Characterization of process-related interfacial dielectric loss in aluminum-on-silicon by resonator microwave measurements, materials analysis, and imaging

Superconducting coplanar waveguide (CPW) resonators¹ in the microwave band are extensively used to read out the state of qubits in the circuit quantum electrodynamics (cQED) architecture.^2–6 They can also be used as tools to investigate the loss in superconducting qubits.^5,7,8 Several loss mechanisms can adversely affect the performance of quantum circuits, including charged two-level systems (TLS), spins, non-equilibrium quasiparticles, radiation, and magnetic vortices.^9–11 Among these, TLS dominate the loss at millikelvin temperatures and single-photon-power excitations, i.e., under the operating conditions of a superconducting quantum processor.¹² TLS loss originates in dielectric amorphous materials,¹³ mainly within the oxide layers located at the substrate–air (SA), metal–air (MA), and substrate–metal (SM) interfaces of the device; see Fig. 1. Coupling to a bath of TLS, or to spurious, individual TLS, results in qubit decoherence and resonator loss.^12,14 Extensive work on analyzing the relative importance of dielectric loss due to TLS at different interfaces has been done by the community.^15–19 In our previous work,²⁰ which includes the simulations of CPW resonators of identical geometry to the devices analyzed in this article, we find that a lossy SM interface dominates the total loss of the device. Once the SM loss is mitigated, losses at the MA and SA become dominant.

The TLS loss can be mitigated by refining the fabrication process. Different wet and dry etching processes have been employed to clean the Si substrate before the metal deposition to decrease the loss at the SM interface.^21–24 The choice of silicon (and sapphire) as popular substrates is due to their low loss tangent values,^15,25,26 which is essential in minimizing the loss in areas with high field concentrations. The bulk loss of the substrate then becomes negligible in comparison with the dielectric loss at the interfaces. Various etching methods, including trenching of the substrate, have been explored to transfer the resonator pattern onto the metal layer while maintaining a low-loss SA interface.^27–31 Both the SA and MA interfaces can be cleaned post-processing, before loading the sample into the cryostat for measurement.^32,33 In addition, the metal surface can be oxidized in situ after the deposition by filling the chamber with pure oxygen.⁸ Passivation of the superconducting metal at the MA interface has also been explored.^34,35 We have combined some of these processes in a standard fabrication process for aluminum-on-silicon (Si–Al), discussed below, and it yields CPW resonators with TLS-limited internal quality factors Q_i ≈ 1 × 10⁶ and qubits with relaxation times T₁ ≈ 100 µs.^36–38 The choice of aluminum as the superconducting material originates from its availability and the vast knowledge about its chemistry and fabrication techniques. A recent demonstration of Al transmon qubits with average relaxation time T₁ > 250 µs highlights the enduring potential of aluminum as a platform for superconducting circuits.²⁰ 

This report presents a systematic evaluation of our CPW resonator fabrication process. We fabricate control resonators based on the standard recipe previously developed in our group⁸ and also other resonators using recipes that deviate from the standard by only one step. We determine the resonators’ Q_i and quantify the associated TLS losses in each fabrication step. In parallel, we use atomic force microscopy (AFM) to study the film surface and transmission electron microscopy (TEM) to investigate the material structures at different interfaces of the resonators. Furthermore, we use a complementary set of techniques for compositional depth profiling, including x-ray photoelectron spectroscopy (XPS), energy dispersive x-ray spectroscopy (EDS), and time-of-flight secondary ion mass spectrometry (ToF-SIMS). XPS is suitable for exploring the chemical composition and the chemical state of the selected elements, while the ability of ToF-SIMS—a technique that has only recently gained popularity for investigating losses in superconducting devices³⁹—to detect traces of individual elements and molecules with part-per-million accuracy can unfold the elemental depth profile of the device. We find that the most critical step to minimize the loss at the substrate–metal interface is the initial removal of the native oxide from the surface of the silicon wafer using hydrofluoric acid (HF). The substrate–air interface loss and, to a degree, the metal–air loss can also be reduced temporarily post-fabrication by the same acid treatment. This has previously been demonstrated on devices made from niobium^10,40 and tantalum.⁴¹ 

