{"id":"db035b36-0b29-439f-9e80-86582a072fa1","arxiv_id":"2506.17474","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Treating finished aluminum-on-silicon superconducting devices with atomic layer etching and deposition reduced two-level-system loss and raised median transmon T1 from about 101 to 196 microseconds.","lead":"A dry surface treatment that etches away native oxide and fabrication residue, then coats aluminum qubit surfaces with a thin clean alumina layer, roughly doubled measured qubit lifetimes in tests on finished devices. The results suggest a packaging-step fix for surface-induced energy loss in aluminum superconducting circuits, if the improvements replicate in larger samples.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The two-fold TLS-loss reduction is not separated from the 300°C thermal budget: no annealed-only resonator control is reported, and the sole annealed-only transmon chip is n=1, so thermal effects could contribute.","rationale":"The manuscript presents a coherent experimental story with independent support for the central claim: the resonator δ_TLS measurement is junction-free, the XPS/PiFM/TEM data show surface chemical changes, and the transmon T1/Q distributions improve after treatment. These are real strengths. However, the causal attribution to ALE+ALD specifically, rather than to the 300°C thermal budget, is the most load-bearing assumption. The absence of an annealed-only resonator control is a concrete gap: the resonator measurements in Section II.A compare untreated versus treated chips, while the only annealed-only device appears in the transmon dataset with n=1. The XPS results indicate that annealing alone changes the surface chemistry, so the thermal effect cannot be dismissed as negligible. The transmon comparison also suffers from a treatment-induced frequency shift and unexplained junction resistance changes, which the authors acknowledge in Section II.D and SI Section III. These issues do not refute the claim, but they do mean the current evidence is conditional rather than definitive. The proposed annealed-only resonator test directly targets the weakest link and would settle whether the observed improvement is due to the ALE/ALD chemistry or to heating during processing.","tokens_in":18211,"tokens_out":10721,"duration_ms":130187,"concrete_test":"Add an annealed-only control group for resonators: process at least five resonator chips through the full 300°C vacuum thermal cycle in the ALD chamber with the same time and pressure profile but with no TMA, HF-pyridine, or H2O precursor exposures, and measure δ_TLS under the identical protocol as Section II.A. If the annealed-only δ_TLS distribution is not statistically distinguishable from untreated, the thermal confound is excluded; if it lies between untreated and ALE+ALD, the headline effect must be renormalized by the anneal-only contribution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that ALE+ALD specifically reduces surface TLS density is underdetermined because the treatment is performed at 300°C and the only thermal control is a single annealed-only transmon chip. Section II.A compares resonator δ_TLS for untreated versus ALE+ALD-treated chips with no annealed-only resonator group, while Section II.B reports only one annealed-only chip. The XPS data in Section II.C show that annealing alone removes Al-O-OH groups and reduces O=C-O carbon, so the thermal budget is not chemically inert. The transmon comparison is further complicated by the +500-800 MHz f01 shift (SI Section III) and by the paper's own statement in Section II.D that 'changes in the junction also occur during the ALE/ALD process' with an unexplained resistance change. If a substantial part of the resonator or transmon gain comes from the 300°C anneal rather than the ALE/ALD chemistry, the attribution of the improvement to reduced TLS defect density on capacitor surfaces would be weakened. The claim is not refuted by the current data, but the control needed to separate the chemistry from the thermal budget is missing for resonators and underpowered for transmons.