Prospect of making XPS a high-throughput analytical method illustrated for a CuxNi1−xOy combinatorial material library

Combinatorial material science crucially depends on robust, high-throughput characterization methods. While X-ray photoelectron spectroscopy (XPS) may provide detailed information about chemical and electronic properties, it is a time-consuming technique and, therefore, is not viewed as a high-throughput method. Here we present preliminary XPS data of 169 measurement spots on a combinatorial 72 × 72 cm2 CuxNi1−xOy compositional library to explore how characterization and evaluation routines can be optimized to improve throughput in XPS for combinatorial studies. In particular, two quantification approaches are compared. We find that a simple integration (of XPS peak regions) approach is suited for fast evaluation of, in the example system, the [Cu]/([Cu] + [Ni]) ratio. Complementary to that, the time-consuming (XPS peak-) fit approach provides additional insights into chemical speciation and oxidation state changes, without a large deviation of the [Cu]/([Cu] + [Ni]) ratio. This insight suggests exploiting the fast integration approach for ‘real time’ analysis during XPS data collection, paving the way for an ‘on-the-fly’ selection of points of interest (i.e., areas on the sample where sudden composition changes have been identified) for detailed XPS characterization. Together with the envisioned improvements when going from laboratory to synchrotron-based excitation sources, this will shorten the analysis time sufficiently for XPS to become a realistic characterization option for combinatorial material science.


Introductions
Combinatorial material science allows screening of large compositional spaces for desired functional properties to discover new materials, for example for device applications. Combinatorial material libraries, like the one we are investigating in this study, can be described as a single substrate holding many experiments. 1 This denition can be realized by creating intentional gradients in composition and/or thickness across the substrate. For example, over 5000 possible compositions can be produced by combining three elements with composition step changes of 10 at % for each element (for more details, see S.I. †). These libraries are usually deposited on relatively largecompared to usual laboratory-scalesubstrates to increase the number of (accessible) experiments per library. Even though this approach can reduce the material costs and deposition time, the increased number of samples/experiments on a single library comes with an enormous expansion of times for measurement and data evaluation, motivating the development of (semi)automated methods. Therefore, high-throughput methodologies, which balance the requirements for measuring robust and relevant material properties with rapid characterization, are needed for efficient combinatorial materials research, 2 explaining the focus on so called standard measurement techniques so far. However, advanced analytics can provide information that is not available from the standard techniques. Thus, X-ray photoelectron spectroscopy (XPS), the method highlighted in our study, provides information about the chemical and electronic (surface) structure of materials that present the initial situation from where the interface properties develop in thin-lm stacks of modern (opto)electronic devices.
In addition to the elemental (surface) composition, XPS can identify chemical species. This insight cannot be (easily) accessed by standard measurement techniques, but is a prerequisite to expedite deliberate device-driven new material discoveries. However, data acquisition and evaluation of classical XPS are time-consuming depending on the measured energy range, needed energy resolution, and numbers of sweeps required to collect high-quality data. Thus, using XPS in combinatorial materials research has been limited. 3,4 The results of the present study illustrate the feasibility of XPS data collection and analysis in combinatorial materials research, by testing the prospects for establishing XPS as a highthroughput characterization tool. The system chosen is a combinatorial oxide library with a compositional spread of Cu and Ni, i.e., mostly comprising double metal oxide (MO) material.
Cu x Ni 1Àx O y is potentially useful as a hole-selective transport layer (HTL) for halide perovskite (HaP) absorbers in solar cells. HaP-based cells have shown unprecedented performance evolution over the past decade, currently reaching power conversion efficiencies exceeding 25%. 5 NiO has a 3.7 eV direct bandgap, a hole mobility of z3 cm 2 V À1 s À1 , and weak optical absorption in the visible wavelength range. 6 Cu 2 O has a smaller bandgap (2.1-2.3 eV) with also weak visible light absorption, but with higher hole mobility, >100 cm 2 V s À1 . 7,8 Combining both binary oxides promises to open a route to tune optoelectronic and structural properties to arrive at a new material, optimized for hole-conduction, electron-blocking 9,10 with ideal interface energetics 11 as HTL in HaP-based optoelectronic devices.

