{"id":"23cecc03-8295-4464-896f-10fb536b5881","arxiv_id":"2608.12021","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A homogenized microlens-array trap system with real-time parallel transport assembles a defect-free 32x32 register of 1024 neutral-atom qubits at 99.3% filling.","lead":"This paper reports a 1024-qubit neutral-atom computing register assembled with a new optical setup that combines homogenized light and parallel moving tweezers. It shows a practical route to scaling atom-based quantum computers while keeping high filling rates.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 1024-qubit defect-free claim rests on a cited 99.8% detection fidelity not re-measured for the full array; a small per-site error could inflate the 99.3% filling and the single-shot image.","rationale":"The reader's weakest assumption, imaging fidelity, is the most load-bearing concern for the paper's central claim. The defect-free 1024-qubit register is presented through fluorescence images, and the per-site detection fidelity is only cited from earlier work rather than re-measured on the full array. The 99.3(2)% filling is a cumulative maximum over repetitions, not a sustained value, so the single defect-free image could in principle be a fortuitous fluctuation even under the cited detection fidelity. The model-based extrapolations in Sec. III.B, where an effective lifetime is fitted to the same data it explains, are a secondary issue; they affect the quantitative comparison of transport protocols but not the direct filling statistic. Data availability restrictions further hinder independent verification. Because the reader already marked the verdict CONDITIONAL with moderate confidence, and the concern identified here is the same one, no change in verdict is warranted. The proposed detection-calibration test would settle whether the defect-free claim is overstated.","tokens_in":14681,"tokens_out":10926,"duration_ms":102424,"concrete_test":"Obtain the raw fluorescence images (or take a new calibration set) and measure per-site false-positive and false-negative rates on the full 43x43 array by comparing images with and without a resonant push-out pulse on selected sites, and by using far reservoir sites that are known to be empty. Then recompute the 1024-site cumulative maximum filling fraction and the defect-free count with the corrected per-site likelihoods. If the corrected filling remains at or above 99.0%, the defect-free claim is supported; if it drops below 99.3% by more than the stated 0.2% uncertainty, the central claim is overstated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim, assembly of a defect-free 1024-qubit register with 99.3(2)% filling (Fig. 1, Sec. III.B), depends on fluorescence images being an unbiased record of site occupancy. The paper does not calibrate detection on the 43x43 array; it cites an 'average fidelity of 99.8(2)%' from prior work [20] in Sec. II.A. For a 1024-site QPU, even a symmetric per-site error of 0.2% gives only about a 13% probability that a truly full array appears defect-free in a single image; if the true fidelity were 99.5%, that probability drops to about 0.6%. Conversely, an uncalibrated false-positive rate on empty sites would inflate the apparent filling. The 99.3(2)% statistic is the average of the maximum filling over 12 rearrangement repetitions, so it is not a direct measurement of a sustained configuration, and the defect-free image could be a rare fluctuation. Because the raw data are withheld, the filling statistics cannot be independently checked. A full-array detection calibration, reporting per-site false-negative and false-positive rates, is needed to support the defect-free claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a micro-optical architecture for neutral-atom qubit arrays that combines diffractive beam shaping with microlens-array (MLA) patterning to create homogenized 43×43 tweezer arrays with more than 3500 sites. The authors demonstrate trap-frequency uniformity of 4.5% in the central 32×32 region, parallelized AOD-based transport with up to 38 simultaneously steered tweezers, and the assembly of a 32×32 (1024-site) quantum processing unit with a reported cumulative maximum filling fraction of 99.32^{+0.19}_{-0.20}% (reported as 99.3(2)% in the abstract/conclusion) and a single-shot fluorescence image of a defect-free 1024-qubit register. A simple exponential-loss model (Eqs. 1–2) is used to quantify the benefit of parallelization and repeated rearrangement sequences, yielding a