betaFeedback

WorksPhysics and Astronomy

Sidewall Metrology Protocol for Superconducting Qubit Devices

Abstract

Sidewall morphology is a potentially important factor influencing superconducting qubit performance, yet its quantitative characterization remains challenging due to the absence of standardized measurement protocols. In this work, we develop a scanning probe microscopy-based measurement framework for the quantitative characterization of superconducting qubit sidewalls. The framework introduces a refined root mean square lateral sidewall roughness descriptor and a new curvature-equivalent radius for quantifying vertical sidewall morphology. The descriptors and measurement procedures are applied to different components of qubit devices, forming the basis of a proposed sidewall characterization protocol. This provides a foundation for quantitative comparisons of fabrication processes across laboratories, enabling future investigations of correlations between sidewall geometry and qubit performance.

Introduction

Superconducting qubits are considered one of the leading technological platforms for building scalable quantum computers [1, 2]. However, achieving reliable and consistent device performance remains a major challenge [3, 4, 5]. Even for devices with the same design and layout, qubit performance often varies both between fabrication batches and within the same chip [6, 7]. Structural properties related to materials and fabrication processes are believed to be important factors influencing the consistency of superconducting qubit performance [4, 8, 9, 10]. Recent studies have highlighted sidewall profiles and angles as structural parameters that may influence the coherence times of superconducting qubit devices [11, 12, 13].

Sidewall profiles and angles are usually investigated using cross-sectional focused ion beam–scanning electron microscopy (FIB-SEM) or high-resolution transmission electron microscopy (HR-TEM). Although these techniques provide high-resolution two-dimensional imaging, obtaining statistically representative quantitative information across devices remains challenging, making it difficult to establish statistically meaningful correlations with qubit performance [13, 14, 15]. In order to draw reliable conclusions about the effect of sidewall properties on qubit lifetime, it is important to extract detailed quantitative information on the structural profile. Scanning probe microscopy (SPM) can potentially provide quantitative information on both sidewall angles and associated morphology with high spatial resolution.

SPM-based metrology has been studied for characterizing semiconductor sidewall structures using tilted or three-dimensional atomic force microscopy (AFM) configurations [16, 17, 18, 19, 20]. However, the focus has primarily been on lateral line-edge roughness (LER), including its variation from the bottom to the top of the sidewall. Quantitative characterization of the full vertical sidewall profile, including local angles and morphological variations, has received comparatively less attention, particularly in the context of superconducting qubit devices, where such geometrical features may be relevant to loss mechanisms.

This work focuses on sidewall morphology and geometry, aiming to develop a framework for accurate, reproducible, and traceable measurements with associated quantitative descriptors. To establish these descriptors, we draw on a literature review and an analysis of relevant length scales and morphological features [13, 21, 22, 23, 24, 25, 26]. The selected descriptors are intended to capture sidewall features that can be measured using AFM and relevant to device variability and loss mechanisms in superconducting qubits. We experimentally demonstrate the measurement and quantitative analysis of these descriptors on the sidewalls of key components of superconducting qubit devices. Based on these experimental results, we propose a sidewall measurement and data analysis protocol that can facilitate future statistical studies correlating sidewall features with qubit performance.

Theoretical Considerations

Transmon qubits, coherence, and resonators

Transmon qubits are defined by the two-state subspace consisting of the ground state ∣0⟩|0\rangle and the first excited state ∣1⟩|1\rangle of a circuit containing a Josephson junction connected to a capacitance (see Figure 1 (a)) in the regime where EJ/EC≫1E_J/E_C \gg1, where EJE_J is the Josephson energy and ECE_C the charging energy [27].

Figure 1. (a) Blueprint of Al-based (blue color is Al) superconducting qubit investigated in the present study (device number 1-7). Each qubit comprises a cross-shaped capacitor pad connected via an Al/AlOx_x/Al Josephson junction to a surrounding Al layer. Each qubit is capacitively coupled (“resonator coupling”) to a readout resonator, and each readout resonator is inductively coupled to a feedline (not shown in the figure). A microwave driveline is capacitively coupled to the qubit island to enable controlled qubit driving (∣0⟩→∣1⟩|0\rangle\to|1\rangle). The five colored lines (two in zoom inset and three in the main figure) show the positions of sidewall profiles shown in Figure 3 and Figure 4. (b) Schematic of the tilted-sample AFM approach enabling quantitative analysis of sidewall geometries, including undercut structures. The red crosses indicate vertical sidewall regions that are not accessible and cannot be reconstructed using the conventional non-tilted approach. (c) Schematic illustrating a Vertical Sidewall Profile (VSP) ZVSP,i(x)Z_{\mathrm{VSP},i}(x) and a Lateral Sidewall Profile (LSP) XLSP,i(y)X_{\mathrm{LSP},i}(y).

The qubit relaxation time (T1T_1) characterizes the duration over which an excited qubit relaxes back into the ground state, dissipating energy into the environment. For independent and Markovian decoherence processes ii, relaxation times add reciprocally, 1/T1=∑i1/T1i1/T_1 = \sum_i 1/T_1^i, making any individual T1iT_1^i a lower bound of the total relaxation time. For transmon qubits, leading channels are commonly believed to be residual two-level systems at interfaces [21, 28], dielectric losses [29, 30], Purcell loss through the readout resonator [31, 32, 33], quasiparticle tunneling [34, 35, 36], and radiative losses [37, 38]. We note that the precise nature of the detrimental two-level systems is the topic of ongoing efforts to understand, but candidates include atomic tunneling defects, surface contaminants and oxides, adsorbates [21], grain defects [23], and localized vibrons [39]. For a broader, qualitative overview of sources limiting the performance of transmon qubits, we refer to Appendix A.

Sidewall morphology is also of importance for superconducting readout resonators, where dielectric and conductor losses may limit the attainable quality factor [25, 26, 40]. The etched sidewalls define mesoscopic metal–substrate and metal–vacuum interfaces that can exhibit enhanced electric-field participation and may constitute relevant sources of dissipation through two-level systems and other interfacial loss mechanisms. Moreover, the sidewall geometry may affect local field enhancement and the effective participation of surface oxides.

Geometrical interfaces as sources of loss

The interfaces of a transmon qubit components (metal-air, metal-substrate, oxide-metal, etc.) each contribute to the dielectric loss relaxation time as

T1dielectric=ω∑ipitan⁡δi,(1)T_1^{\mathrm{dielectric}} = \omega\sum_i p_i \tan\delta_i, \tag*{(1)}

where ω\omega is the qubit frequency, and pip_i and tan⁡δi\tan\delta_i the participation ratio and loss tangent of interface ii, respectively [8]. The participation ratio of the stored electrical energy in interface ii is given by

pi=ϵi2W∫Adr ti∣Ei∣2,(2)p_i = \frac{\epsilon_i}{2W} \int_A \mathrm{d}r\, t_i |E_i|^2, \tag*{(2)}

where ϵi\epsilon_i is the interface dielectric constant, tit_i the interface thickness, Ei\mathbf{E}_i the interface surface electric field, and WW the total capacitor energy. Gauss’ law establishes that the surface charge density (which is directly proportional to the electric field) is inversely proportional to the radius of curvature for a surface protrusion, ρ(rprot)\rho(\mathbf{r}_{\mathrm{prot}}), i.e.,

∣Ei(rprot)∣∝1ρ(rprot),(3)|E_i(\mathbf{r}_{\mathrm{prot}})| \propto\frac{1}{\rho(\mathbf{r}_{\mathrm{prot}})}, \tag*{(3)}

so dielectric losses may in general be expected to be more prominent at surface protrusions with small radii of curvature. The radius of curvature can be directly derived from the measured height profile. The surface topography probed by AFM may therefore be used to shed light on the likely impact of dielectric losses of the qubit. We caution, however, that for a correlation to be established in a statistically sound and meaningful fashion, a comprehensive understanding of the various decoherence channels and their relative contribution must first be established. We also caution that the qubits operate in a low charge density regime governed by superconducting electrodynamics and finite penetration depth (further commented on below) in which the entire geometry (such as capacitor pad spacing and orientation) may eventually decide whether the granularity and surface topography plays a significant or insignificant role in terms of dielectric losses. Recent studies have highlighted the importance and distribution of dielectric losses via parasitic two-level systems located at interfaces, which ultimately may put stringent conditions on surface purity and composition [12, 13, 21, 24, 41].

In many quantum devices, AFM-resolved curvature on the scale of 10-100 nm (see also Sec. VI B) are not merely descriptive of the surface morphology, but also fall in a range over which microwave currents and electromagnetic fields are redistributed. For example, a 200 nm Al film is expected to display a superconducting penetration depth around λ=60\lambda= 60 nm [42] and a coherence length in vicinity to the range of ξ≈100\xi\approx100-250 nm depending on the mean free path.

In addition to the planar surfaces, sidewalls constitute an important part of the metal–substrate–vacuum interface and may contribute disproportionately to dielectric losses [12, 13]. Sidewall morphology, including its inclination, curvature, and roughness, determines both the local electric-field distribution and the interfacial area available for hosting parasitic two-level systems. This may be particularly important in Josephson junctions, in which the electrical field concentrates [22, 43, 44]. Variations in the sidewall profiles arising from the fabrication process may alter the participation of lossy dielectric regions, making quantitative characterization of the sidewall important with respect to the coherence of superconducting circuits [4, 9].

We emphasize that surface roughness should not be viewed as a direct measure of superconducting quality, but rather as a possible proxy for the microscopic film and interface properties that contribute to readout resonator loss. Rougher growth may accompany greater defect and grain-boundary densities, modified oxide/interface chemistry, increased effective interfacial area, or local thickness and field inhomogeneities, all of which can enhance dissipative channels such as two-level systems [45, 46, 47]. Electronic disorder itself is not necessarily detrimental. In disordered or granular superconductors electronic disorder can increase TcT_c or kinetic inductance while preserving low losses [48, 49, 50]. Roughness may serve as an experimentally accessible indicator of the underlying material state. No universal monotonic relation between surface roughness and internal quality factors should be expected across different materials and fabrication processes.

Definition of relevant geometrical descriptors

In this subsection the theory for two sidewall morphology descriptors is laid out, one to characterize the lateral morphology and one to characterize the vertical (cross sectional) morphology. The theory behind the curvature-based probe radius estimation are also given. Additionally, the band pass filtering of the profiles is summarized briefly in Appendix B.

Lateral sidewall profiles

The Lateral Sidewall Profile (LSP), XLSP,i(y)X_{\mathrm{LSP},i}(y), describes the lateral position of the sidewall surface projected onto a plane parallel to the substrate, where yy denotes the coordinate along the nominal sidewall direction, see Figure 1 (b-Figure 1c). The index ii identifies the individual profiles measured at different vertical positions along the sidewall from the foot of the sidewall to the top. Details of reference subtraction, the filtering procedure, and the practical descriptor calculation is provided in Appendix B.

The RMS lateral sidewall roughness, RqLSPR_q^{\mathrm{LSP}}, is defined as the average RMS amplitude of the band pass filtered residual profiles:

RqLSP=1N∑i=1NRq,iLSP,(4)R_q^{\mathrm{LSP}} = \frac{1}{N} \sum_{i=1}^{N} R_{q,i}^{\mathrm{LSP}}, \tag*{(4)}

where Rq,iLSPR_{q,i}^{\mathrm{LSP}} is the RMS amplitude calculated for the individual profile ii, and NN is the number of measured profiles.

Vertical sidewall profiles and morphology

The Vertical Sidewall Profile (VSP), ZVSP,j(x)Z_{\mathrm{VSP},j}(x), is defined as the cross section of the vertical sidewall, see Figure 1 (c), where the index jj identifies the position along the yy-direction of the vertical sidewall profile, and the yy-direction is parallel to the nominal sidewall direction. A residual amplitude descriptor such as RqR_q cannot be meaningfully defined as there is no meaningfully reference profile available. However, according to Eq. (3), dielectric losses may in general be expected to be more prominent at surface protrusions with small radii of curvature. Thus we propose the local vertical sidewall morphology to be characterized by a curvature-equivalent radius, RρR_\rho, deduced from the local radius of curvature of the vertical sidewall profiles, which weigh small radius of curvatures more strongly than large radius of curvatures.

