{"id":"40cd3cc9-6558-41c8-8174-71c8a52f6901","arxiv_id":"2412.09693","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Hydrogenated amorphous silicon parallel-plate resonators show two-level-system noise 5 to 80 times lower than previous amorphous dielectrics and comparable to crystalline substrates.","lead":"Researchers measured the noise from atomic-scale defects in hydrogenated amorphous silicon films inside superconducting microwave resonators and found it far lower than in other amorphous dielectrics and close to crystalline materials. The result could enable compact, multilayer superconducting detectors and qubit circuits using an easily deposited dielectric.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Literature comparison for the a-SiC:H point (K) extrapolates from 10 Hz to 1 kHz using α=-1, contradicting the cited reference's own two-slope spectrum (α=-1 below 100 Hz, α=-0.5 above); this shifts K upward by ~3.2× and can erode the stated 5× improvement margin.","rationale":"I read the paper in good faith. The measurements are extensive, the two-setup cross-check is valuable, and the low level of TLS noise in these a-Si:H PPCs is credible. The reader's concern about fixing α=-0.5 for their own fits is real but bounded; a slope error across 200 Hz-2 kHz shifts the 1 kHz amplitude by at most ~30%. The more load-bearing issue is in the literature comparison: the K point is extrapolated with a single α=-1 slope, even though this paper itself summarizes Kouwenhoven et al. as having α=-0.5 above 100 Hz. That makes the K point about 3.16× too low and directly threatens the '>5' margin in the abstract for other amorphous dielectrics. The fix is straightforward and checkable. Since the qualitative advance likely survives even after correction, the appropriate verdict remains conditional rather than reject; the abstract and the 5-50× range must be revised if the corrected margin falls below 5.","tokens_in":9041,"tokens_out":10222,"duration_ms":107160,"concrete_test":"Recompute the K point in Figure 5 using the two-slope model from Kouwenhoven et al.: apply α=-1 from 10 Hz to 100 Hz and α=-0.5 from 100 Hz to 1 kHz, then compare the red envelope at matched photon number N. If the corrected K point lies within a factor <5 of the envelope, revise the '5-50×' statement and the abstract's '>5' claim (or restrict the claim to a-Si:H). Also re-plot including the three excluded outlier datasets to bound envelope sensitivity.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central quantitative claim that these a-Si:H resonators beat the best prior amorphous-dielectric TLS noise by >5 relies on the Figure 5 comparison. The K point (Kouwenhoven et al., a-SiC:H PPC) is the limiting 'other amorphous dielectric' case: the paper quotes a 5-50× improvement over K, and the abstract's >5 factor is tied to that lower bound. To place K at ν=1 kHz, the authors extrapolate its 10 Hz measured value using α=-1. But the same reference, as summarized earlier in this paper, reports α≈-1 only below 100 Hz and α≈-0.5 above 100 Hz-1 kHz. The correct piecewise extrapolation yields STLS(1 kHz)=STLS(10 Hz)/(10·√10)≈STLS(10 Hz)/31.6, not /100, so the K point should be about 3.16× higher. The stated 5-50× range becomes roughly 1.6-16×, so the '>5' improvement over the best other amorphous dielectric is not established. This is an internal comparison inconsistency, not a question of the authors' own fit quality; the comparison versus a-Si:H (H) is unaffected.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports measurements of two-level-system (TLS) frequency noise in hydrogenated amorphous silicon (a-Si:H) deposited by two PECVD variants and integrated into niobium lumped-element resonators with parallel-plate capacitors. The authors measure fractional-frequency noise PSDs for 18 resonances across five devices at 230 mK, fit a sum of white noise and a TLS term with spectral slope α fixed to −0.5 over 200 Hz to 2 kHz, and extract TLS noise amplitudes at 1 kHz as a function of stored photon number. They cross-check two devices at NIST at 20 mK, finding agreement within about a factor of two, and compare their results to a literature compilation. The central claim is that their a-Si:H resonators improve on the best previous amorphous-dielectric TLS noise by a factor >5 and approach crystalline-dielectric performance.","tokens_in":9320,"tokens_out":5516,"duration_ms":55874,"significance":"If validated, the quantitative claim would establish PPC-based superconducting resonators with a-Si:H as a practical low-noise alternative to CPW and IDC devices, with potential impact on KID design and multi-layer qubit architectures. The paper's strengths include systematic power and temperature sweeps, a cross-laboratory check on the same devices, the use of a well-established readout and calibration procedure, and explicit acknowledgment of limitations such as the 2–3 dBm