speedup factor of 9 and a median rearrangement time of 9.6 ms for 1024 sites.","tokens_in":14909,"tokens_out":8740,"duration_ms":82741,"significance":"If the central claim holds, this is a significant advance in the neutral-atom quantum-computing platform: it demonstrates a scalable, passive-optics route to large, uniform tweezer arrays and kiloqubit-scale defect-free registers with millisecond-scale assembly. The paper's strengths include a transparent experimental model with explicitly stated loss and fidelity parameters, site-resolved characterizations, and clear acknowledgment of reduced efficiency in the first rearrangement sequences. The architecture's modularization of intensity homogenization and patterning is a genuine technical contribution. However, the headline claim of a defect-free 1024-qubit register rests on a detection fidelity cited from prior work, not re-measured on the full array, and on a maximum-over-repetitions filling statistic rather than a sustained configuration; both points need to be addressed before the claim can be accepted at face value.","major_comments":[{"comment":"The central claim of a defect-free 1024-qubit register (Fig. 1(a)) and the reported filling fraction of 99.32% (Fig. 5(a), Section III.B) rest on single-shot fluorescence images being an unbiased record of site occupancy. The only detection-fidelity value quoted in the manuscript, 99.8(2)%, is taken from the authors' earlier work [20] and is not re-measured for the full 43×43 array or for the post-rearrangement conditions (different background, atom temperature, and transport history). The paper does not report per-site false-positive and false-negative rates. This matters because for a truly full 1024-site array, a symmetric per-site error of 0.2% gives only about a 13% probability that a single image appears perfectly filled, while an uncalibrated false-positive rate on empty sites would inflate the apparent filling fraction. I ask the authors to provide an in-situ detection calibration for the full array, quantify false positives and false negatives, and show that the 99.32% cumulative statistic and the defect-free image are consistent with the calibrated fidelity (e.g., by reporting the number of runs that achieved 100% filling and the per-trial success probability).","section":"II.A and III.B, Fig. 1(a), Fig. 5(a)"},{"comment":"The phrase 'sustained near-unity filling fraction of 99.3(2)%' overstates what is measured. Figure 5(a) reports the 'cumulative maximum filling fraction'—the average, over 1000 runs, of the maximum filling achieved during n_r=12 rearrangement sequences for the 1024-site target. This is not a time-averaged or terminal filling fraction, and it does not indicate that the register is 'sustained' at 99.3% outside the brief moments captured by the best image. The paper does not report the filling distribution after each sequence or the fraction of runs that reached 100% at n_q=1024. Please report the per-sequence filling distribution (e.g., mean and standard deviation of filling at the end of each rearrangement sequence, and the success probability of a fully filled image per run) and adjust the abstract and conclusion wording accordingly.","section":"Abstract; Section III.B; Conclusion"},{"comment":"The data availability statement says the data are not publicly available because they contain commercially sensitive information. Given that the paper's headline quantitative claims (the 99.3% filling statistic and the single-shot defect-free image) cannot be independently checked from the manuscript alone, this is a substantial limitation. I request that at least the processed per-site occupancy matrices (or histograms) underlying Figure 5(a) and the single-shot image in Figure 1(a) be deposited in a public repository, or that a detailed description of the image-analysis pipeline and thresholds be provided so the filling statistics can be reproduced from the raw data.","section":"Data Availability"}],"minor_comments":[{"comment":"Typo: 'incorporting' should be 'incorporating'.","section":"II.B"},{"comment":"The word 'occurence' is misspelled as 'relative occurence' in two places in the body text; the correct spelling is 'occurrence'.","section":"III.B (twice) and Fig. 5(b)"},{"comment":"The maximum number of parallel transport tweezers is given as 'up to 50' in the abstract, 'up to 43' in Section III.A, and 'up to 38' in the conclusion and Figure 5(b); please clarify which value refers to the system capability versus the observed maximum in the reported