The curvature-equivalent radius, RρR_{\rho}, is defined as the inverse of the profile-averaged RMS curvature. The RMS-equivalent curvature is first determined for each individual profile and subsequently averaged over all measured profiles.

Small values of RρR_{\rho} indicate the presence of numerous sharp surface features with small local radii of curvature, whereas larger values correspond to smoother sidewalls. The auxiliary minimum radius of curvature, ρmin⁡\rho_{\min}, is defined as the smallest local radius of curvature along the sidewall profile. The detailed definitions of the vertical sidewall profile ZVSP,j(x)Z_{\mathrm{VSP},j}(x), filtering procedure, curvature calculation, RMS averaging, the curvature-equivalent radius and the minimum local radii of curvature are given in Appendix B.

Curvature-Based Probe Radius Estimation

In the measurements, the recorded surface profiles represent the surface topography convolved with the geometry of the AFM probe tip. Although geometrical de-convolution methods based on a known probe geometry exist, traceable reference standards for validating reconstructed probe geometries in the range <10< 10 nm are, to the knowledge of the authors, not available. Therefore, we recommend that the descriptors (measurands) are specified together with the probe geometry used for the measurements, in analogy to the specification of stylus geometry in conventional roughness measurements. Thus, the AFM probe is regarded as an integral part of the measurement procedure. Surface features that cannot be resolved due to the probe geometry, such as hypothetical deep trenches with widths smaller than the probe diameter, are therefore not considered quantifiable with the present measurement approach. This approach is consistent with the selection of an appropriate S-filter prior to further analysis, which defines the lower spatial wavelength limit of the evaluated surface morphology.

In this work, an effective probe radius ρprobeeff\rho_{\mathrm{probe}}^{\mathrm{eff}} was estimated from the local curvature of multiple sharp protrusions on a tip characterization sample; details of the procedure are given in Appendix C.

Algorithm for sidewall region detection

One of the fundamental challenges in processing sidewall profiles is to develop a general method for identifying the sidewall region in the data. Depending on the structure of the sidewall and the way it was fabricated, this transition region can be more or less straight-forward to determine. In Figure 2, we display three methods for sidewall step edge detection. The first method is based on knowledge of the step height and takes as input parameter the AFM-derived height, hAFMh_{\mathrm{AFM}}, obtained from a least-squares fit to the central third of the aluminium pad region. The foot of the sidewall is defined as the position where the profile displays a height of z(x)=ϵh,foothAFMz(x)=\epsilon_{\mathrm{h,foot}}h_{\mathrm{AFM}}, where ϵh,foot=0.02\epsilon_{\mathrm{h,foot}}=0.02 is used for the two examples shown in Figure 2. Similarly, the top is detected at the position where the profile displays a height of z(x)=(1−ϵh,top)hAFMz(x)=(1-\epsilon_{\mathrm{h,top}})h_{\mathrm{AFM}}, with ϵh,top=0.02\epsilon_{\mathrm{h,top}}=0.02. The resulting sidewall region is displayed by the blue bars in Figure 2, and labeled height-based sidewall width. To ensure a robust routine, the foot and top are detected starting from the sidewall midpoint, as defined by the position where z(x)=hAFM/2z(x)=h_{\mathrm{AFM}}/2. In this way, we prevent roughness of the substrate or deposition structure to interfere with the step detection.

Figure 2. (a) Example of a vertical sidewall profile of a Josephson junction electrode with indications of the detected sidewall region by use of the height-based method (shown in blue), summed gradient-based method (shown in magenta) and the gradient-based method (shown in turquoise). (b) Example of a vertical sidewall profile of a resonator coupling with indications of the detected sidewall region by use of the height-based method (shown in blue), summed gradient-based method (shown in magenta) and the gradient-based method (shown in turquoise). The three different methods are described in the text.

In the second method, which is shown by the turquoise bars in Figure 2 and is denoted the gradient-based sidewall detection, the gradient is calculated along the sidewall profile, and the foot is determined as the position, where the local gradient becomes smaller than a value, ϵg,foot=0.05\epsilon_{\mathrm{g,foot}}=0.05, set in the analysis. Depending on the choice of this parameter, the detection sensitivity can be increased or decreased. Likewise, the top of the sidewall is determined as the position where local gradient becomes smaller than the set value ϵg,top\epsilon_{\mathrm{g,top}}. Similar to the first method, we detect foot and top starting from the middle of the sidewall.

In the third method, we instead use the cumulative sum of the squared gradient to detect the foot and top of the sidewall, starting the analysis from the substrate region. In Figure 2, the result of this method is shown by the magenta bars, and the detection threshold is set by ϵsg,foot\epsilon_{\mathrm{sg,foot}} and ϵsg,top\epsilon_{\mathrm{sg,top}} for the foot and top, respectively. In the present example we used ϵsg,foot=0.005\epsilon_{\mathrm{sg,foot}}=0.005 and ϵsg,top=0.02\epsilon_{\mathrm{sg,top}}=0.02. The cumulative-gradient sidewall detection method defines the transition region based on the accumulated change in local sidewall angle. It therefore defines the transition from the distribution of geometrical changes along the profile rather than from an external height reference. This approach enables, e.g., protrusions or overshoots at the sidewall edge to be treated in a consistent manner. Since the method identifies the transition based on the accumulated gradient, the result is inherently influenced by the profile length over which the gradient is integrated, and it should be noted that, as opposed to the first two methods, the third method is more susceptible to roughness of the substrate, which can in principle lead to an overestimation of the sidewall width.

Based on the observed profiles, including those shown in Figure 2, the height-based sidewall detection method provides a robust and reproducible definition of the transition length. However, it requires an independent determination of the step height and is, by definition, not sensitive to the exact transition from the substrate to the sidewall or from the sidewall to the top plane of the pad, even when these are well defined. The gradient-based sidewall detection method also provides a robust and reproducible definition of the transition length, where the transition is defined as the region over which the local sidewall angle changes. Based on the observed profiles, the method is highly sensitive to a well defined transition between e.g. the substrate and the sidewall. Since the method identifies the transition from changes in the local sidewall angle, it is, by definition, independent of the height of the aluminum pad and by experience the transition point between the sidewall and the top plane is not always intuitively correct.

Results and discussion

Lateral sidewall profiles

Figure 3 summarizes the AFM characterization of the lateral sidewall morphology for the investigated superconducting qubits and readout resonators. The average lateral sidewall roughness, RqLSPR_{q}^{\mathrm{LSP}}, ranges from approximately 2–5 nm for the Josephson junction structures to approximately 20 nm for the readout resonators. A sidewall with a visible local pit exhibits an increased roughness value of approximately 30 nm. The profile-to-profile variation of the extracted roughness values within the same image area, quantified by σ(Rq,iLSP)\sigma\left(R_{q,i}^{\mathrm{LSP}}\right), ranges from approximately 0.3–0.6 nm for the qubit components and up to approximately 5 nm for the readout resonators. The extracted roughness values RqLSPR_{q}^{\mathrm{LSP}} and their associated variations demonstrate that the fabricated sidewalls exhibit measurable deviations from the ideal geometry, which can be quantified consistently using the proposed amplitude descriptors.

Figure 3. (a)–(e) AFM height maps of various circuit components and their sidewalls. Orange windows indicate the locations of the in-plane profile variations shown in grayscale images next to the maps. A lateral sidewall profile XLSP,i(y)X_{\mathrm{LSP},i}(y) corresponding to the middle cut of the orange windows are indicated below each grayscale map. (f) Average sidewall roughness calculated from multiple sidewall profiles acquired across seven wet-etched transmon qubit devices. (g) Same as (f) for linear circuit elements. (h) Profile-averaged sidewall roughness calculated for the readout resonators of the same seven devices as measured in (f) and (g), as well as five additional readout resonators from an alternative fabrication process, exemplified in (e) and referred to as type 2 (concerning devices 8–12). In (h) we provide information of the AFM probes used in the measurements to quantify the reproducibility of the results and dependence on the probe. For device 8, we observe that repeated measurements with Probe 1 as well as an additional measurement with Probe 2 give results that agree within the uncertainty. In the case of Probe 3, we invoke different shade and size of the data points to reflect the time that data was taken going from dark/large to bright/small for increasing times. The error bars represent the variation of the observed vertical sidewall profile ZVSP,i(x)Z_{\mathrm{VSP},i}(x) extracted at different ziz_i-positions from the foot to the top of the aluminum pad. The relative difference in sidewall roughness between devices 11 and 12 is comparable to the small variation observed in repeated measurements using the same probe.

Vertical sidewall profiles

Figure 4 summarizes the observed vertical sidewall profiles of a capacitor pad and the smaller Josephson junction (JJ) electrode. A qualitative assessment gives that most profiles on the same chip are similar, however, one profile (Capacitor pad of device 4) is significantly different, indicating that a single profile of one edge is not sufficient to make a general statement about the angle and shape of the sidewall on the chip.

Vertical sidewall profiles and effective radii of curvature

Figure 4. (a) Vertical sidewall profiles ZVSP,if(x)Z^{f}_{\mathrm{VSP},i}(x) of the capacitor edge at the positions marked in the inset. (b) Similar profiles for the Josephson junction element “Cap” (the double-layer structure) at the positions marked in the inset. (c) Single profile of the capacitor edge on device 2. Circular arcs corresponding to the local radius of curvature ρ\rho are shown for the two smallest protrusions and two smallest depressions. (d) Effective radius of curvature Rρ=1/κRMSR_{\rho}=1/\kappa_{\mathrm{RMS}} (see Eq. (16)) of the vertical profiles of JJ electrodes and capacitor pads as a function of device number.

The single profile of the capacitor (device 2) in panel (c) shows a distinct sharp protrusion near the top of the aluminum pad, with an observed local minimum radius of ρOBS=−21.8\rho_{\mathrm{OBS}}=-21.8 nm (negative for protrusions), which is equal to the conservative estimate of the resolution limit of the probes. Corrected for the probe shape, the best estimate of the local radius of curvature is ρ=ρOBS−ρprobeeff=−21.8 nm+10.9 nm=−10.9 nm\rho= \rho_{\mathrm{OBS}} - \rho_{\mathrm{probe}}^{\mathrm{eff}} = -21.8\ \mathrm{nm} + 10.9\ \mathrm{nm} = -10.9\ \mathrm{nm}. The associated measurement uncertainty from probe-to-probe variation is 6.3 nm, see Sec. VI B.

Panel (d) shows the profile-averaged RMS curvature RρR_{\rho} of the capacitor pad and the JJ electrode for the different devices. The electrodes have values of RρR_{\rho} in the range from 35 to 82 nm, while the capacitor pads have values in the range from 51 to 151 nm. The observed variation between sidewall profiles at the same spot ranges from 2 to 10 nm and is in general larger for larger RρR_{\rho}. A short-wavelength cutoff of λS=8\lambda_{\mathrm{S}} = 8 nm and λS=25\lambda_{\mathrm{S}} = 25 nm was applied to the capacitor and electrode profiles, respectively. The filtering strongly attenuates features with spatial wavelengths well below 8 nm and 25 nm, respectively, and therefore limits the possible observation of radii of curvature well below these values.

Sidewall angle distribution

A way to visualize and analyze the sidewall geometry without compromising the sidewall sampling point density obtained in a tilted AFM scan is to bin the local sidewall angles into a (normalized) probability density function (PDF). A symmetry reference point for such an analysis is chosen to be the sidewall midpoint (defined above). For example, an idealized SS-shaped sidewall with a single inflection point about which it permits symmetry, such as z=harctan⁡(x/w)z = h\arctan(x/w), would display identical sidewall angle PDFs on either side of the inflection point. We employ the height-based sidewall detection scheme to define the sidewall foot, top, and midpoint.

