{"id":"278041ac-7ccf-41f5-a1d7-9fcf89ce63de","arxiv_id":"1908.02606","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"HSQ resist left on NbN coplanar waveguide resonators dominates their low-temperature microwave loss, with an inferred loss tangent of 8.0 x 10^-3.","lead":"The paper shows that leftover HSQ, a resist used to pattern very small superconducting wires, is the largest source of microwave energy loss in the tested resonators, with an inferred material loss of about 8 parts per thousand. This matters because superconducting qubits, amplifiers, and photon detectors are limited by such loss, so the result points to a concrete fabrication change: remove the HSQ.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported δ_i_HSQ is inflated if any Nb2O5 grows under HSQ; the claim that HSQ dominates every geometry rests on this unverified assumption and should be tested by oxide characterization.","rationale":"I read the paper as making two claims: a qualitative one, that HSQ increases TLS loss, supported directly by the paired with/without-HSQ measurements in Table I and Fig. 6; and a quantitative one, that HSQ has δ_i = 8.0e-3 and is the dominant loss source in every geometry. The qualitative claim is credible and is not the target of my concern. The quantitative claim, however, is not a direct measurement. It depends on a four-parameter fit of Eq. (1) to eight values, and one of the parameters is strongly coupled to the model's treatment of Nb2O5 under HSQ. The authors explicitly assume no Nb2O5 grows under HSQ because it takes days to form, but the elapsed time between RIE and HSQ spin is not reported, and even a nanometre-scale oxide would change the fitted δ_i_HSQ because δ_i_Nb2O5 is large. At the largest gap, the with-HSQ loss exceeds the no-HSQ loss by only 0.30e-5, so the dominance of HSQ in that geometry is delicate. This is not an internal contradiction or a circular argument; the frequency-shift measurements are independent of the simulation. It is a model-dependence problem in the extracted loss tangent. The reader's verdict of CONDITIONAL is appropriate, and the condition should specifically require oxide characterization and a sensitivity analysis of the fitting parameters. My concern does not move the verdict; it sharpens the condition under which the paper would be accepted.","tokens_in":11519,"tokens_out":8374,"duration_ms":97316,"concrete_test":"Prepare witness samples through the exact process flow up to HSQ spin, stop before HSQ exposure, and measure the NbN surface oxide thickness by XPS depth profiling or cross-sectional TEM/EDX. Then feed the measured thickness into the model and refit Eq. (1), including an under-HSQ Nb2O5 layer of that thickness. If the refit gives δ_i_HSQ below approximately 4e-3, or if the HSQ fractional contribution at g_cpw = 5 µm falls below 50%, the headline quantitative claim is not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The quantitative claim is δ_i_HSQ = 8.0e-3 and its corollary that HSQ dominates loss in every measured geometry (Section VI, Table II, Fig. 7). This value is not directly measured; it is the output of a four-parameter fit of Eq. (1) to the F_TLS δ_i_TLS values in Table I. The fit is stable only relative to a specific loss-bookkeeping assumption stated in the Section IV material-parameter paragraph: \"Because Nb2O5 requires several days to achieve any meaningful thickness, it is assumed that no Nb2O5 is present underneath the HSQ.\" In the no-HSQ samples, the model places 5 nm of Nb2O5 (δ_i = 4.7e-2) on the center conductor; in the HSQ samples it removes that term and replaces it with HSQ. Because the measured with-HSQ loss is only modestly larger than the no-HSQ loss (e.g., 1.66 versus 1.36 × 10^-5 at g_cpw = 5 µm), the fitted δ_i_HSQ must be large enough to overcome the modeled removal of the Nb2O5 term. If native oxide does grow under HSQ, the fitted δ_i_HSQ would generally be lower, and the HSQ contribution in the wider geometries could drop below the other loss terms. The fabrication sequence leaves the NbN exposed between RIE and HSQ spin, with no time specified, so the assumption is not process-locked. No uncertainty or sensitivity analysis is provided to show how much δ_i_HSQ or the dominance claim changes under plausible under-HSQ oxide thicknesses.