Operando study of the preferential growth of SiO₂ during the dry thermal oxidation of Si_0.60Ge_0.40(001) by ambient pressure x-ray photoelectron spectroscopy

Strain engineering of Si and SiGe devices has resulted in enhanced carrier mobilities and improved device performance.^1,2 Although strained engineering was first introduced on 2D field effect transistors, strain engineering using SiGe has been extended to 3D transistor architectures including FinFETs. The electrical passivation of SiGe is much more difficult than for Si due to the poor-quality oxide (mixture of SiO₂ and GeOx) at the SiGe surface compared to the high-quality oxide (SiO₂) at the Si surface. Multiple studies have illustrated that the oxidation kinetics for SiGe is much different than Si,³ where it was proposed that Ge catalyzes the oxidation of SiGe.⁴ As a result, a thin Si cap layer has been used to reduce the interface trap density for SiGe FinFETs. Other approaches to passivate the SiGe interface include low-pressure oxidation,⁵ plasma oxidation,⁶ and ozone oxidation.⁷ These methods were performed at low temperatures to reduce the thickness of the oxide layer and ideally lead to the preferential growth of SiO₂.

X-ray photoelectron spectroscopy (XPS) has been extensively used to study the oxidation of SiGe due to the high surface sensitivity and excellent chemical state resolution.^4,5,7,8 Oxidation state resolved XPS results have indicated that varying the O₂ pressures $(PO2)$ between 0.01 and 100 mbar and the oxidation temperatures can significantly influence the composition of the grown oxide layer.⁵ Using XPS data to estimate the grown oxide thickness, it was found that reducing $PO2$ while increasing the oxidation temperature resulted in preferential growth of SiO₂ over Si_1−xGe_xO₂. An analysis of the Gibbs free energy change versus pressure and temperature were consistent with these observations.^3,5 To date these studies have been performed either ex situ or in situ, where the XPS measurements were performed after an oxidation process under a set of conditions. The ability to obtain XPS data in operando, during the oxidation process, should provide more information on oxidation kinetics and the chemical states at the SiGe interface under realistic conditions.

Recent studies have indicated the ambient-pressure x-ray photoelectron spectroscopy (AP-XPS) could be effectively used to investigate the initial growth of SiO₂ at the Si surface.^9,10 Real-time growth rates of SiO₂ were monitored during oxidation for dry and wet oxidation processes, respectively. In both cases, chemical state analysis of the Si 2p core-level photoelectron spectra illustrated the presence of multiple oxidation states. Dry oxidation results indicated an increase in oxidation rates when $PO2$ was increase from ∼0.0075 to 0.75 mbar at 450 °C.⁹ Wet oxidation results indicated an increase in oxidation rates when the oxidation temperature was increased from 250 to 500 °C for a constant H₂O pressure (P_H2O) of ∼0.0075 mbar, where a significant increase occurred for temperatures above 400 °C.¹⁰ During dry and wet oxidation processes, an initial rapid increase in oxidation rates were observed for temperatures above 400 °C, followed by a slow regime. The processes were found not to match the Deal–Grove model for silicon oxidation,¹¹ which was developed for thicker SiO₂ films and higher oxidation temperatures.

In this study, we characterized the initial stages of SiGe oxidation to further understand the effect of $PO2$ on the composition of the grown oxide. These studies were performed using a laboratory based AP-XPS system where spectra were fit to quantify chemical states from the Si 2p and Ge 3d spectra in operando. Evaluating the experimental data using simulation of electron spectra for surface analysis (SESSA) software package allows further insight into the oxide thickness and composition.

An epitaxial Si_0.60Ge_0.40(001)/Si(001) wafer and a Ge(001) wafer were provided by a semiconductor manufacture. A Si(001) wafer was obtained from Addison Engineering. The substrates were cleaved to obtain samples which were ∼10 × 10 mm². ACS grade acetone (Pharmco), ACS electronic grade isopropyl alcohol (IPA) (Fischer Chemical), and ACS grade 48–51 wt. % hydrofluoric acid (HF) (VWR International) were used to prepare the samples. Prior to experiments, the samples were cleaned using a 2-min. acetone ultrasonic bath, rinsed with IPA, and blown dry with nitrogen. Si_0.60Ge_0.40(001)/Si(001) and Si(001) samples were then exposed to a low-temperature oxygen plasma (PE50 Plasma Cleaner) for 10 min to remove surface carbon contamination. For Ge(001), the O₂ plasma treatment resulted in significant oxide growth which was later difficult to remove. Thus, for Ge(001), the O₂ plasma treatment was replaced with a sequential wet etch which included a 2 min dip in de-ionized water (18.2 MΩcm) followed by a 30 s dip in 10 wt. % H₂O₂. For all samples, a HF wet etch was performed to remove native oxides and passivate the surface with hydrides which can be desorbed by annealing. This was done using a 50:1 H₂O:HF (48–51 wt. %) wet etch for 10 min. Surfaces were then blown dry with ultrahigh purity nitrogen. The samples were mounted onto Ta sample holders by spot welding Ta strips over the edges of the sample, introduced into the load lock, and pumped down to 10⁻⁶ mbar within 5–8 min after the HF etch. To monitor sample temperature, we have used Ta sample holders which have a transferable K-type thermocouple that was spot welded to the Ta plate next to the SiGe substrate. While these procedures will not produce oxygen and carbon free surfaces, the intent of this study is to replicate surfaces similar to what would be encountered using industrial cleanroom manufacturing processes. Thus, some surface contamination was expected.