Figure 5 shows the resulting sidewall angle PDFs on either side of the sidewall midpoint for both a non-linear (a) and a linear circuit element (b). Both AFM profiles reveal partly shallow and wide sidewall angle PDFs, clear asymmetries about the midpoint, and some distributions even being bimodal [51]. These observations should be contrasted with the (more idealized) case of a straight edge with rounded corners, for which one usually expects a unimodal, sharply peaked PDF. The shallowness of the sidewalls (such as characterized by the median value αˉJJ,foot→mid=51∘\bar{\alpha}_{\mathrm{JJ,foot}\to\mathrm{mid}}=51^\circ in Figure 5 (a)) may be expected to have consequences for the electrostatics of the device. Furthermore, the rather large widths of the sidewall angle distributions (such as characterized by the interquartile range IQRJJ,foot→mid=53∘\mathrm{IQR}_{\mathrm{JJ,foot}\to\mathrm{mid}}=53^\circ in Figure 5 (a)) and their nonunimodality problematize simplified characterizations of the geometry, such as by using a single sidewall angle to describe its shape. Across devices and circuit features (even within a single qubit) a large variety of shapes and sidewall angle distributions are observed.

Figure 5. Sidewall angle analysis for a non-linear circuit element (device no. 5) (a) and a linear circuit element (device no. 6) (b). In this analysis we identify the foot, midpoint, and top of the sidewall using the height-based sidewall scheme (cf. Figure 2) with ϵh,foot=ϵh,top=0.02\epsilon_{h,\mathrm{foot}} = \epsilon_{h,\mathrm{top}} = 0.02. Local sidewall angles α\alpha, defined by the angle with the vertical, are binned to produce normalized probability distributions (PDF’s) corresponding to the regions between the foot and midpoint and between the midpoint and the top, shown in the insets. The sidewall angle PDFs may be characterized by their median, αˉ\bar{\alpha}, and interquartile range, IQR, resulting in: αˉJJ,foot→mid=51∘\bar{\alpha}_{\mathrm{JJ,foot}\to\mathrm{mid}} = 51^\circ, IQRJJ,foot→mid=53∘\mathrm{IQR}_{\mathrm{JJ,foot}\to\mathrm{mid}} = 53^\circ, αˉJJ,mid→top=79∘\bar{\alpha}_{\mathrm{JJ,mid}\to\mathrm{top}} = 79^\circ, IQRJJ,mid→top=27∘\mathrm{IQR}_{\mathrm{JJ,mid}\to\mathrm{top}} = 27^\circ for (a), and αˉCoupl,foot→mid=47∘\bar{\alpha}_{\mathrm{Coupl,foot}\to\mathrm{mid}} = 47^\circ, IQRCoupl,foot→mid=28∘\mathrm{IQR}_{\mathrm{Coupl,foot}\to\mathrm{mid}} = 28^\circ, αˉCoupl,mid→top=21∘\bar{\alpha}_{\mathrm{Coupl,mid}\to\mathrm{top}} = 21^\circ, IQRCoupl,mid→top=13∘\mathrm{IQR}_{\mathrm{Coupl,mid}\to\mathrm{top}} = 13^\circ for (b).

The definitions of the sidewall descriptors and the analysis of the measurement results presented in this work form a practical, step-by-step measurement procedure. A summary of the sidewall descriptors and the essential elements defining the AFM measurement protocol are summarized in Table I.

A key objective of a measurement protocol is to enable reproducible and comparable measurements across different laboratories and measurement systems. An integral part of such a protocol is therefore an assessment of the expected variability and the associated uncertainty of the reported sidewall descriptors.

Table II lists the sources of variability and the associated standard uncertainty and include the influence of the probe used, possible contamination, experimental settings and data processing. For a detailed discussion of the source of variability and the associated standard deviation see Appendix F.

Source of variabilityStandard uncertainty uuMitigation strategyVariation ss after mitigation
Probe-to-probe variation (RqLSPR_q^{\mathrm{LSP}})10-20%Use of a single pre-characterized probe (RqLSPR_q^{\mathrm{LSP}})≤4%\leq4\%
Probe-to-probe variation (RρR_\rho)5-8 nmUse of a single pre-characterized probe (RρR_\rho)3-5 nm
Probe degradation and instability5-15 %Probe stability monitoring≤4%\leq4\%
Spatial inhomogeneity of the sampleVariesAveraging over large scan areas or multiple locationsVaries
Particles and contamination≲50%\lesssim50\%Particle removal and separate particle analysis<2%< 2\%
Additional experimental parameters (e.g. applied force)≲20%\lesssim20\%More detailed specification of measurement parameters≤4%\leq4\%
Data processing (outliers, algorithm implementations, sequence)≲10%\lesssim10\%Standardized data-processing procedure≤4%\leq4\%

Table II. Standard uncertainty uu denotes the estimated uncertainty contribution associated with each source of variability for the measurands RqLSPR_q^{\mathrm{LSP}} and RρR_\rho. Residual variation ss denotes the experimentally observed standard deviation after mitigation. The corresponding quantities are denoted u(RqLSP)u(R_q^{\mathrm{LSP}}), u(Rρ)u(R_\rho), s(RqLSP)s(R_q^{\mathrm{LSP}}), and s(Rρ)s(R_\rho).

Sidewall descriptorsProbe specificationMeasurement parametersData processing and analysis
• RqLSPR_q^{\mathrm{LSP}}: Lateral sidewall roughness, Sec. II C<br>• RρR_\rho: Curvature equivalent radius, Sec. II C<br>• ρmin⁡\rho_{\min}: Minimum radius of curvature, App. B<br>• Sidewall angle α\alpha and width, Figure 2• Probe specification is part of measurement, Sec. VI B<br>• Estimation of probe radius ρprobeeff\rho_{\mathrm{probe}}^{\mathrm{eff}}, Eq. (23), App. C<br>• Uncertainty of probe radius u(ρprobeeff)u(\rho_{\mathrm{probe}}^{\mathrm{eff}}), Eq. (25)• Tilt of sample, scan length, scan direction, number of pixels, number of profiles analysed, Sec. VI C<br>• Validity criteria for measurements App. D<br>• Mitigation strategies App. G• Leveling Sec. VI D<br>• High- and low-pass filtering parameters λs\lambda_s, λc\lambda_c, Sec. II C<br>• Correction for probe shape Eq. (24) and App. C 1<br>• Definitions of RqLSPR_q^{\mathrm{LSP}}, Eq. (10), RρR_\rho, Eq. (17), ρmin⁡\rho_{\min}, Eqs. (19), (20)<br>• Sidewall width, Figure 2 and 5<br>• Evaluation of uncertainty Table II, App. G<br>• Limitations and precautions App. C 4 and D

Table 1. Summary of sidewall descriptors and essential elements defining the AFM measurement protocol.

Improving measurement repeatability

Repeatability describes the precision obtained when, for example, the same probe and measurement settings are used throughout a series of measurements. Under these conditions, relative differences in sidewall descriptors can typically be determined much more precisely than the variability observed when different probes or measurement parameters are used. Consequently, repeatability determines the sensitivity with which differences between sidewalls can be resolved over a short time span, whereas probe-to-probe variability and other effects primarily limit the comparability of measurements performed at different times or in different laboratories. Table II lists the mitigation strategies and the residual variation after mitigation. The residual variation s(RqLSP)s(R_q^{\mathrm{LSP}}) after the mitigation strategy has been applied is, in most cases, calculated from experimental observations. Similar residual variations are expected for s(Rρ)s(R_\rho); however, in contrast to s(RqLSP)s(R_q^{\mathrm{LSP}}), the values reported for s(Rρ)s(R_\rho) should be regarded as approximate estimates based on experimental observations and laboratory experience. For a detailed discussion of mitigation strategies for all sources of variation, see Appendix G.

Combined measurement uncertainty

In conventional stylus-based roughness measurements the influence of the mechanical stylus used are usually not corrected for. In other words the measured roughness and associated combined measurement uncertainty are method depend. In this work we suggest a similar approach, that is, the measured result and associated uncertainty should be interpreted together with the shape of the AFM probe and data-processing used. Thus, a possible bias associated with the AFM approach using an AFM probe and the associated data-processing procedure is regarded as part of the measurement method and is therefore not treated as a separate uncertainty contribution.

In this work the combined measurement uncertainty include contributions associated with the use of different probes, probe degradation and instability, possible adsorbed particles and contamination, variations in measurement conditions, and data-processing procedures. In addition to these contributions, the influence of imperfect calibration of the microscope also contributes to the overall measurement uncertainty. Based on calibration measurements using certified reference standards, this contribution is estimated to be below 3%. Based on the discussion and the variability estimates presented in Appendix F, the combined measurement uncertainty (k=1k = 1) for the morphological descriptors investigated in this work is estimated to be in the range of 15-30%.

Conclusion

A protocol for superconducting qubit sidewalls characterization has been developed. The protocol introduces two morphological descriptors; a refinement of the RMS lateral sidewall roughness and a new curvature-equivalent radius for quantifying the vertical sidewall morphology, together with an auxiliary parameter describing the minimum radius of curvature along the profile. In addition, methods for defining the sidewall position and extracting the average sidewall angle have been established.

The proposed descriptors and methods were applied to representative samples and shown to provide consistent and physically meaningful measures of sidewall geometry. Measurement variability, uncertainty, and relevant limitations and precautions were also discussed to establish a systematic framework for sidewall characterization. The proposed protocol is expected to facilitate reproducible sidewall measurements and comparisons of fabrication processes across laboratories. Future work may extend the protocol through a comprehensive uncertainty analysis, additional morphological descriptors, and statistical correlations between sidewall characteristics and device performance.

Methods

Sample fabrication

The transmon devices were fabricated on high-resistivity silicon substrates (>20> 20 kΩ\Omegacm). 200 nm-thick aluminum films were deposited on the substrate using a Plassys MEB550S E-beam evaporator. The qubit capacitors, readout resonators, feedlines, and control lines were patterned by E-beam lithography and subsequently wet-etched by the Transene Type D aluminum etchant. Josephson junctions were formed by Manhattan-style double-angle aluminum evaporation in the same Plassys evaporator, with an intermediate oxidation step. Electrical connections between the Josephson junctions and the remainder of the circuits were defined by e-beam lithography, followed by argon-ion milling and aluminum deposition in the Plassys evaporator. In addition, two chips containing a total of six aluminum readout resonators were fabricated separately, and are referred to as type 2 in Figure 3(e) and regards device number 8-12, shown in Figure 3(h). These resonators were fabricated on double polished C-plane sapphire Al2_2O3_3. Wafers were precleaned in acetone for 5 minutes in a sonic bath followed by oxygen (O2\mathrm{O}_2) plasma cleaning for 5 minutes at a power of 200 W and an O2\mathrm{O}_2 pressure of 1 mbar. For the base layer we deposit 100 nm of aluminium at a base pressure below 10−710^{-7} mbar using an e-beam evaporator with a deposition rate of 0.5 nm/s. To align subsequent lithography steps, we first pattern Ag markers using thermal evaporation and a lift-off process. The resonators and feedline are wet-etched using a single PMMA layer developed using an IPA/water mixture (5:3) for 70 s and etched with ma-D 332 for approximately 4 minutes. The wafer is then profusely rinsed with water and spin dried and remaining resist stripped in 50∘50^\circC acetone for 10 minutes and a sonic bath for 3 minutes. Finally, we rinse the wafer with IPA and spin dry.

Probe specification

In this work, AFM probe with a specified radius of curvature below 10 nm are used (Non-contact probes, spring constant ∼45\sim45 N/m and with the probe located at the cantilever end (see Figure 1 (b) and Ref. [52]). The effective radius of the batch of probes used - including wear - was estimated from images and profiles of a sample designed for AFM probe characterization and found to be

ρprobeeff=10.9 nm.(5)\rho_{\mathrm{probe}}^{\mathrm{eff}} = 10.9\ \mathrm{nm}. \tag*{(5)}

and the associated standard uncertainty was conservatively estimated to be u(ρprobeeff)=6.3u(\rho_{\mathrm{probe}}^{\mathrm{eff}}) = 6.3 nm. Details of the estimation, including representative profiles and images of the AFM probe-characterization sample, are given in Appendix E.

