Introduction to reproducible laboratory hard x-ray photoelectron spectroscopy

X-ray photoelectron spectroscopy (XPS) performed with x rays that have a photon energy higher than 2140 eV is called HAXPES.¹ In recent years, HAXPES (hard x-ray photoelectron spectroscopy) is an increasingly used technique² complementing other laboratory analytical methods, especially traditional XPS. Historically, HAXPES has largely been used and developed using synchrotron radiation because laboratory-based high energy x-ray sources were insufficiently intense to generate spectra on a reasonable timescale. There are now more than 20 synchrotron radiation beamlines dedicated to HAXPES and the popularity of the method means that most of these synchrotron beamlines are over-subscribed. Nevertheless, in the last few years, the development of efficient spectrometers and laboratory sources has generated laboratory-based instruments capable of delivering HAXPES spectra on a timescale sufficiently quickly for practical applications. While these instruments cannot provide the x-ray intensity of synchrotrons, there are several advantages that laboratory HAXPES instruments have over synchrotrons that make them attractive. These include easy access to the instrumentation; highly reliable and reproducible source intensity; the potential to generate calibration schemes so data from different instruments are directly comparable; the generation of data libraries that can be shared between users; and the development of reliable quantification methods.

This article summarizes the current state-of-the-art in laboratory HAXPES and describes some of the advantages and potential uses of the technique. It provides references to resources and data that may be useful to new users of laboratory HAXPES and highlights areas that require some work to make HAXPES reproducible and reliable in the same manner as a well-run XPS instrument.

The main reason to use HAXPES is to increase the kinetic energy of photoelectrons from relatively low binding-energy core levels, which increases the information depth of the emitted photoelectrons. The information depth in XPS is traditionally taken to be the depth from which 95% of the signal originates: estimated as three times the electron effective attenuation length (EAL). The kinetic energy, E_k, is the difference between the x-ray energy and the sum of the core-level binding energy and the spectrometer work function (typically ∼5 eV). For materials ranging from aluminum to gold and when E_k > 1000 eV, the EAL used for thickness measurements is proportional to (E_k)^p, where p is between 0.855 and 0.875.³ The obvious use of this increased information depth is for structural investigations that are not possible with normal x-ray sources. This includes the analysis of buried interfaces such as electronic layers below a surface capping layer and compositional studies in the bulk of materials and interfaces below a contaminated or damaged surface layer. Figure 1 shows an increase in electron effective attenuation length for Si 2p photoelectrons in silicon dioxide using the four most common x-ray sources available on commercial instruments. High-resolution spectra for 25 nm thick silicon oxide on silicon metal are shown. Using the Cr x-ray source, it is possible to detect the metal substrate beneath the oxide overlayer. This illustrates the ability of HAXPES to look at interfaces buried beyond the traditional XPS information depth. The measurement of overlayer thickness can be performed using traditional XPS methods, simply with a larger EAL in the equations.⁴ A potential advantage of HAXPES in this regard is that many elements have photoelectron peaks widely separated in kinetic energy, which permits a variety of depths to be probed. For example, using a Cr Kα source, Ti 1s photoelectrons have a kinetic energy of ∼444 eV with an associated information depth of ∼3.3–4 nm, while Ti 2p photoelectrons have a much higher kinetic energy of ∼4950 eV with an information depth of more than 20 nm. Variable depth analysis can be performed in a single experiment for studying gradients, segregation, and oxidation phenomena. Such information can add certainty to the analysis or test assumptions, such as the lateral or vertical homogeneity of coatings. In this regard, it can also be useful to correlate both XPS and HAXPES data to obtain different information depths from the same sample.

XPS is well established as an analytical tool for the study of heterogeneous catalysts, and advice on how to use it correctly has been covered elsewhere.⁵ Heterogeneous catalysts often consist of metal domains distributed on an oxide support. XPS data can be used with several models to determine the dispersion of the active metal sites on the surface of the support.^6,7 The Davis model⁸ allows XPS to estimate the particle size of the metal sites, independent of the support surface area and loading of the active metal sites.⁵ The Davis model uses a ratio of the intensities of two photoelectron peaks from the same element that have significantly different kinetic energies such that there is sufficient difference in the EALs of the two types of photoelectron. This requirement means that there is a limitation in the number of elements that can be analyzed when using standard XPS due to the lack of appropriately spaced photoelectron features. However, higher energy x-ray sources used in HAXPES increase the number of elements that the Davis model can be applied to. For example, monochromated Ag Lα HAXPES has been used to study gold and platinum particles using the 3d and 4f photoelectrons,^9–11 and reasonable agreement with STM measurements was obtained.