data.","section":"II.A, III.A, IV"},{"comment":"Equation (2) introduces F as the binomial cumulative distribution function; this is stated in the text but should be defined explicitly alongside the equation, especially since the notation \\(\\bar{p}_{\\rm cum}\\) and \\(\\bar{p}_M\\) are easily confused.","section":"III.B, Eq. (2)"},{"comment":"The sentence 'The experimental data for repetitive parallelized transport of Fig. 5(a) are used to calculate the expected values for the other two transport protocols' is a clear disclosure that the model is fit to the same data it explains; this is acceptable for an empirical model, but the authors should state explicitly that this is not a parameter-free prediction and should provide a goodness-of-fit estimate (e.g., residuals or comparison to experimental uncertainties).","section":"III.B, paragraph after Eq. (2)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reports an impressive experimental advance, but the central 'defect-free 1024-qubit register' claim depends critically on a detection fidelity quoted from a prior paper and on a maximum-over-repetitions statistic. I believe this is fixable: the authors can measure the full-array detection fidelity (false positives/negatives), report per-sequence filling statistics, and tone down the 'sustained' language. The data availability statement is a red flag for a claim of this magnitude; I would urge the editor to require at least the processed occupancy data and representative images to be released. The title itself ('Defect-Free ... 1024 ...') is stronger than what a single image can certify at 99.8% detection fidelity; a softened title or a quantitative probability-of-success statement would be more accurate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a genuine experimental step, not a breakthrough in scale alone (others have 3000+ qubits), but the modular microoptical approach is new and the assembly statistics are solid. The one thing I'd want before fully trusting the 'defect-free 1024' headline is a full-array detection calibration.\n\nWhat's new: separating intensity homogenization (top-hat beam shaper) from tweezer patterning (MLA) is a clean architectural idea. It gives a 43x43 array with 4.5% rms trap-frequency uniformity over the central 32x32, which is good. The parallel transport with up to 38 tweezers and real-time intensity control, yielding 5-9x speedup and ~10 ms rearrangement for 1024 sites, is a real advance. The paper is careful with error bars and openly labels its effective-lifetime model as a parameterization rather than a first-principles calculation.\n\nSoft spots, in order of concern. First, the 99.3(2)% filling and the single-shot defect-free image depend on the imaging fidelity of 99.8(2)% cited from prior work [20]. The array here is larger than what was likely characterized before, and a per-site error of a few tenths of a percent can matter at this scale. Re-measuring detection on the full 43x43 array, separating false-positive and false-negative rates, should be a referee requirement. Second, the raw data are withheld—the 'commercially sensitive' justification is unexplained and unusual for a university group. It limits independent verification of the filling statistics. Third, the 99.3% figure is the average of maximum fillings over 12 repetitions, not a sustained configuration measured at a later time; the word 'sustained' in the abstract overstates that. The tau_eff used to compare transport protocols is inferred from the same data it explains, but the authors say so, so I don't count it as a flaw.\n\nOverall the central result is plausible and the architecture is worth knowing. This is for experimentalists working on neutral-atom arrays and for people benchmarking hardware platforms. I'd send it to peer review, but with a request for the detection calibration and ideally the data. If those come through, this will be a useful reference for anyone building kiloqubit neutral-atom processors.","headline":"A credible kiloqubit-scale assembly result on a modular MLA platform, but the defect-free claim needs a full-array detection calibration to fully land.","tokens_in":15457,"tokens_out":3940,"would_cite":true,"duration_ms":38417,"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":"The paper reports rapid assembly of a defect-free 1024-site neutral-atom qubit register with 99.3% filling.","keywords":["neutral atom qubits","optical tweezer arrays","defect-free assembly","microlens arrays","beam shaping","parallelized atom transport","quantum