Fabrication-related impurity study of thin superconducting Al films using x-ray absorption spectroscopy

Superconducting aluminum thin films are integral to many astrophysics detector applications,¹ dark matter, or particle detection experiments,² and as components in certain quantum computing circuits.^3,4 It is known that the performance of such devices can be impacted by the presence of surface or bulk contamination introduced on/in the thin film during device fabrication.^5,6 For instance, the two-level system (TLS) properties of disordered oxides or insulators can introduce unwanted radio frequency (RF) loss and noise into microwave kinetic inductance detectors (MKIDs), as well as decoherence in quantum computing circuits.^7,8 The degree of the impact depends on the chemical state, amount, and location of the contaminants, and the degree of disorder.^7,9 TLS dissipation and noise at low microwave powers can be reduced if we can identify and eliminate surface layers contributing to TLS.¹⁰ We can understand the chemical state of the surface and bulk, and how they are influenced by fabrication steps by looking at the spectral shape of x-ray absorption spectroscopy.¹¹ 

X-ray absorption spectroscopy (XAS) is an extremely powerful method for material characterization, since it is sensitive to the chemical and electronic structures of materials.^12,13 Depending on the energy of the incident x rays and whether photons, Auger electrons, or all electrons are detected, probing depths range from several to hundreds of nanometers. When an x-ray photon hits a sample, it can either be scattered or absorbed by an electron. During the absorption process, a core-electron is excited up to an unoccupied state above the Fermi energy. Therefore, x-ray absorption is a transition between two quantum states from an initial state (with an x ray, a core-electron, and no photoelectron) to a final state (with no x ray, a core-hole, and a photoelectron). The intensity of the x-ray absorption coefficient is proportional to the transition rate, which is governed by Fermi’s golden rule. The energy dependence of x-ray absorption is determined by the distribution of unoccupied states extending from the core-level permitted by dipole selection rules.^14–16 The resulting core-hole can be refilled by an electron at the higher energy levels or valence electrons or previously excited electrons, with the emission of photons (photon-in/photon-out process). A photon detector can be used to collect these photons as a total fluorescence yield (TFY) signal for XAS. A second method of decay of the core-hole is Auger electron yield (AEY). Additionally, total electron yield (TEY) detection includes the secondary-electron cascade. Therefore, we have three methods for detecting x-ray absorption signals: TFY, AEY, and TEY. The TEY mode has a probe depth of 5 nm (soft x rays) to tens of nanometers (hard x rays), limited by the depth from which electrons emerge with enough energy to overcome the work-function barrier. In contrast, TFY detection probes the samples at depths of up to micrometers.¹⁷ These signals come from the different decay routes of refilling the core-hole and, therefore, can be collected simultaneously. Since thin film patterning processes are often conducted via wet and dry etch chemistries, the surfaces of the materials are most influenced by remaining residues and adsorbates, and, thus, we choose the TEY mode of detection. AEY can be too sensitive to the surface conditions, which may limit their usefulness for thin film studies when the surface is not perfectly clean (such as after cleaving devices in a vacuum). In contrast, TFY is more sensitive to the bulk making it less ideal if the surface is the main focus of interest.