Another consequence of the larger information depth is a reduction in the effect of surface contamination on the photoelectron spectrum. The fraction of depth that the adventitious layer occupies with respect to the total information depth is smaller for high energy photoelectrons than for Al Kα photoelectrons. The result is a smaller attenuation of the peak and background signals caused by the presence of a contaminant overlayer. Figure 2 shows an overlay between the survey spectra from a stainless-steel sample using both Al x-ray and Cr x-ray sources. For the higher energy HAXPES source, a much smaller C1s peak due to adventitious carbon is apparent as well as higher relative peak intensities for iron and the “cleaner” background shapes largely due to the absence of interfering Auger electron peaks.

The ability to see beyond the contamination layer is particularly important for materials where carbon is a part of the material. Particularly when it is challenging to estimate how long one should sputter-clean the sample until the adventitious carbon is removed yet no damage to the sample is done. The example shown in Fig. 3 is from the Ti₃AlC₂ MAX phase from which XPS quantification includes a higher contribution from the adventitious carbon to the overall C 1s signal. The lowest kinetic energy XPS peak (Ti 2p) is strongly attenuated by the contamination overlayer resulting in a significant underestimation of the amount of titanium in the sample. A much closer composition to the expected stoichiometry is obtained using HAXPES for which a larger proportion of signal originates from beneath the contamination.

For the same reason, the deeper information depth of HAXPES allows one to probe beyond the depth of damage induced by ion sputtering. To take advantage of this benefit, the probing depth of the x-ray source must be deeper than the altered layer depth induced by ion sputtering. Damage depth is strongly dependent on ion beam sputtering parameters and material being analyzed. For commonly used ion beam settings of monatomic Ar⁺ at 2–5 kV, the damage depth is on the order of 10 nm, which is why there are usually significant signs of sputter-induced damage and altered stoichiometry in XPS spectra for many compound materials. For efficient and accurate chemical analysis of buried layers and interfaces, the photoelectrons analyzed must originate from below the ion beam damage depth. HAXPES information depths can be larger than the altered layer depth and should, therefore, provide a better estimate of the deeper, and hopefully unaltered, material. Nevertheless, a significant fraction of the HAXPES signal will come from the altered layer and careful work is required before quantitative analysis is possible. At this point of time, there is a lack of general guidance and protocols in the literature.

Dubey et al.¹² examined Sb as an interfacial layer between Li₇La₃Zr₂O₁₂ (LLZO) solid-state electrolyte and metallic Li. By using HAXPES, it was possible to analyze the unperturbed chemical state of Sb at the buried Sb/LLZO interface during successive sputtering steps. The probing depths for the photoelectron lines of interest using Cr are between 16 and 19 nm, which are larger than the estimated 5 nm depth of the mixing zone induced by sputtering with 1 kV Ar⁺ ions at an incident angle of 45°, from Monte-Carlo simulations. This shows that successive cycles of sputtering and HAXPES analysis permit unperturbed chemical states to be probed at the buried Sb/LLZO interface, which is not possible by conventional XPS sputter-depth profiling.

The kinetic energy of Auger electron peaks is independent of the x-ray energy. When higher energy x-ray sources are used, the Auger transitions observed in XPS remain at the same kinetic energy, while the photoelectron peaks are shifted to higher kinetic energy. On the binding energy scale, this means that the Auger electron peaks appear to move to higher binding energy. The ability to separate Auger electron peaks from photoelectron peaks is very advantageous for analyzing mixtures of 3d transition metals, such as Mn, Co, and Fe, which have significant overlap between core electron lines and Auger peaks in normal XPS. Another example is the analysis of GaN, which is very difficult in traditional XPS due to the N 1s photoelectron line being coincident with the highly structured Ga LMM Auger electron region. Figure 4 shows an overlay of survey spectra using Al and Cr x-ray sources showing that N 1s is free of overlap in HAXPES data allowing for accurate quantification. This general positive assessment of the removal of photoelectron and Auger electron interference needs to be balanced with the fact that the excitation of deep core-level electrons in HAXPES will give rise to high-kinetic energy Auger electron features, which may also cause interferences, so caution is required.