AP-XPS measurements were performed using a SPECS AP-XPS system with PHOIBOS concentric hemispherical analyzer. A monochromatized Al K_α x-ray source (hν = 1486.6 eV) was used with a near ambient pressure (NAP) cell where the emission angle (θ) was 90° with respect to the surface, and the source axis to analyzer axis angle was 60°. The binding energy was calibrated using a sputter cleaned silver film and measuring the Ag 3d_5/2 core level, which was set to a binding energy (E_b) of 368.27 eV.¹² High-resolution Si 2p, Ge 3d, O 1s, and C 1s spectra were obtained with a pass energy of 35 eV before and after each set of oxidation experiments. Spectra during the oxidation process were obtained with a pass energy of 50 eV. This allowed the collection of a single spectrum in approximately 80 s with an energy resolution that could resolve the oxidation states in the Si 2p and Ge 3d spectra. These spectra were charge corrected using the aliphatic C 1s with E_b set to 284.8 eV.¹³ For the AP-XPS experiments, the samples were heated to 300 °C, and then, oxygen was introduced to the desired pressure in the NAP cell using a manual leak valve. Spectra collections were obtained immediately once the NAP cell pressure reached the desired value.

AP-XPS spectra were fit using literature values for the elemental and associated oxidation states for Si 2p (Ref. 14) and Ge 3d.¹⁵ The Si 2p and Ge 3d spin–orbit split doublets and branching ratios were fixed, and the elemental components were fit with Gaussian/Lorentzian peaks and the oxide components were fit with Gaussian peaks. A Shirley background was used for fitting the spectra.

The National Institute of Standards software SESSA has been used to characterize the mixed SiGe oxide layer.¹⁶ Spectrometer settings (photoelectron emission angle, acceptance angle, etc.) for SESSA were set to experimental values. For our analysis, the atomic density for pure Si_0.60Ge_0.40 (n_SiGe) was set to 4.775 × 10²² atoms/cm³, which was derived from the mass density of 3.6405 g/cm³.¹⁷ Densities of SiGe oxide layers of varying Ge fraction were determined by linear interpolation between that of amorphous SiO₂ (n_SiO2 = 2.25 × 10²² atoms/cm³)¹⁸ and amorphous GeO₂ (n_GeO2 = 2.12 × 10²² atoms/cm³, calculated from the mass density of 3.677 g/cm³).¹⁹ 

High-resolution core-level spectra were performed in order to evaluate the oxidation state of Si and Ge at the interface and to obtain spectra that will assist with peak fitting the AP-XPS data obtained during operando SiGe oxidation experiments. In Figs. 1(a)–1(d), we provide Si 2p, Ge 3d, O 1s, and C 1s spectra, respectively, from an initial SiGe surface and after oxidation at $PO2=10−4,10−2,and1mbar$⁠. Each of the spectra includes the experimental data, which is presented as black dots, while the fit to the experimental data is presented as a solid black line.