Samples 1–4 were prepared following a standard deposition and lithography steps for the patterning of aluminum thin films into MKID resonators. In all of the samples, the aluminum thin films were deposited on silicon substrates by sputtering an Al target in an AJA dc magnetron sputtering system. For samples 1–4, the aluminum was lithographically patterned into a linear array of superconducting coplanar waveguide (CPW) microwave resonators using a standard metal wet etchant, FujiFilm E-6, consisting of a mixture of HNO₃, H₃PO₄, and CH₃COOH. These patterned substrates were diced into chips 3 × 16 mm² in size, with 16 resonators on each. The aluminum forms a CPW ground plane and center conductor. There is a gap of 5 μm between the central Al meandering conductors (10 μm wide) that form both the CPW resonators and a CPW microwave feedline. Microwave quality factors were measured prior to XAS measurements for samples 1–4. To gain further insight and confirmation of the relation between specific process steps and the film states measured by XAS, we also prepared nine additional “control” samples (control samples 1–9) where we either intentionally attempted to keep the film pristine and/or intentionally exposed the films to a range of common micromachining process steps, including specific plasma treatments. Variations in treatment were explored to determine the effect on the resulting chemistry. Control samples 1–4 and 9 were unpatterned and differed mainly in the plasma treatment process steps. Control samples 5–8 included both patterned and unpatterned samples and differed mainly in the lithography steps. Other variations in the samples include the film thickness, the target location in the sputter chamber, the sputter chamber base pressure, the cleaning of the silicon substrate prior to the deposition, and the details of the lithography steps, as listed in Tables I and II.

Measurements of the microwave loss of the CPW resonator samples 1–4 were completed inside a dilution refrigerator microwave testbed at temperatures ranging from 10 to 60 mK. For these measurements, the silicon chip was first mounted inside a gold-plated copper radio frequency (RF) “package” using Ge-varnish applied at the four corners of the die to provide both heatsinking and mechanical support. To route the RF signal used to read out the resonator's transmission response, two abutting RF fanout boards were placed on either end of the die inside the package, each featuring a 50 Ω CPW feedline. Aluminum wirebonds were then used to connect the RF fanout boards to a 50 Ω CPW RF feedline on the device die. These fanout boards consisted of Au-plated copper conductors on a Duroid substrate. Additional RF aluminum wire bond connections were made directly between the device and fanout board ground planes and the gold-plated copper package. Two SubMiniature version A (SMA) type connectors installed into the sides of the gold-plated copper package were used to route the RF signal from the fanout board, with the SMA center pin soldered directly to the fanout board centerline. After installation of the device and wire bonding inside the package, a gold-plated copper lid was installed to provide a light-tight enclosure, with aluminum tape applied over all interfaces to provide additional shielding.

This enclosed device package was then mounted at the cold stage of the dilution refrigerator, where coaxial RF cables were connected to the SMA connectors and routed from the cold stage to the room temperature stage. The RF cables were heat-sunk via DC blocks at the intermediate temperature stages. RF attenuators were also inserted at some intermediate temperature stages to control the power level at the device, damp reflections from cable interfaces, and provide attenuation of thermal radiation down the RF lines. A low-noise cryogenic amplifier was installed at the 4 K stage on the output side of the RF line to provide cold amplification and improve signal-to-noise. To prevent higher temperature thermal radiation from reaching the devices via the RF lines, thermal blocking filters¹⁸ were also inserted at the input and output sides of the device package. A cylindrical Amumetal 80% nickel-iron alloy¹⁹ shield was inserted around the external cryostat to shield it from magnetic fields during the cooldown.

After cooling down to a base temperature, microwave measurements of the complex transmitted S-parameters, S₂₁, through the device as a function of microwave frequency, f, were completed using a vector network analyzer over a range of read power levels. Resonator internal quality factors were then extracted for each resonator at each power level by fitting the measured S₂₁(f) transmission to a Lorentzian model [following Eq. (E.1) in Ref. 20].²⁰ These results are summarized in Table III. By comparing the change in the relative microwave loss (Q_i⁻¹) as a function of read power at very low read power levels (photon occupation numbers, N = 10²–10³), we can evaluate the relative TLS loss contributions in each sample, with a larger difference corresponding to a higher TLS loss for that sample. Sample 2 was measured to have the lowest TLS loss, followed by sample 1, sample 4, and sample 3, respectively. However, we must note the microwave measurements of these samples were completed 1–2 years prior to the XAS measurements. We noted clearly visible corrosion on sample 4 at the time of XAS measurements, which was not visible at the time of the microwave measurements for this sample, indicating that the chemical state of that film and RF properties had changed over the intermediate time period. Now that we have demonstrated the applicability of this XAS measurement technique, a future systematic study would hope to ensure timely microwave and XAS measurements.

