{"id":"ad104e79-1323-4a75-985c-4e507130bc66","arxiv_id":"2608.07676","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A model-based feasibility study claims a ~100 g CMOS image sensor array at a reactor can detect CEvNS within 7 to 50 days using fast frame rates and active shielding.","lead":"This paper proposes using CMOS image sensors as compact neutrino detectors at nuclear reactors, and it argues that a roughly 100-gram silicon detector could see coherent neutrino-nucleus scattering in under 50 days. It matters because reactor neutrino detection is currently done with much larger or slower detector technologies, and CMOS sensors could combine low energy thresholds with fast, active background rejection.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"IDB cluster model assumes spatially independent pixels; correlated readout/dark-current noise at 1000 fps could raise the low-energy background and push the 17-day H1 claim.","rationale":"The paper proposes a concrete CIS detector concept and derives detection times via a self-consistent signal/background model. I agree with the reader's CONDITIONAL verdict: the Table I parameter set is not yet demonstrated in one device, and the 100 DRU flat background is an assumption. My stress-test identifies an additional, more internal gap in the IDB model: the cluster probabilities in Appendix A factorize as products of single-pixel activation probabilities, which assumes no spatial correlations. High-frame-rate CIS readout is known to have correlated row/column noise, and dark current is non-uniform (hot pixels); both violate this assumption. Since the optimal energy threshold (0.065 keVee) is chosen specifically to suppress the IDB contribution below the 100 DRU physical background, a correlated-noise term could shift that threshold upward and directly lengthen the claimed detection times. This does not refute the concept; it means the quantitative claim requires prototype-level validation of the noise covariance and dark-current uniformity before the 17-day (or 7-day) figure can be taken as more than an idealized projection. The reader's weakest_assumption (combination of parameters) is valid and related; my concern is complementary. Therefore the verdict remains CONDITIONAL.","tokens_in":19909,"tokens_out":22569,"duration_ms":233891,"concrete_test":"Build or use a prototype Skipper-CMOS/CIS (or the cited qCMOS/DEPFET devices) at Table I parameters; record raw frames at 1000 fps with the sensor dark (shuttered) at -20 C. Measure the pixel-pixel covariance matrix and dark-current map. Generate synthetic frames using the measured covariance and repeat the Appendix A cluster analysis (or a Monte Carlo cluster finder) for the [0.065,0.158] keVee window. If the correlated-noise IDB rate exceeds 100 DRU, or if it raises the required E_min above 0.072 keVee, the Table II times are invalid. Equivalent analytic test: inject a tunable row-correlated noise of 0.5 e- RMS into the independent-pixel model and check how the E_min for IDB < 100 DRU shifts.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The detection times in Table II (17 days for H1 at 1 e- noise) rely on the Appendix A event-cluster model, which computes the probability of a multi-pixel event as products of single-pixel activation probabilities: n_s = Σ N_{s,v} p^s q^{v(s)} (Eqs. A2-A6). This factorized form is exact only if pixel values are spatially uncorrelated. At 1000 fps with parallel readout, CIS readout noise contains row/column common-mode and power-supply correlated components, and the dark current has hot pixels and defect clusters. Even if the quoted 1 e- σ_RN is the total per-pixel RMS after correction, a residual row-correlated component of, say, 0.3-0.5 e- makes simultaneous threshold crossings in the same row far more likely than p^s, so the true IDB rate near the 0.065 keVee (~17 e-) analysis threshold can be orders of magnitude above Fig. 8. The paper neither states the independence assumption as a limitation nor adds a covariance term. If the true IDB exceeds the flat 100 DRU physical background in the optimal window, E_min must be raised, reducing signal acceptance and increasing the 17-day (and 7-day for 0.2 e-) detection times.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript proposes the use of CMOS image sensors (CIS) for detecting reactor coherent elastic neutrino-nucleus scattering (CEvNS). It develops an active-shielding strategy based on rejecting frames coincident with muon triggers, and an analytical 'polyking' model for intrinsic detector backgrounds from dark current and readout noise. Using a flat 100 DRU physical background and a ~100 g detector with the parameters in Table I (1 Mpix, 15 µm pixels, 725 µm thickness, 1 or 0.2 e- readout noise, 1000 fps), the authors compute 3σ detection times with the Asimov significance formula: 17 days for H1 = 0.014 GW_th/m² at 1 e- noise and about 7 days at 0.2 e- noise. The signal calculation uses the Huber antineutrino flux, the Standard Model CEvNS cross section, and a measured quenching parameterization.","tokens_in":20100,"tokens_out":16800,"duration_ms":149139,"significance":"If the stated assumptions hold, the paper provides the first quantitative sensitivity study for CEvNS