There have been many studies in the 1970s and 1980s that showed that the AP is sensitive to changes in the chemical state when the photoelectron energy is not. This is due to the relationship between the Auger electron energy and the extra-atomic relaxation energy. West and Castle used Zr Lα radiation (2042 eV) to study Auger parameters in silicates.¹⁵ HAXPES sources have also been employed in this regard. Edgell et al.¹⁶ used a monochromatic Ag Lα x-ray source to generate APs from Si KLL, Si 1s, and Si 2p kinetic energies. These were then used to calculate the extra-atomic relaxation energies of the silicon atom in insulating materials. Castle¹⁷ has shown that there is a correlation between the AP and refractive index of aluminosilicates, because the AP is sensitive to the polarizability of the surrounding material. Although APs should be independent of charge correction, Edgell et al. showed¹⁶ that an unoptimized electron flood gun can affect the kinetic energy difference between Si 1s and Si 2p peaks, which could have an impact on the calculation of APs.

Another method of both obtaining thickness information and extending the information depth is through the shape of the inelastic background. The background to the lower kinetic energy of a photoelectron peak contains useful information on the depth distribution of elements in the sample.^18,19 Data analysis relies upon an understanding of the rate at which energetically charged particles lose energy in interactions with the material through which they travel.

This approach, which underpins ion scattering spectroscopies, is not widely used in XPS in part due to the perceived complexity of the analytical problem. Analysis of the data necessitates the use of specialized software such as sessa (Ref. 20) or quases.²¹ In some cases, simplified approaches for specific material systems²² or with limited information content such as Tougaard’s AOS3λ method²³ can be used. The AOS3λ method has been extended to provide a potentially useful analysis of combined XPS and HAXPES data from the same sample.²⁴ A particular challenge in the inelastic background analysis of XPS spectra is the proximity and overlap of peaks from different elements, which makes it difficult to measure background shapes for elemental peaks over an extended range. In HAXPES, the analytical depth of the inelastic background is extended for several reasons: increased information depth, increased kinetic energy range and, importantly, fewer peaks interfering with the energy loss shape. A simple combination of the first two factors suggests that the analytical depth of background analysis should increase more than linearly with the electron kinetic energy. In practice, the useful analytical depth is limited by elastic scattering of electrons, which changes their direction, but for materials made from light elements, the importance of elastic scattering reduces at higher electron kinetic energies.²⁵ Another factor that is not important in XPS is accounting for energy losses from core electrons. This occurs with low probability (typically around 1% of all electron energy losses) but involves a single, large energy loss. When analyzing extended loss structures in HAXPES, the accurate shape of the inelastic background requires the inclusion of these contributions because the frequency of large energy loss events grows in proportion to the number of inelastic scattering events.

Using only photoelectron peaks, the information depth of HAXPES is limited to a few tens of nanometers.²⁶ However, with inelastic background analysis, the information depth can extend from more than 50 nm to over 200 nm^21,27–32 depending upon the electron kinetic energy and the material being analyzed. Figure 5 illustrates the change in background shape and intensity for an 18 nm iridium complex at various depths in an organic matrix. The iridium complex contains 3 at. % Ir and is approximately equivalent to a monolayer of iridium metal in terms of the amount of iridium. Without any overlayer, sharp iridium photoelectron peaks are observed, which are representative of the intrinsic emission line shape from the compound. Within the spectra, there are several hundred eV separation between peaks, which provides a large analysis depth. Note that the samples depicted in Fig. 5 have a 1000 nm organic underlayer and this acts to eliminate the background signal from the silicon substrate, specifically the background arising from the Si 1s peak at ∼1839 eV binding energy. Because low energy electrons from the substrate also act to provide charge compensation for insulating overlayers in monochromatic instruments, these thick samples accumulate charge in the HAXPES instrument and require charge compensation using low energy electrons.

With a 50 nm overlayer, peaks are just observable which is consistent with the expected 3L ≈ 45 nm information depth at these energies. The broad background shape arises from iridium photoelectrons that lose kinetic energy moving through the organic material and, as the overlayer gets thicker, the features move to lower kinetic energy (higher binding energy) and broaden through the statistical nature of the energy loss process. The mean energy loss for electrons during an inelastic collision is ∼30 eV; therefore, the inelastic loss maximum for defined layers such as this iridium layer sample appears at ∼2 eV/nm below the photoelectron peak (after dividing the mean energy loss by the inelastic mean free path). Thus, at 200 nm depth, the maximum of the background “hump” is ∼400 eV from the original photoelectron peak and this accords with the data presented in Fig. 5. The background position and shape can be analyzed more accurately to return information on the depth distribution of the element; in the example shown the quases software package was used. With an appropriate energy loss function, a relative accuracy of ∼10% in depth for the sharp upper boundary of the iridium-containing layer was obtained.