The XAS experiment was performed at Lawrence Berkeley National Laboratory’s Advance Light Source on beamline 6.3.1 (measurement date October 9, 2019).²¹ This is a bending magnet beamline that provides soft x-ray photons with an energy range between 200 and 2000 eV. We used three gratings 240, 600, and 1200 lines/mm for three different ranges of energy 250–400, 400–900, and 900–1700 eV, respectively, to maximize brightness. The brightness of the beamline flux is 1 × 10¹¹–2 × 10⁹ photons/s/0.1%bandwidth. The beam spot size is approximately 300 × 90 μm² (FWHM). The solid samples were mounted on a copper holder using carbon tape. After degassing the samples for about 1 h, they were inserted into the chamber, where the sample holder could be moved along the x, y, and z directions and rotated about the vertical axis, allowing us to choose desired locations on the sample and angles of incidence. These measurements were done at 60° from normal (grazing incidence), where the measurement becomes more surface sensitive. The measurements were conducted in the TEY geometry. Each spectrum shown in this study is an average of four to five individual scans. TEY signals were collected using a picoammeter monitoring the drain current.²² Measurements were taken at room temperature under a mild vacuum (2–5 Torr). Regardless of the detection mode, any element of the periodic table has a characteristic x-ray absorption edge energy, and it is labeled according to the particular core-level from which it is excited. For example, the K-edge corresponds to the excitation of an electron from the 1 s orbital. In this work, we studied impurities via a wide range of energy scans and then focused on the K-edges of certain elements such as the C K-edge, the O K-edge, the F K-edge, and the Al K-edge. We also chose various spots, both on and off the patterned regions of the devices. Hence, we found the location and type of particular impurities in the sample.

The elemental composition of the samples is shown in the wide energy range scan in Fig. 1. The peaks with red vertical dotted lines corresponding to 284.20, 409.9, 543.10, 696.70, and 1559.6 eV are the signature peaks of carbon, nitrogen, oxygen, fluoride, and aluminum, respectively. The peaks represented by gray dotted lines are O, F, and Al second-order K-edge peaks, respectively. The second-order K-edge peak lies at half of the primary energy due to grating. Sample 4 is highly contaminated with fluorine while samples 1, 2, and 3 have moderate contamination. Different chemical structures have been produced due to fluorine contamination, which will be discussed in more detail in Sec. III C.

