{"id":"a81d5359-3401-48c2-9882-d0ea4ed05a78","arxiv_id":"2501.15059","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Electron microscopy of a transmon device reveals 7-20 nm oxide layers and carbon contamination at air-exposed interfaces, a cleaner metal-substrate interface, and suggests these imperfections drive dielectric loss.","lead":"This paper cuts open a superconducting quantum chip and uses electron microscopy to find oxide layers and carbon contamination on the surfaces that touch air. These imperfections are likely places where energy leaks out of qubits, so the work points to specific fabrication steps that could be improved.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Causal attribution gap: observed oxide/carbon layers are not quantitatively tied to TLS-limited loss, so 'significantly contribute to limiting device performance' is under-supported by the data presented.","rationale":"The microscopy is careful and self-aware; the authors repeatedly qualify the quantitative strength ('qualitative trends observed across a limited number of cross-sectional lamellae'). The observations are plausible and consistent with known native AlOx formation and hydrocarbon contamination. The clean metal-substrate interface and the thicker/dirtier oxide at sidewalls and base versus top also match process expectations, so the factual characterization is likely sound. The weakest point is not the microscopy itself but the inference to device-level TLS loss: the paper provides no quantitative estimate of how much loss the observed layers contribute and explicitly disclaims direct correlation. The abstract's 'significantly contribute' is therefore stronger than the demonstrated result. This is a correctable issue: either soften the central claim to 'may host TLS-relevant disorder' or provide TLS loss-budget or control data. Because the paper already presents hedged language in the body, the appropriate verdict remains CONDITIONAL rather than REJECT; the concern narrows the central claim but does not invalidate the characterization. The reader's weakest assumption (representativeness and artifact-freedom of two FIB-prepared lamellae) is real and closely related, but it is secondary: the most load-bearing gap is causal, not statistical.","tokens_in":12095,"tokens_out":4954,"duration_ms":51855,"concrete_test":"Perform TLS-sensitive microwave loss measurements on 8-16 co-fabricated resonators from the same wafer: measure internal quality factor Qi as a function of mean photon number n from about 0.01 to 100 at base temperature to extract the TLS loss tangent (low-power loss minus high-power saturated loss). Then strip the air-exposed AlOx/carbon layers on half the devices (e.g., in-situ argon ion-mill followed by Nb/Ru encapsulation) and re-measure. If the extracted TLS loss tangent does not decrease systematically when the observed metal-air/substrate-air layers are removed, the causal attribution in the abstract fails. A complementary analytical check is to estimate the TLS-limited T1 from the measured oxide/carbon thicknesses and literature loss tangents; if the predicted T1 greatly exceeds the measured 71 microseconds, the observed imperfections are not the dominant limiter.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central inference runs from STEM/EDS/EELS maps (Figs. 3-8) to the abstract's claim that TLS imperfections at critical interfaces significantly contribute to limiting device performance. The load-bearing premise is that the observed 7-20 nm oxide/carbon layers at the metal-air and substrate-air interfaces are a substantial source of the measured microwave loss. The paper's own microwave section states: 'a direct quantitative correlation between atomic-scale material imperfections and individual qubit performance is not established here.' The included microwave data (average T1 = 71 ± 5 µs, resonator Qi around 10^4) are consistent with many loss channels; no power-dependent TLS spectroscopy, no control samples, and no loss-budget estimate from the observed oxide/carbon layer volumes are presented. Without such a link, the observations show that the interfaces contain disorder that plausibly hosts TLS defects, but they do not establish that this disorder significantly limits this device's performance. The body's hedge words 'suggest' and 'may serve as sources' soften the conclusion, but the abstract and conclusion state a stronger causal claim. This is the most load-bearing concern because it is the step from 'dirty interfaces exist' to 'dirty interfaces are what limit coherence here.' Even granting artifact-free, representative lamellae, this causal gap remains.