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a dry post-fabrication surface treatment for aluminum-on-silicon superconducting circuits, combining atomic layer etching (ALE) of the native aluminum oxide with in situ atomic layer deposition (ALD) of a thin Al2O3 cap. The authors measure microwave resonator internal loss and planar transmon T1/Q before and after the treatment, reporting a two-fold reduction in TLS-associated loss in resonators and a comparable improvement in transmons, with median Q = 3.69 ± 0.42 × 10^6 and T1 = 196 ± 22 μs after treatment. They support the interpretation with surface characterization using XPS, PiFM, and TEM-EELS, showing removal of PMMA contamination, reduction of oxyhydroxide species, and a thinner, more stoichiometric surface oxide. A single annealed-only transmon chip is used as a thermal-budget control, and the authors acknowledge that junction properties change during the process.","tokens_in":18450,"tokens_out":2514,"duration_ms":27534,"significance":"If the central claim holds, this is a practically valuable result: a scalable, conformal, Al-compatible surface treatment applied at the end of the fabrication flow that more than doubles transmon coherence, with demonstrated stability over months. The surface analysis is unusually thorough for a device-focused paper, and the direct before/after comparison on the same transmon chips is a strong experimental design. The paper also makes a falsifiable prediction—that surface TLS density is reduced—which is grounded in standard high/low-power loss extraction rather than fitted to the outcome. The main risk, as the authors partly acknowledge, is that the 300 °C thermal budget itself produces some of the chemical changes and device shifts, and the controls presently are insufficient to fully separate the ALE/ALD chemistry from the anneal.","major_comments":[{"comment":"The resonator result lacks an annealed-only control group. The central claim of a two-fold reduction in δ_TLS (Eq. 1) is attributed to the ALE+ALD chemistry, but the process is performed at 300 °C, and the XPS data in §II.C show that annealing alone removes Al-O-OH and reduces O=C-O carbon. Without a resonator chip subjected only to the same thermal budget, the observed reduction in δ_TLS cannot be separated from the anneal's chemical effects. The authors should either add an annealed-only resonator cohort or explicitly weaken the causal claim to 'treatment, including its thermal budget,' and discuss what the resonator data do and do not establish.","section":"§II.A and Methods IV.C"},{"comment":"The transmon attribution rests on a single annealed-only chip, which is underpowered for the variance in T1/Q distributions shown in Figure 2. Moreover, the treatment shifts f01 upward by 500–800 MHz (SI Section III) and the paper states in §II.D that 'changes in the junction also occur during the ALE/ALD process' with unexplained resistance changes. The Purcell curves in SI Section II are modeled, not measured corrections, and the reported median Q and T1 are not compensated for the frequency-dependent Purcell contribution. The causal step from 'ALE+ALD treatment improves T1/Q' to 'surface TLS density in the capacitor is reduced' is therefore load-bearing and not fully tested. The authors should present a quantitative estimate of the maximum Purcell contribution to the reported improvement, or measure TLS density directly (e.g., via temperature-dependent loss or coherent TLS spectroscopy) on treated versus annealed-only devices.","section":"§II.B and SI Sections II–III"},{"comment":"Fluorine incorporation is not assayed, although the ALE process uses HF-pyridine and the proposed mechanism in Eq. (2) forms AlF3 as a surface intermediate. The XPS measurements (Figures 3 and SI-5 through SI-7) do not include a F 1s spectrum, and the paper does not discuss fluorine residues as a possible loss source or as a marker of incomplete ALE. Given that the treatment is claimed to produce a clean, stoichiometric Al2O3 surface, the absence of any fluorine assay leaves an important gap in the surface-chemistry evidence chain. The authors should provide F 1s XPS data or another direct measurement of residual fluorine on treated surfaces.","section":"§II.C and Methods IV.D"},{"comment":"The resonator statistics are reported only as 'three resonators on two devices' for each group, with no per-device error bars on the derived δ_TLS values or on the factor-of-two reduction. Figure 1 shows three representative pairs, but the reader cannot assess the scatter or significance of the improvement. The authors should report all paired values, the uncertainty on each δ_TLS extraction (including the HP/LP fitting errors), and a statistical test of the difference between treated and untreated groups.","section":"§II.A and Methods IV.C"}],"minor_comments":[{"comment":"The