power calibration uncertainty. The main weakness is an internal inconsistency in the literature comparison, which undermines the abstract's '>5' factor for 'other amorphous dielectrics' as currently presented.","major_comments":[{"comment":"The comparison point K (a-SiC:H from Kouwenhoven et al.) is extrapolated from 10 Hz to 1 kHz using α = −1, but the same reference, as described earlier in this paper, reports α ≈ −1 only below 100 Hz and α ≈ −0.5 above 100 Hz to 1 kHz. Using the correct piecewise extrapolation raises the K point by a factor of 100/(10*sqrt(10)) ≈ 3.16, changing the claimed 5–50× improvement over K to roughly 1.6–16×. The abstract's claim of '>5' improvement over other amorphous dielectrics is therefore not established by the presented comparison. The authors should either use the piecewise slope or rephrase the claim to be specific to a-Si:H (point H), which is unaffected by this issue.","section":"Figure 5 and the paragraph beginning 'In order to compare our a-Si:H TLS noise results...'"},{"comment":"The red envelope in Figure 3 is defined by eye, and three outlier datasets are excluded without a stated quantitative criterion. Since the factor '>5' in the abstract is derived from the lower edge of this envelope in the Figure 5 overlay, the authors should provide a reproducible definition of the envelope (for example, a fixed percentile of the datasets or the range containing a stated fraction of resonator datasets) and report how the lower edge and the comparison factors change if the outliers are included.","section":"Figure 3 and the paragraph defining the red envelope"}],"minor_comments":[{"comment":"The fixed slope α = −0.5 is motivated by fits at high feedline readout power, but the paper does not quantify the scatter in α across the different resonators or powers used to set this fixed value. A brief sensitivity check of the extracted amplitude a to plausible variations in α would strengthen the robustness of the reported STLS values.","section":"Figure 2 and the fitting procedure"},{"comment":"The 2–3 dBm power calibration uncertainty is invoked to explain the Caltech–NIST discrepancy. Since the photon-number axis in Figures 3–5 depends directly on the feedline power, the authors should state how this uncertainty propagates to N and whether it affects the placement of the red envelope relative to the literature points plotted at nominal powers.","section":"Figure 4 and the Caltech–NIST comparison"},{"comment":"The sentence 'The direction − − − →B1B2 corresponds to the frequency direction' uses a malformed arrow that is difficult to read; please typeset the vector notation properly.","section":"Figure 1 and calibration description"},{"comment":"The phrase 'We see that the STLS(ν = 1 kHz) results presented in this article are 8 – 80 times lower than previous measurements for a-Si:H (H)' would benefit from specifying whether the range reflects the full spread of the red envelope or a single representative value, given the envelope's finite width.","section":"Results section, the comparison paragraph"}],"recommendation":"major_revision","confidential_remarks":"The experimental work appears careful and the cross-laboratory check is a genuine strength. My main concern is the internal inconsistency in the literature comparison, which is fixable by re-deriving the K point with the piecewise spectral slope or by rewording the claim. I would not reject on this basis; the paper should undergo major revision to make the quantitative comparison reproducible and internally consistent. The manuscript fits the scope of the journal as an instrumentation and measurement paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe headline: Defrance et al. report TLS noise in PECVD a-Si:H parallel-plate resonators that sits close to crystalline-substrate levels, and they back it with a two-lab cross-check (Caltech at 230 mK and NIST at 20 mK). That is the first time an amorphous dielectric PPC has shown noise in this range, and it makes a-Si:H a credible option for multilayer KIDs and qubit circuits. The prior loss-tangent paper (Defrance et al., PRM 2024) provides fabrication context; the noise measurements here are independent of that work.\n\nThe measurement protocol is careful: on-resonance noise time-streams, a calibration tone for the volts-to-fractional-frequency conversion, fits restricted to 200 Hz–2 kHz where PSDs are reproducible, and a fixed spectral slope alpha=-0.5 justified by high-power fits. The cross-laboratory agreement (less than a factor of 2 for six common resonances) is a genuine strength. The spread across 18 resonators is honestly displayed, and the acknowledged 2–3 dBm power calibration uncertainty is stated plainly.