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports atomic-scale structural and chemical characterization of cross-sectional lamellae extracted from a Josephson junction (qubit side) and a spiral inductor (resonator side) of a double-sided superconducting circuit fabricated on intrinsic silicon. Using STEM with EDS and EELS, the authors identify oxide layers approximately 7-20 nm thick and carbon contamination at the metal-air and substrate-air interfaces, while the buried metal-substrate interface appears relatively clean. They attribute these features to the patterning, wet-etching, resist-removal, and ambient-exposure steps of their fabrication process. The paper also reports basic microwave characterization of 16 qubits and 16 resonators from the same wafer (mean T1 = 71 +/- 5 microseconds, mean Qi about 10^4) and argues that the observed interface disorder suggests that TLS imperfections at critical interfaces significantly contribute to limiting device performance.","tokens_in":12251,"tokens_out":5764,"duration_ms":53393,"significance":"The study's strength is its direct, atomic-scale evidence about where oxides and carbon reside in a specific Al/Si fabrication flow, with careful vacuum transfer and combined EDS/EELS analysis. If the observations are representative, they support the emerging picture that air-exposed interfaces are the dominant impurity sites and that the buried MS interface is comparatively clean, with concrete implications for process mitigation (e.g., post-etch cleaning, passivation, controlled ambient exposure). The authors are explicit that their results are qualitative and that no direct quantitative correlation with qubit performance is established, which is appropriate. However, because the abstract and conclusion assert a stronger causal role than the evidence supports, the paper's significance is as a materials case study rather than as a loss-budget or TLS-mechanism demonstration.","major_comments":[{"comment":"The abstract and conclusion claim that TLS imperfections at critical interfaces 'significantly contribute to limiting device performance,' but the microwave section explicitly states that 'a direct quantitative correlation between atomic-scale material imperfections and individual qubit performance is not established here.' The device-level data in Table 2 (mean T1 = 71 +/- 5 microseconds; mean internal quality factor Qi about 10.1 x 10^4) are compatible with many loss channels; no power-dependent TLS spectroscopy, no comparison with control samples, and no loss-budget estimate based on the observed 7-20 nm oxide/carbon layers are presented. This is a load-bearing gap because it is the step from 'dirty interfaces are present' to 'dirty interfaces limit this device's coherence.' Please soften the abstract and conclusion to 'potential sources' or 'plausible contributors,' or add a quantitative link (e.g., estimated TLS loss tangent from the observed layer volumes) that supports the stronger claim.","section":"Abstract; Microwave Characterization"},{"comment":"The authors correctly state that their findings 'represent qualitative trends observed across a limited number of cross-sectional lamellae' and that quantitative analysis 'would not be robust and reproducible.' Yet the central qualitative conclusion about the relative cleanliness of MA, SA, and MS interfaces is based on an unreported number of lamellae; from the described methods it appears that one lamella was prepared from each of the two components. The manuscript does not state how many lamellae were examined, how many regions were analyzed on each, or whether the observed contrasts were reproduced at other locations. Please add this sampling information or explicitly restrict the claims to the examined cross-sections; as written, the reader cannot assess the representativeness on which the general conclusion depends.","section":"Results and Discussion (first paragraph); Materials Characterization (FIB-SEM)"},{"comment":"The FIB workflow uses Ga+ ion milling and Pt deposition from an organometallic precursor. The authors note that carbon in the Pt capping layer originates from the GIS precursor, but no assessment is provided of whether FIB-induced damage, redeposition, or electron/ion-beam-induced chemistry could create or enhance the oxide and carbon signals at the MA and SA interfaces. Because the paper quotes oxide thicknesses to the nanometer level (e.g., 'approximately 10 nm thick' for Line-1 in Fig. 4 and 'approximately 20 nm thick' for Line-1 in Fig. 8), please include a control discussion or literature evidence that the observed layers are not preparation artifacts, or reduce the level of quantitative specificity in these claims.","section":"Materials Characterization (Sample Preparation using FIB-SEM); Results and Discussion (Figs. 4-8)"}],"minor_comments":[{"comment":"'For this reasons' should be 'For this reason.'","section":"Results and Discussion, first paragraph"},{"comment":"Please correct 'emphasizes on the need' and 'emphasize on the fact' to 'emphasizes