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports an experimental and numerical study of two-level-system (TLS) dielectric loss in NbN coplanar waveguide resonators with and without a hydrogen silsesquioxane (HSQ) capping layer on the center conductor. The authors measure the temperature-dependent frequency shift of four resonator geometries (gap widths 0.5–5 µm) at single-photon power and extract the product F_TLS·δ_TLS for each device. They then compute geometric filling factors using 3D finite-element simulations that solve the Maxwell–London equations for the disordered NbN film (penetration depth ≈1 µm), and fit the loss-budget model of Eq. (1) to obtain the intrinsic loss tangents of HSQ, the substrate–metal interface, niobium oxide, and silicon oxide. The central claim, quantified in Table II and stated in Section VI, is that HSQ has δ_i_HSQ = 8.0 × 10⁻³ and, because of its large filling factor, is the dominant TLS loss source in every measured geometry.","tokens_in":11872,"tokens_out":4498,"duration_ms":44808,"significance":"The qualitative result that HSQ increases microwave loss is convincingly established by the paired with/without-HSQ comparison: in each of the four geometries the measured F_TLS·δ_TLS is higher when HSQ is present (Table I), and the frequency-shift method isolates TLS loss from other mechanisms. If the quantitative value for δ_i_HSQ survives further scrutiny, it provides a useful design rule for high-impedance NbN nanowire circuits, where HSQ is commonly left in place. The methodological contribution—using 3D Maxwell–London simulations to account for the large magnetic penetration depth in disordered superconductors when computing participation ratios—is significant and clearly presented. The manuscript also gives credit to prior surface-loss literature and includes sufficient experimental detail for the measurement setup to be reproduced.","major_comments":[{"comment":"The quantitative extraction of δ_i_HSQ depends critically on the stated assumption: 'Because Nb2O5 requires several days to achieve any meaningful thickness, it is assumed that no Nb2O5 is present underneath the HSQ.' This assumption is load-bearing: the model removes the 5-nm Nb2O5 layer from the center conductor in the HSQ samples and replaces it with HSQ, while the measured with-HSQ loss is only modestly larger than the no-HSQ loss (e.g., 1.66 vs 1.36 × 10⁻⁵ at g_cpw = 5 µm). If a native oxide grows under the HSQ, the fitted δ_i_HSQ would be lower, and the claim of HSQ dominance in the wider geometries could fail. The fabrication sequence leaves the NbN exposed between RIE and HSQ spin with no time specified, and no direct oxide characterization or sensitivity analysis is provided. This issue directly affects the central quantitative claim and should be addressed with, e.g., XPS/TEM cross-section or a sensitivity study varying the under-HSQ oxide thickness.","section":"Section IV, material-parameter paragraph; Table II"},{"comment":"The four loss tangents in Table II are obtained by fitting Eq. (1) to the eight F_TLS·δ_TLS values in Table I, yet no uncertainties, covariance, or goodness-of-fit statistics are reported. With four free parameters and eight data points, the fit may not uniquely constrain the individual loss tangents; for instance, δ_i_SM and δ_i_SiO2 both affect the geometry dependence in similar ways. The two-significant-figure value δ_i_HSQ = 8.0 × 10⁻³ therefore conveys a precision that is not supported by the presented analysis. Please report parameter uncertainties (e.g., from a bootstrap or least-squares covariance) and a residual analysis, or explicitly discuss the confidence region for δ_i_HSQ.","section":"Table II and Fig. 6"},{"comment":"The conclusion that HSQ is the dominant TLS loss source 'for all dimensions' goes beyond the direct measurements, which show only that HSQ increases loss; the dominance statement is an inference from the fitted parameters. Because the fitted parameters are sensitive to the assumed layer thicknesses, permittivities, and the no-Nb2O5-under-HSQ assumption (as noted above), the dominance claim for the widest geometry (g_cpw = 5 µm) is not yet robust. A sensitivity analysis showing how δ_i_HSQ and the per-region loss contributions in Fig. 7 change under plausible variations of the model parameters would be required to support this strong claim.","section":"Section VI and Fig. 7"}],"minor_comments":[{"comment":"The abstract contains an incomplete sentence: 'We emphasise that the loss caused by a spin-on-glass resist such as hydrogen silsesquioxane (HSQ), used for ultrahigh lithographic resolution relevant to the fabrication of nanowires, and find that, when used, HSQ is the dominant source of loss...' The phrase beginning 'the loss caused by...' lacks a main verb. Please rewrite.","section":"Abstract"},{"comment":"The fixed ratio of gap to center conductor width is not stated; please give the ratio (or the center-conductor widths) to make the geometry table self-contained.","section":"Section II"},{"comment":"In Section V, 'These results confirms' should be 'These results confirm'.","section":"Section