The oxygen and aluminum K-edge x-ray absorption spectra are plotted in Figs. 2 and 3, respectively. Aluminum oxides form thermodynamically stable α-Al₂O₃, on which about one dozen metastable phases known as “transition aluminas” or “metastable alumina polymorphs” exist, and they are denoted by Greek letters β-Al₂O_3, δ-Al₂O_3, χ-Al₂O_3, γ-Al₂O_3, κ-Al₂O_3, θ-Al₂O_3, ρ-Al₂O_3, and η-Al₂O₃.²³ Each phase has different local geometry, e.g., α-Al₂O₃ is constituted dominantly by AlO₆ octahedra, θ–Al₂O₃ has 50% AlO₆ and 50% AlO₄, and γ-Al₂O₃ has a moderate amount of AlO₄ (15%–45%).²⁴ To identify the phase of oxidized aluminum, we compared our spectra with the reference spectra of commonly occurring phases of Al₂O₃.^25,26 XAS depends on the local geometry of the absorbing atom.²⁷ First, let us look at oxygen as an absorbing atom. The oxygen K-edge XAS spectra of samples 1–3 consist of a tiny pre-edge peak at 532.2 eV, a main peak at 536.0 eV, and a postedge peak near 539.5 eV. Two different reference spectra for α-Al₂O₃ (Refs. 28 and 29) and γ-Al₂O₃ (Refs. 25 and 26) show a matching of the main peak and postedge with each other, despite their differing trends. The main and postedge peaks for samples 1–3 match with the curve for α-Al₂O₃. Additionally, in the case of aluminum as an absorbing atom, we see XAS signals from both metallic aluminum and oxidized aluminum. As seen in Fig. 3(a), the Al K-edge signal of oxidized aluminum has been overwhelmed by the metallic aluminum K-edge signal and looks more metallic as the x-ray probe can penetrate about 5–8 nm. Thus, to reduce the contribution of the metallic signal, we have subtracted the K-edge XAS spectrum of pure Al and compared the subtracted spectra with reference spectra. In the reference spectrum of α-Al₂O₃, two peaks at 1568 and 1567.1 eV are associated with AlO₆ octahedra. In the spectra of θ-Al₂O₃ and γ-Al₂O₃, the peak associated with 1566.0 eV is characteristic of the AlO₄ tetrahedra.²⁴ The lack of a peak at 1566 eV and the presence of two peaks—the first prominent peak at 1567.6 eV and the second peak at 1570.9–1572.0 eV—are signature peaks of α-Al₂O₃. In our case, the metallic aluminum K-edge subtracted spectra consist of two peaks at 1567.5 eV (second vertical line) and 1570.9 eV (third vertical line) indicating the dominating AlO₆ octahedra phase, as well as the downward trend of the curve after the first peak (second vertical line) and matching the final peak at 1579.8 eV (fifth vertical line) showing the α-Al₂O₃ phase. This phase (α-Al₂O₃) is the most stable phase of aluminum oxides.²³ Studies show naturally grown oxides under ambient conditions form α-Al₂O₃.³⁰ 

Even though the main oxygen and aluminum K-edge peak positions for the S1, S2, and S3 samples are similar, the comparative heights of the pre-edge peak of the O K-edge and the relative height of the second peak in the Al K-edge are different [calculation is shown in Fig. 3(b)]. The pre-edge peak of the O K-edge is strongest in S2, which has the lowest TLS loss. This hints that the pre-edge peak of the O K-edge may probe important differences in the α-Al₂O₃ layer, including possible differences in it is crystallinity or amount.^26,31 The order of the relative height of the second Al K-edge peak (at 1571 eV) with respect to the first Al K-edge peak (1566 eV) is S1 > S3 > S2, which indicates S2 has the least amount of oxidized aluminum and is also consistent with our measurement of the lowest TLS loss in this sample.

Even though aluminum films are oxidized with an α-Al₂O₃ phase on the surface in samples 1–3, in the case of sample S4, we also detected an oxygen signal from the Si layer. S4 is the thinnest Al film (10 nm), so silicon under the Al is detectable in TEY. In addition, in the patterned regions, the silicon is exposed and hence a TEY signal from Si is also achievable. The Si K-edge of S4 with the reference spectra for the Si K-edge of pure Si and oxidized (SiO₂) from Ref. 32 have been plotted in Fig. 4.³² The data have been fit with a linear combination of reference spectra of Si and SiO₂, with a fit result of 80.8 ± 0.3% Si and 19.2 ± 0.3% SiO₂ with R = 1.3% (chi-square = 1.72) mismatch.

From the wide range energy scan, we observed that the most significant fluorination occurred in sample 4. Therefore, we chose this sample for a deeper study of fluorination. The Al K-edge spectra at grazing incidence (60° with sample normal) for spots both on and off the device were measured and compared with the Al K-edge of most possible chemicals, as shown in Fig. 5. The experimental Al K-edge reference spectra are taken from Ref. 33 for AlN and from Ref. 34 for AlF₃, and the computed Al K-edge for Al₂SiO₅ and TiAl are taken from Ref. 35. Our Al K-edge shows a strong peak at about 1571.5 eV, which corresponds to AlF₃.³⁴ For spot 1 (an unpatterned Al film region), Al K-edge signal is dominantly from fluorinated aluminum. When we go from spots 1 to 6 (a patterned region), the signal from the metallic Al K-edge peak positioned at 1559.5 eV increases. The corrosion of the aluminum on this sample (which developed after the microwave Q measurement) is less severe “on pattern.”