the need' and 'emphasize the fact,' and 'receptively' to 'respectively.'","section":"Introduction; Conclusion; Microwave Characterization"},{"comment":"The text refers to 'Fig. 2(c)' and 'Fig. 2(d)' for the final thinned lamellae, but the caption lists only panels (a) and (b); please align the references and the figure panels.","section":"Fig. 2"},{"comment":"'Qubits relaxation and coherence times' should be 'Qubit relaxation and coherence times,' and 'with weighted-standard deviations' should be 'with weighted standard deviations.'","section":"Microwave Characterization"},{"comment":"The caption describes parameters but does not state which uncertainties are shown; clarify whether Qi has an associated uncertainty or note that only qubit-time uncertainties are listed.","section":"Table 2 and caption"}],"recommendation":"major_revision","confidential_remarks":"This is a solid materials characterization study whose main defect is the mismatch between the strong causal claim in the abstract/conclusion and the explicitly acknowledged absence of a quantitative correlation. The revision should focus on tempering the causal language or adding a loss-budget estimate, and on clarifying the sampling and artifact-control basis for the qualitative conclusions. No concerns about novelty or citation practice beyond minor formatting issues."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a useful paper, not a transformative one. It gives direct images of oxide layers and carbon residue at the air-exposed interfaces of a 3D cQED transmon, plus a clean Al/Si interface. The microscopy looks careful and the authors are honest about its limits. The main problem is a mismatch between the abstract's causal claim and what the data can support.\n\nThe new part is the specific device: double-sided coaxial cQED with a spiral inductor. Prior TEM work on Nb and Al interfaces exists (Altoé 2022), but these lamellae show the same story in a different architecture. The linescans giving oxide thicknesses of 7–20 nm and the carbon distribution are concrete additions. The methods section is detailed enough to reproduce the sample prep, and the acknowledgment that the carbon in the Pt capping layer comes from the GIS precursor is good self-awareness.\n\nThe soft spots are real but not crippling. The conclusions rest on two lamellae, and the authors themselves say the quantitative analysis is not robust. That is fine for a qualitative study, but then the abstract should not say 'significantly contribute to limiting device performance.' The microwave data are wafer-level averages, not linked to the observed defects; there is no TLS spectroscopy, no control samples, no loss budget. So the causal step is asserted, not demonstrated. That is an overreach, but it sits in the framing, not in the core observations.\n\nWho is this for? Anyone fabricating Al-based superconducting circuits, especially in 3D architectures. It is a useful benchmark for what to expect at MA and SA interfaces, and it strengthens the existing consensus that air-exposed interfaces host the lossy layers. It deserves peer review because the data are reproducible and the field needs more direct interface characterization. Just ask the authors to soften the abstract and conclusion to match the careful hedging in the body, and to either add error bars or explicitly keep the analysis qualitative.\n\nI would send it out with a request for revision, not desk-reject it.","headline":"A careful STEM/EDS/EELS look at qubit interfaces that overreaches in the abstract; the observations themselves are believable and worth having.","tokens_in":12885,"tokens_out":1459,"would_cite":true,"duration_ms":14823,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Atomic-scale maps show lossy oxide and carbon layers sit at air-exposed interfaces, not the buried metal-substrate boundary.","keywords":["superconducting qubits","two-level systems","TLS defects","interface oxidation","carbon contamination","scanning transmission electron microscopy","electron energy loss spectroscopy","dielectric loss"],"falsifier":"Compare these maps with a lamella prepared without ion milling or platinum capping—for example by cryo-fracture or broad-beam polishing—and check whether the 7–20 nm oxide and carbon layers at the air interfaces remain; if they disappear, they were preparation artifacts. Alternatively, on identical devices, strip exactly these surface layers with a short wet etch or in-situ plasma clean and measure resonator internal quality factors: if the quality factor does not move, those layers are not the limiting loss.","tokens_in":11853,"feed_emoji":"🔬","tokens_out":5842,"duration_ms":53189,"temperature":0.7,"pith_summary":"This paper uses atomic-resolution cross-sectional imaging to ask where dielectric loss in a superconducting transmon device physically lives. By