Instrumentation and measurement

The tilted sidewalls were characterized using a high-resolution NX20 AFM (Park Systems, South Korea) with a scanning range of 100 μ\mum ×\times 100 μ\mum ×\times 15 μ\mum in intermittent-contact mode. For optimal imaging of the sidewall morphology and geometrical shape, the scanning plane should ideally be parallel to the sidewall surface. However, the accessible sample tilt is limited by the AFM geometry and sample configuration, which prevent an exact alignment of the apriory unknown angle of the sidewalls. In this work, the samples were mounted on a 50∘50^\circ wedge with the z direction of the AFM aligned roughly perpendicular to the sidewall, see Figure 1 (b).

The sidewalls were measured on seven test qubits distributed over two chips at locations, as indicated in Figure 1. In addition, sidewall profiles of five readout resonators were measured on the two chips containing readout resonators.

The scan length in the fast-scan direction was intended to include at least two long-wavelength cutoffs when possible. For the linear components (capacitor pads, coupling structures, and readout resonators), the long- and short-wavelength cutoff limits were selected as λC=8 μm\lambda_C = 8\ \mu\mathrm{m} and λS=25\lambda_S = 25 nm, respectively. The cutoff wavelength λC\lambda_C and λS\lambda_S were selected based on the physical considerations discussed in Sec. III B and taking the possible measurement range of the AFM into account also. Sampling per profile should be as high as possible, in this work 4096 pixels per lines. The number of profiles per area investigated are at least 32.

For the Josephson junction electrodes, the scan length was limited by the feature dimensions and was typically 2–3 μ\mum, maintaining a sampling interval well below the Nyquist criterion. The long- and short-wavelength cutoff limits were selected as λC=0.25 μm\lambda_C = 0.25\ \mu\mathrm{m} and λS=25\lambda_S = 25 nm for the RqR_q analysis and λS=8\lambda_S = 8 nm for the RpR_p analysis. For the Josephson junction cap (that is, the 3D overlap structure), this length scale is comparable to the total width of the sidewall and no additional long-wavelength cutoff is applied. For the cap a short-wavelength cutoff λS=8\lambda_S = 8 nm was chosen. Selected values are based on the physical considerations discussed in Sec. III B. Further considerations regarding measurement validity and the acceptance criteria, particularly relevant to the reliable determination of the radius of curvature, are given in Appendix D.

Data Processing

For general AFM data processing and calculation of lateral sidewall roughness, SPIP [53] was used. A python-based analysis routine was developed for the calculation of the curvature-equivalent radius, sidewall width, and sidewall angle distribution.

For the lateral sidewall profiles, a least-squares linear reference profile was subtracted from each measured profile prior to filtering and calculation of the descriptors. All profiles were subsequently filtered using a Gaussian S-filter with a short-wavelength cutoff λS\lambda_S and where appropriate an L-filter with a long-wavelength cutoff λC\lambda_C described in Sec. VI C, followed by calculation of the lateral roughness.

The profile-averaged RMS curvature, κRMS\kappa_{\mathrm{RMS}}, is independent of the rotation of the Cartesian coordinate system due to the use of the geometric curvature definition.

Acknowledgments

The author acknowledges the support of DFM, PTB, and NQCP facilities in carrying out this work. In some sections of the article, an AI tool was used for language and grammar corrections. SAK acknowledges funding from Danish Agency for Higher Education and Science, DIANA Quantum Center (DIANAQ), The Danish Chips Competence Center (DKCCC), DIRAC: A national CERN lighthouse initiative, Innovation Fund Denmark (IFD) projects: DeQID, QLIGHT, European Metrology projects: MetSuperQ (Project No. 23FUN08), No-QTeS (Project No. 23NRM04), DINAMO (Project No. 24DIT02). MK gratefully acknowledge support from the Novo Nordisk Foundation (grant no. NNF22SA0081175), the NNF Quantum Computing Programme (NQCP), Villum Foundation through a Villum Young Investigator grant (grant no. 37467), the Innovation Fund Denmark (grant no. 2081-00013B, DanQ), the U.S. Army Research Office (grant no. W911NF-22-1-0042, NHyDTech), by the European Union through an ERC Starting Grant, (grant no. 101077479, NovADePro), and by the Carlsberg Foundation (grant no. CF21-0343). Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of Army Research Office or the US Government. Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them. MB and LG acknowledge funding by the German Federal Ministry of Research, Technology and Space (BMFTR) within the framework programme “Quantum technologies – from basic research to market” (contract number: 13N16158) and by the European Partnership on Metrology, co-financed from the European Union’s Horizon Europe Research and Innovation Programme and by the Participating States (contract number: 23FUN08 MetSuperQ).

Author Contributions

S.A.K, J.G. conceived the idea, experimental plans. J.G, O.D and E.P.D performed the experiments. J.G, H.S.G, A.T.R. performed data analysis. Z.S., A.L.G., L.W.C. M.K., M.B. and L.G. provided transmon and resonator samples. J.G, H.S.G, A.T.R., S.A.K prepared the manuscript with the contribution from all other authors. S.A.K supervised the project.

Competing financial interests

The authors declare no competing financial interests.

Data availability

The data supporting the findings in the study can be made available upon reasonable request to the corresponding authors.

Supplementary Information

Appendix A: Sources of qubit decoherence and dephasing; Appendix B: Definition of relevant geometrical descriptors; Appendix C: Curvature-Based Probe Radius Estimation (CPRE); Appendix D: AFM measurement validity and acceptance criteria; Appendix E: Experimental estimation of the effective probe radius; Appendix F: Sources of measurement variability; Appendix G: Improving measurement repeatability; Appendix H: Relation between ISO Gaussian filtering and standard deviation formulation.

ParameterImpact on T1T_1Impact on T2T_2
Capacitor geometryInterface and radiative lossCharge noise sensitivity
Josephson junction sizeTwo level system and Josephson energy EJE_JFlux noise if looped
Resonator capacitive couplingPurcell decayPhoton shot noise
Dielectric interfacesDielectric losses and heat dissipationTwo-level system induced frequency noise
Chip packagingRadiative and IR lossesStabilization of thermal environment
Filtering and shieldingCut stray radiation, reduce quasiparticle generation, reduce radiative lossesReducing dephasing via environment

Table III. Qualitative summary of key transmon qubit features and their qualitative impact on the relaxation time (T1T_1) and the dephasing time (T2T_2).

Sources of qubit decoherence and dephasing

Complementary to the qubit relaxation time, T1T_1, the qubit dephasing time (T2T_2) characterizes how long a superposition of the qubit’s basis states retains its relative phase before phase fluctuation in the environment cause the phase information to decay (even if the population of the states remains unchanged). It therefore measures the timescale over which quantum interference effects persist.

Table A provides a qualitative summary of key parameters that influence transmon qubit performance. Typical sources that limit coherence include dielectric loss, Purcell decay, quasiparticle tunneling, radiative loss, two-level systems, and infrared radiation. Typical important sources of dephasing include charge noise, flux noise, photon shot noise, and 1/f1/f noise.

Definition of relevant geometrical descriptors

Profile filtering

All profiles analyzed are first corrected by subtraction of a reference line if applicable. Then a short-wavelength cutoff filter is applied with the wavelength limit λS\lambda_S. This operation retain all spatial wave lengths longer than λS\lambda_S while suppressing shorter wavelengths. Thus this will suppress noise and possible influence of the detailed probe structure on a length scale below λS\lambda_S. Where applicable also a long-wavelength cutoff filter is applied with the wavelength limit λC\lambda_C. This filter retain spatial wave lengths shorter than λC\lambda_C while suppressing longer wave lengths. The intention is to remove the waviness of the surface morphology and only include the spatial bandwidth possibly relevant for the variation or losses. This selection of bandwidth is in analogy to roughness measurements on a longer scale.

The lateral sidewall profile

The Lateral Sidewall Profile (LSP), XLSP,i(y)X_{\mathrm{LSP},i}(y), see Figure 1, describes the lateral position of the sidewall projected onto a plane parallel to the substrate, where yy denotes the coordinate along the nominal sidewall direction. The lateral sidewall profile RqLSPR^{\mathrm{LSP}}_q is obtained by projecting the measured AFM surface from the tilted measurement coordinate system onto the substrate plane. The index ii identifies the individual profiles in the set of measured profiles and corresponds to their vertical position along the sidewall, from the foot to the top of the sidewall. For the measurement geometry used in this work, this projection is approximated by

XLSP,i(y)≈ZAFM,i(xAFM,y)cos⁡θ,(6)X_{\mathrm{LSP},i}(y) \approx Z_{\mathrm{AFM},i}(x_{\mathrm{AFM}},y)\cos\theta, \tag*{(6)}

where θ\theta is the angle between the substrate normal and the AFM measurement direction, as illustrated in Figure 6. The amplitude-based lateral sidewall roughness are calculated relative to the designed target geometry, Xref(y)X_{\mathrm{ref}}(y), which may be straight or curved. For a straight target geometry, a least-squares linear reference profile is subtracted from each measured XLSP,i(y)X_{\mathrm{LSP},i}(y), yielding the residual profile

Figure 6. Schematic of the tilted-sample AFM approach enabling quantitative analysis of sidewall geometries, including undercut structures. The red crosses indicate vertical sidewall regions that are not accessible and cannot be reconstructed using the conventional non-tilted approach. The sampled data of a sidewall profile are acquired in a rotated coordinate system (xAFM,zAFM)(x_{\mathrm{AFM}}, z_{\mathrm{AFM}}) relative to the substrate plane.

ΔXLSP,i(y)=XLSP,i(y)−Xref(y),(7)\Delta X_{\mathrm{LSP},i}(y) = X_{\mathrm{LSP},i}(y) - X_{\mathrm{ref}}(y), \tag*{(7)}

which represents the fabrication-induced deviations from the ideal sidewall geometry.

Before calculating the sidewall descriptors, the residual profiles ΔXLSP,i(y)\Delta X_{\mathrm{LSP},i}(y) are filtered using high- and low-pass filters to remove measurement artefacts and spatial frequencies outside the defined measurement bandwidth. In analogy with conventional surface texture measurements, the spatial bandwidth is defined by an S-filter cutoff wavelength, λS\lambda_S, and an L-filter cutoff wavelength, λC\lambda_C, where λS\lambda_S defines the short-wavelength limit and λC\lambda_C defines the long-wavelength limit of the measured profile. The selected cutoff wavelengths are determined by the physically relevant length scales of the sidewall geometry. The resulting filtered residual profiles, ΔXLSP,if(y)\Delta X_{\mathrm{LSP},i}^{f}(y), are used for calculation of the sidewall descriptors.

The Gaussian filter width is selected to provide 50% amplitude transmission at the corresponding cutoff wavelength:

σ=0.1874λ,(8)\sigma= 0.1874\lambda, \tag*{(8)}

where λ\lambda corresponds to either the short-wavelength cutoff λS\lambda_S or the long-wavelength cutoff λC\lambda_C. Further details are given in Appendix H.

The RMS roughness, RqR_q, is selected as a descriptor of the residual lateral sidewall amplitude variation because it provides a well-established statistical measure of profile amplitude. The exponent of two in the RMS definition emphasizes larger deviations while providing a simple and widely used metric for comparison of surface amplitude variations.

The individual RMS lateral sidewall roughness values, Rq,iLSPR_{q,i}^{\mathrm{LSP}}, are calculated from the filtered residual lateral sidewall profiles, ΔXLSP,if(y)\Delta X_{\mathrm{LSP},i}^{f}(y):

Rq,iLSP=1L∫0L(ΔXLSP,if(y))2 dy.(9)R_{q,i}^{\mathrm{LSP}} = \sqrt{\frac{1}{L}\int_{0}^{L}\left(\Delta X_{\mathrm{LSP},i}^{f}(y)\right)^2\,\mathrm{d}y}. \tag*{(9)}

The reported RMS lateral sidewall roughness, RqLSPR_q^{\mathrm{LSP}}, is defined as the average RMS amplitude of the measured profiles across the sidewall:

RqLSP=1N∑i=1NRq,iLSP.(10)R_q^{\mathrm{LSP}} = \frac{1}{N}\sum_{i=1}^{N}R_{q,i}^{\mathrm{LSP}}. \tag*{(10)}

where NN is the number of measured LSP profiles along the vertical extent of the sidewall and LL is the length of each individual sidewall profile.