register","acousto-optic deflectors"],"falsifier":"Re-image the same 1024-site register twice without moving or losing atoms and count site-by-site disagreements; if the per-site detection fidelity over the full array is below the assumed 99.8%, the reported filling fraction and defect-free certification would need to be revised downward.","tokens_in":14482,"feed_emoji":"⚛️","tokens_out":9943,"duration_ms":93971,"temperature":0.7,"pith_summary":"The paper reports a two-dimensional array of 1024 laser-trapped rubidium atoms on a regular 32×32 grid, with 99.3(2)% of sites occupied and a median rearrangement time around 10 ms. The central claim is that splitting the creation of homogeneous, high-power tweezer arrays into two separate optical functions—a beam shaper that flattens the laser profile and a microlens array that forms the trap pattern—removes the main scaling bottleneck, while parallelized acousto-optic transport tweezers with real-time intensity control move atoms quickly enough to beat vacuum-limited losses. If correct, this puts neutral-atom quantum registers above the kiloqubit level in a single plane, with 4.5% trap-frequency uniformity and assembly speeds suited to quantum computing and simulation.","feed_headline":"1024 neutral-atom qubits assembled defect-free in 10 ms","feed_subtitle":"A 32×32 grid of single atoms holds at 99.3% filling, pushing neutral-atom quantum computing past the kiloqubit mark.","key_machinery":"The load-bearing mechanism is the modular microoptical light-shaping chain: a diffractive beam shaper converts a Gaussian beam into a flat-top profile with 94.5% efficiency, and that uniform illumination feeds two interleaved microlens arrays that form homogeneous 43×43 trap grids; a separate dynamic port adds acousto-optic-deflector tweezers whose multi-frequency intensities are synthesized in real time so that parallelized transport trajectories keep constant trap depth. On the algorithm side, the rearrangement scheme reduces the two-dimensional problem to successive one-dimensional passes, uses a greedy nearest-assignment to map available atoms to empty target sites while preserving order, and parks surplus atoms in structured reservoir bands around the target pattern. The authors quantify assembly with a survival model $\\bar p = \\exp(-t/\\tau_{\\mathrm{eff}})$ and a binomial cumulative formula for the expected maximum filling after $n_r$ rearrangement repeats.","core_discovery":"The authors demonstrate a regular 32×32 quantum processor filled to 99.3(2)% and repeatedly rearranged to a defect-free pattern by combining homogenized microlens-array traps with parallelized multi-tweezer transport. Two interleaved 43×43 microlens arrays, illuminated by diffractive top-hat beam shapers, produce more than 3500 sites with 3.65(3) µm spacing, 0.86(7) µm waist, and a radial trap-frequency spread of only 4.5% over the central 32×32 region. Up to 38 steerable tweezers work simultaneously, executing deterministic row/column moves via a greedy nearest-assignment algorithm; a 1024-atom target pattern is assembled with a median rearrangement time of 9.6 ms, versus 88.1 ms for single-tweezer transport, a factor-9 speedup. Repeating the sequence up to 12–50 times raises the cumulative maximum filling toward 99.3–100%, and a binomial model with effective lifetime $\\tau_{\\mathrm{eff}}$ connects transport duration, repetition count, and filling fraction.","pith_inferences":["Our inference: if the reported filling and transport time reproduce, the natural next measurement is an in situ detection-fidelity test on the full 32×32 array; a per-site fidelity below the assumed 99.8% would directly lower the certified qubit count.","Our inference: the repeated-rearrangement routine acts as a virtual atom-lifetime extension, so combining it with the measured 10 s static lifetime suggests a route toward defect-free assembly of 5000- or 10000-site registers, an extrapolation the paper does not make.","Our inference: the real-time multi-frequency acousto-optic control could be extended to curved or crossing transport trajectories, enabling cyclic rearrangement schedules instead of row/column passes, and possibly to dual-species registers by swapping beam-shaping parameters and array pitch."],"forward_implications":["A neutral-atom quantum processor can now start from a 1024-site register with 99.3% filling, so atom assembly is no longer the dominant obstacle at the kiloqubit scale.","The 4.5% rms trap-frequency spread over the processor region means qubits