V"},{"comment":"The caption states that the curves are offset by 15 kHz, but does not explain which curves are offset; please clarify the offset convention.","section":"Fig. 2 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is well written and the experimental comparison is clean, but the quantitative claim needs a sensitivity analysis before acceptance. The central issue is the unverified under-HSQ oxide assumption; this is a single assumption that the authors could address with a modest additional experiment or a focused sensitivity study. I believe the paper is within the scope of the journal and the contribution is worth publishing after major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, what's new: a clean paired comparison of otherwise identical NbN CPW resonators with and without an HSQ cap, plus a 3D finite-element Maxwell-London treatment of the filling factors for a disordered superconductor whose penetration depth is comparable to the device dimensions. That is a real advance over the usual electrostatic participation-ratio simulations, and the quantitative output—δ_i_HSQ = 8 × 10^-3 and the fitted loss tangents for the substrate-metal, Nb2O5, and SiO2 layers—is the sort of number device engineers will want. The qualitative result, that HSQ is a serious loss source and its removal would improve high-impedance nanowire resonators, is well supported by the data: the with-HSQ loss is higher in every geometry.\n\nThe soft spots are all around the bookkeeping assumptions. Four loss tangents are fitted to eight data points, Table II has no uncertainties, and the fit leans on specified layer thicknesses, permittivities, and the dirty-limit NbN model. The stress-test note is on target: the explicit assumption that no Nb2O5 grows under the HSQ is the load-bearing one. The fabrication sequence leaves the NbN exposed after RIE with the time before HSQ spin unspecified; if a few nm of niobium oxide form there, the fitted δ_i_HSQ is overestimated and the claim that HSQ dominates in the wider geometries may not hold. The paper would be materially stronger with a sensitivity sweep over under-HSQ oxide thickness and with uncertainties on the fitted loss tangents. None of this undermines the central qualitative conclusion, but it does mean the headline number should be read as process- and model-dependent.\n\nThe citation pattern is appropriate, and the authors are transparent about the assumption, which I credit. This is a serious paper that deserves a proper referee round. I would accept it for review, and I'd encourage the editor to ask for the uncertainty analysis before publication.","headline":"A genuinely useful paired measurement and a new Maxwell-London filling-factor method, but the headline HSQ loss tangent is model-dependent and needs a sensitivity analysis before I would trust the exact number.","tokens_in":12407,"tokens_out":2811,"would_cite":true,"duration_ms":27579,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that the lithography resist HSQ, when left on finished NbN nanowire resonators, is the dominant source of two-level-system microwave loss.","keywords":["superconducting resonators","two-level systems","dielectric loss","hydrogen silsesquioxane (HSQ)","NbN nanowires","kinetic inductance","filling factor","Maxwell-London simulation"],"falsifier":"Take finished resonators that still carry HSQ, remove the HSQ with a dedicated strip process, and remeasure the single-photon TLS loss; if the loss does not drop by the amount predicted from $F_{\\mathrm{HSQ}} \\times 8.0 \\times 10^{-3}$, the dominance claim is wrong. Alternatively, measure the HSQ loss tangent directly in a resonator whose filling factor is fixed by geometry alone, such as a lumped-element capacitor with the HSQ as the known dielectric.","tokens_in":11300,"feed_emoji":"⚡","tokens_out":6739,"duration_ms":67348,"temperature":0.7,"pith_summary":"This paper aims to identify and quantify where two-level-system (TLS) dielectric loss comes from in high-impedance superconducting nanowire resonators, and it points to the spin-on-glass resist hydrogen silsesquioxane (HSQ). The authors built nominally identical NbN coplanar waveguide resonators with and without a 30 nm HSQ layer on the center conductor, with gaps from 5 µm down to 500 nm, and found that the HSQ-covered resonators always lose more. Combining the measured loss with 3D finite-element simulations of the electric fields, they extracted an intrinsic loss tangent for HSQ of $8.0 \\times 10^{-3}$, about four times the value typically used for silicon oxide, and showed that this one layer dominates the total loss in every geometry studied. If the paper is right, removing or avoiding HSQ after patterning would