Hidden Heterogeneity in Cross-Platform Josephson-JunctionBenchmarking:A Case Study in Statistical Confounding withArchitecture-Dependent Decoherence Physics
Abstract Cross-platform comparison of superconducting quantum devices requires statistical care because heterogeneous architectures implement qualitatively different transport mechanisms, couple to distinct microscopic loss channels, and typically operate in different temperature regimes. We present a methodological case study showing how aggregation of data across platforms can mask qualitatively different within-architecture temperature dependencies when platform identity is confounded with operating temperature. To enable controlled comparison, we introduce two scaling parameters: a dimensionless normalized temperature \(T^{*} = k_{B}T/(hf)\) and a resistance-like decoherence parameter \(A = (hf/eI_{c})(\ln Q/2\pi)\). We develop a detailed physical argument for why the baseline value of \(A\) is expected to be architecture-dependent, focusing on the distinct roles of Cooper-pair tunneling in SIS junctions, Andreev bound states in SNS junctions, ferromagnetic layer effects in SIsFS devices, and confinement effects in nanowire junctions. Applying this framework to published data from five distinct junction architectures (Transmon SIS, conventional SIS, SIsFS, planar SNS, and nanowire SNS; \(N = 23\) points from five independent studies), we find that the pooled dataset exhibits a weak positive correlation between \(A\) and \(T^{*}\) (\(r = 0.085\), \(p = 0.70\)). However, when stratified by architecture, within-group correlations reveal qualitatively different temperature dependencies: SIsFS shows strong positive dependence (\(308.4 \pm 22.6\,\Omega\), \(p = 0.047\)), planar SNS shows strong positive dependence (\(205.6 \pm 11.6\,\Omega\), \(p = 0.036\)), while transmons show near-zero dependence (\(-7.2 \pm 17.3\,\Omega\), \(p = 0.688\)). Two architectures show positive slopes that reach statistical significance despite small samples (\(n=3\) each), though these results are sensitive to individual data points (see Appendix C). Analysis of variance confirms that architecture explains the dominant share of variance (ANOVA \(F_{4,18} = 670.7\), \(p < 0.0001\); ICC \(\approx 0.98\)). Decomposition of \(A\) reveals that the \(1/I_c\) term accounts for the factor-of-6.6 variation across architectures (\(r = -0.94\) with \(\log I_c\)), while the \(\ln Q\) term varies by only 30\%. We argue that these observations are physically expected from architecture-specific transport and loss mechanisms, and we outline experimental designs with overlapping temperature sweeps that could disentangle temperature dependence from platform identity in future benchmarking efforts. All data, extraction uncertainties, and analysis code are provided in the appendices to ensure reproducibility.
Abstract Cross-platform comparison of superconducting quantum devices requires statistical care because heterogeneous architectures implement qualitatively different transport mechanisms, couple to distinct microscopic loss channels, and typically operate in different temperature regimes. We present a methodological case study showing how aggregation of data across platforms can mask qualitatively different within-architecture temperature dependencies when platform identity is confounded with operating temperature. To enable controlled comparison, we introduce two scaling parameters: a dimensionless normalized temperature (T^{} = k_{B}T/(hf)) and a resistance-like decoherence parameter (A = (hf/eI_{c})(\ln Q/2\pi)). We develop a detailed physical argument for why the baseline value of (A) is expected to be architecture-dependent, focusing on the distinct roles of Cooper-pair tunneling in SIS junctions, Andreev bound states in SNS junctions, ferromagnetic layer effects in SIsFS devices, and confinement effects in nanowire junctions. Applying this framework to published data from five distinct junction architectures (Transmon SIS, conventional SIS, SIsFS, planar SNS, and nanowire SNS; (N = 23) points from five independent studies), we find that the pooled dataset exhibits a weak positive correlation between (A) and (T^{}) ((r = 0.085), (p = 0.70)). However, when stratified by architecture, within-group correlations reveal qualitatively different temperature dependencies: SIsFS shows strong positive dependence ((308.4 \pm 22.6,\Omega), (p = 0.047)), planar SNS shows strong positive dependence ((205.6 \pm 11.6,\Omega), (p = 0.036)), while transmons show near-zero dependence ((-7.2 \pm 17.3,\Omega), (p = 0.688)). Two architectures show positive slopes that reach statistical significance despite small samples ((n=3) each), though these results are sensitive to individual data points (see Appendix C). Analysis of variance confirms that architecture explains the dominant share of variance (ANOVA (F_{4,18} = 670.7), (p < 0.0001); ICC (\approx 0.98)). Decomposition of (A) reveals that the (1/I_c) term accounts for the factor-of-6.6 variation across architectures ((r = -0.94) with (\log I_c)), while the (\ln Q) term varies by only 30%. We argue that these observations are physically expected from architecture-specific transport and loss mechanisms, and we outline experimental designs with overlapping temperature sweeps that could disentangle temperature dependence from platform identity in future benchmarking efforts. All data, extraction uncertainties, and analysis code are provided in the appendices to ensure reproducibility.