The vertical sidewall profile

Since no target geometry is available for the Vertical Sidewall Profile (VSP), see Figure 1, a residual amplitude descriptor such as RqR_q cannot be meaningfully defined. Instead, descriptors based on the local radius of curvature provide a measure of the interface geometry, quantifying regions with small radii of curvature that are expected to concentrate electric fields and thereby influence energy dissipation.

The vertical sidewall is characterized by a set of individual measured profiles ZVSP,i(x)=ZAFM,i(xAFM)Z_{\mathrm{VSP},i}(x)=Z_{\mathrm{AFM},i}(x_{\mathrm{AFM}}), where ii denotes the position of the profile along the sidewall in the nominal yy-direction. Each profile represents the vertical sidewall surface in a cross-sectional plane perpendicular to the nominal sidewall direction, with xx representing the coordinate normal to the sidewall. Prior to calculating the curvature, each individual profile is filtered using a Gaussian S-filter with a short-wavelength cutoff of λS\lambda_{\mathrm{S}} to suppress measurement noise and spatial wavelengths that cannot be reliably resolved due to the finite AFM probe geometry. The resulting filtered individual profile is denoted ZVSP,if(x)Z_{\mathrm{VSP},i}^{f}(x).

The local observed curvature κi(x)\kappa_i(x) is calculated from the filtered vertical sidewall profile ZVSP,if(x)Z_{\mathrm{VSP},i}^{f}(x). For a profile expressed in Cartesian coordinates, the local curvature is defined as

κi(x)=ZVSP,if′′(x)(1+(ZVSP,if′(x))2)3/2.(11)\kappa_i(x)=\frac{Z_{\mathrm{VSP},i}^{f\prime\prime}(x)}{\left(1+\left(Z_{\mathrm{VSP},i}^{f\prime}(x)\right)^2\right)^{3/2}}. \tag*{(11)}

where ZVSP,if(x)Z_{\mathrm{VSP},i}^{f}(x) is the filtered individual vertical sidewall profile expressed in the observed AFM coordinate system.

For surface features where both the probe apex and the sample feature can locally be approximated by circular arcs, tip–sample convolution can be approximated by an additive relationship between the observed radius of curvature, the effective probe radius, and the sample radius of curvature:

ρOBS(x)≈ρ(x)−ρprobeeff.(12)\rho_{\mathrm{OBS}}(x)\approx\rho(x)-\rho_{\mathrm{probe}}^{\mathrm{eff}}. \tag*{(12)}

Here, ρOBS(x)\rho_{\mathrm{OBS}}(x) is the signed radius of curvature obtained from the AFM measurement, ρprobeeff\rho_{\mathrm{probe}}^{\mathrm{eff}} is the effective probe radius and is defined as a positive quantity, see Sec C, and ρ(x)\rho(x) is the signed radius of curvature of the sample feature. With the coordinate convention used here, ρ(x)\rho(x) is negative for surface protrusions and positive for depressions or holes. The approximation can be applied to both convex protruding and concave features, provided that the local geometry is accessible to the probe and that the curvature contribution from the probe can be represented by an effective radius. Limitations associated with hypothetical extremely sharp concave features that cannot be accessed by the probe are discussed below.

The local radius of curvature corrected for the probe radius is therefore given by

ρ(x)≈ρOBS(x)+ρprobeeff=1κOBS(x)+ρprobeeff,(13)\rho(x)\approx\rho_{\mathrm{OBS}}(x)+\rho_{\mathrm{probe}}^{\mathrm{eff}}=\frac{1}{\kappa_{\mathrm{OBS}}(x)}+\rho_{\mathrm{probe}}^{\mathrm{eff}}, \tag*{(13)}

where κOBS(x)\kappa_{\mathrm{OBS}}(x) is the signed local curvature calculated from the measured AFM profile. The corresponding probe-radius-corrected local curvature is

κ(x)=1ρ(x)=κOBS(x)1+ρprobeeffκOBS(x).(14)\kappa(x)=\frac{1}{\rho(x)}=\frac{\kappa_{\mathrm{OBS}}(x)}{1+\rho_{\mathrm{probe}}^{\mathrm{eff}}\kappa_{\mathrm{OBS}}(x)}. \tag*{(14)}

To avoid reporting protrusion radii with magnitudes smaller than the effective probe radius, corrected radii satisfying ρ(x)≥−ρprobeeff\rho(x)\geq-\rho_{\mathrm{probe}}^{\mathrm{eff}} are assigned the limiting value −ρprobeeff-\rho_{\mathrm{probe}}^{\mathrm{eff}}. The corresponding probe-radius-corrected local curvature is subsequently obtained as κi(x)=1/ρi(x)\kappa_i(x)=1/\rho_i(x). An RMS-equivalent mean curvature, κRMS,i\kappa_{\mathrm{RMS},i}, with units of inverse length, is defined for each individual profile as the root-mean-square value of the local curvature along the profile. To account for the actual geometry of the curved profile, the averaging is performed along the arc length:

κRMS,i=(∫x1x2κi(x)2[1+(ZVSP,if′(x))2]1/2 dx∫x1x2[1+(ZVSP,if′(x))2]1/2 dx)1/2.(15)\kappa_{\mathrm{RMS},i}=\left(\frac{\int_{x_1}^{x_2}\kappa_i(x)^2\left[1+\left(Z_{\mathrm{VSP},i}^{f\prime}(x)\right)^2\right]^{1/2}\,\mathrm{d}x}{\int_{x_1}^{x_2}\left[1+\left(Z_{\mathrm{VSP},i}^{f\prime}(x)\right)^2\right]^{1/2}\,\mathrm{d}x}\right)^{1/2}. \tag*{(15)}

The profile-averaged RMS curvature is defined as the arithmetic mean of the values obtained from the NN individually evaluated vertical sidewall profiles:

κˉRMS=1N∑i=1NκRMS,i.(16)\bar{\kappa}_{\mathrm{RMS}}=\frac{1}{N}\sum_{i=1}^{N}\kappa_{\mathrm{RMS},i}. \tag*{(16)}

The curvature-equivalent radius, RpR_p, is defined as the inverse of the profile-averaged RMS curvature:

Rp=1κˉRMS.(17)R_p=\frac{1}{\bar{\kappa}_{\mathrm{RMS}}}. \tag*{(17)}

Small values of RpR_p indicate the presence of many sharp features, where the surface profile locally matches the curvature of associated small circles. Small local radii of curvature contribute to κRMS\kappa_{\mathrm{RMS}} and therefore reduce RpR_p, independent of whether they correspond to protruding or recessed surface features.

The minimum radius of curvature

The local radius of curvature, ρ(x)\rho(x), can take both positive and negative values. For a profile described by z(x)z(x), the local radius of curvature is given by

ρ(x)=(1+(dzdx)2)3/2d2zdx2.(18)\rho(x) = \frac{\left(1+\left(\frac{\mathrm{d}z}{\mathrm{d}x}\right)^2\right)^{3/2}}{\frac{\mathrm{d}^2z}{\mathrm{d}x^2}}. \tag*{(18)}

With the coordinate convention used here, negative values of ρ(x)\rho(x) correspond to protrusions, whereas positive values correspond to depressions or holes. The minimum radius of curvature is therefore defined separately for the two types of features. For protrusions, the descriptor is the negative radius with the smallest absolute value,

ρmin⁡prot=−min⁡ρ(x)<0∣ρ(x)∣,(19)\rho_{\min}^{\mathrm{prot}}=-\min_{\rho(x)<0}|\rho(x)|, \tag*{(19)}

whereas for depressions or holes it is defined as

ρmin⁡dep=min⁡ρ(x)>0ρ(x).(20)\rho_{\min}^{\mathrm{dep}}=\min_{\rho(x)>0}\rho(x). \tag*{(20)}

Thus, ρmin⁡prot\rho_{\min}^{\mathrm{prot}} remains negative, while ρmin⁡dep\rho_{\min}^{\mathrm{dep}} is positive. The ρmin⁡\rho_{\min} descriptor serves two purposes. When the measurement quality is sufficient to ensure reliable curvature determination, ρmin⁡\rho_{\min} may be interpreted as an estimate of the smallest local radius of curvature present on the sidewall, thereby identifying the sharpest resolvable geometrical feature of the measured profile. Furthermore, ρmin⁡\rho_{\min} is used to determine the Curvature-Based Probe Radius Estimation, see Appendix C.

Curvature-Based Probe Radius Estimation (CPRE)

Blind probe reconstruction can estimate the probe geometry from measured surface topography, but for the probe shapes within few nanometers from the apex, which is the relevant length scale for this work, the reconstruction depends critical on the treshold values for the auxiliary parameter to suppress noise in the image used to reconstruct the probe shape [54, 55, 56]. Sharp features [57] and sharp grating profiles [58] have also been used to estimate an effective AFM probe radii.

In the present work, an approach based on the local curvature of multiple sharp protrusions is used to obtain a conservative upper bound, referred to as Curvature-Based Probe Radius Estimation (CPRE). Rather than reconstructing the complete probe geometry, CPRE estimates an effective probe radius from measurements on a tip characterization sample and uses this estimate to correct the measured local radii of curvature. For an ideal round probe the estimated effective probe radius would be equal to the radius of the round probe. The method provides a practical probe radius correction together with an associated measurement uncertainty suitable for quantitative AFM measurements.

Theory

In the CPRE approach, a dedicated tip-characterization sample containing sharp features is measured using the AFM probe under validation. For a single averaged measured profile, six local minima of the radius of curvature associated with protrusions are identified. The radius of curvature at the jj-th local minimum is denoted by

ρobs,jprot=ρobs(xj),ρobs(xj)<0,j=1,…,6,(21)\rho_{\mathrm{obs},j}^{\mathrm{prot}}=\rho_{\mathrm{obs}}(x_j), \qquad\rho_{\mathrm{obs}}(x_j)<0, \qquad j=1,\ldots,6, \tag*{(21)}

where xjx_j denotes the position of the jj-th local minimum. The upper-bound estimate for the effective probe radius is then determined from the average absolute radius of the m=6m=6 local minima,

ρprobeUB=1m∑j=1m∣ρobs,jprot∣.(22)\rho_{\mathrm{probe}}^{\mathrm{UB}}=\frac{1}{m}\sum_{j=1}^{m}\left|\rho_{\mathrm{obs},j}^{\mathrm{prot}}\right|. \tag*{(22)}

The upper-bound estimate for the effective probe radius is based on the observed radius of curvature of protrusions, ρobsprot\rho_{\mathrm{obs}}^{\mathrm{prot}}. The corresponding radius of curvature for depressions, ρobsdep\rho_{\mathrm{obs}}^{\mathrm{dep}}, is not used because it cannot provide an upper bound for the effective probe radius.