For our experiments, we used high-resistivity intrinsic Si (100) wafers and aluminum superconducting films. The standard fabrication process started with wafer cleaning, by submerging the wafer in a mixture of ammonium hydroxide, hydrogen peroxide, and deionized (DI) water (NH₄OH:H₂O₂:H₂O, 1:1:5)—known as standard clean 1 (SC1)—at 80 °C for 10 min. The wafer was then rinsed with DI water and subsequently dipped in a 2% aqueous solution of hydrofluoric acid for 1 min, followed by another DI water rinse. Within a few minutes, we loaded the cleaned wafer into the load-lock of the evaporator Plassys MEB550S to minimize the re-oxidation of the Si surface.⁴⁷ Next, we heated the substrate, in situ, to 300 °C for 10 min. After about 20 h, the wafer cooled down to room temperature (∼20 °C) and the chamber base pressure reached 4 × 10⁻⁸ mbar. We evaporated 150 nm of Al at the deposition rate of 1 nm s⁻¹, followed by static oxidation of the Al surface at the pressure of 10 mbar for 10 min.

Following the film deposition, the resonators were fabricated by means of optical lithography and wet etching (in aluminum etchant, i.e., mixture of phosphoric, nitric, and acetic acids) and measured in a dilution refrigerator at about 10 mK. Besides the standard sample (No. 1), we measured five additional samples (Nos. 2–6), where for systematic evaluation, we fabricated each sample by varying only one step of the standard process. The list of the samples and the type of variation are summarized in Table I. We fabricated sample No. 1 using the standard process, while for sample No. 2, we did not remove the native oxide of the silicon substrate in HF prior to metal evaporation. For sample Nos. 3–5, we skipped the SC1 step, the substrate preheat, and the in situ oxidation of aluminum, respectively, and finally, we prepared sample No. 6 using a short (4 h) pumping time (reaching a pressure of 6 × 10⁻⁸ mbar) prior to metal evaporation.

Subsequently, we investigated the influence of varying the conditions of the film deposition itself. For wafer No. 7 (includes sample Nos. 7a–7c), we followed the same conditions as for sample No. 1, with the exception that the SC1 cleaning was omitted and the Al film was sputtered instead of being evaporated: after cleaning the substrate using HF dip, the wafer was immediately transferred to the heated load-lock (80 °C) of a sputter tool (DCA MTD620), which was pumped down for 40 min until a pressure below 5 × 10⁻⁷ mbar was reached. Thereafter, the wafer was transferred to the deposition chamber. Here, the substrate was heated to 300 °C for 10 min and pumped for 16 h, at which point the base pressure of the deposition chamber was reached at 2.2 × 10⁻⁸ mbar. Al was then deposited by direct-current (DC) magnetron sputtering in argon (Ar) plasma, followed by in situ oxidation of the film surface. In the supplementary material, we also present an investigation into the effects on the resonator Q_i resulting from the deposition rate, substrate temperature during deposition, and high-temperature substrate preheating.

We used three samples, Nos. 7a–7c, to investigate the effect of a short dip in 2% HF post-fabrication, with the aim to remove the surface oxides of both the Al film and the Si substrate, and with them any adsorbates left behind by the fabrication process. This dip was only 15 s long due to the tendency of Al to be etched in this solution, and the samples were rinsed in DI water afterward. The effect of this procedure is shown on sample No. 7b, while No. 7a is kept as the control. Sample No. 7c underwent a vapor HMDS deposition at 100 °C immediately after HF treatment, where the hydrophobic HMDS monolayer deposited this way could be expected to protect the Si from re-oxidizing .²² More details on sample Nos. 7a–7c are given in Table II. Due to the tendency of both Al and Si to regrow oxides under ambient conditions, sample Nos. 7b and 7c were mounted inside the dilution refrigerator, with the pump-down preceding the cooldown started within 2.5 h after the HF dip.

The XPS experiment was performed using a PHI 5000 VersaProbe III system equipped with a monochromatic aluminum x-ray source (i.e., Al Kα). To aid in depth profiling, we performed a stepwise iterative procedure of Ar⁺ ion sputtering to etch away the Al layer followed by an XPS measurement, until the SM interface was reached. We scanned selected binding-energy ranges based on the elements of interest, including oxygen (O 1s), aluminum (Al 2p), and silicon (Si 2p), in order to conduct a qualitative analysis of the chemical state(s) of individual elements.