Abstract Cross-platform comparison of quantum devices requires careful statistical methodologyto avoid spurious conclusions. We present a methodological case study demonstrating howSimpson’s paradox can arise in meta-analyses of superconducting circuit data. We introducetwo physically motivated scaling parameters—a dimensionless normalized temperature T∗ =kBT/hf and a resistance-like decoherence parameter A = (hf/eIc)(ln Q/2π)—and applythem to a secondary analysis of published data from Transmon, SNS, and SINIS devices(N = 10 data points from 3 studies).Across the pooled dataset, we observe a strong correlation (r = 0.910) between A and T∗.However, we demonstrate that this correlation is a statistical artifact arising from completeconfounding between temperature and device type: the three platforms operate in nonoverlappingtemperature regimes, and within-group correlations are consistently negative(though not statistically significant given the small sample sizes of n = 3–4 per group).Mixed-effects modeling indicates that platform identity accounts for the dominant share ofvariance in A (ICC ≈ 0.98), though we caution that variance component estimates from onlyk = 3 groups carry substantial uncertainty.Our analysis serves as a methodological warning for cross-platform data aggregation inquantum physics, illustrating how pooled correlations can reverse direction when data aredisaggregated—a hallmark of Simpson’s paradox. The proposed scaling framework may proveuseful for future benchmarking, but we emphasize that the current data are insufficient totest any hypothesis of universal thermodynamic scaling. We outline experimental conditionsnecessary for such tests, including overlapping temperature ranges and controlled singledevicesweeps.This work is intended as a methodological case study rather than a test of aphysical scaling law
We present a systematic study of the temperature-dependent transport properties of Nb- and Al-based superconductor--insulator--thin superconductor--ferromagnet--superconductor (SIsFS) Josephson junctions. By varying the superconducting material and the thickness of the intermediate superconducting layer, we identify three distinct transport regimes: a tunnel-dominated SIS-like regime, a temperature-driven crossover, and a complex weak-link regime. A unified Usadel-based description connects these regimes through the self-consistent pair potential of the intermediate s-layer. While for Al-based devices an SIS-like transport regime is observed, a self-consistent calculation of the s/F/S proximity electrode, combined with perturbative tunnel coupling to the outer electrode, accounts for Nb devices with different thickness of the superconducting interlayer. The calculation places the 10-nm-interlayer sample in a proximity-supported, strongly suppressed s state, close to the SInFS limit in which the interlayer is effectively driven normal (\textit{n}). A reduced nonlinear long-junction description captures the same limiting physics. The evolution of I_cR_N(T) thereby provides a direct experimental signature of the three transport regimes. Beyond the fundamental interest, the results establish practical design guidelines: Nb junctions with sufficiently thick s interlayers are well suited for high-speed cryogenic memory elements and tunable superconducting parametric amplifiers, while Al junctions are natural candidates for coherent superconducting quantum circuits
In superconducting quantum circuits, decoherence improvements are frequently obtained through process interventions that simultaneously modify surface chemistry, microstructural topology, and device geometry, leaving mechanistic attribution structurally underdetermined. Predictive materials engineering requires measurable structural statistics to be separated from geometry-dependent coupling coefficients into independently testable factors. We introduce the concept of classical and quantum microstructure. In that context, we formulate a channel-wise separable framework for decoherence in superconducting transmon qubits in which each loss channel is described by a reduced prescriptor. Here, a channel-specific microstructural state variable is determined independently of device geometry, and a geometry-dependent coupling functional is computable from field solutions without reference to surface chemistry. We derive this product form from a spatially resolved kernel representation and establish a perturbative separability criterion that defines the regime where independent variation of the variables is valid. The framework specifies five prescriptor classes for dominant loss pathways in transmon-class devices. Falsifiability is operationalized through a pre-committed 2x2 experimental protocol in which the variables must satisfy independent ratio checks within propagated uncertainty. A Minimum-Dataset Specification standardizes reporting for cross-laboratory inference. Part I establishes the conceptual and mathematical architecture; coordinated experimental validation is reserved for Part II.