Assuming that the minimum radius of curvature of the characterization sample and the effective probe radius are of comparable magnitude, the effective probe radius can be estimated as

ρprobeeff=12ρprobeUB.(23)\rho_{\mathrm{probe}}^{\mathrm{eff}} = \frac{1}{2}\rho_{\mathrm{probe}}^{\mathrm{UB}}. \tag*{(23)}

The measured local radius of curvature is subsequently corrected for the probe contribution according to

ρi(x)={ρobs,i(x)+ρprobeeffif ρobs,i(x)+ρprobeeff≤−ρprobeeff,−ρprobeeffif −ρprobeeff<ρobs,i(x)+ρprobeeff<0,ρobs,i(x)+ρprobeeffif ρobs,i(x)+ρprobeeff≥0.(24)\rho_i(x) = \begin{cases} \rho_{\mathrm{obs},i}(x) + \rho_{\mathrm{probe}}^{\mathrm{eff}} & \text{if }\rho_{\mathrm{obs},i}(x) + \rho_{\mathrm{probe}}^{\mathrm{eff}} \leq-\rho_{\mathrm{probe}}^{\mathrm{eff}}, \\ -\rho_{\mathrm{probe}}^{\mathrm{eff}} & \text{if }-\rho_{\mathrm{probe}}^{\mathrm{eff}} < \rho_{\mathrm{obs},i}(x) + \rho_{\mathrm{probe}}^{\mathrm{eff}} < 0, \\ \rho_{\mathrm{obs},i}(x) + \rho_{\mathrm{probe}}^{\mathrm{eff}} & \text{if }\rho_{\mathrm{obs},i}(x) + \rho_{\mathrm{probe}}^{\mathrm{eff}} \geq0. \end{cases} \tag*{(24)}

where ρi(x)\rho_i(x) represents the probe-corrected local radius of curvature of profile ii. To avoid reporting protrusion radii with magnitudes smaller than the effective probe radius, corrected values for which ρi(x)≥−ρprobeeff\rho_i(x) \geq-\rho_{\mathrm{probe}}^{\mathrm{eff}} are assigned the limiting value −ρprobeeff-\rho_{\mathrm{probe}}^{\mathrm{eff}}.

Uncertainty

The uncertainty associated with the limited knowledge of the effective probe radius is evaluated conservatively by assuming a rectangular distribution over the interval from zero to the upper bound of the probe radius of curvature, ρprobeUB\rho_{\mathrm{probe}}^{\mathrm{UB}}:

u(ρprobeeff)=ρprobeUB23,(25)u\left(\rho_{\mathrm{probe}}^{\mathrm{eff}}\right) = \frac{\rho_{\mathrm{probe}}^{\mathrm{UB}}}{2\sqrt{3}}, \tag*{(25)}

where 232\sqrt{3} is the standard deviation factor for a rectangular distribution.

Traceability

The traceability of the correction term ρprobeeff\rho_{\mathrm{probe}}^{\mathrm{eff}} is established through the calibration of the AFM measurement system. The traceability of the CPRE approach does not require the characterization sample to be a certified reference material, since the correction term is derived from the calibrated AFM measurement and the corresponding probe-characterization procedure.

CPRE treats the unknown contribution from the finite radius of curvature of the characterization sample as part of the uncertainty evaluation. Consequently, the method remains traceable through the AFM calibration, while the uncertainty reflects the limited knowledge of the actual sharpness of the characterization sample. A characterization sample with reduced sharpness will result in a larger uncertainty; however, the estimated probe radius remains valid within the associated uncertainty limits.

Limitation

The simple correction model ρi(x)=ρobs,i(x)−ρprobeeff\rho_i(x) = \rho_{\mathrm{obs},i}(x) - \rho_{\mathrm{probe}}^{\mathrm{eff}} represents a local curvature correction based on an effective probe radius. It does not attempt to reconstruct inaccessible geometric information. In particular, hypothetical extremely sharp concave features that cannot be accessed by the probe are not assessed, as no assumed detailed knowledge of the probe shape can recover information that is absent from the measured AFM profile. This limitation is analogous to conventional surface roughness measurements, where probe convolution effects could also be present but are not normally considered. Because the correction term ρprobeeff\rho_{\mathrm{probe}}^{\mathrm{eff}} has an associated standard uncertainty u(ρprobeeff)u(\rho_{\mathrm{probe}}^{\mathrm{eff}}), corrected radii of curvature smaller than ρprobeUB\rho_{\mathrm{probe}}^{\mathrm{UB}} may be numerically obtained. However, such estimates cannot be distinguished from the probe-radius upper bound within the uncertainty introduced by the probe-radius correction.

AFM measurement validity and acceptance criteria

Amplitude-modulation AFM can exhibit apparent height shifts (often in the order of less than a nm) and other imaging artifacts [59, 60, 61] associated with the nonlinear tip-sample interaction and transitions between attractive- and repulsive-dominated oscillation states, which can produce changes in amplitude and phase. Such dynamical effects may influence the measured profile and hence the derived radius of curvature, particularly near regions with small radius of curvature. Therefore, all measurements were checked for signs of instability and related artifacts [59, 60, 61]. As acceptance criteria for measurement validity the following were used: (i) no distinct jump in the phase along the profile, in particular near regions with small radius of curvature, (ii) the phase variation was required to be smooth and remain within ±5∘\pm5^\circ, (iii) the difference between trace and retrace profiles was required to remain within the acceptable measurement variation, and (iv) the derived radius of curvature was required to be insensitive to reasonable changes in drive frequency and amplitude setpoint.

Experimental estimation of the effective probe radius

The tip characterization sample used consisted of a thin film of nanoscale sharp protrusions [62], reported to be niobium with protrusion radii less than 5 nm [63]. The sample was imaged under the same measurement conditions as for the sidewall. A conservative upper limit for the observed ρobs\rho_{\mathrm{obs}} was estimated based on a profile averaged over 93 individual profiles of a used probe, where the average radius of the six smallest observed radius of curvature of the averaged profile was found to be 21.7(0.3) nm calculated according to Eq. (22), see Figure 7. Averaging 93 profiles reduces random noise and suppresses short-term drift, making the estimated average a conservative upper limit. Based on the above measurement and other measurements for different probes, the upper bound ρprobeUB\rho_{\mathrm{probe}}^{\mathrm{UB}} for the effective probe radius for the batch of probes used was conservatively estimated to be 21.7 nm. This gives an effective probe radius for the batch of probe used according to Eq. (23) of

Figure 7. Profile of the tip characterization sample recorded by AFM. The top panel shows a profile averaged over 93 individual profiles recorded with a used probe. The green elongated circular arcs indicate the six protrusions with the smallest radii of curvature, and the green elongated arcs indicate the six depressions with the smallest radii of curvature. The lower panel shows the local observed radius of curvature 1/κOBS(x)=ρOBS(x)1/\kappa_{\mathrm{OBS}}(x)=\rho_{\mathrm{OBS}}(x). The smooth variation of the local observed radius of curvature indicates consistent and reliable data.

ρprobeeff=10.9 nm.(26)\rho_{\mathrm{probe}}^{\mathrm{eff}} = 10.9\ \mathrm{nm}. \tag*{(26)}

The associated uncertainty is conservatively estimated according to Eq. (25) to be u(ρprobeeff)=6.3u(\rho_{\mathrm{probe}}^{\mathrm{eff}})=6.3 nm. The observed average minimum radius of curvature on each of n=128n=128 tilted profiles in an image (see Figure 8) was found to be

Figure 8. Analysis of an ensemble of profiles of a tilted tip characterization sample. The upper panel shows the smallest observed radii of curvature of the protrusions, ρobsprot\rho_{\mathrm{obs}}^{\mathrm{prot}}, and depressions, ρobsdep\rho_{\mathrm{obs}}^{\mathrm{dep}}, for each of the 128 individual profiles as a function of the profile position along the yy-direction. The lower panel shows the value of κOBS\kappa_{\mathrm{OBS}}, calculated from the observed profile according to Eq. (15), for each profile position along the yy-direction.

ρ‾obsprot=1n∑i=1n∣ρobs,iprot∣=8.5 nm.(27)\overline{\rho}_{\mathrm{obs}}^{\mathrm{prot}} = \frac{1}{n}\sum_{i=1}^{n}\left\lvert\rho_{\mathrm{obs},i}^{\mathrm{prot}}\right\rvert= 8.5\ \mathrm{nm}. \tag*{(27)}

This confirms that the upper bound ρprobeUB=21.7 nm>ρ‾obsprot\rho_{\mathrm{probe}}^{\mathrm{UB}} = 21.7\ \mathrm{nm} > \overline{\rho}_{\mathrm{obs}}^{\mathrm{prot}} is a conservative choice, allowing small variations of probe shape, wear and the possible influence of drift and noise on each profile with a good margin.

The experimentally estimated effective radius is larger than the manufacturer’s specification of ρ<10 nm\rho< 10\ \mathrm{nm}. This is consistent with the possibility of probe wear during use, which may increase the effective radius of curvature, and with the conservative approach outlined here.

Treatment of sharp protrusions

For very sharp protrusions, the probe-radius correction in Eq. (14) can become unreliable if 1+ρprobeeffκOBS1 + \rho_{\mathrm{probe}}^{\mathrm{eff}}\kappa_{\mathrm{OBS}} approaches zero, i.e., if κOBS\kappa_{\mathrm{OBS}} approaches −1/ρprobeeff-1/\rho_{\mathrm{probe}}^{\mathrm{eff}}. To avoid this unreliable regime while maintaining the contribution of sharp protrusions to the curvature descriptor, a conservative lower limit of 21.8 nm21.8\ \mathrm{nm} is applied on the observed radius of protrusions. Thus, if the observed protrusion radii satisfy −21.8 nm<ρOBS,i(x)<0-21.8\ \mathrm{nm} < \rho_{\mathrm{OBS},i}(x) < 0, it is assigned the limit value ρOBS,i(x)=−21.8 nm\rho_{\mathrm{OBS},i}(x) = -21.8\ \mathrm{nm} corresponding to κOBS,i(x)=−0.0459 nm−1\kappa_{\mathrm{OBS},i}(x) = -0.0459\ \mathrm{nm}^{-1}. With ρprobeeff=10.9 nm\rho_{\mathrm{probe}}^{\mathrm{eff}} = 10.9\ \mathrm{nm}, the resulting corrected curvature is limited to κi(x)=−0.0917 nm−1\kappa_i(x) = -0.0917\ \mathrm{nm}^{-1}. Positive observed radii, corresponding to depressions, are not affected by this limiting procedure. The limit is conservative; several radii of around 21.8 nm21.8\ \mathrm{nm} and below have been reliably measured on the tip characterization sample. In the hypothetical case that all observed protrusion radii were smaller than 21.8 nm21.8\ \mathrm{nm}, the corrected radius would still be limited to −10.9 nm-10.9\ \mathrm{nm}. Taking into account the associated measurement uncertainty u(ρprobeeff)=6.3 nmu(\rho_{\mathrm{probe}}^{\mathrm{eff}}) = 6.3\ \mathrm{nm}, a radius significantly smaller than 21.8 nm21.8\ \mathrm{nm} cannot be measured.

Sources of measurement variability

Probe-to-probe variation

An important source of variability in the sidewall descriptors is probe-to-probe variation; i.e., different probes may yield different descriptor values for the same sidewall geometry. Although probes may have the same nominal tip radius (e.g., ρprobe<10 nm\rho_{\mathrm{probe}} < 10\ \mathrm{nm}), variations in the fabrication process result in differences in the actual probe geometry. The standard uncertainty associated with the above variation is estimated to be 10–20% for RqSPR_q^{\mathrm{SP}}. For the probe used in this work, the effective probe radius is assigned a conservatively estimated standard uncertainty of u(ρprobeeff)=6.3 nmu(\rho_{\mathrm{probe}}^{\mathrm{eff}}) = 6.3\ \mathrm{nm}, (see App. Sec. F) which contributes to the uncertainty of RρR_{\rho} through the proberadius correction given in Eq. (11) to (17). In general the standard uncertainty associated with variation in the probe radius is estimated to be 5 - 8 nm.

Probe degradation and instability

Another source of variability is probe degradation and contamination. Probe wear or material transfer from the sample surface during scanning may alter the effective probe geometry and measurement response. Based on experience from numerous measurements, most probes are suitable for approximately one day of measurements. The standard uncertainty associated with the above variation is estimated to be 5–15% for both RqLSPR_{q}^{\mathrm{LSP}} and RρR_{\rho}.

Spatial inhomogeneity of the sample

The reported descriptors represent averages over the investigated microscopic regions (<100 μm<100\ \mu\mathrm{m} in lateral size). The descriptor values may vary between different microscopic regions due to local inhomogeneity in the fabrication process. This larger-scale variation is not captured by a single measurement and is therefore not included in the reported descriptor values. It should be evaluated separately and it should be stated separately for what area(s) the measurement are valid. The variation associated with with sample inhomogeneity for the superconducting elements investigated has been in the range of 5–15% for both RqLSPR_{q}^{\mathrm{LSP}} and RρR_{\rho}.

