Band alignments at Ga2O3 heterojunction interfaces with Si and Ge

Amorphous Ga2O3 thin films were deposited on p-type (111) and (100) surfaces of silicon and (100) germanium by atomic layer deposition (ALD). X-ray photoelectron spectroscopy (XPS) was used to investigate the band alignments at the interfaces using the Kraut Method. The valence band offsets were determined to be 3.49± 0.08 eV and 3.47± 0.08 eV with Si(111) and Si(100) respectively and 3.51eV± 0.08 eV with Ge(100). Inverse photoemission spectroscopy (IPES) was used to investigate the conduction band of a thick Ga2O3 film and the band gap of the film was determined to be 4.63±0.14 eV. The conduction band offsets were found to be 0.03 eV and 0.05eV with Si(111) and Si(100) respectively, and 0.45eV with Ge(100). The results indicate that the heterojunctions of Ga2O3 with Si(100), Si(111) and Ge(100) are all type I heterojunctions.

the Ga 2p spectrum, indicative of the preferential removal of oxygen, thus leaving a sub-stoichiometric oxide or elemental Ga; finally, their discussion on the formation of an interfacial SiO2-x layer is limited by the issues raised above.
The band offsets of Ga2O3 with Ge have not been reported previously, with only that found for Ga2O3(Gd2O3) having been reported, where ∆ # = 2.35 ± 0.1eV 13 , however in this case the Kraut method is used incorrectly, as the thickest sample used is only 3nm and therefore still sensitive to the interface. Previous reports of photoelectron spectroscopy on Ga2O3(Gd2O3) on Ge have found evidence of an interfacial oxide layer formed upon deposition of the dielectric layer 14,15 , with the notable exception of Chu et al. 13 , however the binding energy reported is considerably larger (at a binding energy of approximately 30eV) than that expected for elemental Ge. Such an energy suggests that the wafer may in fact be oxidized, however accurate determination of the oxidation state is rendered impossible since the data had not been deconvoluted.
In this work, we describe the use of the Kraut method 16 to investigate the band alignments of Ga2O3 with Si(111), Si(100) and Ge(100) substrates. As the Kraut method requires a thick, bulk-like sample, a thin, interfacial sample, where the core-levels of the substrate can be observed and a clean substrate, thick (31nm) and thin (3nm) Ga2O3 layers were synthesized on the aforementioned substrates. The use of a thin sample also allows for the quality of the interface to be investigated in a non-destructive manner. In contrast to Chen et al. 12 , we use a monochromatic x-ray source, thus avoiding the issue of x-ray satellites.
Ga2O3 layers were deposited by atomic layer deposition (ALD) on clean p-type Si(111) and Ge(100) wafers in an Oxford Instruments Plasma-OpAL reactor. The precursors used for deposition were triethylgallium (TEGa) and O2 plasma and with a substrate temperature of 250 o C. 58 ALD cycles were used to obtain thin samples, and 580 cycles for the thick samples.
Prior to deposition, the Si substrates were cleaned by 10 minutes immersed in methanol in a sonic bath, followed by 10 minutes immersed in acetone in a sonic bath. The Si substrates were then immersed in 5% HF for 5 minutes with agitation before being rinsed with distilled water and dried by N2 gas. The Ge substrates were cleaned by 10 minutes immersed in methanol in a sonic bath, followed by 10 minutes immersed in acetone in a sonic bath and dried by N2 gas. X-ray photoelectron spectroscopy (XPS) measurements were performed in a standard UHV chamber, with a base pressure below 2×10 -10 mbar, with the main residual gas being hydrogen. A monochromatic Al Kα SPECS XR 50 M source (ℎ = 1486.6 eV) operating at a nominal power of 250W was used in conjunction with a PSP Vacuum Technology Ltd Resolve 120 MCD5 electron energy analyzer. The spectrometer was calibrated to the positions of the Ag 3d5/2 and Fermi level of a clean polycrystalline Ag foil. By fitting the Fermi-Dirac distribution to the Ag Fermi level, the resolution of the analyzer is found to be 0.37±0.05 eV.
Inverse photoemission spectroscopy measurements were performed using a PSP Vacuum Technology BaO cathode dispenser electron source and an isochromat NaCl photon detector, both at 45° to the sample normal. Energy calibration was performed using the known energy position of the lowest unoccupied molecular orbital of multilayer C60 deposited in situ and the spectrometer resolution determined from a polycrystalline Ag foil was 1.00±0.10 eV.
The deposited films were confirmed to be amorphous by x-ray diffraction with a Rigaku SmartLab x-ray diffractometer (ℎ = 8047.8 eV). Spectroscopic ellipsometry was used to determine the thicknesses of the deposited films with a Horbia Jobin Yvon MM-16 spectrometer with a range of 430 -850nm. The thicknesses of the deposited films were found to be 30.05±0.18nm and 3.05±0.10nm for the thick and thin layers respectively. Ellipsometry spectra were fitted using a Cauchy model adopted from Ref 17 .