A detailed discussion of inelastic background analysis is beyond the scope of this paper; however, the example presented here demonstrates the capability to measure the depth distribution of submonolayer concentrations of elements more than 100 nm deep into a sample. Such analysis is not necessarily damage-free, but it is certainly much less damaging than sputter depth profiling. What is not possible is a detailed chemical state analysis, which is a major strength of XPS. In effect, the information obtained is similar to medium-energy ion scattering and the fundamental principles, based upon the slowing down of charged particles traveling through a material, are almost identical.

In principle, the total electron intensity in the peaks and the background can be used to measure the absolute amount of an element within the extended information depth; however, there are several areas that require further work before this is possible. These include accurate calibration of the intensity scale of HAXPES instruments, consideration of electron angular emission including x-ray polarization and nondipole effects,^33,34 and the incorporation of the kinetic energy and material dependency of inelastic mean free paths and elastic scattering into analysis packages. Finally, it is rare that energy loss functions used in electron spectroscopy incorporate the losses associated with core-level excitation because core-level losses generally comprise less than 1% of all inelastic losses. Ignoring core-level losses is acceptable in XPS but may not be when an extended loss structure over many hundreds of eV is analyzed. For the 200 nm overlayer the photoelectrons have undergone, on average, 13 inelastic loss events. Therefore, the fraction of electrons having at least one energy loss due to either C 1s or O 1s scattering (with energy losses greater than 285 and 532 eV, respectively) will exceed 10%. Core-level losses will start to have a significant effect on the shape and intensity of background loss features for layers buried more than ∼10λ beneath the surface.

The essential difference between XPS and HAXPES is that XPS, using Al Kα and Mg Kα x-ray sources, has been commercialized and standardized over many decades, whereas HAXPES is in the early stages of commercialization and standardization. Therefore, there are significant resources in terms of reference energies and intensities that permit the calibration of instruments and ensure comparability in data. In this section, we outline a roadmap to the calibration of HAXPES instruments.

Many experiments undertaken using XPS rely on the determination of photoelectron binding energies; it is critical that the binding energy scale has been calibrated and that it is linear across the whole binding energy range of the spectrometer. This should be done with a multipoint calibration procedure using photoelectron peaks of well-established binding energies, which should cover the extremes of the binding energy scale. For XPS instruments that use Al Kα (1486.7 eV) and Mg Kα (1253.6 eV) x-ray sources, the ISO 15472 method is used. This employs the binding energies of the Au 4f_7/2 (83.96 eV), Ag 3d_5/2 (368.21 eV), and Cu 2p_3/2 (932.62 eV) photoelectron peaks, which provide calibration points at the upper, middle, and lower extremes of the kinetic energy scale, respectively. Because the energy range of the kinetic energy analyzers used in laboratory-based HAXPES instruments far exceeds 932.62 eV, the method in ISO 15472 becomes less applicable. This has been addressed in the paper by Siol et al.,³⁵ where the authors proposed using Ag 3d_5/2 and Cu 2p_3/2 photoelectrons as calibration points at low binding energy and Cu L₃M_4,5M_4,5 and Ag M₄N_4,5N_4,5 x-ray induced Auger electrons at high binding energy. The use of an x-ray induced Auger electron as the calibration point for the high binding energy scale has a distinct advantage as it has a kinetic energy that is independent of the x-ray energy used. Thus, Leadley described a method³⁶ for checking the binding energy scale linearity of HAXPES instruments using the binding energy position of the Ag 3d_5/2 photoelectron peak and the kinetic energy of the Ag M₄N_4,5N_4,5 x-ray induced Auger peak. This is applicable to all HAXPES instruments because the Ag M₄N_4,5N_4,5 peak will always be located at the high binding energy of the scale, irrespective of the x-ray energy used.