Particle and contamination

The descriptors can also be influenced by particles and other surface contamination originating from the fabrication process or the measurement environment. Such features may significantly affect device functionality, and their presence should be documented and quantified. However, particles may consist of different materials introduced after fabrication, and it remains unclear whether they should be included in descriptors intended to describe the designed geometry of the sidewall. Furthermore, particles and contamination cannot be meaningfully described by the sidewall descriptors focused on average amplitude and surface curvature. The recommended approach is therefore to minimize or remove the influence of particles and contamination when determining the sidewall descriptors defined in this work. Ideally, new descriptors should be defined and recorded to quantify the presence and amount of particles and contamination; however, this is outside the scope of the present work. The standard uncertainty associated with the above variation is estimated to be up to 50 % for both RqLSPR_{q}^{\mathrm{LSP}} and RρR_{\rho} before the measurement is classified as an outlier.

Additional experiment parameters

General measurement conditions reported in the literature, but not specifically addressed in this paper, may contribute to the observed variation in the descriptors. These include the applied probe force, thermal stability, environmental vibrations, and, importantly, the adjustment of the feedback loop, which effectively acts as an additional filtering process. The standard uncertainty associated with the above variation is estimated to be up to 20% for both RqLSPR_{q}^{\mathrm{LSP}} and RρR_{\rho}.

Data processing

General aspects of data processing reported in the literature, but not specifically addressed in this paper, may also contribute to variation in the descriptors, in particular the treatment of outliers, the exact numerical implementation of the calculations, and the application of filters, including, e.g., the order in which high- and low-pass filtering is performed. The standard uncertainty associated with the above variation is estimated to be up to 10% for both RqLSPR_{q}^{\mathrm{LSP}} and RρR_{\rho}. For a summary of sources of variability, see Table II.

Improving measurement repeatability

This subsection describes procedures for improving repeatability and enabling accurate comparison of sidewall descriptors. Using the same probe, short-term changes in probe condition (timescales of less than one day) can be monitored and mitigated by performing reference measurements on a reference sample before and after a series of measurements. If the reference measurements remain within predefined limits, the measurements performed between the reference measurements can be considered to have a low relative variation.

Using clean surfaces without particles or contamination is the preferred approach to minimize variation from this source. In this work it is documented that this is possible, and it is estimated that the influence of particles and contamination on the images shown is close to zero %. Alternatively, suspected particles and contamination can be identified and removed during image processing, for example by excluding affected regions. With this approach, the residual variation can also be limited to approximately 1-2 % (not shown). Further improvement in repeatability can be achieved by maintaining identical measurement conditions, including applied force, feedback parameters, and other instrument settings. Finally, consistent data processing using identical algorithms and processing sequences can reduce additional numerical variation.

Figure 3(h) shows a total of 10 measurements performed with three different probes on four different samples. The average observed standard deviation, s(RQLSP)=4.1%s(R_{Q}^{\mathrm{LSP}}) = 4.1\%, obtained using the mitigation strategies described above, represents a conservative upper limit for the contribution from each individual source of variation, including probe-to-probe variation, probe degradation, experimental parameters, and data processing.

Relation between ISO Gaussian filtering and standard deviation formulation

The Gaussian S- and L-filters used in this work follow the ISO 16610-21 (formerly ISO 11562) definition, where the filter is specified by its cutoff wavelength, λ\lambda. The normalized spatial weighting function is

G(x)=1αλexp⁡[−π(xαλ)2],(28)G(x) = \frac{1}{\alpha\lambda}\exp\left[-\pi\left(\frac{x}{\alpha\lambda}\right)^2\right], \tag*{(28)}

where α=0.4697\alpha= 0.4697. This definition corresponds to a Gaussian filter with 50% amplitude transmission at the cutoff wavelength.

The Gaussian filter can equivalently be expressed in terms of its standard deviation,

G(x)=1σ2πexp⁡(−x22σ2).(29)G(x) = \frac{1}{\sigma\sqrt{2\pi}}\exp\left(-\frac{x^2}{2\sigma^2}\right). \tag*{(29)}

Equating the exponents of the two Gaussian expressions gives

σ=αλ2π≈0.1874λ.(30)\sigma= \frac{\alpha\lambda}{\sqrt{2\pi}} \approx0.1874\lambda. \tag*{(30)}

Thus, the ISO Gaussian filter is mathematically equivalent to the standard Gaussian parameterization, with the filter width expressed by the standard deviation rather than the cutoff wavelength.