For Fig. 1(a), only the Si 2p_3/2 component was included for the Si⁰, Si¹⁺, Si²⁺, Si³⁺, and Si⁴⁺ oxidation states which are located at E_b = 99.4, 100.4, 101.2, 102.0, and 103.2 eV, respectively. For Fig. 1(b), only the Ge 3d_5/2 component was included for the Ge⁰, Ge¹⁺, Ge²⁺, Ge³⁺, and Ge⁴⁺ oxidation states which are located at E_b = 29.4, 30.2, 31.2, 32.3, and 33.3 eV, respectively. The dominant species for the initial Si 2p spectra, Fig. 1(a), were both Si⁰ and Si¹⁺, where the Si¹⁺ component was likely due to exposure to air when transferring the sample into the AP-XPS system. After oxidation, the Si⁴⁺ component becomes more intense, where the relative intensity of the Si⁴⁺ component was found to decrease in relative intensity compared to the Si⁰ component for decreasing $PO2$⁠. The dominant species for the initial Ge 3d spectra, Fig. 1(b), were both Ge⁰ and Ge¹⁺, where the Ge¹⁺ component was again likely due to exposure to the air when transferring the sample into the AP-XPS system. After oxidation, the Ge⁴⁺ component becomes more intense, where the Ge⁴⁺ component was found to significantly decrease in relative intensity compared to the Ge⁰ component for decreasing $PO2$⁠. Compared to the Si⁴⁺ component, we found that the Ge⁴⁺ component was significantly reduced after oxidation at $PO2=10−4mbar$ as opposed to oxidation at 10⁻² and 1 mbar. For Fig. 1(c), the O 1s spectra were fit by two components, where the peaks at E_b = 532.4 and 534.2 eV correspond to O 1s for Si_1−xGe_xO₂ and the Ge L₃M₂₃M₂₃ Auger, respectively. For Fig. 1(d), the C 1s was fit using a single component with E_b = 284.8 eV. This corresponds to aliphatic C on the surface. The O 1s and C 1s spectra from the initial sample indicate that the surfaces were partially oxidized and have surface carbon contamination prior to performing the oxidation experiments.

An analysis of spectra in Figs. 1(a) and 1(b) indicates that the dominant oxidation states in the grown oxide were Si⁴⁺ and Ge⁴⁺, respectively. To track oxide growth, we performed AP-XPS during the oxidation of the SiGe surface. We have fit all of the spectra using fitting parameters determined from fitting the spectra in Figs. 1(a) and 1(b). In Figs. 2(a) and 2(b), we plot the ratio of the Si⁴⁺/Si⁰ and Ge⁴⁺/Ge⁰ obtained from fitting the Si 2p and Ge 3d spectra, respectively, at each $PO2$ versus oxidation time.

These results indicate that the Si⁴⁺/Si⁰ and Ge⁴⁺/Ge⁰ intensity ratios increase for longer oxidation times and also increase for higher $PO2$⁠. There also appears to be at least three regions, where the Si⁴⁺/Si⁰ and Ge⁴⁺/Ge⁰ intensity ratios initially rapidly increase; followed by a second transitionary region; and finally, a quasisaturated regime. Based on the larger Si⁴⁺/Si⁰ ratios compared to the Ge⁴⁺/Ge⁰ ratios for identical conditions, it was apparent that Si was preferentially oxidized compared to Ge at all $PO2$⁠. We also found that the Ge⁴⁺/Ge⁰ intensity ratios significantly decrease at $PO2=10−4mbar$ compared to the Si⁴⁺/Si⁰ intensity ratios. These results suggest that there was even further preferential oxidation of Si compared to Ge at lower $PO2$⁠. It is generally accepted that the interface oxidation reaction is limited in the very thin oxide (<25 Å) regime due to either breaking of Si–Ge bonds²⁰ or O₂ dissociation.²¹ The preferential oxidation of Si in the rapid regime can, thus, be explained by the thermodynamic instability of GeO₂ relative to SiO₂ causing oxygen to preferentially react with silicon at the surface.²² 

Using the Si 2p and Ge 3d experimental data and values for the inelastic mean free paths, the thickness and composition of the grown oxide can be determined. Initially, we used Eq. (1) to estimate the thickness of Si_1−xGe_xO₂ using Si 2p photoelectron peak intensities,

where λ_ox is the Si 2p photoelectron mean free path in the oxide, R_exp is the experimental Si 2p intensity ratio of oxidized silicon and unoxidized silicon I_ox/I_Si for a given experimental spectrum, and R₀ is the Si 2p intensity ratio of I_ox,∞/I_Si,∞ which is determined by obtaining spectra from thick thermal oxides and H-terminated Si(001) surfaces.²³ This method assumes that the ratio R₀ for thick thermal oxide grown on H-terminated Si(001) is equal to R₀ for thick thermal SiO₂ oxide on Si_1−xGe_x(001), as this intensity ratio may change based on the Ge fraction in the alloy. To obtain simulated values, we have used the SESSA software package. For the analysis, we calculated photoemission spectra where we varied both the thickness and composition of the Si and Ge in the Si_1−xGe_xO₂ film. Simulations were initially performed with a value of x = 0.40 for Si_1−xGe_xO₂, and then the oxide thicknesses were varied between 2 and 50 Å. The value of x was then decreased by trial and error until a reasonable match was achieved when we compared the simulated and experimental Si⁴⁺/Si⁰, Ge⁴⁺/Ge⁰, Si⁴⁺/Ge⁴⁺, O/Ge⁰, and O/Si⁰ intensity ratios. In Figs. 3(a)–3(c), we show the Si⁴⁺/Si⁰ and Ge⁴⁺/Ge⁰ intensity ratios versus oxide thickness for oxidations performed at three $PO2$⁠. The simulated data were obtained using SESSA and were compared to the experimental data calculated with (a) λ_ox = 3.57 nm and R₀ = 0.31; (b) λ_ox = 3.56 nm and R₀ = 0.26; and (c) λ_ox = 3.54 nm and R₀ = 0.22.