The XPS spectra were analysed using the CasaXPS software. Core-levels were fitted with pseudo-Voigt functions atop a Shirley background. Valence band maxima (VBM) positions were found by linear extrapolation to the background. The errors on core-level binding energies and the VBM were determined to be ±0.05 eV. Since the mean-free-path of the photoelectrons is dependent on the kinetic energy of the emitted photoelectron, core-levels with large separations in kinetic energy will have different probing depths and hence different sensitivities to the interface potential. The Ga 3p, Ge 3p and Si 2p core-levels were therefore used for Kraut method analysis as these peaks occur in a narrow range of binding energies, ensuring that the photoelectrons originated from similar depths.
The Kraut method, uses the binding energies of the core levels and the VBM of the three samples to determine the valence band offset (VBO), ∆ # , as shown in equation (1) ∆ # = ∆ 67 + ( : is the energy difference between the core levels of the deposited film and the substrate for the 3nm sample, and ξ denotes the VBM for the relevant sample.  Fig. 1(a) show the Ga 3p region for the thick, 30nm Ga2O3 sample. Fig. 1(b) shows the valence band measured by XPS and the conduction band, as measured by IPES. The VBM was found at -3.48±0.05 eV, whilst the CBM occurred at 1.15±0.14 eV relative to the Fermi level. By combining these values, the bandgap of the 30nm Ga2O3 film was determined to be 4.63±0.14 eV, consistent with the 4.4-4.9eV band gap range previously reported for amorphous Ga2O3 films [18][19][20][21] .  The data show that the surfaces of the substrates were clean of any contamination. The binding energies and their differences from the VBM obtained from these figures are given in table I. The binding energy of the Si 2p3/2 core-level is consistent with values reported in the literature for p-type Si wafers [22][23][24][25] . The separation between the Si 2p3/2 and the VBM is found to be 98.83±0.08 eV, consistent with that reported previously [25][26][27][28] .  Fig. 3(a) shows the Si 2p and Ga 3p region for the 3nm Ga2O3/Si(111) sample, whilst Fig. 3(b) shows the Ge 3p and Ga 3p region for the 3nm Ga2O3/Ge(100) sample.
As XPS is a surface-sensitive technique, varying the take-off angle will change the effective probing depth for a particular photon energy. The dependence of the mean-free-path of the photoelectrons in the material on the take-off angle is given by = H sin , where λ0 is the inelastic mean-free-path (IMFP) of an electron with the same kinetic energy with a take-off angle normal to the sample surface, and is calculated from H = MNOH P Q + 0.54√ , 29,30 where E is in eV and H is in Å,. Using this relation, the IMFP is calculated to be 2.0nm. By measuring the core levels at multiple take-off angles, the presence of any band-bending close to the interface can be determined. The Ga 3p3/2, Si 2p3/2 and Ge 3p3/2 binding energies were determined for 30º, 60º and 90º take-off angles. If band-bending were present at the interface, it would be expected that the difference in the core-level binding energies, ∆ 67 would vary with different take off angles. In fact, no such angular dependence is seen, as shown in Fig. 4(a)-(c). It can, therefore, be concluded that band bending at the interface is negligible for these systems. If bandbending were present, then the FWHM of the core-levels would also be expected to broaden 31,32 , in fact, no changes in the FWHM are observed, further confirming the absence of band-bending.
The Debye length is characteristic of the length scale over which the effect of the interfacial potential is felt. For crystalline β-Ga2O3 nanowires, the Debye length is found to be 40.5nm 33 . However, as the Debye length is proportional to the defect density 34 , the Debye length of the films described here will be considerably smaller, as the films described in this work are amorphous and hence defect-rich. Similar behaviour has been observed in a-IGZO thin film transistors 34 , where the defect density is approximately 2 × 10 MT WO .

TABLE I -The binding energies positions of the core-levels and valence band minima and band gaps for each substrate and the bulk like Ga2O3. Also shown are the calculated valence band and conduction band offsets.
Substrate  (1), the VBO for the amorphous Ga2O3 film on Si(111) was determined to be 3.49±0.08 eV consistent with that found for β-Ga2O3 on Si(111) by Chen et al. 12 and contrasting to that reported by Guo et al. 9 . The VBO for the amorphous Ga2O3 film on Ge(100) is likewise determined to be 3.51±0.08 eV.
Using the bandgap determined in Fig. 1(b) combined with literature bandgaps for Si(1.11 eV) and Ge (0.67 eV), the conduction band offsets (CBO) are found to be 0.07±0.14 eV and 0.45±0.14 eV for the Si(111) and Ge(100) samples respectively. The core-level energies, valence band maxima, band gaps and band offsets are summarised in table I. A schematic diagram of the band alignments is given in Fig. 4.

Fig. 4 -A schematic band alignment diagrams showing the XPS determined band alignments for ALD-grown Ga2O3 on a variety of common semiconductor substrates.