\n\nSoft spots are in the comparison, not the raw data. The abstract's \">5 improvement over the best other amorphous dielectric\" rests on point K (Kouwenhoven et al., a-SiC:H). The authors extrapolate that 10 Hz measurement to 1 kHz with alpha=-1, but the same reference—and their own text—reports alpha≈-1 below 100 Hz and alpha≈-0.5 above. The correct piecewise extrapolation raises K by about 3.2x, dropping the quoted 5–50x margin to roughly 1.6–16x. The \">5\" claim is not established for a-SiC:H; it may still hold for a-Si:H (point H), but the abstract's broader phrasing is too strong.\n\nThe red envelope in Fig. 3 is hand-drawn, and three outlier datasets are excluded without a quantified effect on the comparison—a minor issue compared to the K-point extrapolation, but worth tightening. The fixed alpha=-0.5 is reasonable; checking against free-alpha fits at low power would reassure that the extracted amplitude isn't absorbing slope variation.\n\nWho is this for? Anyone building superconducting resonators on amorphous dielectrics. The core data are valuable and likely reproducible. The comparison needs fixing: re-derive K with the piecewise slope, or soften the abstract claim. I'd send it to review; a competent referee will catch the same issue, and the authors can fix it in revision.","headline":"New a-Si:H TLS noise result is likely a real advance, but the >5 improvement claim depends on a faulty extrapolation of the a-SiC:H comparison point.","tokens_in":9818,"tokens_out":2464,"would_cite":true,"duration_ms":24409,"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":"Hydrogenated amorphous silicon capacitors show TLS noise comparable to crystalline substrates, beating prior amorphous dielectrics by a factor of 5 or more.","keywords":["two-level systems","TLS noise","hydrogenated amorphous silicon","parallel-plate capacitors","superconducting resonators","kinetic inductance detectors","PECVD","frequency noise"],"falsifier":"A direct measurement of the fractional-frequency noise power spectral density below 100 Hz on these same a-Si:H resonators would test the fixed-slope assumption: if the noise rises above the extrapolated −0.5 slope line, the reported $S_{\\mathrm{TLS}}$ at 1 kHz would still stand but the integrated noise below 100 Hz would be higher than modeled. Alternatively, fabricating the same a-Si:H film into an interdigitated capacitor resonator and comparing $S_{\\mathrm{TLS}}$ at fixed photon number would test whether the low noise is intrinsic to the film or an artifact of the parallel-plate geometry.","tokens_in":8868,"feed_emoji":"🔬","tokens_out":3530,"duration_ms":33365,"temperature":0.7,"pith_summary":"The paper reports measurements of two-level-system (TLS) noise in superconducting lumped-element resonators whose capacitors are made from hydrogenated amorphous silicon (a-Si:H) deposited by PECVD. The authors claim that the TLS noise of two a-Si:H recipes is more than five times lower than the best previously reported values for a-Si:H and other amorphous dielectrics, and is comparable to noise levels seen on crystalline substrates. If correct, this removes a key obstacle to using parallel-plate capacitors in kinetic inductance detectors and other superconducting circuits, enabling multilayer architectures without the noise penalty of amorphous dielectrics. The claim is based on fractional-frequency noise measurements between 200 Hz and 2 kHz, fit to a sum of TLS noise with spectral slope −0.5 and white noise.","feed_headline":"Amorphous silicon resonator noise drops to crystal levels","feed_subtitle":"PECVD-deposited a-Si:H capacitors cut TLS noise more than fivefold, opening multilayer detector designs.","key_machinery":"The measurement chain: each device is a 50 Ω coplanar-waveguide feedline inductively coupled to six lumped-element LC resonators, each made of niobium with an inductor and two a-Si:H parallel-plate capacitors in series. Noise time-streams are rotated to the frequency-dissipation basis using a two-tone calibration, and fractional-frequency noise power spectral densities are fit to a model $a\\nu^{\\alpha} + b$ with $\\alpha$ fixed to −0.5. The extracted amplitude $a$ at 1 kHz is the TLS noise level, compared across stored power (photon number) and against literature values by overlaying the measurement envelope on a reproduced figure from a review.","core_discovery":"The two-level-system (TLS) fractional-frequency noise measured at 1 kHz in niobium lumped-element resonators with a-Si:H parallel-plate capacitors is 8–80 times lower than previous a-Si:H microstrip results, 5–50 times lower than a-SiC:H, more than 100 times lower than silicon nitride, and comparable to resonators on crystalline silicon and sapphire. These results were obtained with two PECVD a-Si:H recipes, and the low noise is consistent with the previously reported low loss tangent of the same films. The authors conclude that a-Si:H PPCs are a viable low-noise alternative to coplanar-waveguide and interdigitated-capacitor architectures for superconducting resonators.","pith_inferences":["If the a-Si:H TLS noise is set by the