with CIS and identifies a potentially compact, fast detector for reactor monitoring and safeguards. Its strengths include the use of standard external inputs (Huber flux, SM cross section, measured quenching), a transparent analytical background model with concrete falsifiable predictions, and a clear presentation of the significance calculation. The active-shielding-by-frame-rejection concept is a useful adaptation of rare-event techniques to high-frame-rate pixel sensors. However, the central detection-time claims rest on three unvalidated assumptions: the absence of correlated readout/dark-current noise in the polyking model, the simultaneous technological feasibility of all Table I parameters, and a flat 100 DRU physical background. These issues make the headline numbers conditional, but the framework itself is sound and the paper could become convincing with targeted additions.","major_comments":[{"comment":"The polyking cluster probabilities are factorized as products of single-pixel activation probabilities, p^s q^{v(s)}, which is exact only if pixel values are spatially independent. At 1000 fps with column-parallel readout, CIS noise typically contains row/column common-mode and power-supply correlated components, and dark current is spatially correlated through hot pixels and defect clusters. A residual correlated component of even 0.3-0.5 e- RMS would make simultaneous threshold crossings far more likely than p^s, potentially raising the IDB by orders of magnitude at the 0.065 keVee analysis threshold and thereby invalidating the 17-day (H1, 1 e-) and 7-day (0.2 e-) detection times in Table II. The manuscript neither states the independence assumption as a limitation nor adds a covariance term or a bound on the allowed correlated component; this must be addressed, for example by extending the model with a correlation term or by validating it on measured CIS dark-frame noise images.","section":"Section II.C / Appendix A, Eqs. (A2)-(A6)"},{"comment":"The detector parameters in Table I are drawn from separate demonstrations rather than from a single device. The 725 µm fully depleted thickness is demonstrated in a CMOS-DEPFET operating at ~9.8 e- noise and 1.1 MHz; sub-electron noise is shown in a 64x64 DEPFET array or in Skipper-CMOS with 18 µm epitaxial thickness; no cited sensor simultaneously achieves 1 Mpix, 15 µm pixels, 1000 fps, sub-electron noise, and a dark current of 1.5 e-/pix/s. The statement that the design is 'technologically feasible' is therefore an extrapolation. Because the headline detection times are conditional on the simultaneous realization of all Table I parameters, the paper should either cite a device that meets all specifications or clearly reframe the results as a parametric sensitivity study with a technology-development roadmap.","section":"Section IV / Table I"},{"comment":"The physical background is assumed to be a flat 100 DRU, justified by Skipper-CCD Compton measurements and by the expectation that active shielding removes the neutron-induced exponential low-energy component. No measurement exists for the proposed CIS technology in the sub-100 eV region, and a flat spectrum is not guaranteed: surface backgrounds, residual Cherenkov photons, or gamma-induced events could introduce a non-flat spectral shape. Since the detection times in Table II scale directly with the background rate, the paper should include a sensitivity scan over the background rate and spectral shape (e.g., 30, 100, and 300 DRU, optionally with an exponential low-energy component) to show how the 17-day and 7-day claims would change if the background assumption is violated.","section":"Section II.B / Sections III.B and IV"}],"minor_comments":[{"comment":"The statement that the active-shielding inefficiency 'increases the observation time by 20%' is only true for the 1000 fps case; at 200 fps the dead-time fraction is about 63%, which is correctly reflected in Table II but not in the text. Please specify the frame rate when quoting this number.","section":"Section IV"},{"comment":"The energy resolution uses sigma_RN = 2 e- for all scenarios, even when the detector readout noise is taken as 0.2 e- or 1 e- in Table II. This is a conservative choice, but it is inconsistent with the detector parameters used elsewhere and could underestimate the signal acceptance for the low-noise configurations; a justification or a scan over this value would be helpful.","section":"Appendix B, Eq. (B6)"},{"comment":"The derivation would be clearer if the probability for a given total charge Q were written explicitly as a sum over partitions of the product of per-pixel charge probabilities; Eqs. (A2)-(A6) alone give only the probability of a cluster of a given size and do not show how the charge distribution enters the calculation.","section":"Appendix A"},{"comment":"The muon rate of 200 Hz is adopted as 'representative' without an explicit scaling from the veto geometry; since the live-time calculation depends on this rate, a more detailed estimate or a range of plausible values would strengthen the analysis.","section":"Section II.B"},{"comment":"References [8] and [66] are the same article, and [10] and [65] also appear to be the same work; please consolidate to avoid duplicate