TEM bright field (BF) imaging was performed using a FEI Tecnai T20 microscope operated at 200 kV. TEM cross section samples were prepared using a FEI Versa 3D focused ion beam scanning electron microscope (FIB-SEM). Scanning transmission electron microscopy (STEM) and energy dispersive x-ray spectroscopy (EDS) analyses were carried out using a JEOL Monochromated ARM 200F TEM, which is equipped with a Schottky field emission electron source, a double-Wien monochromator, a probe Cs corrector, an image Cs corrector, and a double silicon drift detector (SDD) for EDS.

In addition, we performed the ToF-SIMS measurement using an IONTOF 5 system, with a bismuth ion (Bi⁺) gun for imaging and a cesium ion (Cs⁺) sputter gun for depth profiling. The mass spectrum of the secondary ions emitted from the sample was collected at each sputtering step and converted to an elemental depth profile. Note that the XPS and ToF-SIMS measurements were performed either on chips from the same wafers as the resonators were fabricated on, or on chips from wafers that were prepared following identical cleaning and film growth procedures.

Each sample contains up to eight resonators (Fig. 1), whose frequencies range between 4 and 8 GHz. The frequencies are varied by changing the length of the resonators, while keeping the cross-sectional dimensions w and g identical. We measured the power-dependent S₂₁ of the feedline using a vector network analyzer (VNA) and calculated $FδTLS0$ by fitting the results with the TLS loss model described in Eq. (3). Figure 2(a) presents Q_i as a function of ⟨n⟩ for a representative resonator of a comparable frequency, one resonator each for sample Nos. 1–6. At low ⟨n⟩, Q_i becomes scattered. This is likely due to the temporal instability of TLS, as the TLS can switch in and out of resonance with the device on timescales comparable to the measurement time due to the large number of averages required at low ⟨n⟩.^36,48,49 For samples fabricated with the standard process, on average, we find Q_i = 1.1 × 10⁶ when ⟨n⟩ = 1, and $FδTLS0=0.87×10−6$⁠. For the other samples, the values of Q_i together with the fitting parameters of the TLS loss model [Eq. (3)] are shown in Table I. The parameter $FδTLS0$ from the fit is presented as a box plot in Fig. 2(b), where each point in this plot is from an individual resonator. The bounds of the box represent the upper and lower quartiles of the statistical distribution, and the whiskers extend to the farthest data points lying within 1.5× of the interquartile range from the box bounds. The loss due to other parameters δ_other, also presented in Table I, is about 2 × 10⁻⁷ for all samples.

The highest level of TLS loss is observed in sample No. 2 (no HF), a fivefold increase compared to that of the standard sample (No. 1); see Fig. 2(b). Using TEM, we find a clean SM interface of the standard sample, whereas there exists a layer of ∼1.5 nm-thick oxide at the SM interface of No. 2 [Figs. 3(a) and 3(b)]. The XPS spectra for O 1s, Al 2p, and Si 2p at the SM interface of sample Nos. 1 and 2 are shown in Figs. 3(c)–3(e), respectively. Although the oxide layer at the SM interface of No. 2 was supposedly comprised of native oxide atop the Si substrate (SiO₂), the oxidation–reduction reaction at room temperature between Al and SiO₂ has turned the oxide into an aluminum oxide type.⁵⁰ The signature of this oxide is observed through the XPS spectra of O 1s and Al 2p. The Al 2p spectrum shows that metallic Al, with the binding energy position at 72.8 eV, is the dominating chemical state, while a small shoulder positioned at 75.3 eV corresponds to the oxidized Al(III) state. The major peak of O 1s at 532 eV gives an energy difference of 456.8 eV with Al (III). This implies that the interface oxide is mostly in the form of aluminum oxide. Still, as the oxygen content is rather low and close to the XPS detection limit, i.e., 1.0 at. %, it is difficult to determine exactly in which form the oxide has developed; see Figs. 3(c) and 3(d). The characteristic peak of Si 2p at 99.4 eV indicates that silicon is mostly in its elemental state. The secondary peak at 104.0 eV refers mostly to the Al 2p plasmon loss peak instead of the overlapping Si(IV) state, with reference to both the energy difference and the area ratio compared to the metallic Al 2p peak; see Figs. 3(d) and 3(e).