Nevertheless, there is a shortcoming of using the x-ray induced Auger peak as a reference point for highly accurate work, which is the dependence on the accuracy of the x-ray energy. Unfortunately, x-ray energy can fluctuate slightly due to the monochromator settings, which might give the impression of changes in binding energy scale linearity. Thus, it would be preferable to use a photoelectron peak binding energy that is present at the high binding energy end of the scale. Ideally, this photoelectron should be generated from an electrically earthed high purity metal foil that can be ion etched to remove native oxides and adventitious contamination. In addition, the photoelectron peak should have a kinetic energy greater than 200 eV to avoid the secondary electron cascade and be sufficiently sharp that a well-defined maximum value can be found. In the case of HAXPES instruments using Ag Lα (2984.3 eV) x rays, this could be the Au 3d_5/2 photoelectron peak that has a binding energy of 2206.7 eV.³⁷ For HAXPES instruments using Cr Kα (5414.8 eV) or Ga Kα (9251.7 eV) x rays, the Ti 1s photoelectron peak (4966 eV) or the Cu 1s photoelectron peak (8979 eV) could be used.

The recoil energy is subtracted from the kinetic energy of the electron and, therefore, increases the binding energy scale position of the peak. Because the atom will be bonded to other atoms, the recoil will be coupled into vibrational states and the probabilistic occupation of those states will lead to a broadening of the photoelectron line,⁴¹ which is evident in the higher energy B 1s spectrum in Fig. 6. The shift of 0.13 eV demonstrated in that figure agrees with Eq. (3) and indicates a binding energy position for B 1s of ∼188.31 eV for Al Kα XPS and a recoil-induced shift of ∼0.06 eV even for this “normal” XPS value.

Figure 7 plots the consequences of Eq. (3) on the binding energy scale positions of the same peaks measured at different photoelectron energies. The horizontal dashed line represents the ∼0.05 eV uncertainty for a well-calibrated XPS instrument analyzing conductive samples. It is notable that, up to 10 keV photoelectron energy, the predicted recoil of silver is within this limit and, therefore, for elements heavier than silver, the recoil effect in HAXPES is only consequential for the most accurate work. For transition metals, such as iron, the recoil effect becomes important for photoelectrons above 5000 eV and for silicon, the effect is important above 2500 eV. For lighter elements, such as carbon and lithium, recoil energy should be observable even in the XPS energy range.

There are consequential requirements concerning HAXPES energy referencing. Fixed reference energies, which are independent of the photon energy, cannot be strictly justified, but the use of fixed binding energy scale positions for elements heavier than silver will be useful until the necessary standards are available. For “softer” HAXPES instruments with photon energies less than ∼5000 eV, the use of fixed reference energies for elements heavier than iron is similarly justifiable.

The practice of using calibrated Al Kα binding energies to fix the binding energy scale position of the same photoelectron peaks in HAXPES requires care. For conductive samples in electrical contact with the spectrometer, there is no need to perform energy referencing if the spectrometer work function is known. For light elements in a conductive sample, it may be possible to use Eq. (3) to adjust the binding energy scale position of the peak to compare XPS and HAXPES positions, but such an approach has not been tested.

Energy referencing in XPS measurements of insulators has been performed using the peak positions of adventitious carbon, evaporated gold, and implanted argon atoms. None of these methods are satisfactory even in XPS, generally resulting in an uncertainty of more than 0.5 eV in the binding energy scale. For HAXPES, both the recoil-induced shift and the small cross section of C 1s at high x-ray energies increase the uncertainty in the C 1s peak position to an even more unacceptable level. Current practice is to use both XPS and HAXPES sequentially on the sample, using whatever charge reference method is preferred by the analyst for the XPS data and then energy reference the HAXPES data to an appropriate peak in the XPS data assuming that the binding energy position does not change, or changes in accordance with Eq. (3). Then, the energy reference is no worse than XPS, but still makes the mistake of trying to reference insulator binding energies to the Fermi level of the spectrometer. For XPS and especially for HAXPES, there is a clear and ongoing need to find improved methods for charge compensation and energy referencing for insulators. Refer to a recently published paper on neutralizing insulators for more guidance.⁴² 

Quantitative elemental analysis using XPS is well understood, relying upon the use of sensitivity factors and knowledge of the instrument energy-dependent transmission characteristics.⁴ For XPS using traditional Al Kα and Mg Kα x-ray sources, accurate reference spectra for clean copper, silver, and gold are available to characterize the instrument transmission and calibrate the intensity scale of spectra.⁴³ Currently, no publicly accessible intensity calibration procedure is in place to calibrate the intensity scale of HAXPES instruments except for monochromated Ag Lα instruments that have an x-ray source to analyze a lens angle of ∼60°.⁴⁴ Reference spectra of copper, silver, and gold for this type of analyzer have been made publicly available as electronic data.⁴⁵ Nevertheless, most manufacturers have developed procedures to account for instrument transmission in different operating modes and have libraries of sensitivity factors appropriate to their instrument and calibration procedures. The comparability of intensity calibration procedures between manufacturers and their relationship to an accurate calibration remains unclear at this stage.