Synthesis
The Cu x Ni 1Àx O y combinatorial material library was deposited by pulsed laser deposition (PLD). The commercially available uorine-doped SnO 2 (FTO) coated glass substrate (72 Â 72 mm 2 ) was washed with deionized water, cleaned in an ultrasonic bath with soap, rinsed with ethanol, and washed again with deionized water. Subsequently, the washed glass substrate was treated using an Ar Plasma (PLASMA-PREEN II-862, Plasmatic Systems, Inc.) for 4 min. The substrate was then placed in the PLD chamber (Neocera) together with the target materials Cu 2 O (99.9% pure, Kurt J. Lesker Company) and NiO (99.9% pure, Kurt J. Lesker Company) and pumped down to a base pressure of 9 Â 10 À5 mbar before starting the deposition. The target-substrate distance was kept at 70 mm, the temperature of the substrate was set to 500 C, and Ar gas was owed into the chamber, resulting in a deposition pressure of 5 Â 10 À1 mbar. Laser pulse for ablation was done using KrF excimer laser (248 nm, CompexPro, Coherent). To realize a Cu-Ni compositional gradient of the MO lm over the whole substrate, rst Cu 2 O was ablated with 50 pulses on one side of the sample; then the substrate was rotated by 180 , and NiO was ablated with 50 laser pulses across from the Cu 2 O deposition sample position. This sequence was repeated 600 times for a total of 30 000 pulses per target and sample position. The laser energy density was tuned to 2.13 J cm À2 , with a beam spot size of 0.033 cm 2 , and a repetition rate of 8 Hz. Binary Cu 2 O and NiO samples were deposited using the same setup, parameters, and glass substrates, without target and sample position alteration for reference, i.e., each target was ablated separately for 30 000 pulses, on two different substrates. Aer deposition, the sample was taken out in a nitrogen (N 2 ) ushed glove bag attached to the PLD chamber and sealed in N 2 -lled vacuum bags for transport from the Bar-Ilan University (BIU), to the Helmholtz-Zentrum Berlin für Materialien und Energie GmbH (HZB). There, the samples were introduced into a N 2 -purged glovebox directly attached to the (ultra-high vacuum) UHV-backbone of the Energy Materials In Situ Laboratory Berlin (EMIL), which can handle samples of up to 6 in diameter or 100 Â 100 mm 2 in size. 12 The 72 Â 72 mm 2 Cu x Ni 1Àx O y material library was mounted in N 2 -atmosphere on a custom-made sample holder and transferred into the UHV system to the XPS surface analysis system.

Characterization
XPS measurements of the combinatorial materials library and corresponding reference samples were performed in EMIL using the surface analysis system employing a nonmonochromatized PREVAC RS40B1 Mg K a /Al K a twin anode Xray source and a ScientaOmicron Argus CU electron analyser. The samples were studied with Al K a excitation at a pressure of <5 Â 10 À8 mbar. The energy scale was calibrated using a clean Au foil, setting the Au 4f 7/2 peak to binding energy (BE) of 84.00 eV. The measurement time to acquire the most prominent core level spectra of all elements, Ni (Ni 2p), Cu (Cu 2p), O (O 1s), with a pass energy of 30 eV and sufficient signal-to-noise ratio, was around two hours per spot. The (full width at half maxima) area illuminated by the twin anode X-ray source in that setup (see Fig. 1) was z2 cm 2 . 13 The required spatial resolution was realized by selecting the aperture of the Argus CU analyser (A4) to limit the eld of view to z4 mm 2 . In that way, the 72 Â 72 mm 2 library sample could sensibly be divided into 13 Â 13 (in total 169) different measurement spots with a centre distance of 5 mm between each.