We also studied fluorination in the same sample using fluorine as an x-ray-absorbing atom. The fluorine K-edge of sample 4 has been plotted in Fig. 6. The experimental reference spectra for α-AlF₃, β-AlF₃, and t-AlF₃ are from Ref. 36; –CF₂–CF₂– from Ref. 37; and TiF₄ from Ref. 35. The two main peaks at 684 and 688.5 eV of all the AlF₃ phases match with each other, but the t-AlF₃ and the α-AlF₃ phases have one more peak at 680 and 695 eV, respectively. This indicates that the fluorine is chemically absorbed and bonded with the aluminum in sample 4, forming a layer of β-AlF₃ on the surface. Sample 4 was O₂ ashed in a March reactive ion etching (RIE) equipped with six different gases: O₂, sulfur-hexa-fluoride (SF₆), CF₄, CHF₃, Ar, and N₂O for use as plasma etchants. Thus, the significant carbon, nitrogen, and fluorine K-edges are not necessarily unexpected given that the sample had potential exposure to these contaminants.

The fluorine signal for the S1–S3 samples is present, but weaker than in S4. In Fig. 7, we can see the relative strength of the fluorine signal with respect to the Al signal. The order of strength is S1 > S3 > S2 indicating S2 has the lowest amount of fluorine. This also correlates with our measurement of the lowest TLS loss in S2.

We wanted to determine the impact of different process steps, including a standard silicon substrate wafer cleaning processes, SF₆ plasma etching, passivation with a fluorocarbon, and exposure to photoresist adhesion promoters during the lithography process. As shown in Tables I and II, there are potentially three processes that could expose the Al film to fluorination: a SF₆ plasma etch (as in CS2), the deep reactive ion etch (DRIE) passivation process involving a C₄F₈ plasma (as in CS4), and a HF dip prior to the Al deposition (done for all samples CS1–CS9). Among these, CS1 and CS9 are as-deposited thin films with no subsequent processing, and, therefore, we can expect fluorination only from fluorine residue during the substrate cleaning process. To examine these other potential fluorine causes, we also chose to examine CS2 and CS4. CS2 is exposed to the plasma of SF₆, and CS4 is passivated by fluorocarbon from one cycle of DRIE. The wide-band XAS spectra of the four samples and nine control samples are plotted in Fig. 6, which shows the elemental presence of chemicals. The control samples CS1 and CS9 do not have any trace of fluoride, which indicates that the aluminum thin films are not absorbing fluoride from the residue of HF. However, the control samples, CS2 (red line) and CS4 (blue line), both have a strong fluoride signal. CS3 (green line) has been thickly oxidized via exposure to an oxygen (O₂) plasma for 12 min, and as expected, it has a strong oxygen signal.

Small differences in the spectra near the K-edges can be seen clearly in the normalized stack plots of K-edges of the O, F, Al, and Si shown in Fig. 8 (note that here we have not included the control samples with a weak K-edge signal). We can see the differences in the F K-edge of CS2 and CS4. CS2 has formed AlF₃ since both the F and Al K-edges of CS2 are identical to those of S4. It is known that aluminum and aluminum oxides are prone to react with SF₆. The reaction of an SF₆ plasma with Al₂O₃ and Al can occur following the reaction: 3SF₆+ 2Al₂O₃= 3SO₂F₂ + 4AlF₃ (Ref. 38) and Al + F₂= AlF₃ (where the F⁻ ions dissociate from the SF₆ during this process³⁹). In the case of CS4, the F K-edge is from C₄F₈. The plasma formed from C₄F₈ deposits a Teflonlike smooth passivating film consisting of a network of long linear fluorocarbon (CF₂)_n chains with only a little cross-linking.⁴⁰ As a result of a stoichiometric reaction between aluminum and fluorinated polymers such as poly(tetrafluoroethylene) and their co-polymers, elemental carbon, and aluminum trifluoride can be produced as the following the equation [2Al + 3(–CF₂–)_n→ 3C + 2AlF₃].^41,42 Even though Al is susceptible to absorbing fluorine from the fluorocarbon by Al in the case of CS4, there is no absorption of fluorine by aluminum from the passivating fluorocarbon since CS4’s Al K-edge is similar to the Al K-edges observed in the untreated samples (CS1 and CS9). The oxygen K-edge along with the nitrogen K-edge of CS4 is very feeble, indicating surface protection by the fluorocarbon is effective in preventing oxidation of the aluminum surface. In CS4, the fluorine K-edge should be from the fluorocarbon and the carbon K-edge should be from both a fluorocarbon and adventitious carbon. Though the control samples (CS3, 6–8) are not intentionally fluorinated, it is possible fluorocarbon was introduced during the oxygen ashing step if the chamber had previously been used for fluorine-plasma steps.