cutting thin lamellae through a Josephson junction and a spiral resonator and mapping elements with scanning transmission electron microscopy plus X-ray and electron-energy-loss spectroscopy, the authors find 7–20 nm aluminum-oxide layers and carbon contamination at the metal-air and substrate-air interfaces, while the buried metal-substrate interface is relatively clean. They interpret these air-exposed imperfections as the microstructural sources of two-level-system loss that limit qubit coherence and resonator quality factors. If true, it redirects mitigation effort: protect or clean the exposed surfaces rather than assuming the metal-substrate junction is the problem. The paper is careful to call the findings qualitative trends, since they rest on a limited number of lamellae.","feed_headline":"Atomic maps trace qubit loss to surface oxide, not buried interface","feed_subtitle":"STEM maps show 7–20 nm oxide and carbon on exposed surfaces, pointing fabrication fixes away from the buried interface.","key_machinery":"The load-bearing tool is cross-sectional scanning transmission electron microscopy combined with two spectroscopies: energy-dispersive X-ray spectroscopy (EDS) and electron-energy-loss spectroscopy (EELS). FIB-milled lamellae through the junction and resonator are thinned to electron transparency, and one-nanometer-sliced linescans reconstruct elemental profiles across each interface, distinguishing oxygen, aluminum, silicon, carbon, and platinum. The O K-edge EELS analysis, with background removed by power-law fitting and cross sections computed from generalized oscillator strengths, lets the authors identify aluminum oxide layers and carbon contamination at specific interfaces rather than inferring them indirectly from microwave loss.","core_discovery":"The central discovery is that the critical interfaces are not equally bad. Oxygen and carbon maps from two lamellae, one through the Josephson junction and one through the spiral inductor, show pronounced oxide formation at the metal-air and substrate-air surfaces, with thicknesses around 7–20 nm and higher carbon concentration at the base and sidewalls of the etched aluminum film, while the metal-substrate boundary shows no relative oxygen signal and sharp elemental transitions. The tunnel barrier of the junction, formed by in-situ static oxidation, shows lower carbon contamination than the air-formed oxide layers. The authors connect these observations to the fabrication sequence: wet etching and photolithography preferentially contaminate sidewalls and base, and ambient exposure grows the surface oxide, so the resulting amorphous dielectric layers are plausible hosts for two-level-system defects that dissipate microwave energy. The stated conclusion is that two-level-system imperfections at these air-exposed interfaces significantly contribute to limiting device performance.","pith_inferences":["A testable prediction follows: if these air-interface layers are the loss source, then removing them with a gentle surface clean or capping them before ambient exposure should produce a measurable rise in resonator internal quality factor on identically fabricated devices.","The spatial pattern—thicker oxide at the base and sidewalls than at the top—suggests the wet-etch step, not just ambient oxidation, is a key variable; comparing the same process with dry etching or added plasma ashing would isolate its contribution.","Because the conclusions rest on two lamellae, a robust extension would be statistical sampling: prepare several lamellae from different chips and correlate per-lamella oxide and carbon thickness with qubit T1 measured on the same chip, something the current qualitative study explicitly does not attempt.","If the metal-substrate interface is genuinely clean, then published improvements that target only that boundary may be less impactful than surface passivation of air-exposed interfaces, an ordering that could be tested by comparing device performance under the two treatment families."],"forward_implications":["If the air-exposed interfaces are the dominant two-level-system host, then cleaning or passivating metal-air and substrate-air surfaces should raise qubit T1 and resonator internal quality factors more than improvements to the buried metal-substrate interface.","The 7–20 nm surface-oxide thicknesses set a scale: any protective cap or surface treatment must prevent regrowth of these layers, not just remove them once.","Higher carbon at the base and sidewalls of etched aluminum implicates wet-etch and lithography residues as a controllable contamination source, so altering those steps should reduce loss non-uniformity across a wafer.","A clean metal-substrate interface in both lamellae means substrate-air loss and metal-air loss, not metal-substrate loss, are where fabrication-process changes can be