citations.","section":"References"},{"comment":"The saw-tooth structure in the left panel is not explained in the text; a sentence describing its origin would help the reader interpret the model behavior.","section":"Figure 3"},{"comment":"The phrase 'set at the value to 0.215 keV' should be reworded to 'set to the value of 0.215 keV' for clarity.","section":"Section III.A"}],"recommendation":"major_revision","confidential_remarks":"The paper is a conceptual sensitivity study for a not-yet-built detector. The main risk to the central claim is the unstated independence assumption in the polyking background model; if correlated readout noise is present at the level commonly seen in high-speed CIS, the quoted detection times could be optimistic by a significant factor. The paper would be substantially strengthened by a measurement-based validation of the IDB model and by framing the feasibility of the Table I parameters as a roadmap rather than an existing device. I support publication after the major comments are addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look, but read the numbers as projections from a parameter list that has never been assembled in one device. The genuinely new piece is the combination of fast CIS frame readout with a Poisson veto—reject frames likely to contain a muon—and the polyking-based analytic model for intrinsic detector background from dark current and readout noise. That model is real and internally consistent: up to size-4 clusters, it enumerates polyking geometries and produces event-rate spectra from single-pixel activation probabilities. The paper is also honest about what it has not done: Table I is explicitly described as technologically feasible, not demonstrated.\n\nThe physics inputs are standard and handled correctly: reactor flux, SM CEvNS cross-section, silicon quenching factor, Fano noise, and the Asimov significance formula. The figures and Table II follow directly from the stated model, and the exposure scaling is clean. The paper deserves referee time.\n\nThe soft spots are the assumptions carrying the central claim. First, the 100 DRU flat physical background is an extrapolation from Skipper-CCD and CONUS measurements, not a measured CIS background at this threshold and frame rate. Second, the polyking model factorizes single-pixel activation probabilities (Eqs. A2–A6), which assumes spatially uncorrelated pixel outputs. At 1000 fps with parallel readout, row/column common-mode and hot pixels are the norm; a residual correlated component of a few tenths of an electron could multiply the low-energy IDB rate and push the threshold back up, lengthening the 17-day claim. The paper neither states the independence assumption nor adds a covariance term. That is the load-bearing caveat. Minor: Appendix B quotes sigma_RN = 2 e- for the signal smearing while Table I and II use 1 e- and 0.2 e-; probably conservative, but inconsistent.\n\nThe 17-day and 7-day numbers should be read as conditional sensitivity projections, which is what the paper mostly does. The active-shielding idea and the IDB model are worth preserving. I would send it to peer review, with the expectation that the authors be pressed on the covariance issue and on obtaining a measured low-energy CIS background. The right audience is the detector and reactor-neutrino community; I would cite this if I were writing a proposal on compact CEvNS monitors.","headline":"A coherent feasibility study with a genuinely new frame-veto idea and a real analytic IDB model; the discovery-time numbers are projections that hinge on unvalidated parameter combinations and a spatial-independence assumption.","tokens_in":20709,"tokens_out":2651,"would_cite":true,"duration_ms":29395,"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":"A ~100 g CMOS image-sensor array could detect reactor coherent neutrino scattering at 3σ in about 17 days.","keywords":["CEvNS","CMOS image sensors","reactor neutrinos","coherent elastic neutrino-nucleus scattering","intrinsic detector background","polykings","readout noise","active shielding"],"falsifier":"Build a prototype with the Table I specifications and place it at a reactor site with a measured muon veto; if the measured dark current exceeds 1.5 e-/pix/s, the readout noise at 1000 fps exceeds 1 e-, the sensitive thickness falls below 725 microns, or the event rate in [0.065, 0.158 keVee] exceeds the predicted CEvNS signal plus the 100 DRU flat background, then the claimed 17-day detection time is falsified.","tokens_in":19624,"feed_emoji":"📷","tokens_out":7740,"duration_ms":69100,"temperature":0.7,"pith_summary":"This paper argues that a compact detector built from commercial CMOS image sensors can observe reactor CEvNS, the coherent scattering of antineutrinos off whole silicon nuclei, in weeks rather than years. The authors claim that with roughly 100 g of sensitive silicon, one-electron readout noise, 1000 frames per second, and a flat 100 DRU background, a 3-sigma detection is reachable in 17 days at about 16 meters from a 3.6 GW thermal reactor; with sub-electron (0.2 e-) noise the time falls to about 7 days. The reason this matters is that CEvNS is a Standard-Model process that is also a sensitive probe of new physics, and compact silicon