substantially reduce microwave loss in nanowire-based qubits, amplifiers, and detectors.","feed_headline":"HSQ resist loss is 8x10^-3 and dominates NbN resonator loss","feed_subtitle":"A common nanowire lithography resist, left on the device, dominates two-level-system loss in high-impedance superconducting circuits.","key_machinery":"The argument is carried by geometric filling factors, $F_k = U_k / U_{\\mathrm{total}}$, the fraction of stored electric energy in each lossy region, computed from 3D finite-element solutions of the Maxwell–London equations. Because the disordered NbN film has a magnetic penetration depth of roughly $1\\,\\mu\\mathrm{m}$, comparable to the device dimensions, the superconductor is modelled by a complex permittivity in the dirty local limit, and electrostatic simulations are shown to give the wrong filling factors. These $F_k$ feed the linear decomposition $1/Q_{\\mathrm{TLS}} = \\sum_k F_k \\delta^i_k$, which converts measured total TLS loss into the intrinsic loss tangent of each region and singles out HSQ as the largest term.","core_discovery":"On the paper's own terms, the central discovery is that HSQ is the dominant TLS loss channel in superconducting NbN coplanar waveguide resonators whenever it is left on the device, and that its intrinsic loss tangent is $\\delta^i_{\\mathrm{HSQ}} = 8.0 \\times 10^{-3}$. This is roughly four times the loss tangent of the silicon-oxide interfaces that have historically been blamed for surface losses, so the standard assumption that a spin-on-glass resist behaves like silicon oxide would seriously underestimate its impact. The dominance is not because HSQ is extraordinarily lossy per volume, but because it sits directly above the center conductor, where the electric field is strongest, so its geometric filling factor is large and grows as the resonator gap shrinks. The paper establishes this by fabricating pairs of resonators that differ only in the HSQ layer, measuring the temperature dependence of the resonance frequency to isolate TLS loss, and using full-wave simulations to convert the measured total loss into per-material loss tangents.","pith_inferences":["The extracted HSQ loss tangent inherits the assumed layer stack; if the actual oxide thicknesses or the absence of $\\mathrm{Nb_2O_5}$ under HSQ are wrong, the fitted value would compensate. A geometry with a single well-known lossy layer would separate those uncertainties.","Because developed HSQ is porous amorphous silica, its loss tangent likely depends on exposure dose, development, and baking; the reported value may represent one process condition, so a processing study could find lower-loss HSQ variants.","The same measurement-and-simulation workflow could rank other nanofabrication resists and hard masks by their predicted TLS loss, giving nanowire developers a screening tool before device fabrication."],"forward_implications":["Leaving HSQ on a finished nanowire resonator sets a floor on the achievable internal quality factor; stripping HSQ after e-beam patterning should raise $Q$ at single-photon powers.","At sub-micron gaps, the top dielectric's filling factor grows fastest, so high-impedance resonator design must treat the dielectric layer on top of the conductor as a first-order loss term, not a correction.","Electrostatic filling-factor calculations are not reliable for disordered-superconductor devices whose dimensions are comparable to the magnetic penetration depth; Maxwell–London simulation is needed to separate loss channels.","The extracted loss tangents of HSQ, substrate-metal interface, niobium oxide, and silicon oxide can be reused as material parameters to predict TLS loss in future nanowire circuits before fabrication."],"supporting_citations":[{"why":"Supplies the NbN nanowire superinductor fabrication and measurement context, including the claim that TLS dominates loss in thin disordered films.","marker":"[4]"},{"why":"Provides the participation-ratio method for surface losses in coplanar waveguide resonators and the result that metal-air interfaces contribute little.","marker":"[25]"},{"why":"Gives the baseline surface-dielectric loss tangent near $2.6 \\times 10^{-3}$ that the HSQ value is compared against.","marker":"[26]"},{"why":"Supplies the trenched-resonator interface-loss model and the substrate-metal layer representation used in the simulation.","marker":"[28]"},{"why":"Provides the TLS defect model and reference intrinsic loss tangents for dielectric films at single-photon energies.","marker":"[34]"},{"why":"Supplies the frequency-shift-versus-temperature fitting formula used to extract $F_{\\mathrm{TLS}} \\delta^i_{\\mathrm{TLS}}$.","marker":"[35]"},{"why":"Supplies the complementary resonator