film's intrinsic defect density, further reduction might be possible by tuning deposition temperature or hydrogen content, potentially approaching single-crystal limits.","The fixed-slope fitting assumption ($\\alpha = -0.5$) should be tested at audio frequencies below 100 Hz, where recent measurements on other amorphous dielectrics show a transition to a steeper slope.","A direct side-by-side measurement of the same a-Si:H film in both parallel-plate and interdigitated capacitor geometries on one setup would confirm that the low noise is intrinsic to the film rather than a geometric effect.","The comparison to literature relies on one dataset extrapolated from 10 Hz to 1 kHz; a direct measurement at 1 kHz on that same material would sharpen the claimed improvement."],"forward_implications":["Parallel-plate-capacitor kinetic inductance detectors can be built with a-Si:H without the previously expected TLS noise penalty, shrinking detector footprint and reducing stray-light sensitivity.","Multilayer superconducting circuits, including microstrip resonators and qubit architectures, become feasible with a deposited dielectric that approaches crystalline noise levels.","The factor-of-5-plus improvement over the best prior amorphous dielectrics warrants revisiting applications where amorphous dielectrics were ruled out by TLS noise.","The result strengthens the connection between low loss tangent and low TLS noise in a-Si:H, supporting the tunneling model's fluctuation-dissipation predictions."],"supporting_citations":[{"why":"Supplies the measurement method and comparison data points for TLS noise in coplanar waveguide resonators on crystalline substrates.","marker":"[1]"},{"why":"Provides the semiempirical model for TLS noise scaling with power and temperature that the analysis uses.","marker":"[2]"},{"why":"The review whose figure is reproduced and augmented for the literature comparison of TLS noise levels.","marker":"[9]"},{"why":"Recent a-SiC:H parallel-plate capacitor TLS noise data with the frequency-slope measurement used to extrapolate to 1 kHz.","marker":"[14]"},{"why":"Provides silicon nitride parallel-plate capacitor TLS noise data used in the comparison.","marker":"[15]"},{"why":"The companion paper with the same a-Si:H recipes and devices, supplying the low loss tangent and the photon-number calibration.","marker":"[23]"}],"fun_headline_variants":["a-Si:H capacitors match crystal TLS noise","TLS noise in a-Si:H cut fivefold over best","Amorphous silicon rivals crystals for TLS noise","Low TLS noise in a-Si:H beats other films","PECVD a-Si:H cuts TLS noise to crystal levels"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes that between 200 Hz and 2 kHz the measured frequency noise is exactly a TLS component with spectral slope −0.5 plus white noise, so that a single amplitude extracted from each fit represents the TLS level.","fun_headline_variants_meta":{"raw":{"variants":["a-Si:H capacitors match crystal TLS noise","TLS noise in a-Si:H cut fivefold over best","Amorphous silicon rivals crystals for TLS noise","Low TLS noise in a-Si:H beats other films","PECVD a-Si:H cuts TLS noise to crystal levels"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000327,"raw_usage":{"total_tokens":1759,"prompt_tokens":807,"completion_tokens":952,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":423,"completion_tokens_details":{"reasoning_tokens":876}},"tokens_in":423,"tokens_out":952,"duration_ms":8641,"temperature":1.0,"reasoning_tokens":876,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T16:50:08.441719+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct measurement of the fractional-frequency noise power spectral density below 100 Hz on these same a-Si:H resonators would test the fixed-slope assumption: if the noise rises above the extrapolated −0.5 slope line, the reported $S_{\\mathrm{TLS}}$ at 1 kHz would still stand but the integrated noise below 100 Hz would be higher than modeled. Alternatively, fabricating the same a-Si:H film into an interdigitated capacitor resonator and comparing $S_{\\mathrm{TLS}}$ at fixed photon number would test whether the low noise is intrinsic to the film or an artifact of the parallel-plate geometry.","supporting_citations":[{"cited_title":"Gao , author M","cited_arxiv_id":null,"evidence_quote":"Provides the semiempirical model for TLS noise scaling with power and temperature that the analysis uses."},{"cited_title":"Kouwenhoven , author G","cited_arxiv_id":null,"evidence_quote":"Recent a-SiC:H parallel-plate capacitor TLS noise data with the frequency-slope measurement used to extrapolate to 1 kHz."},{"cited_title":"Cryogenic microwave performance of silicon nitride and amorphous silicon deposited using low-temperature ICPCVD","cited_arxiv_id":"2409.09301","evidence_quote":"Provides silicon nitride parallel-plate capacitor TLS noise data used in the comparison."}],"review_version":1}