The development of quantum computers, capable of surpassing classical computing in solving specific complex problems, hinges on breakthroughs in materials science, precision device fabrication, and the optimization of quantum architectures [1]. Among the leading candidates for quantum computation, superconducting qubits, particularly those based on transmon designs, have shown promising advances with coherence times reaching hundreds of microseconds [2]. However, to fully unlock the potential of scalable quantum computing, further enhancements in coherence are necessary. The intricate structure-coherence relationship within these devices, especially at critical interfaces such as metal-air, metal-dielectric, and metal-substrate in Josephson junctions (JJs), poses a significant challenge due to the potential for two-level-system loss channels. Here, we detail our pioneering efforts to understand these structure-coherence relationships utilizing advanced electron microscopy techniques. Focusing on the Al-AlOx and Al-Si substrate interfaces in Josephson Junctions, we employ a combination of low-loss and core-loss electron energy loss spectroscopy (EELS) and four-dimensional scanning transmission electron microscopy (4D-STEM) conducted at room and ultra-low temperatures (sub-6 K). This approach allows for an unprecedented examination of the chemical composition, bonding environments, and nanoscale structures of key surfaces and interfaces critical to the quantum coherence. Figure 1 presents a typical superconducting architecture at the chip level; each superconducting qubit consists of a Josephson Junction in parallel with an ultra-low-loss capacitor to create a resonant structure in the 3-6GHz range. JJs are highlighted in the area boxed in Fig. 1A and can be observed in detail in Fig. 1B. Lamella cross-sections (as shown in Fig. 1C) were prepared using a dual-beam focused ion beam (FIB) FEI-SEM Helios Nanolab. The lift-off was performed using 30 kV Ga+ ions, and the final cleaning step was carried out at 2 kV to remove surface damage on the area of interest, resulting in a final sample thickness of 50-100 nm. Transmission electron microscopy (TEM) and scanning transmission electron microscopy (STEM) data were collected on a probe-corrected JEOL ARM200 S/TEM operating at 200 kV. The convergent angle for ADF-STEM imaging was 25 mrad. Electron energy loss spectra (EELS) were acquired at 200 kV using a Gatan GIF Quantum on a K2 pixelated detector. Data processing (background subtraction, plural scattering removal, and signal mapping) was conducted using Gatan GMS software. Four-dimensional STEM (4D-STEM) data were collected by a K3 and STELA cameras. The convergent angles were 0.5-1.5 mrad, and the camera length was 6-10 cm. The step size was 5-10 Å. Merging microscopic insights with electrical data and theoretical frameworks, we aim to reveal how atomic-scale irregularities affect coherence times. Our results highlight key decoherence mechanisms in superconducting qubits and suggest specific improvements for qubit design and manufacturing. By linking electron microscopy findings to device performance, we outline strategies to reduce decoherence, essential for achieving reliable quantum computing. This work bridges materials science and quantum engineering, driving efforts in making scalable quantum computers forward [3]. A) Microscopic view of a superconducting transmon qubit, highlighting its symmetric design with central island and capacitor pads. B) SEM image of a JJ displaying their Al top and bottom contacts. C) STEM image of a JJ showing its key components and interfaces: Al top and bottom contact, AlOx tunnel barrier, Si substrate and Al-Si interface. A) EELS chemical maps of two interfaces, Al/Al and Al/Si using K-edges of oxygen, aluminum and silicon of a Josephson junction. It is possible to detect a formation of oxides at both interfaces. Plots showing Al and Si K-edge at 1560 eV and 1839 eV, respectively, and the ELNES are shown on the right side (Scale bar in A – 20 nm). B) Analysis of the disordered Al phases at the interfaces utilizing ePDF.