To quantify soft and hard x-ray photoelectron spectroscopy data, a background is first subtracted from the experimental spectrum. The integrated peak intensity is then divided by a relative sensitivity factor (RSF) to normalize for various physical and instrumental factors that affect the signal. RSFs can be derived from first principles for an accurately calibrated instrument or determined empirically from reference materials for each instrument. ISO 18118:2015 “SCA-XPS Guide to the use of experimentally determined RSFs for the quantitative analysis of homogeneous materials” has details on different types of sensitivity factors.

The best option is to determine sensitivity factors for the specific instrument, but this is a time-consuming task requiring high purity and clean reference materials. If there is no transmission correction procedure for the instrument, then the sensitivity factor determination needs to be repeated on a regular basis. Recently, theoretical RSFs have been generated by the National Physical Laboratory (NPL) and they are applicable to true, calibrated XPS spectra. Theoretical RSFs are essentially the product of a calculated photoionization cross section, typically using Hartree–Slater or Dirac–Fock methods, and the estimated inelastic mean free path of the generated electrons. To account for instrumental parameters such as the angle of emission of the detected electrons relative to the photon source, various asymmetry parameters need to be considered,³⁴ as well as elastic scattering of electrons in the solid. Theoretically derived quantification parameters assume that either: all the photoelectron intensity, main peak plus shake-ups, is captured; or that all elements lose the same fraction of intensity to shake-up processes. Cant et al. described a method of calculating average matrix relative sensitivity factors (AMRSFs) for any given instrumental setup with an excitation source energy between 1.5 and 10 keV.³³ 

Figure 8 shows Si 1s and Si 2p high-resolution HAXPES spectra from pure Si, nitride, and oxide. The fraction of signal that is lost to final state effects is listed in each panel of the figure for each compound. One can see that, for an element in a compound, the fraction of a photoelectron signal originating from a given orbital that goes into the main peak varies with the chemical state of the atom. The fraction may also vary from core level to core level for a given atom in a given chemical state. There are often profound differences in shake-up intensity for different elements in a compound, a clear example being the shake-up intensity of the Cu 2p peaks from CuO, where the shake-up intensity is at least half of the total intensity compared to the O 1s peak in which obvious satellite peaks cannot be seen.

The first practical limitation of theoretical RSFs is that one should acquire a much wider region than the main peak. For Si 2p, for example, instead of acquiring ∼20 eV energy window, one should acquire 60 eV range or even more. For higher energy sources, where some transitions, such as Si 2p, have very low photoelectric cross sections, it may become impractical from a time of acquisition standpoint. Another challenge is spectral overlap, which becomes much more probable when a wider acquisition region is used. For example, when the materials under investigation include both aluminum and silicon atoms, the Si 2s and 2p peaks sit upon extended signals coming from the Al 2s and 2p peaks, respectively.

For experimentally derived RSFs, there is an assumption that the fraction of the total intensity captured in the main peak is the same for the compound under evaluation as it was for the reference compound used to produce the RSF. If this is not true, then the experimental RSF determined is strictly valid only for the specific chemical state of the atom in the reference compound. For routine XPS and HAXPES analysis, it is common to use empirical sensitivity factors and the peak areas defined only by the main peak and excluding shake-up structure. Ideally, empirical sensitivity factors should be established using reference compounds with similar chemistries. For example, ionic liquids are a useful reference material for establishing the RSFs of organic compounds.^46–48 Ionic liquids have a clean surface, contain many elements, are stable to vacuum and x rays, and have some electrical conductivity.