characterization method and the same TLS frequency-shift fit.","marker":"[36]"},{"why":"Provides measured dielectric-loss data for amorphous thin films, including the niobium-oxide parameters used in the fit.","marker":"[43]"},{"why":"Supplies the empirical complex-permittivity description of the superconductor used to model NbN in the finite-element simulation.","marker":"[52]"}],"fun_headline_variants":["HSQ resist loss 4x silicon oxide, tops NbN resonator loss","HSQ spin-on-glass loss tangent 8e-3 dominates resonator loss","Geometric scaling? HSQ resist, not oxide, sets TLS loss","HSQ resist on hot spot: 4x lossier than silicon oxide"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The fitted HSQ loss tangent assumes the device is exactly the modelled stack of uniform dielectric layers with the stated thicknesses and permittivities, with no niobium oxide under the HSQ; if those assumptions miss the real field distribution, the extracted HSQ loss changes even though the measured total loss is unchanged.","fun_headline_variants_meta":{"raw":{"variants":["HSQ resist loss 4x silicon oxide, tops NbN resonator loss","HSQ spin-on-glass loss tangent 8e-3 dominates resonator loss","Geometric scaling? HSQ resist, not oxide, sets TLS loss","HSQ resist on hot spot: 4x lossier than silicon oxide"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000662,"raw_usage":{"total_tokens":3032,"prompt_tokens":956,"completion_tokens":2076,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":572,"completion_tokens_details":{"reasoning_tokens":1994}},"tokens_in":572,"tokens_out":2076,"duration_ms":16506,"temperature":1.0,"reasoning_tokens":1994,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:39:19.269955+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take finished resonators that still carry HSQ, remove the HSQ with a dedicated strip process, and remeasure the single-photon TLS loss; if the loss does not drop by the amount predicted from $F_{\\mathrm{HSQ}} \\times 8.0 \\times 10^{-3}$, the dominance claim is wrong. Alternatively, measure the HSQ loss tangent directly in a resonator whose filling factor is fixed by geometry alone, such as a lumped-element capacitor with the HSQ as the known dielectric.","supporting_citations":[{"cited_title":"High Kinetic Inductance NbN Nanowire Superinduc- tors,","cited_arxiv_id":null,"evidence_quote":"Supplies the NbN nanowire superinductor fabrication and measurement context, including the claim that TLS dominates loss in thin disordered films."},{"cited_title":"Surface loss simulations of superconducting coplanar waveguide resonators,","cited_arxiv_id":null,"evidence_quote":"Provides the participation-ratio method for surface losses in coplanar waveguide resonators and the result that metal-air interfaces contribute little."},{"cited_title":"Sur- face participation and dielectric loss in superconducting qubits,","cited_arxiv_id":null,"evidence_quote":"Gives the baseline surface-dielectric loss tangent near $2.6 \\times 10^{-3}$ that the HSQ value is compared against."},{"cited_title":"Analysis and mitigation of interface losses in trenched superconducting coplanar waveguide resonators,","cited_arxiv_id":null,"evidence_quote":"Supplies the trenched-resonator interface-loss model and the substrate-metal layer representation used in the simulation."},{"cited_title":"Microwave dielectric loss at single pho- ton energies and millikelvin temperatures,","cited_arxiv_id":null,"evidence_quote":"Provides the TLS defect model and reference intrinsic loss tangents for dielectric films at single-photon energies."},{"cited_title":"Experimental evidence for a surface distribution of two- level systems in superconducting lithographed microwave resonators,","cited_arxiv_id":null,"evidence_quote":"Supplies the frequency-shift-versus-temperature fitting formula used to extract $F_{\\mathrm{TLS}} \\delta^i_{\\mathrm{TLS}}$."},{"cited_title":"Properties of supercon- ducting planar resonators at millikelvin temperatures,","cited_arxiv_id":null,"evidence_quote":"Supplies the complementary resonator characterization method and the same TLS frequency-shift fit."},{"cited_title":"Measurement of dielec- tric losses in amorphous thin films at gigahertz frequen- cies using superconducting resonators,","cited_arxiv_id":null,"evidence_quote":"Provides measured dielectric-loss data for amorphous thin films, including the niobium-oxide parameters used in the fit."},{"cited_title":"Empirical model of the microwave properties of high- temperature superconductors,","cited_arxiv_id":null,"evidence_quote":"Supplies the empirical complex-permittivity description of the superconductor used to model NbN in the finite-element simulation."}],"review_version":1}