{"id":"4cb59a0d-5782-4e87-ae6a-04aa60d09fcd","arxiv_id":"2502.10508","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A multi-amplifier sensing CCD was read out with two new modes, reaching about 0.45 electrons of noise in a region of interest.","lead":"This paper tests two faster ways to read out a new type of CCD camera that can count single photons or single electrons. The methods cut wasted readout time and still reach sub-electron noise in a small target region.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 0.45e- ROI noise figure may be biased low by estimating the median bias from the same dark frames used to evaluate the noise; the paper reports no independent residual-noise check.","rationale":"The paper's most important claim is the demonstration of sub-electron readout noise in a ROI at ~0.45e-. The reader's weakest assumption identified the median bias subtraction and the independence/equal-variance assumptions in Eq. (2) as the key reliance. I agree that the bias subtraction is the load-bearing point, but I go further: the paper does not describe the noise measurement procedure at all, which raises a concrete risk that the same dark frames are used both to build the bias and to evaluate the noise, biasing the residual variance low. This is an overfitting effect independent of whether the bias is visually clean. The reader's secondary points about the missing baseline comparison for the continuous mode and missing error bars are valid but less central. My proposed split-half or leave-one-out test would settle whether the 0.45e- number is robust. Since the paper can still be accepted if the authors provide this independent residual-noise measurement, the conditional verdict remains appropriate; no change to the reader's recommendation is needed.","tokens_in":9400,"tokens_out":10727,"duration_ms":99481,"concrete_test":"Acquire (or re-analyze) a set of dark frames taken with the ROI readout (40 samples, 7 amplifiers). Split the frames into two groups. Construct the median bias image from group A, subtract it from every frame in group B, and compute the pixel-by-pixel standard deviation in the ROI of group B. Compare this fully independent residual noise to the reported 0.45e- and to the Eq. (2) prediction. A leave-one-out variant is also informative: for each frame, subtract the median bias of the remaining N-1 frames and compute the standard deviation of the held-out residuals. If the independently evaluated noise stays within about 10% of 0.45e-, the bias-subtraction concern is resolved; if it is substantially higher, the headline number is an artifact of the estimation procedure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim of Section 4 is the sub-electron ROI noise of about 0.45e-. This value is obtained after subtracting a pixel-by-pixel median bias image built from 20 dark frames, but the paper never states how the 0.45e- number was measured, how many frames were used for the evaluation, or whether those frames were the same darks used to construct the bias. If the bias and the noise are estimated from the same stack, the median is not independent of the residual fluctuations; it absorbs a fraction of the frame-to-frame variance, biasing the measured standard deviation low by a factor that can approach sqrt(1-1/N) for N frames. In addition, the clock-induced baseline shifts are assumed to be identical between the bias darks and the data frames. This is supported only by visual inspection ('no evident residuals'), not by a numerical residual map or a split-half validation. Either effect could make the reported noise lower than the true readout noise in a real ROI exposure, directly undermining the headline claim of 0.45e- sub-electron readout. The independence and equal-variance assumptions of Eq. (2) are secondary: the agreement with the predicted 0.448e- is reassuring but cannot substitute for an independent measurement of the residual noise after bias subtraction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript reports two optimizations of the readout of Multi-Amplifier Sensing Charge-Coupled Devices (MAS-CCDs): a continuous readout that eliminates pre-scan overhead by dumping consecutive rows into the serial register before the previous row has fully cleared, and a region-of-interest (ROI) readout that concentrates multiple non-destructive samples per amplifier in a targeted area. For the ROI mode, the authors quote a readout noise of approximately 0.45e- for Ns=40 samples and Na=7 amplifiers, and they state that this is consistent with Eq. (2) when starting from a measured single-sample noise of 7.5e-. For the continuous mode, they report a per-pixel standard deviation around 50 ADU and conclude that the method has no negative impact on readout noise. The paper is an experimental demonstration with limited detail on the noise-evaluation methodology.","tokens_in":9791,"tokens_out":8493,"duration_ms":77957,"significance":"If fully validated, the ROI readout would be a meaningful step toward high-speed sub-electron imaging with thick fully depleted silicon sensors, relevant to quantum imaging, astronomical spectroscopy, and rare-event searches. The paper's strength is that the ROI noise is compared with the closed-form expectation of Eq. (2) using a separately measured single-sample noise, which is a testable consistency check rather than a hidden fit. However, the lack of a specified measurement procedure, missing uncertainties, and the reliance on visual inspection for residual baselines make the central quantitative claim currently under-supported. The results are promising, but the experimental validation needs to be strengthened before the claims can be accepted