expected to matter.","The low carbon in the in-situ oxidized tunnel barrier supports keeping the junction oxidation step in vacuum or controlled gas, since that interface appears comparatively pure."],"supporting_citations":[{"why":"Establishes the prevailing attribution of two-level-system loss to structural and chemical defects at interfaces, the hypothesis the paper tests directly.","marker":"[15]"},{"why":"Provides the dielectric-loss-from-two-level-systems mechanism in Josephson qubits that motivates looking for material imperfections at critical interfaces.","marker":"[14]"},{"why":"Supplies evidence that two-level-system defects are distributed at surfaces and interfaces in superconducting resonators, which the paper's air-interface findings extend.","marker":"[22]"},{"why":"Develops the method for determining interface dielectric losses separately, which the paper's metal-air, metal-substrate, substrate-air distinction follows.","marker":"[17]"},{"why":"Demonstrates localization and mitigation of loss in superconducting circuits, used by the paper to argue that surface oxide and contamination are the controllable loss source.","marker":"[28]"},{"why":"Shows substrate surface engineering for silicon/aluminum resonators, the benchmark for the paper's finding that the metal-substrate interface is clean.","marker":"[19]"}],"fun_headline_variants":["Qubit loss traced to surface oxide, not buried interface","Nanoscale maps reveal oxide at qubit surfaces as main decoherence source","Surface oxide, not buried interface, blamed for superconducting qubit loss","STEM links qubit decoherence to air-exposed oxide layers","Fabrication fix: avoid air-formed oxide on qubit surfaces"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conclusions assume that the two FIB-prepared lamellae fairly represent the device interfaces and that the milling and capping steps did not create or erase the oxide and carbon features; the paper itself notes the findings are qualitative trends from a limited number of lamellae.","fun_headline_variants_meta":{"raw":{"variants":["Qubit loss traced to surface oxide, not buried interface","Nanoscale maps reveal oxide at qubit surfaces as main decoherence source","Surface oxide, not buried interface, blamed for superconducting qubit loss","STEM links qubit decoherence to air-exposed oxide layers","Fabrication fix: avoid air-formed oxide on qubit surfaces"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000197,"raw_usage":{"total_tokens":1321,"prompt_tokens":860,"completion_tokens":461,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":476,"completion_tokens_details":{"reasoning_tokens":369}},"tokens_in":476,"tokens_out":461,"duration_ms":4426,"temperature":1.0,"reasoning_tokens":369,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T14:39:06.889199+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare these maps with a lamella prepared without ion milling or platinum capping—for example by cryo-fracture or broad-beam polishing—and check whether the 7–20 nm oxide and carbon layers at the air interfaces remain; if they disappear, they were preparation artifacts. Alternatively, on identical devices, strip exactly these surface layers with a short wet etch or in-situ plasma clean and measure resonator internal quality factors: if the quality factor does not move, those layers are not the limiting loss.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the prevailing attribution of two-level-system loss to structural and chemical defects at interfaces, the hypothesis the paper tests directly."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the dielectric-loss-from-two-level-systems mechanism in Josephson qubits that motivates looking for material imperfections at critical interfaces."},{"cited_title":"The physics of superconducting microwave resonators (California Institute of Technology, 2008)","cited_arxiv_id":null,"evidence_quote":"Supplies evidence that two-level-system defects are distributed at surfaces and interfaces in superconducting resonators, which the paper's air-interface findings extend."},{"cited_title":"Determining interface dielectric losses in superconducting coplanar-waveguide resonators","cited_arxiv_id":null,"evidence_quote":"Develops the method for determining interface dielectric losses separately, which the paper's metal-air, metal-substrate, substrate-air distinction follows."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Demonstrates localization and mitigation of loss in superconducting circuits, used by the paper to argue that surface oxide and contamination are the controllable loss source."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows substrate surface engineering for silicon/aluminum resonators, the benchmark for the paper's finding that the metal-substrate interface is clean."}],"review_version":1}