The XPS spectra indicate no trace of O 1s in sample No. 1 at the Si–Al interface; see Fig. 3(c). However, the depth profile of No. 1 from the ToF-SIMS measurement in Fig. 4(a) features a small peak from oxygen at the SM interface. Since the oxide was not observed in TEM either, we speculate that the trace oxide in No. 1 is either local or in the form of a discontinuous layer along the grain boundaries.²⁰ The oxygen signal at the SM interface in No. 2 is more intense and agrees well with the TEM and XPS results; see Fig. 4(b). The intensity of the O peak in No. 2 is almost an order of magnitude higher than that of the other samples, in agreement with the relatively higher concentration of O at the SM interface of this sample and its correspondingly high TLS loss level; see Fig. 2.

Surface treatment with SC1 and HF has been a prominent cleaning method in the semiconductor industry for several decades.⁵¹ The SC1 solution cleans the surface of the Si substrate from most organics and some metallic contamination by trapping them into an oxide layer on the surface, which is then removed by HF. If SC1 is not performed, some contamination remains on the substrate surface, resulting in 15% higher $FδTLS0$ in No. 3 compared with No. 1 [Fig. 2(b)], even though the oxide layer, as the main source of TLS loss, has already been removed by the HF solution.

Another process extensively used to ensure a good quality SM interface is the heating of the substrate under vacuum.^52–54 The elevated temperature of the substrate can desorb moisture and volatile molecules from the surface.⁵⁵ In some cases, the substrate undergoes annealing temperatures of above 700 °C for surface reconstruction to achieve a better atomic transition between the metal and substrate.^24,56 In our experiments, the sample with no preheating process (No. 4) showed a 61% increase in $FδTLS0$ compared with sample No. 1. However, one of the resonators on sample No. 4 shows a significantly higher loss than the remaining resonators. Discarding this outlier resonator, the mean $FδTLS0$ of sample No. 4 falls to 1.1 ± 0.07 × 10⁻⁶—a 26% increase in TLS loss compared to the standard sample (No. 1). We observe no discernible difference in the oxygen concentration at the SM interface of sample No. 4 compared to the standard sample (No. 1) in the ToF-SIMS measurement shown in Fig. 4. It might be that the change in the oxygen level is below the observable limit, or it might not be the desorption of oxidizing species behind the increased performance of the devices on sample No. 1 when the preheat step is included.

The TLS loss of sample No. 5 is just marginally higher (10%) than the standard sample; see Fig. 2(b), and one may conclude that in situ oxidation of the Al film is unnecessary. However, upon SEM inspection of No. 5, we found dark spots distributed over the Al film—resembling voids—as shown in Fig. 5(a). The TEM image of Fig. 5(c) shows the cross section of one of these voids. For further investigation, we prepared a second Al film on the Si substrate, where we skipped the in situ oxidation after evaporating the film. We observed no voids in the Al immediately after deposition. Next, we heated a sample to 160 °C for 5 min to create a similar condition to that of baking the photoresist during resonator fabrication. We inspected the sample daily; the voids appeared on the piece after about two days.

The SEM image of sample No. 5 shows residues on the silicon substrate within the CPW gap where Al was wet etched; see Fig. 5(a). These residues remain on the silicon surface after wet etching of the Al layer using aluminum etchant, even if we etch a sample that is not exposed to heating after metal deposition. A longer etch time does not remove the residues either. The lateral size of these residues varies from a few tens of nanometers up to 200 nm, and the thickness is measured to be ∼2 nm, as shown in the AFM image of Fig. 5(b). Cross-sectional EDS analysis in TEM [see Figs. 5(d) and 5(e) and the supplementary material] shows the composition distribution of a residue flake in the etched Al area. There is an intermixing of Al, Si, and O in the flake. A ToF-SIMS analysis of the etched Al areas of sample Nos. 1 and 5 shows a significantly stronger presence of the residues of Al and Al oxide on sample No. 5; see Fig. S7 in the supplementary material.