Aer XPS characterization a slight change of sample colour (from transparent to brownish) was observed. Thus, it cannot be excluded that the extended exposure of the Cu x Ni 1Àx O y library to X-rays (in total: several 100 hours) altered some of its properties. However, no reduction or 'metallization', as reported for other metal oxides like WO 3 14 and MoO 3 , 15 was observed during the measurements. In contrast, the Cu 2p data of measurement spots 7 and 163 (shown in S.I. Fig. 1a and b †) which were collected with a sample illumination time difference of approximately 100 hours, show no 'metallization' but do display a higher Cu(II)/Cu(I) ratio for the sample spot that was illuminated longer.
When characterizing libraries, location of the measurement and moving along the sample reproducibly to a specic position is critical. This is assured by a computer-controlled stepping motor-equipped manipulator and combination with the custom-made sample holder, specically designed for the 72 Â 72 mm 2 sample (see Fig. 1). In this way, the library can be moved in three dimensions and rotated by 360 in the X-Zplane with highly reproducible positions given in mm for X-, Yand Z-direction. Due to space constraints in the analysis chamber, the library was subdivided into eight separate measurement regions ( Fig. 2; see S.I. for more details †).
Two different approaches were used to quantify the XPS data. The rst (coarse) one is based on determining the area under the photoemission peaks by fast integration. To do so, two steps are needed before the integration. First, the K a -satellite peaks 16 are subtracted from each spectruma process that can be easily scripted/automated. Then a linear background is subtracted from the measured Ni and Cu 2p core-level spectra. The Ni and Cu 2p intensities (I integration Ni , I integration with l Cu and l Ni representing the inelastic mean free paths of the Ni and Cu 2p photoelectrons in the Cu x Ni 1Àx O y material, approximated with the l values of (Cu 2 O) x (NiO) 1Àx (see Table  S.I. 2 and related discussion for more details †). s Cu and s Ni being the element-specic photoionization cross-sections of Cu 2p and Ni 2p, respectively 17 (s tot in Table S.I. 3 †), and TF the analyzer transmission function for these photoelectrons (Table  S.I. 3 †). Even with the best attempts to adjust for the impact of different l and s values on the derived intensities, these corrections have signicant uncertainties. Thus, we focus on relative considerations rather than on absolute values for our analyses in the following. The second, more rened evaluation approach, involves detailed tting of the Cu 2p and Ni 2p core levels, aer properly accounting for satellite peaks, to differentiate between different elemental species. Before tting, the K a -satellite peaks are subtracted from each spectrum and the background is accounted for. In the case of the Cu 2p, the doublet sits on a shoulder of the O KLL Auger spectrum, which can be well described using a 6 th -order polynomial function (see Fig. S.I. 2a †). For practical reasons, however, for the evaluation of our data, two linear functions (see Fig. S.I. 2b and related discussion for more details †) are used to account for the O KLL related background. In addition to this background model, the background of the Cu 2p spectrum itself is accounted for by an "active" Shirley-type background (see S.I., Python Script †). Finally, the spectra were t with four Voigt proles (i.e., two doublets each representing the spin-orbit split 3/2 and 1/2 components of the 2p peaks), representing the main components ascribed to Cu(I) and Cu(II) and six Voigt proles for the complex satellite structure le to the main peaks, with two proles each for the Cu(I) and Cu(II) contribution to the 3/2 spin-orbit component. And one prole each for the Cu(I) and Cu(II) contributions to the 1/2 spin-orbit component, as described in literature. 6,18 For the Ni 2p, no background correction beyond the Shirleytype background is needed, but additional to the Ni 2p K asatellites also the K b -satellite peaks of the Cu 2p spectra must be taken into account (further discussion about this in the following section).