Basic quantification obtained using any RSF assumes homogeneous-equivalent atomic concentrations. This assumption of homogeneity may not be true, particularly over the significant information depths measured in a HAXPES experiment. Therefore, the interpretation of quantification results, especially for materials with vertical heterogeneities—surface segregation, gradients, oxidation layers, core-shell structures—must be taken with care, because information depth for the individual peaks used for quantitative results is significantly different. This is, of course, well known in traditional XPS and a different form of reporting, for example, thicknesses for vertical heterogeneity, is often preferred. Additionally, different forms of analysis, such as a variable take-off angle, can be helpful. In this regard, the combination of XPS and HAXPES offers the possibility of providing greater insight into the depth distribution of chemical species.⁴⁹ 

Sample charging is a common problem in XPS analysis; it arises when the flux of electrons out of a region of the sample is not matched by the corresponding influx of electrons. As regions of the sample become positively charged, the net influx of electrons increases. Provided that there is a constant and sufficient supply of electrons from an electron source, usually an earthed, conductive substrate or an electron flood gun, then a steady state will be achieved. This means that different areas of a heterogeneous sample may charge by different amounts, a phenomenon known as differential charging. Differential charging typically results in the broadening and shift of peaks on the binding energy scale. The effect is obvious when the same chemical states exist in regions of different charges, but less obvious when regions of different charges are chemically distinct. Differential charging is difficult to distinguish from inbuilt charge differentials arising, for example, in electrochemical systems or at semiconductor interfaces.

The same effects occur in HAXPES analysis, although the scale lengths for electron transport are different. There are only a few reports on charge compensation in HAXPES and it is unclear whether different strategies from XPS should be applied.^30,50 Nevertheless, if there is an intention to measure nonconducting samples then some form of charge compensation, such as an electron flood gun, is required.

Many laboratory-based HAXPES instruments use the same highly sensitive detection system as the monochromated Al Kα x rays. However, the x-ray intensity from the higher energy monochromated x rays is often significantly lower than the Al Kα source. For example, on an instrument using a dual Ag Lα/Al Kα monochromated x-ray source, the signal intensity from the Ag Lα source is approximately 50-fold lower than from the Al Kα source.⁴⁴ This means that the HAXPES data have to be acquired for longer time periods so that reasonable spectral data are acquired. To achieve similar statistics between XPS and HAXPES data, the dose of x rays must be similar. For this reason, the time taken for HAXPES analysis on most laboratory instruments is greater than for XPS.

Any XPS data acquisition sequence should include a wide scan survey spectrum to determine the elements present at the sample surface at the start and end of the experiment.⁵¹ HAXPES data acquisition may require an increased x-ray dose compared to XPS if the full kinetic energy range of electrons is acquired to achieve the same quality survey spectrum. When this is combined with the lower signal intensity, the acquisition of survey spectra will take a considerably longer time when compared to using monochromated Al Kα x rays. For example, when a survey spectrum acquired using Al Kα x rays takes at least 2 min, on the same instrument a full survey spectrum acquired using Ag Lα will take more than 30 min. A typical acquisition sequence also includes high-resolution spectra to determine peak shapes and chemical shifts; HAXPES data acquisition could take many hours depending on the number of elements of interest and the intensity of the x-ray source.

More transitions are accessible using high energy x-ray sources and one may be tempted to choose transitions with the highest photoelectric cross sections. However, transitions with high cross sections are usually closer in binding energy to the energy of x-ray source; hence, they have lower kinetic energy and, thus, come from more shallow information depths. For example, for Cr Kα x-ray source, the cross section of Hf 3d is at least 15 times higher than that of Hf 4f, while the EAL for Hf 3d is ∼2 times smaller than for Hf 4f. When choosing the appropriate transition, the 4f peaks may seem natural to an XPS user; however, the cross section for 4f peaks decreases rapidly with increasing x-ray energy and at these energies, the nearby 5p and 5s peaks will be of similar, or greater, intensity. A useful choice in this case may be the 4d or even 4p peaks that are about five to ten times more intense than the 4f peaks and have a similar information depth.

Although it is often considered that XPS does not cause significant sample damage, there are some samples that are susceptible to surface modification during an XPS experiment; for example, the reduction of metal oxides and dehalogenation of organic compounds.^52,53 This is mostly related to exposure of the sample to the XPS instrument’s charge neutralization system; however, fine focused x-ray spots can also increase the rate of sample surface modification. Thus, it is desirable to reduce these effects. This has become easier with conventional laboratory-based XPS instruments using monochromated Al Kα x rays, as many modern instruments have sensitivities that allow data acquisition in short timescales.