definition of the effective quality factor is unclear as typeset: 'Q = 2π f01 T1 4' appears to contain a spurious superscript or missing denominator. Please state explicitly whether Q = 2π f01 T1 or Q = 2π f01 T1 / 4, and use a consistent notation in the abstract, main text, and SI.","section":"§II.B, Eq. (2)"},{"comment":"The e-PDF and e-CDF are described as empirical distributions, but the reported medians and MADs presumably come from a bootstrap or resampling procedure. Please state the resampling method and the number of samples used for the error bars.","section":"§II.B, Figure 2(c)"},{"comment":"The main text says three resonators on two devices for both untreated and treated samples, but the caption of Figure 1 says 'three representative resonators.' Please clarify whether these are the same three resonators, and whether the two devices each contribute multiple resonators.","section":"§IV.C, 'three resonators on two devices'"},{"comment":"The statement that residual PMMA forms a continuous layer of at least ~2 nm thickness on the untreated surface is supported by a shielding test described only loosely in the text. Please provide the test details or a reference for the thickness estimate.","section":"§II.C, PiFM discussion"},{"comment":"The sentence 'While our study does not disentangle the individual contributions of polymeric residue removal and Al oxide layer modification' is an important limitation that should be placed earlier in the paper, ideally in the abstract or introduction, so that readers do not overinterpret the causal mechanism.","section":"§II.D"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for an applied physics journal and is likely to be of interest to the superconducting-qubit community. The main concern is the missing thermal-budget control for the resonator data and the underpowered single-chip control for the transmon data, which together make the central causal attribution to ALE/ALD chemistry (rather than the anneal) somewhat fragile. I would encourage the editor to ask for the additional control measurements or a substantial weakening of the causal wording. The fluorine assay is also easy to add and would materially strengthen the surface-chemistry story."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this is a genuinely new process step—post-fabrication, in situ ALE+ALD on fully wired devices, covering top and sidewall surfaces—and the surface chemistry work is solid. The claim that it roughly halves TLS loss is supported by the transmon statistics, not by the resonator data alone. What is not yet nailed down is whether the improvement comes from the ALE/ALD chemistry specifically or partly from the 300°C anneal that comes with it.\n\nThe paper does several things well. The XPS, PiFM, and TEM-EELS data are internally consistent: the treatment removes PMMA residue, replaces the native oxide with a thinner, stoichiometric Al2O3 layer, and leaves the Si surface uncapped. The transmon dataset is reasonably sized—five chips measured before and after, plus six untreated chips—and the median Q shift from 1.9e6 to 3.7e6 is a real effect. The stability over months is also a nice practical point. The authors are candid about the junction resistance changes and do not overclaim a mechanism they did not isolate.\n\nThe soft spots are real but not fatal. The biggest is the one the stress-test note flags: there is no annealed-only resonator control at all, and the annealed-only transmon chip is n=1. Since the XPS shows that annealing alone removes Al-O-OH and reduces O=C-O, the thermal budget is not chemically inert. The transmon comparison is further muddied by the +500–800 MHz frequency shift, which changes Purcell loss and moves the qubit to a different TLS sampling window; the modeled Purcell curves are a reasonable correction, but they are not a substitute for measurement at matched frequencies. Two smaller gaps: fluorine residues from HF-pyridine are not assayed by XPS, and the resonator result rests on three pairs with no error bars on the reduction. None of these refute the central claim; they just mean the causal attribution to ALE+ALD over the anneal is underdetermined.