of Se p y orbitals atM (see figure 10 and [37]). Due to hybridization of the bands along M ( -direction), the band is a mixture of both constituents. Along in the GaSe bulk, hybridization between p z and p y orbitals is possible since both transform identically under the symmetry operations of the point group , which are the identity, a two-fold rotation along y (C2), and reflections at the xy-(σ h ) and yz-planes (σ v ), respectively [42]. In the surface layer, however, neither C 2 nor σ h are symmetry operations. Due to the lower symmetry, it is expected that hybridization between the different orbitals is still possible. The same holds for the corresponding direction in bulk Si ([112]), which has only σ v as the higher symmetry element. It is thus expected that the p y and p z can still hybridize as also observed for the outermost Se atoms (see preceding paragraph). The apparent absence of the band at the Si(1 1 1):GaSe surface is therefore attributed to a low-photoemission intensity. (1 1 1) An alternate approach to the Si(1 1 1):GaSe surface from that of the basic building block of bulk GaSe is to consider it as an ideal termination of bulk silicon. As mentioned above, the Si(1 1 1):GaSe surface is isoelectronic to the well-studied Si(1 1 1):H and Si(1 1 1):As passivated surfaces (see figure 1), consisting of bulk silicon plus either (Si-Si-H), (Si-As) or (Ga-Se). All three terminations contribute nine valence electrons that may give rise to states separate from those of bulk Si. We can separate these into the two electrons in an orbital near the surface (the Si-H bond, the As lone-pair or the Se lone-pair, respectively), the six electrons in the surface bilayer bonds (Si-Si, Si-As or Ga-Se), and the electron contributed to the bond between the bilayer and the substrate (Si-Si, Si-Si or Si-Ga). We thus expect five fully occupied states near the surface for these nine electrons plus one electron from the silicon atom beneath, which bonds to this surface layer. Indeed, the normal emission data at variable photon energy for Si(1 1 1):GaSe (figures 5 and 6), in addition to the two bulk Si states dispersing from to L, show five states that do not disperse with photon energy (figure 6), but do exhibit dispersion parallel to the surface with the symmetry of the surface Brillouin zone ( figure 9). An additional weak shoulder at a binding energy around 5 eV in figure 6 appears only at high photon energy (above the Si L-edge) and may be an inelastic loss feature or defect state, as it also does not disperse with k . Given the strong feature (C) in bulk GaSe near this energy associated with the Ga-Ga bond, a candidate Institute of Physics ⌽ DEUTSCHE PHYSIKALISCHE GESELLSCHAFT defect state is a Ga Si substitutional defect, leading to a Ga-Ga bond. These defects are the most commonly seen with STM [15].

Electronic states of passivated Si
We consider first the electrons physically closest to the surface, in s-p z orbitals of either the Si-H bond, the As lone-pair or the Se-lone pair. Comparison of these systems is simplest at thē K-point, where the bulk Si bands have the largest gaps. Near the zone centre, the lone-pair state largely overlaps the silicon valence band maximum and is difficult to distinguish. The energy of this state at theK-point relative to E VBM is a sensitive measure of the interface dipole between the underlying bulk Si and the surface [46]. AtK, this state for Si(1 1 1):H has binding energy 4.8 eV below the Si valence band maximum (E VBM ) [19], lying in an empty pocket of the projected bulk Si bands. For Si(1 1 1):As, the lone pair state atK is 1.8 eV below E VBM [21], or about 0.5 eV above the highest Si band atK. The equivalent state for Si(1 1 1):GaSe lies right at the edge of the projected bulk Si bands, at an energy about 2.3 eV below E VBM . The equivalent state for Si(1 1 1):AlSe has a similar energy to that for Si(1 1 1):GaSe [40].
The difference in the surface dipole between a Si-As bilayer and a Ga-Se bilayer can be thought of as moving a layer of protons up towards the surface, turning Si 4+ into Ga 3+ and As 5+ into Se 6+ , which should decrease the electrostatic potential energy of an electron on the surface. The equivalent operation shifting a layer of protons in the next bilayer down, comparing a buried Si-Si bilayer to a buried Ga-As bilayer (i.e., Si-Si ≡ Si-Si ≡ As versus Si-Ga ≡ As-Si ≡ As), has been calculated to cause a −0.4 eV difference in the lone-pair energy atK [46], consistent with the −0.5 eV difference we observe between Si(1 1 1):As and Si(1 1 1):GaSe. Si(1 1 1):H, on the other hand, may be generated in a thought experiment by moving a layer of protons from Si(1 1 1):As outward to the other side of the state in question (Si-Si ≡ Si-Si ≡ As versus Si-Si ≡ Si-Si ≡ Si-H), for a smaller electrostatic shift, but in the same direction, as for Si(1 1 1):GaSe. The much lower energy of the state for Si(1 1 1):H is attributed to the large bonding interaction between the Si and H orbitals. The surface Si bilayer is essentially a bulk environment, and the Si-H bond is close in polarity and energy to that of the Si-Si bond.
We next discuss the 'back-bond' states in the surface bilayer, which we associate with the states of binding energy ∼2-3 eV at¯ that disperse downwards towardsM andK. For Si(1 1 1):As, the only difference between the predicted bulk Si transitions and the observed surface band structure for states in this energy range is a small deviation near the pockets in the projected bulk band structure nearM andK [21]. This has been attributed to the breaking of symmetry along the Si-As bond relative to the bulk Si-Si bond pulling the states into the bulk gap. Some surface resonances are found in this energy region for Si(1 1 1):H, mainly near theK point, but there is no clear deviation from bulk energies and dispersions. For Si(1 1 1):GaSe, on the other hand, the 'back-bond' states show a much larger resemblance to those of bulk GaSe (discussed above) than bulk Si, which is to be expected due to the very different polarity of the bonds and energies of the atomic states contributing to them.
Finally, the states bonding the surface layer to the substrate might be expected to be barely changed from bulk Si for Si(1 1 1):H and Si(1 1 1):As, and somewhat modified for Si(1 1 1):GaSe. Indeed, all three surfaces show a weakly dispersing state at a binding energy around 8 eV that may be associated with this bond.

Summary and conclusions
The electronic structure of the Si(1 1 1):GaSe van der Waals surface termination has been studied by photoelectron spectroscopy and scanning tunnelling spectroscopy. A surface band gap of 2.1 eV is detected, which is very close to the bulk GaSe band gap. The large surface band gap, which is approximately symmetric with respect to the Si bulk gap, is the origin for the good electronic passivation of this surface. A large number of electronic bands is determined at the surface. Their dispersion with emission angle is similar to those of bulk GaSe bands, which have been measured for comparison. The electronic structure of the Si(1 1 1):GaSe surface has consequently been discussed in terms of a molecular orbital description of the GaSe bulk band structure, taking hydridization effects into account. This procedure is justified for two main reasons. Firstly, the arrangement of the Ga and Se atoms at the surface are almost identical to those of a half-sheet of crystalline GaSe and the topmost silicon atoms also occupy nearly ideal Si lattice sites. Secondly, the non-polarity of the Si-Ga bond, which is introduced by the polar Ga-Se bond, leads to an effective electronegativity of Ga close to those of Si. The non-polarity of the Si-Ga bond at the surface is evident from the absence of a surface core-level shift of the Si 2p level and from the binding energies of the valence states attributed to the Si-Ga bond, which is very close to those of the Ga-Ga bond states in bulk GaSe. The states may also be interpreted in terms of a passivated termination of bulk silicon, and were compared with Si(1 1 1):H and Si(1 1 1):As to identify the states in terms of the surface bonding and lone-pair levels.