References

  1. [1]M. Kjaergaard, M. E. Schwartz, J. Braumüller, P. Krantz, J. I.-J. Wang, S. Gustavsson, and W. D. Oliver, Superconducting qubits: Current state of play, Annual Review of Condensed Matter Physics 11, 369 (2020).DOI
  2. [2]O. Ezratty, Perspective on superconducting qubit quantum computing, The European Physical Journal A 59, 94 (2023).DOI
  3. [3]C. R. H. McRae, G. M. Stiehl, H. Wang, S.-X. Lin, S. A. Caldwell, D. P. Pappas, J. Mutus, and J. Combes, Reproducible coherence characterization of superconducting quantum devices, Applied Physics Letters 119 (2021).DOI
  4. [4]I. Siddiqi, Engineering high-coherence superconducting qubits, Nature Reviews Materials 6, 875 (2021).DOI
  5. [5]J. J. Burnett, A. Bengtsson, M. Scigliuzzo, D. Niepce, M. Kudra, P. Delsing, and J. Bylander, Decoherence benchmarking of superconducting qubits, npj Quantum Information 5, 54 (2019).DOI
  6. [6]Van Damme, J., Massar, S., Acharya, R., Ivanov, T., Perez Lozano, D., Canvel, Y., Demarest, M., Vangoidssenhoven, Y., Hermans, Y., Lai, J., & al., . et . (2024). Advanced CMOS manufacturing of superconducting qubits on 300 mm wafers. Nature, 634, 74.
  7. [7]Kim, D.-Y., Kwon, H., Han, D. S., Kim, K., Kim, H. K., Lee, C., Kim, G., Lee, J.-Y. M., Kim, D., Kim, E.-A., & al., . et . (2026). Impact of fabrication history induced disorder on Josephson junction reproducibility and qubit performance. Scientific Reports.
  8. [8]J. M. Martinis, Surface loss calculations and design of a superconducting transmon qubit with tapered wiring, npj Quantum Information 8, 26 (2022).DOI
  9. [9]T. Stavenga, S. Khan, Y. Liu, P. Krogstrup, and L. DiCarlo, Lower-temperature fabrication of airbridges by grayscale lithography to increase yield of nanowire transmons in circuit QED quantum processors, Applied Physics Letters 123 (2023).DOI
  10. [10]Khan, S. A., Lampadaris, C., Cui, A., Stampfer, L., Liu, Y., Pauka, S. J., Cachaza, M. E., Fiordaliso, E. M., Kang, J.-H., Kornevchuk, S., & al., . et . (2020). Highly transparent gatable superconducting shadow junctions. ACS Nano, 14, 14605.
  11. [11]Quintana, C., Megrant, A., Chen, Z., Dunsworth, A., Chiaro, B., Barends, R., Campbell, B., Chen, Y., Hoi, I.-C., Jeffrey, E., & al., . et . (2014). Characterization and reduction of microfabrication-induced decoherence in superconducting quantum circuits. Applied Physics Letters, 105.
  12. [12]Kim, R. H., Haeuser, S. J., Park, J.-M., Chan, R. K., Oh, J.-S., Koschny, T., Zhou, L., Kramer, M. J., Murthy, A. A., Bal, M., & al., . et . (2025). Terahertz near-field imaging of sidewall-induced losses in superconducting qubits. Applied Physics Letters, 127.
  13. [13]A. A. Murthy, M. Bal, M. J. Bedzyk, H. Cansizoglu, R. K. Chan, V. Chandrasekhar, F. Crisa, A. Datta, Y. Deng, C. D. Matute Diaz, V. P. Dravid, D. A. Garcia-Wetten, S. Garattoni, S. Ghimire, D. P. Goronzy, S. de Graaf, S. Haeuser, M. C. Hersam, P. Hopkins, D. Isheim, K. Joshi, R. Kim, S. Kolachina, C. J. Kopas, M. J. Kramer, E. O. Lachman, J. Lee, P. G. Lim, A. Lunin, W. Mah, J. Marshall, J. Y. Mutus, J.-S. Oh, D. Olaya, D. P. Pappas, J.-m. Park, R. Prozorov, R. dos Reis, D. N. Seidman, Z. Sung, M. Tanatar, M. J. Walker, M. Wisne, J. Wang, H. Wu, L. Zhou, S. Zhu, A. Grassellino, and A. Romanenko, Identifying Materials-Level Sources of Performance Variation in Superconducting Transmon Qubits, arXiv e-prints, arXiv:2503.14424 (2025).arXiv
  14. [14]M. Alghadeer, S. D. Fasciati, S. Cao, M. Piscitelli, M. C. Spink, D. G. Hopkinson, M. Danaie, S. C. Speller, P. J. Leek, and M. Bakr, Characterization of nanostructural imperfections in superconducting quantum circuits, Materials for Quantum Technology 5, 035201 (2025).arxiv.org/abs/2501.15059
  15. [15]J. Verjauw, R. Acharya, J. Van Damme, T. Ivanov, D. P. Lozano, F. A. Mohiyaddin, D. Wan, J. Jussot, A. M. Vadiraj, M. Mongillo, M. Heyns, I. Radu, B. Govoreanu, and A. Potočnik, Path toward manufacturable superconducting qubits with relaxation times exceeding 0.1 ms, npj Quantum Information 8, 93 (2022).DOI
  16. [16]R. Kizu, I. Misumi, A. Hirai, and S. Gonda, Line edge roughness measurement on vertical sidewall for reference metrology using a metrological tilting atomic force microscope, Journal of Micro/Nanolithography, MEMS, and MOEMS 19, 014003 (2020).DOI
  17. [17]Y. Martin and H. K. Wickramasinghe, Method for imaging sidewalls by atomic force microscopy, Applied Physics Letters 64, 2498 (1994).DOI
  18. [18]S.-J. Cho, B.-W. Ahn, J. Kim, J.-M. Lee, Y. Hua, Y. K. Yoo, and S.-i. Park, Three-dimensional imaging of undercut and sidewall structures by atomic force microscopy, Review of Scientific Instruments 82, 023707 (2011).DOI
  19. [19]R. Kizu, I. Misumi, A. Hirai, and S. Gonda, Line edge roughness measurement on vertical sidewall for reference metrology using a metrological tilting atomic force microscope, Journal of Micro/Nanolithography, MEMS, and MOEMS 19, 014003 (2020).DOI
  20. [20]S.-B. Yoo, S.-H. Yun, A.-J. Jo, S.-J. Cho, H. Cho, J.-H. Lee, and B.-W. Ahn, Automated measurement and analysis of sidewall roughness using three-dimensional atomic force microscopy, Applied Microscopy 52, 1 (2022).DOI
  21. [21]Lisenfeld, J., Bilmes, A., Megrant, A., Kroll, J., Klimov, P., Weiss, G., Martinis, J. M., & Ustinov, A. V. (2019). Electric field spectroscopy of material defects in transmon qubits. Npj Quantum Information, 5, 105.
  22. [22]N. D. Korshakov, D. O. Moskalev, A. A. Solovyeva, D. A. Moskaleva, E. S. Lotkov, A. R. Ibragimov, M. V. Androschuk, I. A. Ryzhikov, Y. V. Panfilov, and I. A. Rodionov, Aluminum Josephson junction microstructure and electrical properties modified by thermal annealing, Scientific Reports 14, 26066 (2024).DOI
  23. [23]M. Hegedüs, R. Banerjee, A. Hutcheson, T. Barker, S. Mahashabde, A. V. Danilov, S. E. Kubatkin, V. Antonov, and S. E. de Graaf, In situ scanning gate imaging of individual quantum two-level system defects in live superconducting circuits, Science Advances 11, eadt8586 (2025).DOI
  24. [24]M. P. Bland, F. Bahrami, J. G. C. Martinez, P. H. Pretegaard, B. M. Smitham, A. Joshi, E. Hedrick, A. Pakpour-Tabrizi, S. Kumar, A. Jindal, R. D. Chang, A. Yang, G. Cheng, N. Yao, R. J. Cava, N. P. de Leon, and A. A. Houck, Millisecond lifetimes and coherence times in 2D transmon qubits, Nature 647, 343 (2025).DOI
  25. [25]Bhatia, E., Du, Y., Koirala, K. P., Liu, M., Macty, J., Ramayakkara, F., Ponce, S. S., Rao, S. P., Rebar, D. J., Sushko, P. V., VanDevender, B. A., Wang, C., Warner, M. G., & Xiao, Z. (2026). Tantalum Damascene Coplanar Waveguide Resonators Fabricated Using 300 nm Scale Processes. In arXiv e-prints. 2604.22086
  26. [26]Z. Lu, M. Song, J. Yuan, Z. Liu, Y. Xu, M. Jiang, Z. Sun, Z. Li, Q. Xiu, H. Gao, and Y. Wei, Study on Etching Methods for Highly Vertical Sidewall Profiles in Niobium Films, IEEE Transactions on Applied Superconductivity 36, 6001907 (2026).DOI
  27. [27]R. Manenti and M. Motta, Quantum Information Science (Oxford University Press, 2023).DOI
  28. [28]E. Hedrick, F. Bahrami, A. C. Pakpour-Tabrizi, A. Joshi, Q. Rumman Rahman, A. Yang, R. D. Chang, M. P. Bland, A. Jindal, G. Cheng, N. Yao, R. J. Cava, A. A. Houck, and N. P. de Leon, Quantifying surface losses in superconducting aluminum microwave resonators, MRS Bulletin (2026).arxiv.org/abs/2603.13183
  29. [29]S. Eun, S. H. Park, K. Seo, K. Choi, and S. Hahn, Shape optimization of superconducting transmon qubits for low surface dielectric loss, Journal of Physics D: Applied Physics 56, 505302 (2023).DOI
  30. [30]N. S. Smirnov, E. A. Krivko, A. I. Solovyova, A. I. Ivanov, and I. A. Rodionov, Wiring surface loss of a superconducting transmon qubit, Scientific Reports 14, 7326 (2024).DOI
  31. [31]E. M. Purcell, Spontaneous Emission Probabilities at Radio Frequencies, in Confined Electrons and Photons: New Physics and Applications, edited by E. Burstein and C. Weisbuch (Springer US, Boston, MA, 1995) pp. 839–839.DOI
  32. [32]A. A. Houck, J. A. Schreier, B. R. Johnson, J. M. Chow, J. Koch, J. M. Gambetta, D. I. Schuster, L. Frunzio, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, Controlling the Spontaneous Emission of a Superconducting Transmon Qubit, Phys. Rev. Lett. 101, 080502 (2008).pubmed.ncbi.nlm.nih.gov/18764596
  33. [33]G. Beaulieu, J.-Z. Chen, M. Scigliuzzo, O. Benhayoune-Khadraoui, A. A. Chapple, P. A. Spring, A. Blais, and P. Scarlino, Fast, High-Fidelity Transmon Readout with Intrinsic Purcell Protection via Nonperturbative Cross-Kerr Coupling, PRX Quantum 7, 033017 (2026).DOI
  34. [34]P. Joyez, P. Lafarge, A. Filipe, D. Esteve, and M. H. Devoret, Observation of parity-induced suppression of Josephson tunneling in the superconducting single-electron transistor, Phys. Rev. Lett. 72, 2458 (1994).pubmed.ncbi.nlm.nih.gov/10055885
  35. [35]P. J. de Visser, J. J. A. Baselmans, P. Diener, S. J. C. Yates, A. Endo, and T. M. Klapwijk, Number Fluctuations of Sparse Quasiparticles in a Superconductor, Phys. Rev. Lett. 106, 167004 (2011).DOI
  36. [36]P. Kamenov, T. DiNapoli, M. Gershenson, and S. Chakram, Suppression of quasiparticle poisoning in transmon qubits by gap engineering, arXiv e-prints , arXiv:2309.02655 (2023).arxiv.org/abs/2309.02655
  37. [37]A. Basharin, Toward Large Nonradiating Qubits: When Quantum Engineering Meets Photonics, Advanced Quantum Technologies 9, e00587 (2026).DOI
  38. [38]De Dominicis, F., Roy, T., Mariani, A., Bal, M., Bonomo, C., Casali, N., Colantoni, I., Crisa, F., Cruciani, A., Ferroni, F., Helis, D. L., Pagnanini, L., Pettinacci, V., Pilipenko, R., Pirro, S., Puiu, A., Ressa, A., Romanenko, A., Vignati, M., … Cardani, L. (2026). Evaluating radiation impact on transmon qubits in above and underground facilities. EPJ Quantum Technology, 13, 47.
  39. [39]A. C. Tyner, J. T. Heath, T. C. Thann, V. P. Michal, P. Krogstrup, M. K. Svendsen, and A. V. Balatsky, Identification of Soft Modes in Amorphous Al₂O₃ via First-Principles, Advanced Quantum Technologies 8, e2500170 (2025).DOI
  40. [40]S. J. K. Lang, I. Eisele, A. Maiwald, E. Music, L. Schwarzenbach, C. Moran-Guizan, J. Weber, D. Zahn, T. Mayer, R. N. Pereira, and C. Kutter, Surface optimization of superconducting aluminum resonators for robust quantum device fabrication, Phys. Rev. Appl. 26, 014043 (2026).DOI
  41. [41]Purkayastha, A., Sharma, A., Patel, P. J., Chen, A.-H., Dempsey, C. P., Asodekar, S., Sinha, S., Tomasian, M., Pendharkar, M., Palmstrøm, C. J., Hocevar, M., Zuo, K., Hatridge, M., & Frolov, S. M. (2025). Transmon qubit using Sn as a junction superconductor. In arXiv e-prints.arXiv
  42. [42]D. López-Núñez, A. Torras-Coloma, Q. Portell-Montserrat, E. Bertoldo, L. Cozzolino, G. A. Ummarino, A. Zaccone, G. Rius, M. Martínez, and P. Forn-Díaz, Superconducting penetration depth of aluminum thin films, Superconductor Science and Technology 38, 095004 (2025).DOI
  43. [43]Zeng, L. J., Nik, S., Greibe, T., Krantz, P., Wilson, C. M., Delsing, P., & Olsson, E. (2015). Direct observation of the thickness distribution of ultra thin Al/AlOx/Al Josephson junctions. Journal of Physics D: Applied Physics, 48, 395308.
  44. [44]J. R. Rhee, Dependence of magnetic tunnel junction properties on tunnel barrier roughness, Journal of Magnetism and Magnetic Materials 304, e300 (2006), international Symposium on Spintronics and Advanced Magnetic Technologies and International Symposium on Magnetic Materials and Applications 2005.DOI
  45. [45]E. Hedrick, F. Bahrami, A. C. Pakpour-Tabrizi, A. Joshi, Q. Rumman Rahman, R. D. Yang, M. P. Chang, A. Bland, A. Jindal, G. Cheng, N. Yao, R. J. Cava, A. A. Houck, and N. P. de Leon, Quantifying surface losses in superconducting aluminum microwave resonators, arXiv e-prints (2026), arXiv:2603.13183 [quant-ph].arxiv.org/abs/2603.13183
  46. [46]Karuppannan, S. K., Kommannobaina, N. M., Utama, A. N., & Mukherjee, M. (2025). Impact of surface roughness on consistent resonator performance. Advanced Materials Interfaces, 12, 2500020.
  47. [47]Almansouri, M., Alhuwymel, T. F., AlYou sef, I., Shafi, A., Albogime, I., Getmanov, A., Hajr, A., & Shamim, A. (2026). Low loss superconducting resonators enabled by aluminum microstructural engineering and dielectric trimming. In arXiv e-prints.arXiv
  48. [48]S. A. Khan, S. Marti-Sanchez, D. Olsteins, C. Lampadaris, D. J. Carrad, Y. Liu, J. Quinones, M. Chiara Spadaro, T. Sand Jespersen, P. Krogstrup, et al., Epitaxially driven phase selectivity of Sn in hybrid quantum nanowires, ACS nano 17, 11794 (2023).DOI
  49. [49]D. C. Mattis and J. Bardeen, Theory of the Anomalous Skin Effect in Normal and Superconducting Metals, Physical Review 111, 412 (1958).DOI
  50. [50]V. Gupta, P. Winkel, N. Thakur, P. van Vlaanderen, Y. Wang, S. Ganjam, L. Frunzio, and R. J. Schoelkopf, Low-loss lumped-element inductors made from granular aluminum, Phys. Rev. Appl. 23, 054067 (2025).arxiv.org/abs/2411.12611
  51. [51]The physical origin of a bimodal sidewall angle distribution may for example be a two-step taper, faceting, a change in etch mechanism partway up the wall, a re-entrant or bowed profile, or micro-trenching near the foot.. (n.d.). The physical origin of a bimodal sidewall angle distribution may for example be a two-step taper, faceting, a change in etch mechanism partway up the wall, a re-entrant or bowed profile, or micro-trenching near the foot..
  52. [52]ATEC-NC, N. U. (2026). Tapping mode afm probe with real tip visibility.
  53. [53]Surf, D. (2026). Image Metrology.
  54. [54]J. S. Villarrubia, Algorithms for Scanned Probe Microscope Image Simulation, Surface Reconstruction, and Tip Estimation, J Res Natl Inst Stand Technol. , 425 (1997).pubmed.ncbi.nlm.nih.gov/27805154
  55. [55]F. Tian, X. Qian, and J. Villarrubia, Blind estimation of general tip shape in AFM imaging, Ultramicroscopy 109, 44 (2008).DOI
  56. [56]E. E. Flater, G. E. Zacharakis-Jutz, B. G. Dumba, I. A. White, and C. A. Clifford, Towards easy and reliable AFM tip shape determination using blind tip reconstruction, Ultramicroscopy 146, 130 (2014).DOI
  57. [57]Rao, A. (2009). Controlled Manipulation of Nanoparticles: Scanning Probe Measurements and Modelling of Trajectories and Dissipative Effects. In University of Basel.
  58. [58]P. Boonpuek and J. R. Felts, Determination of the radii of coated and uncoated silicon AFM sharp tips using a height calibration standard grating and a nonlinear regression function, Beilstein Journal of Nanotechnology 14, 1200 (2023).DOI
  59. [59]Kyhle, A., Sørensen, H., Zandbergen, J. B., & Bohr, J. (1998). Contrast artifacts in tapping tip atomic force microscopy. Applied Physics A: Materials Science & Processing, 66, S329.
  60. [60]B. J. Eves and R. G. Green, Limitations on accurate shape determination using amplitude modulation atomic force microscopy, Ultramicroscopy 115, 14 (2012).DOI
  61. [61]Temiryazev, A. G., Krayev, A. V., & Temiryazev, M. P. (2021). Two dynamic modes to streamline challenging atomic force microscopy measurements. Beilstein Journal of Nanotechnology, 12, 1226.
  62. [62]NioProbe, C. (2026). Aura NanoDevices Inc., NioProbe Product Sheet, Technical product sheet.
  63. [63]H.-C. Liu, G. A. Dahlen, and J. R. Osborne, Critical Dimension Atomic Force Microscopy for Sub-50-nm Microelectronics Technology Nodes, in Applied Scanning Probe Methods VIII: Scanning Probe Microscopy Techniques, NanoScience and Technology, edited by B. Bhushan (Springer, 2008) pp. 31–75.DOI

Paper details

vlnc:W_axejadfu
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

an AI tool

View statement
Looking for related papers…
Contents