Due to the thermal expansion mismatch between Al and Si, heating or cooling can generate stress in the Al film (α_Al = 23.1 × 10⁻⁶ °C⁻¹ and α_Si = 2.6 × 10⁻⁶ °C⁻¹⁵⁷). The textured (polycrystalline) form of evaporated Al and its low melting point make it prone to recrystallization. Stress can promote the recrystallization and formation of defects in the film. The development of hillocks and voids in pure Al films on Si (or even silicon oxide) has been a known challenge in the CMOS industry.⁵⁷ The oxide on the surface of Al might help keep the grains together and partially prevent the creation of defects. We speculate that different types of aluminum oxide grow atop the Al film as a result of in situ oxidation (exposure to pure oxygen at low pressures) or ambient air oxidation (exposure to oxygen and moisture at atmospheric pressure). Oxidation in the ambient air might introduce more defects and porosity in the aluminum oxide, making it less effective in protecting the film. This can result in the formation of voids and defects in the Al film.

With the shorter pumping time used for sample No. 6, the base pressure and the substrate temperature prior to deposition were 6 × 10⁻⁸ mbar and 50 °C (passively cooled down from 300 °C), respectively. The Q_i and TLS loss of the resonators on sample No. 6 are comparable to those of No. 1 (Fig. 2). The ToF-SIMS results of No. 6 show only a negligible change of O intensity at the SM interface in Fig. 4(b), meaning that longer pumping time is not necessary.

For the resonators etched out of a sputtered Al film, we achieved a comparable quality factor to those on the evaporated film; see Fig. 6 and Table II. The average TLS loss $FδTLS0$ obtained for sample No. 7a is about 0.86 × 10⁻⁶, which is comparable to that of No. 1. This is despite the increased surface roughness of the sputtered film compared to the evaporated film, which may increase the loss at the MA interface. The other, non-TLS related loss mechanisms are lower for the sputtered film No. 7a (smaller δ_other). Since δ_other is a combination of several loss mechanisms, pinpointing the exact cause of the lower loss needs further study. A fraction of this may be an effect of lowered quasiparticle loss due to an increased superconducting gap in the sputtered film, as the superconducting critical temperature T_c, collected from 4-point probe DC transport measurements, is 1.2 K for the sputtered film and 0.91 K for the evaporated (standard) film. In contrast, the residual-resistance ratio RRR is 3 for the sputtered film and 8 for the evaporated one, indicating a higher crystalline quality of the evaporated film. We remark that the quality factors in sample No. 7 were obtained after optimizing the sputtering process—for details of the samples, also their quality factors, surface roughness and atomic force micrographs, see the supplementary material.

A short dip in HF post-fabrication reduces the TLS loss, doubling the Q_i achieved at low photon numbers, as shown in Fig. 6 and Table II. Since it took us about 2.5 h to place the HF-treated samples in vacuum (at the mixing chamber inside the dilution refrigerator), we anticipate the regrowth of aluminum oxide at the MA interface, while it takes longer for the silicon oxide on the substrate to fully regrow.⁴⁰ Therefore, we think that the improvement in the loss after the HF treatment is mainly due to the reduced contribution of the SA interface, while a minor improvement comes from the reduced loss at the MA interface: as discussed above, the partial etch of Al can help with removing the adsorbates on the aluminum surface.

A subsequent passivation of the silicon substrate by HMDS in sample No. 7c slightly degrades Q_i compared to No. 7b, although $FδTLS0$ is still below one-half that of the untreated sample No. 7a; see Table II. We note that due to the decreased resonance linewidth of the higher-quality resonators, single-photon levels are not reached in the measurement power span. Therefore, in Table II, we compare the Q_i of the resonators at ⟨n⟩ ∼ 10. The decreased TLS loss of the HF-treated samples becomes comparable to the other losses δ_other, resulting in the low-power Q_i being limited by the other loss mechanisms as well.