If sample damage is caused by the charge neutralizer and not by the x rays, then the longer acquisition times noted previously imply greater damage. Nevertheless, there are interdependencies. Charge neutralization acts to replace electrons lost from the sample by photoemission processes. Therefore, for low energy electron charge neutralization of insulators, the flux of electrons will depend upon the flux of x rays and the accumulated damage will not simply depend upon acquisition time. If the sample is electrically conducting, then it is preferable to avoid the use of charge neutralization if it is likely to damage. HAXPES sample damage requires a more thorough investigation before authoritative guidance can be given.

For XPS, it is good practice to repeat a survey spectrum and high-resolution spectra of elements to identify sample damage if the sample is suspected to undergo chemical damage. A procedure for estimating the amount of damage and correcting for unintended degradation is given in ISO 18554:2016, which requires multiple sequential acquisitions. For some radiation sensitive materials, such as organic halides, damage occurs very quickly and, to capture the initial composition, sequential and fast acquisitions of the halide core-level region may be required before any other acquisition is performed. For HAXPES analysis, similar precautions should be taken, and the same procedure may be useful.

There are many databases and reference works available for understanding HAXPES data, many of which have been available from the earliest days of XPS. The most used database for photoionization cross sections is the 1973 work of Schofield;⁵⁴ more recent calculations have been carried out by Sabbatucci and Salvat⁵⁵ and Trzhaskovskaya and Yarzhemsky.^56,57 There are absolute differences of ∼10% between the photoionization cross sections provided in the different databases, which depend upon the theory used. Only the Trzhaskovskaya and Yarzhemsky database provides information on the angular distribution of ejected electrons which, along with the instrument geometry³⁴ and transmission function, is important for quantitative HAXPES. These databases provide values of photoionization parameters at specific x-ray energies, which often do not coincide with the experimental x-ray energy.

The EAL of electrons, L, depends upon both the electron kinetic energy, E, and the material through which the electron is passing. For HAXPES, with electron kinetic energies above 1000 eV, the kinetic energy dependence is somewhat simpler than for XPS in a specified material. A power law L ∝ E^0.865 is suitable for film thickness measurements, while L ∝ E^0.825 is suitable for the calculation of intensities from homogeneous materials.³ In both cases, the exponent may vary by ∼0.01 depending upon the material, whereas the constant of proportionality depends strongly upon the material. This separation of the energy and material dependency of EALs justifies the use of AMRSFs whereby a sensitivity factor is applicable to a wide range of different materials but also noting the issues with shake-up intensity mentioned earlier.

Using these databases, the photoionization cross sections, asymmetry factors, and elastic scattering effects^25,58,59 have been interpolated³³ and encoded into an Excel spreadsheet, which can be used to calculate theoretical AMRSFs for any HAXPES instrument with an x-ray energy up to 10 keV. Note that these theoretical sensitivity factors include shake-up and shake-off intensities along with the main peak intensity and, therefore, there is still a need to acquire reference spectra for the most accurate work.

The journal Surface Science Spectra hosts a growing collection of laboratory HAXPES spectra of elements and compounds. These include metallic elements for Ag Lα instruments^60–62 and for Cr Kα instruments,^63–88 binary compounds for Ag Lα instruments^89–92 and Cr Kα instruments;^93–112 and organic compounds such as ionic liquids are also available.¹¹³ Currently, there are few Ga Kα spectra,^114,115 and it is anticipated that this number will increase.

In this article, we have described the main uses of laboratory-based HAXPES and detailed some of the current challenges to generating useful and reproducible data, as well as some of the available information available to help address those challenges. HAXPES provides useful additional information to XPS and, importantly, permits the subsurface of materials and the properties of buried layers to be analyzed. For many advanced electronic devices, the ability to look beyond ∼10 nm is important and permits analysis through contaminated, damaged, or passivation layers. Nevertheless, the information is rarely quantitative at this stage and work needs to be done on the calibration of HAXPES instruments, the availability of data and resources, and the development of procedures to analyze the data.

A.G.S. acknowledges funding from the National Measurement System of the UK Department of Science, Innovation and Technology. The authors thank David Cant (NPL), Caterina Minelli (NPL), and Michaeleen Pacholski (Dow) for helpful advice and comments.

The authors have no conflicts to disclose.

Kateryna Artyushkova: Writing – original draft (equal); Writing – review & editing (supporting). Stuart R. Leadley: Writing – original draft (equal); Writing – review & editing (supporting). Alexander G. Shard: Writing – original draft (equal); Writing – review & editing (lead).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.