Abstract Several papers on physics-informed noise modeling for superconducting qubits report gains from embedding known physics into machine-learning models, but we found little discussion in that literature of the specific mechanism by which such gains arise, or of the conditions under which embedding a known relation could instead hurt. We do not claim this gap is unaddressed anywhere in the broader machine-learning literature — bias-variance behavior of physics-informed models and the importance of multi-seed evaluation are both studied elsewhere in adjacent settings, discussed in Section 2 — only that we did not find this specific mechanism characterized for noisy-input extrapolation in the quantum-noise-modeling papers we reviewed. We derive a leading-order (small-σ) bias-variance decomposition for anchoring a regression model to an exact analytic noise relation (here, the T1-relaxation-limited dephasing rate 1/(2T1)): to leading order in the relative input-noise magnitude σ, the exact plug-in estimator has a small positive relative bias of σ² and a variance scaling with the square of the anchored quantity's absolute magnitude. We validate both predictions in a controlled single-variable Monte Carlo experiment: the simulated relative bias (0.65%) matched the leading-order prediction (0.64%) closely, and the simulated variance ratio between two test points matched its prediction to within about 1%. We then test whether this single-variable picture transfers to a more realistic two-feature extrapolation task — predicting the excess, non-relaxation-limited dephasing rate of a held-out device from noisy T1/T2 telemetry — using calibration parameters grounded in five real, independently re-verified IBM Heron/Eagle/Osprey processor snapshots, including a cross-generation transfer test (training on three Heron-class devices, T1 ≈ 137–189µs, evaluating on both a lower-coherence Osprey-class device, T1 ≈ 88µs, and a higher-coherence Eagle-class device, T1 ≈ 263µs). Run once, the isolated single-variable picture does not predict the sign of the effect in this more realistic setting: physics-anchoring loses to a plain ridge baseline on the low-T1 device. Repeating the experiment across 20 independent random seeds, with a one-sample t-test against zero, shows why: the low-T1 effect is statistically indistinguishable from zero (mean − 2.5%, 95% CI [− 10.5%, + 5.4%], t(19) = − 0.67, p = 0.51, 12 of 20 seeds favoring physics-informed), while the high-T1 benefit is statistically robust (mean + 19.7%, 95% CI [+ 16.7%, + 22.7%], t(19) = 13.7, p < 0.0001, 20 of 20 seeds favoring physics-informed). A follow-up collinearity-manipulation experiment, itself repeated across 15 seeds per condition, initially appeared non-significant when analyzed by pooling all seed-condition pairs into one regression; a corrected per-seed-slope analysis, accounting for the repeated-measures structure induced by reusing seeds across jitter levels, instead finds a small but statistically significant positive trend at both held-out devices (p = 0.021 and p = 0.011). All experiments are grounded in real, independently re-verified device numbers but built from synthetic telemetry — documented in full, equation by equation, in Section 6 — in a network-isolated environment without access to live quantum hardware; results should be read as a methodological, cautionary study on synthetic data, not a validated hardware finding, and we make no claim that any of the specific numeric effect sizes would reproduce on real devices. We report this honestly, including a citation error we found and fixed in our own earlier draft and a headline finding (the low-T1 reversal) that did not survive statistical testing, because both are more useful to a careful reader than a clean story would have been. Complete reproduction code is included with this submission.
We have measured the current-voltage $(I\ensuremath{-}V)$ characteristics of small-capacitance single Josephson junctions at low temperatures (T=0.02−−0.6K), where the strength of the coupling between the single junction and the electromagnetic environment was controlled with one-dimensional arrays of dc superconducting quantum interence devices (SQUIDs). The single-junction $I\ensuremath{-}V$ curve is sensitive to the impedance of the environment, which can be tuned in situ. We have observed Coulomb blockade of Cooper-pair tunneling and even a region of negative differential resistance, where the zero-bias resistance ${R}_{0}^{\ensuremath{'}}$ of the SQUID arrays is much higher than the quantum resistance ${R}_{K}\ensuremath{\equiv}{h/e}^{2}\ensuremath{\approx}26\mathrm{k}\ensuremath{\Omega}.$ The negative differential resistance is evidence of the coherent single-Cooper-pair tunneling within the theory of current-biased single Josephson junctions. Based on this theory, we have calculated the $I\ensuremath{-}V$ curves numerically in order to compare then with the experimental ones at ${R}_{0}^{\ensuremath{'}}\ensuremath{\gg}{R}_{K}.$ The numerical calculation agrees with the experiments qualitatively. We also discuss the ${R}_{0}^{\ensuremath{'}}$ dependence of the single-Josephson-junction $I\ensuremath{-}V$ curve in terms of the superconductor-insulator transition driven by changing the coupling to the environment.