across the array experience nearly identical trapping conditions, simplifying uniform gate and Rydberg interaction control.","A factor-9 speedup from parallel transport reduces a complete 1024-qubit rearrangement to about 10 ms, allowing many rearrangement attempts within the reported vacuum-limited atom lifetime.","Because the beam shaper and microlens array are separate modules, larger registers can be targeted by increasing laser power and redesigning the shaper rather than by re-engineering the trap-array optics."],"supporting_citations":[{"why":"Introduces the microlens-array micro-optical approach that this paper scales up.","marker":"[5]"},{"why":"Supplies the model-based real-time synthesis of acousto-optic tweezer patterns used for intensity-matched multi-tweezer transport.","marker":"[19]"},{"why":"Provides the prior supercharged 1000-qubit MLA array and the 99.8% imaging fidelity assumed for defect certification.","marker":"[20]"},{"why":"Establishes reservoir-based deterministic loading that the supercharging step relies on.","marker":"[24]"},{"why":"Supplies the 99.9% transport-efficiency benchmark used in the projection toward near-unity single-shot filling.","marker":"[30]"},{"why":"Provides the continuous-operation large-scale atom-array context and motivates repeated assembly against finite lifetime.","marker":"[31]"},{"why":"Establishes the atom-by-atom 2D array assembly method that the parallelized row/column rearrangement builds on.","marker":"[51]"},{"why":"Establishes the 1D atom-by-atom assembly technique and the single-tweezer baseline for transport speed.","marker":"[52]"},{"why":"Documents the intermodulation and calibration limits of multi-tweezer acousto-optic operation that the real-time control solves.","marker":"[53]"}],"fun_headline_variants":["1024 neutral-atom qubits defect-free in 10 ms","Kiloqubit neutral-atom register built in under 10 ms","Fastest defect-free assembly of 1024 atom qubits","32×32 atom grid filled to 99.3% in milliseconds","Parallel tweezers assemble 1024 qubits 9× faster"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claim's load-bearing premise is that the fluorescence image used to certify the defect-free register detects each atom with the 99.8% single-atom fidelity measured in the authors' earlier work; if full-array detection is less reliable, the 99.3% filling and the defect-free single shots would overstate the true qubit count.","fun_headline_variants_meta":{"raw":{"variants":["1024 neutral-atom qubits defect-free in 10 ms","Kiloqubit neutral-atom register built in under 10 ms","Fastest defect-free assembly of 1024 atom qubits","32×32 atom grid filled to 99.3% in milliseconds","Parallel tweezers assemble 1024 qubits 9× faster"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000464,"raw_usage":{"total_tokens":2354,"prompt_tokens":1014,"completion_tokens":1340,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":630,"completion_tokens_details":{"reasoning_tokens":1247}},"tokens_in":630,"tokens_out":1340,"duration_ms":9343,"temperature":1.0,"reasoning_tokens":1247,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:18:55.035171+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-image the same 1024-site register twice without moving or losing atoms and count site-by-site disagreements; if the per-site detection fidelity over the full array is below the assumed 99.8%, the reported filling fraction and defect-free certification would need to be revised downward.","supporting_citations":[{"cited_title":"Dumke, M","cited_arxiv_id":null,"evidence_quote":"Introduces the microlens-array micro-optical approach that this paper scales up."},{"cited_title":"Mittenb¨ uhler, L","cited_arxiv_id":null,"evidence_quote":"Supplies the model-based real-time synthesis of acousto-optic tweezer patterns used for intensity-matched multi-tweezer transport."},{"cited_title":"Pause, T","cited_arxiv_id":null,"evidence_quote":"Establishes reservoir-based deterministic loading that the supercharging step relies on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the 99.9% transport-efficiency benchmark used in the projection toward near-unity single-shot filling."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the continuous-operation large-scale atom-array context and motivates repeated assembly against finite lifetime."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents the intermodulation and calibration limits of multi-tweezer acousto-optic operation that the real-time control solves."}],"review_version":1}