\n\nFor whom: this is a useful paper for anyone working on Al-based qubit surface engineering or on post-fabrication passivation. It deserves a serious referee—the process is novel, the characterization is credible, and the weaknesses are addressable with more controls (annealed-only resonators, F 1s XPS, more annealed-only transmons). I would send it to peer review, and I would probably cite it if I were working in this area.","headline":"A well-characterized post-fabrication surface treatment for Al qubits with a plausible two-fold TLS loss reduction, but the thermal-budget control is too thin to fully separate the chemistry from the anneal.","tokens_in":19044,"tokens_out":2270,"would_cite":true,"duration_ms":26076,"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":"A post-fabrication dry etch-and-coat treatment doubles the energy-relaxation time of aluminum superconducting qubits to a median of 196 microseconds.","keywords":["superconducting qubits","transmon","two-level system loss","atomic layer etching","atomic layer deposition","aluminum on silicon","dielectric loss","coherence time"],"falsifier":"Measure the single-photon TLS loss tangent on junction-free resonators after the full ALE+ALD recipe, after an annealed-only control, and with XPS assay of residual fluorine: if the annealed-only chip reproduces the twofold loss reduction, or if residual fluorine correlates with the loss change, the surface-TLS attribution would collapse.","tokens_in":17979,"feed_emoji":"⚛️","tokens_out":9688,"duration_ms":91559,"temperature":0.7,"pith_summary":"This paper tries to show that a dry, conformal post-fabrication treatment—atomic layer etching followed by atomic layer deposition (ALE+ALD)—can undo the surface damage that limits aluminum-on-silicon superconducting circuits. Across resonators and planar transmon qubits, the treatment halves the dielectric loss attributed to two-level-system (TLS) defects and roughly doubles median qubit quality factor and energy-relaxation time: treated transmons reach median $Q = 3.69 \\pm 0.42 \\times 10^6$ and $T_1 = 196 \\pm 22~\\mu\\mathrm{s}$, up from $1.91 \\pm 0.18 \\times 10^6$ and about $101~\\mu\\mathrm{s}$ before treatment, with the gain persisting over months. If correct, this gives aluminum-based qubits a scalable, aluminum-compatible route to longer coherence without redesigning the circuit.","feed_headline":"Dry etch-and-coat step doubles aluminum qubit lifetimes","feed_subtitle":"It strips residues and native oxide, then caps the metal in clean alumina, halving two-level-system loss.","key_machinery":"The load-bearing mechanism is the ALE+ALD process itself: self-limiting half-cycles of HF-pyridine and trimethylaluminum etch Al2O3 at about 0.5 Å/cycle, and in-situ TMA/H2O ALD regrows about 1 nm of stoichiometric Al2O3 at 1 Å/cycle, all at 300 °C. This replaces the lossy native oxide with a clean cap while the thermal step weakens PMMA adhesion, allowing the etchant to remove resist residues from aluminum and silicon alike. The argument works by showing that these chemical changes track the measured reduction in TLS loss.","core_discovery":"The authors claim that thermal ALE with alternating trimethylaluminum and HF-pyridine, followed in situ by ALD of a ~1-nm stoichiometric Al2O3 cap, removes the oxygen-rich native oxide and polymeric fabrication residues from all exposed aluminum and silicon surfaces of a finished device, reducing the TLS defect density that dominates single-photon loss. The direct evidence is a roughly 50% drop in $\\delta_{\\mathrm{TLS}}$ for treated coplanar-waveguide resonators and a matching twofold rise in transmon $Q$ and $T_1$, with no such gain from a 300 °C anneal alone. Surface characterization (XPS, PiFM, TEM-EELS) shows thinner, more stoichiometric alumina and greatly reduced PMMA contamination, which the authors tie to the loss reduction; they are explicit that the study correlates material quality with device performance rather than isolating every contribution.","pith_inferences":["If the surface-TLS interpretation is right, extending the etch chemistry to also strip native silicon oxide from exposed substrate areas could push the substrate-air interface to even lower loss.","The observed anticorrelation between qubit $Q$ and the treatment-induced frequency shift suggests that if the anneal's junction shift could be suppressed, treated devices would show larger gains at high frequencies.","The 1-nm ALD cap is the new surface that sees the electromagnetic field; testing alternative low-loss ALD dielectrics in the same process flow would directly show whether the cap's own TLS contribution can be lowered further."],"forward_implications":["Median energy-relaxation time of treated Al-on-Si transmons rises to $T_1 = 