To distinguish between Ni(II) and Ni(III), a different approach is needed. To separate the contributions to the Ni spectra, a different approach is needed. In this case it was assumed that only two contributions -Ni(II) and Ni(III) oxidewere present in the data set, and two sets of nine proles each are used to represent each oxide respectively. There is some uncertainty in representing each oxidation state by a single spectrum in this way, since it neglects the possible contributions of Ni(II) or Ni(III) species with spectral shapes that differ from those of the oxides (e.g., hydroxides). However, if such contributions are signicant they will be reected in the t residues, and the tting procedure can be modied through the introduction of the appropriate representative spectra either as an additional component or as a replacement of one of the two initial components. Four Voigt proles (i.e., two spin-orbit split doublets) are used for the main components and ve Voigt proles to represent the multiplet structure for one oxidation state of Ni. 19 These ts show that the intensity of the Ni 2p 3/2 to 2p 1/2 peaks is not xed to a ratio of 2 : 1 (as expected from their multiplicity), due to the impact of multiplet splitting. 20 Examples of ts of Cu and Ni 2p photoemission spectra are shown in Fig. S.I. 1. † For quantication, the areas under the tted 2p 3/2 spectra, including corresponding satellites, are used ðI fit Cu 2p 3=2 Þ: To derive the composition, an expression like (eqn (1)) can be employed, replacing I integration X by I fit X 2p 3=2 using the cross section of the chosen core level (s 3/2 in Table S.I. 2 †). A comparison of the results based on the integration and t approaches is discussed in detail below to evaluate the possibility of using the former for fast quantication of large XPS data sets collected on material libraries. Fig. 3 shows the background-subtracted Ni 2p (a) and Cu 2p (b) core level spectra of the 169 different measurement spots of the Cu x Ni 1Àx O y library. It can be observed clearly that the Ni 2p and Cu 2p peak intensities follow the nominal NiO-Cu 2 O gradient (indicated by the color code) produced by the deposition process, as expected. The shape of the Ni 2p is strongly inuenced by multiplet splitting. For high Ni contents, the (green) spectra agree well with the reference spectrum of a NiO sample (shown as a black line in Fig. 3a). However, the spectra are slightly shied to a higher BE compared to the reference. For lower Ni contents, the shapes of the Ni 2p spectra start to deviate signicantly from that of NiO, with the difference becoming increasingly pronounced for decreasing Ni contents (see normalized Ni 2p spectra in Fig. S.I. 3a †). However, close inspection of the spectra reveals that the spectral change is mainly caused by an increasing contribution of an Al K b excitation satellite of the Cu 2p spectra (see detailed discussion in conjunction with Fig. S.I. 3b and c †) with decreasing Ni, (i.e., increasing Cu) content. Aer correcting this additional background effect, we nd a (mainly) unvarying spectral shape of the Ni 2p spectra.

Results and discussion
The peak positions of the (blue) Cu 2p spectra in Fig. 3b, representing Cu 2 O-rich regions, show an absence of pronounced Cu(II)-related satellite features at 940-945 eV. 18 Comparison to the Cu 2 O reference spectrum (in black, at slightly lower BE) also conrms that Cu is mainly in the +1 oxidation state (i.e., Cu(I)), as expected considering the use of Cu 2 O as precursor material in the PLD process. However, close inspection of the data reveals that with increasing Ni content, a broadening of the 2p 3/2 peak at $933 eV and a relative increase of the Cu(II)-related satellite intensity occurs (see normalized Cu 2p spectra in Fig. S.I. 4 and example ts Fig. S.I. 1 †). This result indicates a change in the chemical composition and oxidation state of Cuparticularly in the NiO-rich regime, as also supported by the ts shown in Fig. S.I. 1. † However, the decrease of the Cu(I)/Cu(II) ratio is not related to the increase in Cu(II), but rather to the decreasing overall Cu content towards the NiO-rich region. Note that great care has been taken to minimize air exposure of the Cu x Ni 1Àx O y library aer deposition.
