Reconstructed Concentration Depth Profiles from XPS

XPS was utilized to gain qualitative chemical data from the uppermost 10 nm of the surface chemistry of layered thin-film materials.  We illustrate how a more surface-sensitive approach of angle-resolved XPS is used to probe only the topmost 1–3 nm of a material.  The removal of surface contamination is shown to affect the subsequent Maximum Entropy Method (MEM) reconstructed depth profile, with gentle sputter cleaning using the Gas Cluster Ion Source (GCIS) improving the decription of the layer structure.

Introduction

In many industries, thin-film technology applications are of commercial importance and are often used to influence the physical and chemical properties of bulk materials. In collaboration with IMEC, we demonstrate a multi-technique investigation of layered thin film and ultra-thin film coatings using a model system for gate oxide structures.

A combination of techniques allows one to understand material composition and how subtle disparities in chemistry and stoichiometry can affect substrate properties to enhance its application specificity.

Angle-Resolved X-ray Photoelectron Spectroscopy (ARXPS) is often used to analyse thin films. However, determining the depth distribution of elements from this data to construct a reconstructed depth profile is difficult. The established Maximum Entropy Modelling (MEM) method can resolve this issue by improving confidence in measuring elemental and chemical state depth distribution in thin films.

Here we detail development of a standard approach to characterize real-life thin-film material, the contamination effect on the calculated MEM model fit can also be investigated by removing the adventitious carbon overlayer using the Gas Cluster Ion Source (GCIS) and reacquiring ARXPS data.

Experimental

Using a state-of-the-art AXIS spectrometer fitted with a cluster ion source, the layered thin film materials were analyzed. To mitigate against the loss of photoelectrons and subsequent charge build-up, the co-axial charge neutralizer was used. For each element, survey spectra were acquired over a sizable energy range of 0 to 1350 eV, while spectra with a high resolution were acquired over a small energy range.

For layered thin-film materials, data analysis is complicated by the fact that there is signal from several layers within the sampling depth of the technique. ARXPS explores changes in photoelectron peak intensity with grazing angles and MEM provides method of data assessment, being an established method that uses a simple statistical model for quantitative analysis. For ease of use, it is incorporated into the ESCApe software and after recent work by K. Macak¹, this allows the addition of important parameters to refine the MEM model fit, such as stoichiometry, density and number/thickness of layers.

The parameters are important in layered thin-film materials since they allow one to fit different chemical states for elements with differing compositions and densities, rather than presuming regularity throughout the layers. In samples with repeating, identical layers or when there are layers of the same element with a different chemical state, such as the Si substrate and SiO₂ layer(s) seen in these instances, this is vital.

A wide range of materials may be sputter cleaned or depth profiled without any chemical damage to the surface using a combination of cluster size and acceleration voltage of the GCIS.  To remove surface contamination the GCIS was used with 5 keV Ar₂₀₀₀⁺, providing a low etch rate for gentle sputter cleaning.

Results and Discussion

Ideal Reference Sample

Provided by IMEC, an ideal thin film reference sample was primarily used to create and improve a standard analysis workflow. Prepared and characterized by T. Conard et al² and with a well-determined layered structure, this sample comprised 2 nm hafnium oxide on 1 nm silicon oxide on a silicon substrate. The survey spectrum attained was as expected, but demonstrated a C 1s peak as a result of adventitious carbon contamination.

Figure 1. ARXPS Si 2p spectra for ideal reference sample (as received).

To determine elemental and chemical state concentrations as a function of depth, an  ARXPS experiment was performed. Figure 1 shows the ARXPS Si 2p spectra for the ideal reference sample, where the SiO₂ peak increases with the grazing angle, as anticipated. In agreement with the layer thicknesses expected for this thin film sample, the resulting model fit from MEM provides a calculated restructured concentration depth profile.

Figure 2. MEM model fit for ideal reference sample (as received).

The analyzed MEM model fit is complicated by the contamination of adventitious carbon. Essential for removing carbon contamination without causing damage to the substrate, the sample was cleaned using 5 keV Ar₂₀₀₀⁺ large, low energy clusters. The lack of chemical damage caused by the cluster ions is in contrast to monatomic ions, as shown in Figure 3, which compares Hf 4f spectra before (blue) and after (red) argon cleaning with 0.5 keV monatomic and 5 keV Ar₂₀₀₀⁺  cluster modes.