at face value.","major_comments":[{"comment":"The manuscript does not state how the quoted 0.45e- ROI noise was measured, how many images were used for the evaluation, or whether those images are the same frames used to construct the 20-image median bias. If the noise is evaluated on the same stack that defines the bias, the median absorbs a fraction of the frame-to-frame variance and biases the measured standard deviation low; even in the simplest case of N=20 frames the bias is about 2.5%, and it can be larger if the clock-induced baseline shifts are correlated between frames. More importantly, the statement that there are 'no evident residuals of the baseline fluctuations' is based on visual inspection only. Provide an independent residual-noise measurement, for example by building the bias from one half of the dark frames and evaluating the noise on the other half, and report the statistics of the residual map.","section":"Section 4, Figure 7"},{"comment":"The claimed agreement between the measured 0.45e- and the expected 0.448e- computed from Eq. (2) with Ns=40, Na=7, and a single-sample noise of 7.5e- is quoted without any uncertainties on either quantity. With two-significant-figure inputs, the agreement cannot be assessed. The authors should report the statistical uncertainty on the per-pixel standard deviation, the systematic uncertainty from the pixel-gain calibration, the uncertainty on the single-sample noise, and, if possible, a check that the amplifier noises are actually independent and equal-variance as Eq. (2) assumes.","section":"Section 4, Eq. (2)"},{"comment":"The paper states that the seventh amplifier 'was not giving any charge signal, so we decided to directly exclude the channel from the analysis without performing a thorough debugging of the electronic chain.' This post hoc exclusion makes the demonstration effectively a seven-amplifier result, not an eight-amplifier architecture, and it leaves open the possibility that the failed channel indicates a broader systematic issue. Please explicitly state that Na=7 was used in all quoted noise values, and provide evidence that the remaining seven amplifiers have consistent noise and gain, for example by listing the single-sample noise measured per amplifier.","section":"Section 4"},{"comment":"The conclusion that the continuous readout 'has proven to have no negative impact on the readout noise' is not supported by the presented evidence. The histogram of per-pixel standard deviations with a mean around 50 ADU is not compared with the corresponding noise of a standard readout of the same sensor, and the ADU-to-electron conversion is not given in the text. Please provide a quantitative comparison between the continuous and standard readout modes and state the calibration used to convert ADU to electrons.","section":"Section 3, Figures 4(e)-(f), Conclusions"}],"minor_comments":[{"comment":"In the abstract and Section 1, 'This sensor is called the Multiple-Amplifier Sensing Charge-Coupled Device (MAS-CCD) can perform...' should read '...which can perform...'; similarly, 'The second technique explore' should be 'The second technique explores'.","section":"Abstract and Section 1"},{"comment":"Figure 4's caption contains two entries labeled (d) and the in-text references do not match the caption; please renumber the panels (a)-(f) consistently.","section":"Figure 4 caption"},{"comment":"In Section 3, 'This also allows for a virtual over-scan with the same number o pixels' should be 'number of pixels'.","section":"Section 3"},{"comment":"In Section 3, 'The first row in each image, contains empty pixels' has an unnecessary comma.","section":"Section 3"},{"comment":"In Section 4, 'The later technique was explored in the past' should be 'The latter technique'.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is written in the style of a conference proceedings and would benefit from a fuller description of the noise-measurement procedure before it can be considered a journal-level demonstration of sub-electron ROI readout. The core idea is credible and the agreement with Eq. (2) is encouraging, but the missing experimental details and uncertainties are exactly the kind of thing that a careful referee should insist on. I do not see a concern about citation or novelty; the references to prior MAS-CCD work are appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main thing to know: this is a credible, short paper from the MAS-CCD group showing two readout tricks—continuous readout that overlaps row transfers to skip pre-scan pixels, and ROI readout that concentrates multi-amplifier averaging on a target sub-array. The central claim, sub-electron noise around 0.45e- in the ROI with 40 samples across seven amplifiers, is consistent with the standard sigma0/sqrt(Ns*Na) formula, and the single-photon image at 1.2e- mean is a nice demonstration. The continuous mode's combined 16-channel noise of about 1e- is what you'd expect. For people trying to speed up Skipper-class devices, this is useful.\n\nWhat is new: the continuous readout mode is not in the earlier MAS-CCD papers I know; the ROI extension to multiple amplifiers is a natural step but new for this architecture. The paper is clearly written, and the authors are transparent about the seventh amplifier being excluded without a full debug.