196 \\pm 22~\\mu\\mathrm{s}$, with median $Q = 3.69 \\pm 0.42 \\times 10^6$; some devices exceed $9 \\times 10^6$.","The same treatment halves TLS loss in junction-free coplanar waveguide resonators across 5.2–6.2 GHz, indicating the effect is not specific to qubit junctions.","Improvements persist for at least 8–9 months after treatment, so the clean surface does not quickly re-oxidize or re-contaminate in storage.","Because the treatment applies to fully fabricated devices and covers sidewalls as well as top surfaces, it extends beyond capping strategies that leave exposed superconductor sidewalls lossy.","The process is compatible with aluminum, unlike common wet oxide etchants, so it can be applied where buffered oxide etchant or tri-acid treatments would damage the metal."],"supporting_citations":[{"why":"Supplies the thermal ALE chemistry (trimethylaluminum plus HF-pyridine) that removes the native aluminum oxide layer.","marker":"[27]"},{"why":"Supplies the TMA/water ALD process used to regrow the thin stoichiometric Al2O3 cap in situ.","marker":"[28]"},{"why":"Defines the high-power/low-power quality-factor subtraction used to quantify TLS-induced loss $\\delta_{\\mathrm{TLS}}$.","marker":"[23]"},{"why":"Documents the sidewall-loss limitation of pre-fabrication capping approaches, motivating treatment of fully fabricated devices.","marker":"[18]"},{"why":"Establishes PMMA bonding to surface hydroxyls, the mechanism invoked for residue removal during the anneal and etch.","marker":"[45]"},{"why":"Shows PMMA adsorption on oxidized aluminum and silicon, supporting the contamination analysis.","marker":"[46]"}],"fun_headline_variants":["Dry etch-and-cap boosts aluminum qubit lifetimes twofold","Etch-and-coat step halves qubit loss, doubling coherence times","Surface treatment doubles aluminum qubit T1 and Q","Dry etching plus ALD cap curbs TLS loss, lifts qubit lifetimes","Etch-and-coat finish doubles superconducting qubit coherence"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The causal claim assumes that the lifetime gain comes from fewer surface TLS defects in the capacitor, not from the treatment's simultaneous effect on the Josephson junction (it shifts qubit frequency by 500–800 MHz), from the 300 °C anneal, or from unmeasured etching residues such as fluorine.","fun_headline_variants_meta":{"raw":{"variants":["Dry etch-and-cap boosts aluminum qubit lifetimes twofold","Etch-and-coat step halves qubit loss, doubling coherence times","Surface treatment doubles aluminum qubit T1 and Q","Dry etching plus ALD cap curbs TLS loss, lifts qubit lifetimes","Etch-and-coat finish doubles superconducting qubit coherence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000704,"raw_usage":{"total_tokens":3163,"prompt_tokens":922,"completion_tokens":2241,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":538,"completion_tokens_details":{"reasoning_tokens":2155}},"tokens_in":538,"tokens_out":2241,"duration_ms":15477,"temperature":1.0,"reasoning_tokens":2155,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:07:48.098652+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the single-photon TLS loss tangent on junction-free resonators after the full ALE+ALD recipe, after an annealed-only control, and with XPS assay of residual fluorine: if the annealed-only chip reproduces the twofold loss reduction, or if residual fluorine correlates with the loss change, the surface-TLS attribution would collapse.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the TMA/water ALD process used to regrow the thin stoichiometric Al2O3 cap in situ."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the thermal ALE chemistry (trimethylaluminum plus HF-pyridine) that removes the native aluminum oxide layer."},{"cited_title":"Calusine, A","cited_arxiv_id":null,"evidence_quote":"Defines the high-power/low-power quality-factor subtraction used to quantify TLS-induced loss $\\delta_{\\mathrm{TLS}}$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents the sidewall-loss limitation of pre-fabrication capping approaches, motivating treatment of fully fabricated devices."},{"cited_title":"Papirer, J.-M","cited_arxiv_id":null,"evidence_quote":"Establishes PMMA bonding to surface hydroxyls, the mechanism invoked for residue removal during the anneal and etch."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows PMMA adsorption on oxidized aluminum and silicon, supporting the contamination analysis."}],"review_version":2}