detectors could make reactor-neutrino monitoring practical at surface sites. The paper supplies an active-shielding scheme based on discarding frames that coincide with cosmic-muon triggers, and an analytical model of the detector's own dark-current and readout-noise background.","feed_headline":"A 100-gram CMOS camera could spot reactor neutrinos in 17 days","feed_subtitle":"Fast, low-noise silicon pixels plus muon-frame vetoing could make compact reactor monitoring practical.","key_machinery":"The argument rests on an analytical model of intrinsic detector background (IDB) that treats dark current as a Poisson process per pixel and readout noise as Gaussian, then convolves the two into a per-pixel probability density. To turn pixel probabilities into event rates, the model counts connected clusters of pixels above a four-sigma threshold using polykings, the shapes a cluster of up to four pixels can make on a square lattice, so the probability of a spurious n-electron event is a polynomial in the per-pixel activation probability and the number of empty neighbors. A second piece is the CEvNS signal calculation, which convolves the reactor antineutrino spectrum with the Standard-Model coherent cross section, applies a silicon quenching factor and a Fano-limited Gaussian resolution, and evaluates discovery significance with the profile-likelihood ratio formula for small Poisson counts. The active-shielding machinery uses Poisson statistics of cosmic muons: at 1000 fps, a frame coincident with a muon trigger is rejected, leaving more than 90 percent live time while removing muon-induced neutron backgrounds.","core_discovery":"The central claim is that CIS technology, long used for cameras, can be repurposed as a low-threshold, high-frame-rate neutrino detector, combining the single-electron sensitivity previously demonstrated in silicon sensors that read each pixel multiple times with the temporal resolution that serial-readout devices lack. The authors show that the combination of 725-micron fully depleted silicon, 1 Mpix, 15-micron pixels, about 100 g mass in 256 sensors, 1 to 0.2 electron readout noise, and 1000 fps yields an instrumental background that can be pushed above a 0.064 keVee threshold, leaving a window near [0.065, 0.158 keVee] where the CEvNS signal exceeds the 100 DRU flat physical background. In the strong-flux scenario H1 = 0.014 GWth/m2, the profile-likelihood significance reaches 3 sigma in 17 days, and in 7 days at 0.2 e- noise; in a lower-flux scenario H2, about ten times longer is needed. The paper presents this as evidence that a feasible CIS-based detector could deliver the first compact, surface-operable reactor CEvNS observation and enable reactor monitoring applications.","pith_inferences":["The polyking cluster-counting model applies to any pixelated low-threshold imager, not just CIS, so the same formalism could quantify spurious event rates from dark current and readout noise in other silicon detector families and give a common yardstick for comparing technologies.","The paper samples but does not solve the frame-rate tradeoff: for fixed dark current and readout noise, there is likely an optimal frames-per-second value that balances muon-veto dead time, per-frame dark-current accumulation, and readout-noise sampling.","If future devices reach the assumed dark current at lower temperature and sub-electron noise, the threshold could drop below 0.027 keVee, where the H1 signal rate is still substantial; the model would then need to include multi-pixel partial-charge-collection effects before claims at those thresholds solidify.","The 100 DRU flat background is an assumption carried over from other silicon and germanium measurements; a dedicated background measurement with a thick CIS would be the natural next step, since the IDB model alone cannot validate that number."],"forward_implications":["At H1 = 0.014 GWth/m2, a 100 g CIS detector with 1 e- noise, 1.5 e-/pix/s dark current, and 1000 fps reaches 3-sigma CEvNS detection in 17 days; at 200 fps the same detector needs about 44 days.","Lowering readout noise to 0.2 e- shortens the 1000 fps detection time to about 7 days, because the energy threshold can drop from about 0.072 keVee to about 0.027 keVee while keeping the instrumental background below the signal window.","Frame rejection based on coincident cosmic-muon triggers keeps live time above about 90 percent at 1000 fps and suppresses the muon-induced neutron background that otherwise dominates surface reactor sites.","For reduced-flux or longer-baseline scenarios (H2 = 0.004 GWth/m2), the required observation time is roughly an order of magnitude larger, around 150 days for the reference detector.","At 1 e- noise and 200 fps, the detector is already suitable for reactor monitoring and low-energy response validation; sub-electron noise extends sensitivity to beyond-Standard-Model searches such as neutrino magnetic moments and light mediators."],"supporting_citations":[{"why":"Supplies the reference reactor-site observation of CEvNS with a germanium detector, establishing the experimental environment and the active-veto need that motivate the CIS approach.","marker":"[8]"},{"why":"Demonstrates single-electron and single-photon sensitivity