The Ni 2p and Cu 2p spectra measured on individual library spots show a BE shi depending on the composition. Moving from the Cu 2 O-rich to NiO-rich area, the Ni 2p 3/2 BE decreases from 855.0 eV to 854.7 eV (zÀ0.3 eV) and the Cu 2p 3/2 BE increases from 932.8 eV to 933.3 eV (z+0.5 eV). The

corresponding [Cu]/([Cu] + [Ni]
) ratio for all 169 probed spots, as derived with the fast integration and detailed t approaches, using eqn (1), are shown in Fig. 4a   increases. This leads to a "negative intensity", which articially decreases the integral-derived area. The problem can be prevented by choosing a suitable background for each spectrum; however, this intervention violates the desired "hands-off" approach for fast quantication. Another approach would be to consider rows 14 and 15 as  a 'separate' region of interest employing an optimized background correction for this Cu 2 O-poor region. Alternatively, one could exploit statistical methods for data evaluation. Using e.g., Grey relational analysis, we could show that spectra deconvolution is less affected by background effects. 21 A thorough t analysis of the data can reveal additional chemical structure information, e.g., different species and oxidation states and it also allows to more exibly consider changing complicated background contributionslike the O KLL related background in the case of Cu 2p. The integration approach, then again, is expected to be robust and signicantly faster and thus assumed to be more relevant to efficiently evaluate large data sets, as expected to be generated for even more complex combinatorial material libraries or during operando/in situ experiments where fast feedback can also be used in experiment control.
These results on data evaluation schemes can be optimized to show the potential of XPS also in combinatorial materials research. However, for XPS to become a valid high-throughput analysis tool, data acquisition has to be signicantly accelerated. In the current case, the measurement time alone amounted to 200 hours per core level in total. A straightforward way to reduce measurement time is to use a more brilliant light source than the laboratory-based twin anode X-ray source that was used here. Using, e.g., the so X-ray branch of the twocolour beamline of EMIL 22 instead would increase the overall X-ray photon ux by a factor of 30. Considering the focused beam spot (of approx. 30 mm Â 25 mm), the photon ux density would be enhanced by almost 5 orders of magnitude (see S.I.: Photon ux †). However, note that with high-ux densities, beam-induced artifacts (i.e., beam damage) might become an issue for irradiation-sensitive samples. In any case, it seems feasible to signicantly reduce the measurement time to a few (or even below) 1 hour for the 169 spot library as measured here. In addition, the integration quantication approach does not require collecting data with high energy resolution, so a fast sweep with high pass energy or even a survey spectrum could be enough to get most of the compositional information. Fully exploiting the focused beam spot would then allow to increase the number of probed spots, if fast changing sample properties should require this.
Automated spectra processing schemes have to be developed to decrease data evaluation times that allow for 'real-time' data processing and evaluation. Using the integrated approach can (only) be a start. This may enable 'on-the-y' analysis, where, during the measurement, spots of interest are automatically preselected and further investigated. The detailed t analysis can then be done for data of selected spots of interests, further reducing data acquisition time.

Conclusions
We have presented XPS data collected for a large-scale (72 Â 72 mm 2 ) Cu x Ni 1Àx O y combinatorial material library. The measurements were performed to evaluate to what extent XPS can be optimized to become a valid high-throughput method for obtaining relevant material properties of such combinatorial libraries. For that, 169 spots on the large-scale sample were characterized by using the moderate spatial resolution of a state-of-the-art commercially available electron analyzer. Two different quantication approaches were presented. A relatively coarse, but fast and robust approach based on peak area determination by integration and a more detailed, but more resource-consuming approach based on area determination by peak t. Both procedures reveal a clear [Cu]/([Cu] + [Ni]) composition gradient along the sample. The deviation between the composition derived by the integration and t approach is (except for two points out of 169) under 10% absolute, suggesting that the former can be used in fast quasi 'real-time' data evaluation paving the way for 'on-the-y' analysis and automated measurement spot selection in the future. Together with the fact that it seems feasible to signicantly reduce measurement time by using more brilliant light sources, the road for XPS to become a real high-throughput analysis tool for combinatorial materials research is now wide open.

Conflicts of interest
There are no conicts to declare.