The Si K-edges observed in samples CS5, CS6, CS7, and CS8 are similar to the Si K-edges of the as-deposited control samples, CS1 and CS9. For samples CS6, CS7, and CS8, the hexamethyldisilazane (HMDS) C₆H₁₉NSi₂ is removed by three different methods: O₂ plasma ashing, O₂ ashing in the AXIC system, and O₂ ashing in the RIE system. In the case of CS5, the HMDS has purposely been applied to the Al film but not removed via any O₂ ashing steps. We have not observed any Si absorption from the HMDS monolayer by Al, neither from the Al K-edge nor from the Si K-edge. Additionally, CS5 does not have F K-edge. CS5 has not been inserted in the AXIC and the March RIE chambers, so it is not contaminated by fluorine, that is why it does not have the F signal, unlike CS6, CS7, and CS8.

This paper discusses the study of impurities introduced by common microfabrication processing steps in aluminum thin films on silicon substrates by applying x-ray absorption spectroscopy. We have studied the chemical structure and phases of aluminum oxides formed in thin film Al samples. From the oxygen and aluminum K-edges in the XAS spectra, we have determined that thin aluminum films patterned via standard lithography and wet-etch processes are highly vulnerable to oxidization with α-Al₂O₃ phases of aluminum native oxide on the surface and fluorination in the form of β-AlF₃. We have determined that the substrate cleaning process using an HF:H₂O 1:10 solution immediately prior to the aluminum film deposition does not contribute to residues and absorbates. Similarly, the aluminum surface was effectively passivated by fluorocarbon by preventing aluminum from oxidizing without fluorine absorption. The aluminum is always prone to fluorinate; the sample with SF₆ plasma is fluorinated as AlF₃. We also see correlations between measured low TLS microwave loss and the XAS spectra detection of low relative amounts of α-Al₂O₃ and β-AlF₃ and a stronger O K-edge pre-edge peak. Hence, impurities and absorbates in Al thin films have been documented using XAS techniques and may have future applications for understanding the performance and reproducibility of thin film devices.

This research was supported by NASA EPSCoR (Award No. 80NSSC22M0173). We acknowledge the use of the research facilities at the Lawrence Berkeley National Laboratory. The authors would like to thank Alpha T. N’Diaye, a research scientist at Advance Light Source on beamline for helping with the operation of the beamline 6.3.1; and Mona Mirzaei and Nicholas Costen at NASA Goddard for their assistance with the fabrication and packaging of the Al samples.

The authors have no conflicts to disclose.

Ghadendra B. Bhandari: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (equal); Methodology (lead); Project administration (supporting); Software (lead); Writing – original draft (lead); Writing – review & editing (lead). Thomas R. Stevenson: Conceptualization (equal); Investigation (equal); Project administration (equal); Resources (equal); Writing – review & editing (equal). Emily M. Barrentine: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Writing – review & editing (equal). Larry A. Hess: Methodology (equal). Mikel B. Holcomb: Conceptualization (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal).

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