In Fig. 3(a), we find that the simulated Si⁴⁺/Si⁰ and Ge⁴⁺/Ge⁰ intensity ratios, for $PO2=1mbar$⁠, provide a good fit to the experimental data for an oxide thickness of 33 Å and an oxide composition of Si_0.71Ge_0.29O₂. Likewise, for Figs. 3(b) [and 3(c)], we find that the simulated Si⁴⁺/Si⁰ and Ge⁴⁺/Ge⁰ intensity ratios, for $PO2=10−2mbar$ (10⁻⁴ mbar), provide a good fit to the experimental data for an oxide thickness of 28 Å (17 Å) and an oxide composition of Si_0.83Ge_0.17O₂, (Si_0.95Ge_0.05O₂), respectively. For the two experiments run at higher $PO2$⁠, we find that at ∼25–30 Å oxide thickness, there was a change in the Ge⁴⁺/Ge⁰ intensity ratio slope, where the ratio was found to increase compared to the SESSA prediction using a constant oxide composition. We also find that the (Si⁴⁺/Si⁰)/(Ge⁴⁺/Ge⁰) intensity ratio slope also increased in the same regime. This suggests that there may be an increase in the Ge composition in Si_1−xGe_xO₂ for thicknesses above ∼25–30 Å.

To show the change in oxidation rate depending on experimental conditions, it is useful to compare the oxide thickness versus time. In Fig. 4, we show the oxide thickness versus time for $PO2=1$⁠, 10⁻², and 10⁻⁴ mbar at 300 °C. In all three cases, we find that there was an initial rapid regime (dashed line included to guide the eye). This was followed by an intermediate transitionary regime, which was followed by a quasisaturation slow regime (dashed line included to guide the eye).

The growth rates can be determined for the rapid and quasisaturation slow regimes and compared between three $PO2$⁠. These results are given in Table I. We find that the rapid region was very sensitive to $PO2$ where the growth rate increases by 2.5× and 4.8× when increasing $PO2$ from 10⁻⁴ to 10⁻² mbar and 10⁻⁴ to 1 mbar, respectively. Likewise, the quasisaturation slow region was much less sensitive to $PO2$ where the growth rate increases by 1.5× when increasing $PO2$ from 10⁻⁴ to 10⁻² mbar and 10⁻⁴ to 1 mbar. Furthermore, the transition from the rapid and intermediate transitionary regime occurs at ∼7, 10, and 15 Å for $PO2=10−4$⁠, 10⁻², and 1 mbar, respectively. Oxides grown on Si have been found to have a very thin compositional transition layer which is usually estimated as being less than ∼10 Å thick.^24,25 The rapid oxidation regime observed in our experiments are at or below this thickness, and the changes in the oxidation rates in this regime may be related to the relative oxygen solubility in this compositional transition layer.

It has previously been shown that SiGe oxidizes at a faster rate than Si at low partial pressures.⁴ Ultraviolet photoelectron spectroscopy measurements have also indicated that the oxidation of Ge occurs initially through the formation of GeO.²⁶ The oxidation of Ge was found to be very sensitive to $PO2$ and temperature, where the low-temperature (<520 °C) and high-pressure (>1 atm) regime was found to result in more dense GeO₂.²⁷ To compare the oxidation rates between Si_0.60Ge_0.40(001), Si(001), and Ge(001), we have performed AP-XPS at 300 °C and $PO2=1mbar$⁠. In Figs. 5(a) and 5(b), we show the Si⁴⁺/Si⁰ and Ge⁴⁺/Ge⁰ intensity ratios for Si_0.60Ge_0.40(001) and Si(001) and Si_0.60Ge_0.40(001) and Ge(001), respectively.