\n\nSoft spots: the paper never states how the 0.45e- number was measured—how many frames, what region, and whether the same dark frames used to build the median bias were also used for the noise estimate. If they are the same stack, the median absorbs some variance and biases the result low by a factor like sqrt(1-1/N). The text reads as though separate darks were used, but it doesn't say so, and there is no independent residual-noise map or split-half check. That is the main thing I want fixed. The agreement with Eq. (2) is a consistency check, not an independent verification. The continuous readout conclusion says 'no negative impact on readout noise,' but the evidence is a histogram with mean around 1e- and no direct before/after comparison; plausible, but under-supported. Missing error bars on the headline noises is a minor presentation issue. The amplifier exclusion is a real caveat but not damning.\n\nBottom line: for detector people this deserves a serious referee. The core idea holds, the numbers are roughly consistent, and the weak points are missing details rather than load-bearing errors. I would send it out and ask for an explicit noise-evaluation description plus a residual-noise check after bias subtraction.","headline":"A useful, incremental MAS-CCD readout paper with a headline ROI noise number that needs a few missing measurements spelled out.","tokens_in":10291,"tokens_out":3061,"would_cite":true,"duration_ms":29256,"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":"Two readout modes for multiple-amplifier CCDs cut wasted time and still reach sub-electron noise in a targeted region.","keywords":["MAS-CCD","Skipper CCD","sub-electron readout noise","region of interest readout","continuous readout","non-destructive readout","single-photon counting","fully depleted CCD"],"falsifier":"Measure the per-pixel standard deviation from a stack of dark images in the columns where the readout switches between one sample and 40 samples; if the noise there exceeds the averaging-law prediction after bias subtraction, the bias correction is adding noise, and if the covariance between any two amplifiers is nonzero, the square-root gain from combining amplifiers is optimistic.","tokens_in":9236,"feed_emoji":"📷","tokens_out":6049,"duration_ms":53603,"temperature":0.7,"pith_summary":"This paper reports two readout strategies for the Multiple-Amplifier Sensing Charge-Coupled Device (MAS-CCD), a silicon imager whose serial register holds several non-destructive floating-gate amplifiers in a row. The first strategy, continuous readout, dumps each new row into the serial register before the previous row has fully cleared the amplifier chain, eliminating the time spent on pre-scan pixels. The second, region-of-interest readout, takes many samples per pixel in a targeted sub-area and only one sample elsewhere. The authors show that the first method does not degrade readout noise, and that the second reaches about 0.45 electrons of noise with 40 samples per amplifier combined across seven amplifiers, making a projected signal of 1.2 electrons per pixel clearly visible. These modes matter because they attack the main drawback of Skipper-CCD technology, its slow readout, without giving up single-quantum sensitivity.","feed_headline":"Sub-electron noise in a CCD region while the rest reads fast","feed_subtitle":"New readout modes for multiple-amplifier CCDs cut wasted time and reveal a 1.2-electron-per-pixel signal.","key_machinery":"The load-bearing object is the MAS-CCD output stage: a chain of amplifiers (8 or 16 in these sensors) capacitively coupled to the same serial register, so the same charge packet can be measured non-destructively by each amplifier while still being transferred onward. Pixel values are formed by averaging all samples from all amplifiers, giving a noise that falls as the inverse square root of the product of the number of amplifiers and the number of samples per amplifier, provided the amplifiers have independent, equal-variance noise. Continuous readout is a timing scheme that overlaps rows in the serial register; region-of-interest readout is a per-pixel variation of the number of samples. Both techniques are carried by the same non-destructive measurement mechanism, and both rely on a median bias image to absorb the baseline offsets that the clock changes introduce.","core_discovery":"Sub-electron readout noise can be obtained in a MAS-CCD without paying the readout-time penalty that limits Skipper-CCD use. In continuous readout, the next row is loaded into the serial register while the previous row is still being measured, so no time is spent on pre-scan pixels; after combining sixteen channels the per-pixel standard deviation is about one electron, which is the expected value at that sampling. In region-of-interest mode, a 360 by 270 pixel region is read with 40 non-destructive samples per amplifier while the rest of the image uses a single sample per amplifier. Combining the seven working amplifiers yields about 0.45 electrons of noise in that region, matching the averaging law with a single-sample noise of 7.5 electrons, and a projected object averaging 1.2 electrons per pixel becomes visible. The baseline offsets caused by the clock changes are removed with a median bias image, leaving no evident residue.","pith_inferences":["Combining continuous readout with region-of-interest sampling would attack both overheads at once; the paper demonstrates each separately but does not test the joint mode.","Because each amplifier sees the same pixel through a different gain, region-of-interest images could be used to measure inter-amplifier gain variations per pixel, a calibration the paper does not report.","A quantitative map of the residual noise after median bias subtraction, rather than visual inspection, would show whether the clock-transition columns pay any hidden noise cost.","The independence assumption behind the averaging law could be checked directly by computing the covariance between amplifier pairs on dark images; a nonzero correlation