As the Hf 4f spectra look identical before and after cleaning, the 5 keV Ar₂₀₀₀⁺ cluster mode causes no damage to the substrate. In contrast, the 0.5 keV Ar⁺ monatomic mode, obviously damages the substrate as the Hf 4f spectra dramatically changes even when operating with a low-energy monatomic mode.

Figure 3. Comparison of Hf 4f spectra pre– (blue) and post– (red) cleaning with 0.5 keV Ar⁺ monatomic and 5 keV Ar₂₀₀₀⁺ cluster modes.

The elimination of carbon contamination was confirmed by a survey spectrum and then the ARXPS experiment was repeated. As shown in figure 4, the resulting MEM model fit shows the removal of this carbon overlayer, calculating a reconstructed depth profile to give layer thicknesses in agreement with that projected for this thin-film material.

Figure 4. MEM model fit for ideal reference sample after clus-ter cleaning with GCIS.

A More Complex Reference Sample

Due to the success of the workflow for the analysis and characterization of the ideal reference sample, it was applied to a slightly more complex reference sample with a supplementary silicon oxide layer. Prepared by T. Conard et al.,² the well-defined layered thin-film structure comprised a 1 nm SiO₂/2 nm HfO₂/1 nm SiO₂/ Si substrate. The resulting survey spectrum acquired was as expected, but again demonstrated a C 1s peak due to the carbon contamination.

To determine elemental and chemical state concentration as a function of depth, an ARXPS experiment was executed and figure 5 displays the ARXPS Si 2p spectra for this more complex, ideal reference sample. Here, as expected, the SiO₂ peak increases with the grazing angle while the elemental Si peak decreases when decreasing the information depth.

Figure 5. ARXPS Si 2p spectra for more complex ideal reference sample (as received).

As seen in figure 6, the subsequent model fit from the MEM software provides a calculated reconstructed concentration depth profile that appropriately models the two identical SiO₂ layers. As a consequence of only having spectral information for this deeper layer at the initial, less grazing angles, the fit is less accurate for the deeper of the two SiO₂ layers.

To achieve more signal from the deeper SiO₂ layer during the more grazing angles of the ARXPS experiment and improve the MEM model fit, the gentle removal of the 1 nm thick carbon overlayer using the GCIS is exceptionally beneficial.

Figure 6. MEM model fit for more complex idea reference sam-ple (as received).

Similar to before, the adventitious carbon overlayer was removed using the GCIS using the low-energy, large cluster 5 keV Ar₂₀₀₀⁺ mode and the removal of carbon contamination was confirmed by a survey spectrum; once more the ARXPS experiment was repeated.

The resulting MEM model fit, as defined in Figure 7, reveals the removal of the carbon overlayer, thus calculating a reconstructed depth profile to give layer thicknesses in agreement with that anticipated for this thin-film material. Overall, particularly at the deeper SiO₂ interface, the MEM model fit is enhanced.

Figure 7. MEM model fit for more complex ideal reference sample after cluster cleaning with GCIS.

This standard analysis workflow is promisingly successful in characterizing two samples containing layered thin films. Therefore it can be confidently used as a procedure to analyze real-life thin film materials.

Conclusions

To characterize ideal reference thin-film materials that are used as model structures for gate oxides, XPS was utilized. The sampling depth for conventional monochromated Al Kα X- rays is typically cited to be 10 nm which can be reduced to 1–3 nm using ARXPS. MEM was used to determine a reconstructed depth profile from the resultant angle-resolved data. This was found to be in agreement with the layer thicknesses determined by complementary techniques for these thin film materials.

The success of this analysis workflow has allowed the development into a standard protocol to analyze real life materials and ensure confidence in measuring elemental and chemical state depth distribution in thin films. Removal of contamination by gentle cleaning using the GCIS was found to vastly improve the resulting MEM model fit, most notably for deeper layers where the information is restricted to the initial, less grazing angles.

References and Further Reading

1. K. Macak, Surface and Interface Analysis, 2011, 43, 1581-1604.
2. T. Conard et al., Journal of Vacuum Science and Technology A, 2012, 30, 031509.

This information has been sourced, reviewed and adapted from materials provided by Kratos Analytical, Ltd.

For more information on this technique, please visit Kratos Analytical, Ltd.