We find that under these oxidation conditions that SiGe oxidizes much more rapidly than either Si(001) or Ge(001). These results also indicate that Si(001) oxidizes much more rapidly than Ge(001). Prior studies have indicated that the oxidation of Ge(001) occurs through the formation of GeO, and that GeO desorbs from the surface at 400 °C.²⁴ For our AP-XPS analysis, we did not observe Ge²⁺ during the oxidation process and we kept the oxidation temperature below the GeO desorption temperature. As a result, under the experimental conditions studied, we find that the Ge(001) surface was relatively inert to oxidation, which was quite different compared to the Si_0.60Ge_0.40(001) and Si(001) surfaces.

It is useful to compare the observed experimental results for SiGe oxidation to those predicted by the Gibbs free energies of formation.^3,5 Three basic reactions can be considered depending on the temperature and oxygen partial pressure. These reactions include,

In the following discussion, we are using a similar analysis as Song et al.⁵ although it was modified for the experimental conditions used in our study. Gibbs free energy of formation were estimated using thermochemical data from published values.²⁸ In Fig. 6(a), we show the Gibbs free energy of formation versus both temperature and $PO2$⁠. The reaction corresponding to Eq. (2) is given by a solid black line. There was very little oxygen pressure dependence for this reaction, and as a result, we are only showing the data for $PO2=103mbar$⁠. Equation (3) was much more sensitive to $PO2$ and we show the effect of changing $PO2$ between 10⁻⁴ and 10³ mbar. Finally, Eq. (4) was also found to be insensitive to $PO2$ and is indicated by the dashed black line. The direct formation of SiO₂ from Eq. (2) has by far the most negative Gibbs free energy change, indicating that SiO₂ is the most stable material for the oxidation of SiGe. Depending on $PO2$ and temperature, it was found that the Gibbs free energy for Eq. (3) crosses values of Eq. (4). This indicates that GeO₂ can react with Si at reduced $PO2$ to form SiO₂. In Fig. 6(b), we plot the points where Eq. (3) crosses Eq. (4) in Fig. 6(a) to indicate where we would expect to have the formation of pure SiO₂ versus Si_1−xGe_xO₂. The data was then plotted as a phase diagram with $PO2$ versus temperature.

From Fig. 6(b), one can see that decreasing $PO2$ while increasing the oxidation temperature will result in the formation of pure SiO₂. These trends have been reported previously and help provide guidance on how to form SiO₂ at the SiGe interface.⁵ We have also included our AP-XPS results in Fig. 6(b). These results are consistent with the Gibbs free energy analysis, where there was a reduction in the amount of Ge incorporated in the Si_1−xGe_xO₂ films when $PO2$ was reduced at a given temperature. These studies suggest pure SiO₂ could be formed by both increasing the oxidation temperature and reducing $PO2$⁠. However, it should be noted that going to temperatures above 600 °C may result in GeO desorption and this additional reaction should be included in the analysis.²⁹ 

We have demonstrated that chemical-state resolved AP-XPS is effective to monitor the initial stages of SiGe oxidation. These operando studies allow us to evaluate the oxidation rates, oxide composition, and chemical state formation in real time. We have found that the Si_1−xGe_xO₂ film composition was very sensitive to $PO2$ and have compared these results to expectations from a Gibbs free energy analysis for SiGe oxidation. We find that reducing $PO2$ at 300 °C results in a significant reduction in the amount of Ge incorporated in the oxide film, which is consistent with the Gibbs free energy analysis. These data suggest a path for the formation of high-quality dielectrics with reduced interface trap density for SiGe FinFETs. Future studies should be performed over a greater temperature range and lower $PO2$ to determine optimal conditions to form pure SiO₂. The formed dielectrics should be electrically characterized. AP-XPS experiments performed at higher photon energies, available at synchrotron radiation sources, will also allow thicker oxide films to be quantified due to the larger photoelectron mean free path for higher kinetic energy electrons.

Part of this research was conducted at the Northwest Nanotechnology Infrastructure, a National Nanotechnology Coordinated Infrastructure site at Oregon State University which is supported in part by the National Science Foundation (NSF) (Grant No. NNCI-2025489) and Oregon State University. Acquisition of the ambient-pressure x-ray photoelectron spectroscopy/ambient-pressure scanning tunneling microscopy system was supported by the National Science Foundation-Major Research Instrumentation program (Grant No. DMR-1429765), the M. J. Murdock Charitable Trust, Oregon BEST, Oregon Nanoscience and Microtechnologies Institute, and Oregon State University.

The data that support the findings of this study are available from the corresponding author upon reasonable request.