Similar measurements were also performed on a Si(100) substrate. The difference in the VBO for the Si substrates are within the experimental error of the XPS system. The VBO obtained from both the Si(111) and Si(100) samples are consistent with that found by Chen et al. 12 . The CBO obtained from the Si(111) and (100) are different to those obtained by Chen et al. 12 , these differences are likely due to the different methods of evaluating the band gap. Chen et al. use the photoelectron energy loss spectrum from the O 1s peak, rather than the combined VBM-CBM separation used in this work. The error on this method can be estimated to be of the order of 0.6-1.0 eV from Figure 1b of Chen et al. 12 . An interesting question would be whether the doping level affects the band alignment. By considering the position of the VBM with respect to the Fermi level, we can conclude that the Si wafers used in this work are more heavily p-type doped than those used by Chen et al. 12 , however the VBOs found are consistent, suggesting that the doping level has little effect on the band offsets.
Another important variable in the performance of semiconductor devices is the quality of the interface and the existence of any interfacial layers between the substrate and the deposited film. The substrate is frequently observed to be partially oxidized after deposition of a binary metal-oxide upon Si or Ge 27,35-43 , but is typically less than 3nm in thickness 10 . Fig. 3a shows the interfacial Si 2p and Ga 3p spectra. A second set of peaks attributed to silicon occur at 3.5eV above the Si 0 2p3/2 consistent with that expected for the Si 4+ oxidation state of SiO2 [44][45][46][47] . A similar layer was also observed for 3nm Ga2O3/Si(100). Paskaleva et al. 10 have likewise observed the formation of an interfacial SiO2 layer by ellipsometry and C-V measurements when depositing β-Ga2O3 on Si(100) by the closely related chemical vapour deposition technique. Chen et al 12 . do not observe this oxide formation, however the interpretation of their data is limited as the spectra presented are at low resolution and are not deconvoluted. The interpretation of their data is further complicated by the Ar + ion etching process used to produce the interfacial dataset. Here, the Ga 2p peaks are clearly asymmetric, suggesting the formation of either a non-stoichiometric oxide (Ga2O3-x, GaO), or possibly metallic Ga 0 . These features would obscure the signal from the Si 2p for the interfacial oxide, as the Ga 3p peaks would be shifted to a similar binding energy as Si 2p from SiO2. Interpretation of the spectra in Chen et al. 12 is further complicated by their use of a non-monochromatic xray source, resulting in x-ray satellite features. For a Mg source such as the one they used, these occur 8.4 eV and 10.1 eV away from the main peak on the low binding energy side 48 . An alternative explanation could be in the differences between the growth techniques as Chen et al. 12 used pulsed laser deposition to deposit their films, rather than the chemical techniques described here and by Paskaleva et al. 10 .
Similarly, figure Fig. 3(b) shows the XPS spectrum for the interfacial Ga2O3/Ge(100) sample in the Ge 3p and Ga 3p region. Here, the formation of GeO2 occurs analogous to the formation of SiO2 above. The inset in figure Fig. 3(b) shows the Ge 3d region. As well as the GeO2 previously mentioned, there is also a doublet shifted from the elemental Ge peaks by 0.80eV. This corresponds to an oxidation state of Ge +47 .
The thickness of the interfacial oxide can be estimated by the process established by Hill et al. [49][50][51] , where the thickness of the native oxide, d, is given by equation (2). Io is the native oxide peak intensity, and Is is the substrate peak intensity.
Hence, by using equation (2), the native oxide thickness is found to be 0.73nm, 0.75nm, and 0.89nm for the Si(111), Si(100) and Ge(100) substrates respectively. It should be noted that as this analysis neglects the 3nm Ga2O3 overlayer, the signal from the bulk Si/Ge is reduced relative to that of the native oxide, the calculated thicknesses are an upper limit, and the actual thickness of the native oxide will be less than this. Whilst, it would be expected that the interfacial oxide may have an effect on the band alignment, the band offsets reported here are consistent with those reported by Chen et al. 12 who do not observe the interfacial native oxide. This would suggest that a thin interfacial oxide, as observed here, has little effect on the band alignment, and hence the device performance.
To summarize, amorphous Ga2O3 films are grown on Si and Ge substrates by ALD. The valence band offsets are determined by XPS to be 3.49±0.08 eV, 3.47±0.08 eV and 3.51±0.08 eV for the Si(111), Si(100) and Ge(100) substrates respectively. IPES is also used to investigate the conduction band of a thick Ga2O3 film, by combining the IPES with VB-XPS, the band gap is found to be 4.63±0.14 eV. The resulting conduction band offsets are found to be 0.03, 0.05 and 0.45 eV respectively, resulting in a type I heterojunction in all cases. In all cases a thin, native oxide is observed after deposition. The determination of the band alignments to this accuracy, has important implications for devices based on Ga2O3/Si and Ga2O3/Ge heterojunctions, including MOSFET and MOSCAP type devices.