in a silicon sensor, providing the low-threshold capability the CIS concept builds on.","marker":"[9]"},{"why":"Documents a commercial CMOS camera with sub-electron readout noise and low dark current, supporting the feasibility of the assumed detector parameters.","marker":"[29]"},{"why":"Demonstrates 0.2 e- nondestructive readout in a pixelated silicon device, supporting the sub-electron-noise assumption used in the fast detection scenarios.","marker":"[47]"},{"why":"Introduces nondestructive sub-electron readout implemented in a CMOS image sensor process, the key building block for combining low threshold with parallel readout.","marker":"[49]"},{"why":"Provides the measured muon rate and active-veto performance at a reactor site that the frame-rejection live-time calculation uses.","marker":"[55]"},{"why":"Shows flat low-energy background behavior above ground with low-threshold silicon sensors, supporting the 100 DRU flat-background assumption.","marker":"[58]"},{"why":"Gives the profile-likelihood formula used to compute median discovery significance for small Poisson counts in the single-bin counting experiment.","marker":"[69]"},{"why":"Defines the polykings or polyominoes used to enumerate connected-pixel background event shapes in the analytical IDB model.","marker":"[73]"},{"why":"Provides the low-energy silicon quenching-factor parameterization used to convert nuclear recoil energy to measured ionization energy.","marker":"[74]"}],"fun_headline_variants":["100g CMOS imager sees reactor neutrinos in 17 days","CMOS sensors turn cameras into reactor neutrino detectors","Reactor CEvNS with CMOS: 100g silicon, 17-day detection","CMOS image sensors: compact and fast for reactor neutrinos"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole timing estimate hangs on one premise: a single CMOS detector can be built that simultaneously has 725-micron fully depleted silicon, roughly one million 15-micron pixels, 1 (or 0.2) electron readout noise at 1000 frames per second, and a dark current of 1.5 electrons per pixel per second, even though each of these specifications has only been seen separately in different devices.","fun_headline_variants_meta":{"raw":{"variants":["100g CMOS imager sees reactor neutrinos in 17 days","CMOS sensors turn cameras into reactor neutrino detectors","Reactor CEvNS with CMOS: 100g silicon, 17-day detection","CMOS image sensors: compact and fast for reactor neutrinos"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001218,"raw_usage":{"total_tokens":5035,"prompt_tokens":994,"completion_tokens":4041,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":610,"completion_tokens_details":{"reasoning_tokens":3967}},"tokens_in":610,"tokens_out":4041,"duration_ms":25233,"temperature":1.0,"reasoning_tokens":3967,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T00:23:25.932951+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build a prototype with the Table I specifications and place it at a reactor site with a measured muon veto; if the measured dark current exceeds 1.5 e-/pix/s, the readout noise at 1000 fps exceeds 1 e-, the sensitive thickness falls below 725 microns, or the event rate in [0.065, 0.158 keVee] exceeds the predicted CEvNS signal plus the 100 DRU flat background, then the claimed 17-day detection time is falsified.","supporting_citations":[{"cited_title":"Bernstein, N","cited_arxiv_id":null,"evidence_quote":"Supplies the reference reactor-site observation of CEvNS with a germanium detector, establishing the experimental environment and the active-veto need that motivate the CIS approach."},{"cited_title":"Fernandez Moroni, P","cited_arxiv_id":null,"evidence_quote":"Demonstrates single-electron and single-photon sensitivity in a silicon sensor, providing the low-threshold capability the CIS concept builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents a commercial CMOS camera with sub-electron readout noise and low dark current, supporting the feasibility of the assumed detector parameters."},{"cited_title":"Aschauer, P","cited_arxiv_id":null,"evidence_quote":"Demonstrates 0.2 e- nondestructive readout in a pixelated silicon device, supporting the sub-electron-noise assumption used in the fast detection scenarios."},{"cited_title":"Hansen, H","cited_arxiv_id":null,"evidence_quote":"Introduces nondestructive sub-electron readout implemented in a CMOS image sensor process, the key building block for combining low threshold with parallel readout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the measured muon rate and active-veto performance at a reactor site that the frame-rejection live-time calculation uses."},{"cited_title":"Quinn, M","cited_arxiv_id":null,"evidence_quote":"Shows flat low-energy background behavior above ground with low-threshold silicon sensors, supporting the 100 DRU flat-background assumption."},{"cited_title":"Cowan, K","cited_arxiv_id":null,"evidence_quote":"Defines the polykings or polyominoes used to enumerate connected-pixel background event shapes in the analytical IDB model."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the low-energy silicon quenching-factor parameterization used to convert nuclear recoil energy to measured ionization energy."}],"review_version":1}