In conclusion, our investigations show that concerning the loss at the substrate–metal interface, the surface treatment with SC1 and HF is the most effective method to reduce the associated TLS loss in superconducting Al-based CPW resonators fabricated on Si substrates. ToF-SIMS characterization shows one order of magnitude higher intensity of oxygen at the SM interface if the native oxide is not removed from the substrate. In comparison with the standard sample, this leads to an increase in the TLS loss by five times. In addition, we found that avoiding preheating of the substrate prior to deposition can increase the TLS loss by about 60%.

The in situ oxidation of the Al film led to a marginal increase in the loss; however, observation of voids in the Al film and residues in the gap suggests that this step is necessary for not facing random and unwanted open circuits in the device. Finally, despite the higher pre-deposition base pressure and temperature, the TLS loss of the sample with short pumping was still comparable to the standard process, reaffirming the insignificance of the bulk material quality and composition compared with the surfaces and interfaces as the source of TLS loss.⁵⁸ 

With regard to the substrate–air interface, and, to a lower degree, the metal–air interface, a post-fabrication HF dip of resonator chips reveals that removing the surface oxides of Al and Si can decrease the TLS loss by about three times. As the effect of the oxide strip is temporary, until an oxide has regrown, it is desirable to extend this effect for practical applications. We show that immediate passivation of the Si by HMDS using vapor deposition can slightly degrade the Q_i achieved after the HF dip, although still keeping the loss well below that of an untreated sample. Further studies into the longevity of the protective effect of the HMDS layer are necessary.

The supplementary material includes details of the Q factor extraction, a discussion of the sputtering parameter optimization based on microwave measurements and materials analysis, a demonstration of the post-fabrication treatment by HF + HMDS on a sputtered sample, and additional materials characterization of the sample that has not undergone in situ oxidation (sample No. 5).

This work was funded by the Knut and Alice Wallenberg (KAW) Foundation through the Wallenberg Center for Quantum Technology (WACQT) and by the EU Flagship on Quantum Technology HORIZON-CL4-2022-QUANTUM-01-SGA Project No. 101113946 OpenSuperQPlus100. The authors acknowledge the use of the Nanofabrication Laboratory at Chalmers University of Technology. The financial support from Swedish Research Council (VR) and Swedish Foundation for Strategic Research (SSF) for the access to ARTEMI, the Swedish National Infrastructure in Advanced Electron Microscopy (Grant Nos. 2021-00171 and RIF21-0026) are also acknowledged. The XPS, ToF-SIMS, and TEM measurements were performed at the Department of Industrial and Materials Science, Chemistry and Chemical Engineering and the Chalmers Material Analysis Laboratory (CMAL), respectively. The authors acknowledge the valuable feedback and comments from Sandoko Kosen, Henrik Frederiksen, Niclas Lindvall, Daniel Pérez Lozano, and the fabrication team at Quantum Technology laboratory (QT), Chalmers.

The authors have no conflicts to disclose.

Janka Biznárová and Lert Chayanun contributed equally to this work.

Lert Chayanun: Formal analysis (equal); Investigation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (supporting). Janka Biznárová: Formal analysis (equal); Investigation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Lunjie Zeng: Investigation (supporting); Writing – review & editing (supporting). Per Malmberg: Investigation (supporting); Writing – review & editing (supporting). Andreas Nylander: Investigation (supporting); Writing – review & editing (supporting). Amr Osman: Investigation (supporting); Writing – review & editing (supporting). Marcus Rommel: Investigation (supporting); Writing – review & editing (supporting). Pui Lam Tam: Formal analysis (supporting); Investigation (supporting); Visualization (supporting); Writing – review & editing (supporting). Eva Olsson: Investigation (supporting); Writing – review & editing (supporting). Per Delsing: Formal analysis (supporting); Funding acquisition (equal); Writing – review & editing (supporting). August Yurgens: Formal analysis (supporting); Writing – review & editing (supporting). Jonas Bylander: Conceptualization (equal); Formal analysis (supporting); Funding acquisition (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal). Anita Fadavi Roudsari: Conceptualization (equal); Formal analysis (supporting); Investigation (supporting); Supervision (equal); Visualization (supporting); Writing – original draft (equal); Writing – review & editing (equal).

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