would make the square-root gain in amplifier count optimistic."],"forward_implications":["The same target noise can be reached in roughly the square root of the number of amplifiers less time than with a single-amplifier Skipper CCD, because each amplifier contributes an independent measurement of the same charge.","Continuous readout removes the pre-scan overhead for all rows except the first, so full-frame images cost less readout time with no measured noise increase.","Region-of-interest readout lets an observer preview an image at low resolution and then integrate only where a faint object sits, keeping the rest of the field fast.","The combination of seven amplifiers and 40 samples per pixel resolves an average signal of 1.2 electrons per pixel, demonstrating single-quantum sensitivity in a thick, fully depleted sensor.","The noise in the region of interest matches the averaging prediction, so the two techniques preserve the noise-scaling advantage that motivated the MAS-CCD design."],"supporting_citations":[{"why":"Introduces the non-destructive floating-gate readout concept that the MAS-CCD extends.","marker":"[8]"},{"why":"Provides the fully depleted, high-resistivity silicon device technology used for the sensors.","marker":"[9]"},{"why":"Demonstrated sub-electron readout noise in a Skipper CCD, setting the benchmark this work compares against.","marker":"[12]"},{"why":"Demonstrated single-electron and single-photon sensitivity with a Skipper CCD, motivating the noise target.","marker":"[13]"},{"why":"Describes the Low Threshold Acquisition controller used to read out the sensors.","marker":"[14]"},{"why":"Presents the prior MAS-CCD single-quantum measurement results and the architecture used here.","marker":"[26]"},{"why":"Characterizes a sixteen-amplifier sensing CCD, the source of the 16-channel sensor in this work.","marker":"[27]"},{"why":"Introduced the region-of-interest smart-readout technique for Skipper-CCDs that is extended here to the MAS-CCD.","marker":"[29]"}],"fun_headline_variants":["MAS-CCD sub-electron noise without the readout time overhead","Parallel CCD readout: sub-electron noise, no wasted time","Sixteen amplifiers combine to see single electrons","Region-of-interest CCD readout resolves 1.2-electron signals"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The noise formula assumes the working amplifiers have independent, equal-variance noise and that the median bias image removes all clock-induced baseline shifts without adding any noise; if either fails, the reported 0.45-electron noise would not be as low as claimed.","fun_headline_variants_meta":{"raw":{"variants":["MAS-CCD sub-electron noise without the readout time overhead","Parallel CCD readout: sub-electron noise, no wasted time","Sixteen amplifiers combine to see single electrons","Region-of-interest CCD readout resolves 1.2-electron signals"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000575,"raw_usage":{"total_tokens":2775,"prompt_tokens":1064,"completion_tokens":1711,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":680,"completion_tokens_details":{"reasoning_tokens":1642}},"tokens_in":680,"tokens_out":1711,"duration_ms":13135,"temperature":1.0,"reasoning_tokens":1642,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T18:18:06.871797+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the per-pixel standard deviation from a stack of dark images in the columns where the readout switches between one sample and 40 samples; if the noise there exceeds the averaging-law prediction after bias subtraction, the bias correction is adding noise, and if the covariance between any two amplifiers is nonzero, the square-root gain from combining amplifiers is optimistic.","supporting_citations":[{"cited_title":"Ultra low-noise charge coupled device,","cited_arxiv_id":null,"evidence_quote":"Introduces the non-destructive floating-gate readout concept that the MAS-CCD extends."},{"cited_title":"Fully depleted, back-illuminated charge-coupled devices fabricated on high-resistivity silicon,","cited_arxiv_id":null,"evidence_quote":"Provides the fully depleted, high-resistivity silicon device technology used for the sensors."},{"cited_title":"Sub-electron readout noise in a Skipper CCD fabricated on high resistivity silicon,","cited_arxiv_id":null,"evidence_quote":"Demonstrated sub-electron readout noise in a Skipper CCD, setting the benchmark this work compares against."},{"cited_title":"Single-electron and single-photon sensitivity with a silicon Skipper CCD,","cited_arxiv_id":null,"evidence_quote":"Demonstrated single-electron and single-photon sensitivity with a Skipper CCD, motivating the noise target."},{"cited_title":"Low threshold acquisition controller for Skipper charge-coupled devices,","cited_arxiv_id":null,"evidence_quote":"Describes the Low Threshold Acquisition controller used to read out the sensors."},{"cited_title":"Single-quantum measurement with a multiple-amplifier sensing charge-coupled device,","cited_arxiv_id":null,"evidence_quote":"Presents the prior MAS-CCD single-quantum measurement results and the architecture used here."},{"cited_title":"A sixteen multiple-amplifier-sensing ccd and characterization techniques targeting the next generation of astronomical instruments,","cited_arxiv_id":null,"evidence_quote":"Characterizes a sixteen-amplifier sensing CCD, the source of the 16-channel sensor in this work."},{"cited_title":"Smart-readout of the skipper-ccd: Achieving sub-electron noise levels in regions of interest,","cited_arxiv_id":null,"evidence_quote":"Introduced the region-of-interest smart-readout technique for Skipper-CCDs that is extended here to the MAS-CCD."}],"review_version":1}