{"id":"3e0e575e-5444-4413-9906-c0f360adc01b","arxiv_id":"2508.14174","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A prototype MIT-LL CCD with a Stanford ASIC achieves 2.31 e- read noise and 121 eV FWHM at 5.9 keV while enabling readout up to 5 Mpixels/s.","lead":"Engineers report that a prototype X-ray CCD read with a custom ASIC reaches 2.31 electron read noise, 121 eV energy resolution at 5.9 keV, and 5 megapixel-per-second readout using new clock drivers and an automated bias search. The work supports future X-ray missions such as AXIS, which need fast, low-noise cameras for faint cosmic sources.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 2.31 e- read-noise and bias-optimization claim rest on an uncalibrated ADU-to-electron conversion; Table 3 shows the ADU-optimal bias worsens spectral FWHM at 173 K, suggesting the apparent noise improvement may be a gain artifact.","rationale":"The reader's weakest assumption identifies essentially the same load-bearing issue: the uncalibrated ADU-to-electron conversion and the use of ADU-domain noise as the optimization metric, which Table 3 shows can conflict with spectral FWHM. The paper is otherwise clearly written, the hardware development is plausible, and the speed/noise numbers are internally consistent in direction, but the absence of a gain calibration means the headline '2.31 e-' and the claimed benefit of the bias scan cannot be independently verified from the manuscript. The proposed check—measuring gain from the Fe-55 centroid at each bias point and recomputing the noise in electrons—would settle whether the optimization is real or a gain artifact. Since the reader already rendered CONDITIONAL with low confidence, and this concern is addressable rather than fatal, no change to the verdict is needed.","tokens_in":10403,"tokens_out":5041,"duration_ms":59845,"concrete_test":"Re-run the 173 K bias scan on a CCID-93, and at each of the 400 grid points measure both the overscan ADU standard deviation and the Fe-55 Mn K-alpha centroid to derive the conversion gain in e-/ADU. Convert the ADU noise to electrons using this gain, and take spectra at the ADU-optimal bias and at the default bias. If the electron-referred noise ranking changes, or if the ADU-optimal point still yields FWHM >= 120.8 eV while the default yields <= 118.5 eV, then the bias-optimization claim is not supported by the current data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central AXIS-performance claim depends on the reported read noise and on the bias-optimization procedure, but the conversion from overscan ADU scatter to electrons is never calibrated. Section 4 says the algorithm minimizes the standard deviation of the 50-column overscan in ADU and 'assum[es] that the relationship between ADU and electron noise is relatively well behaved over the range of parameters under test.' That assumption is directly tested by the paper's own Table 3: at 173 K, the optimized bias lowers the reported noise from 2.39 to 2.31 e- but increases the Fe-55 FWHM from 118.5 to 120.8 eV. If conversion gain varies across the RGH/RGL/OG/RD grid, then an ADU-noise minimum can simply select a low-gain operating point rather than a low-charge-noise one. Without a per-point gain measurement (e.g., from the Fe-55 peak centroid or a charge-injection transfer curve), the 2.31 e- value and the claimed advantage of the optimization over the default bias are not established. In addition, the 5 MPixels/s figure is reported separately from the low-noise spectral result: at 5 MHz the best noise in Table 2 is 3.84 e-, and no optimized FWHM at 5 MPixels/s is presented, so the summary's 'level of performance required' claim pairs numbers achieved at different operating points.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports the development and characterization of a fast, low-noise CCD readout chain intended for the AXIS mission: an MIT-LL CCID-93 CCD, a Stanford MCRC V1.0 readout ASIC, integrated fast summing-well and reset-gate clock drivers, and an automated scan over the RG, OG, and RD bias voltages. At a 2 Mpixels/s serial transfer rate and 173 K, using the optimized bias point, the authors quote a read noise of 2.31 e- and a FWHM of 121 eV at 5.9 keV. The onboard clock circuitry enables serial transfer rates up to 5 Mpixels/s, where the best reported (pre-optimization) noise is 3.84 e-. The bias-scan algorithm minimizes the standard deviation of a 50-column overscan region in ADU over a four-dimensional grid. The central claim is that the combination of detector, ASIC, clock drivers, and bias optimization achieves the speed and noise performance required for AXIS.","tokens_in":10751,"tokens_out":4333,"duration_ms":46422,"significance":"If the quantitative claims hold, this is a valuable engineering milestone: it demonstrates an ASIC-based, low-power, low-footprint readout chain for fast X-ray CCDs and a practical automated procedure for tuning multiple output-stage biases, which would be useful for populating 16 channels of the CCID-100. The paper is measurement-focused and provides statistical uncertainties on the reported noise and FWHM values. The main advertised advances—5 Mpixels/s clocking and an automated multi-dimensional bias optimization—are real hardware contributions. However, the numerical result that is central to the mission-readiness claim (2.31 e-, 121 eV at high speed) is not yet fully supported because the ADU-to-electron conversion is not calibrated and the speed and low-noise results are not demonstrated simultaneously. The significance of the paper as a mission-enabling demonstration is therefore conditional on additional measurements.","major_comments":[{"comment":"The conversion from overscan ADU scatter to electrons is never described. Section 4 states that the scan assumes the ADU-to-electron relationship is 'relatively well behaved over the range of parameters under test,' but no gain calibration (e.g., Fe-55 peak centroid vs. known charge, or a charge-injection transfer curve) is reported. All noise values in electrons therefore rest on an unstated conversion factor. The paper's own Table 3 tests the assumption: at 173 K, the optimized bias lowers the reported noise from 2.39 to 2.31 e- but increases the Fe-55 FWHM from 118.5 to 120.8 eV. If conversion gain varies across the RGH/RGL/OG/RD grid, an ADU-noise minimum can select a low-gain operating point rather than a low-charge-noise one. Without a per-point gain measurement, the 2.31 e- value and the claimed advantage of the optimization over the default bias are not established.","section":"§4, Tables 2 and 3"},{"comment":"The abstract and Summary claim that the 'level of performance required of the AXIS mission can be achieved,' but this appears to combine results obtained at different operating points. The 2.31 e- noise and 121 eV FWHM are measured at 2 Mpixels/s, while the 5 Mpixels/s capability is characterized separately. Table 2 shows that at 5 MHz the best noise is 3.84 e- (FWHM 125.8 eV), and the table note says those values were obtained 'prior to performing a bias parameter optimization.' No optimized FWHM or spectrum at 5 Mpixels/s is reported. The paper should state the quantitative AXIS requirement and demonstrate or explicitly bound the simultaneous speed/noise performance rather than pairing the fastest speed with the lowest noise from different configurations.","section":"§5, Table 2, Summary"},{"comment":"The optimization criterion is the overscan standard deviation in ADU, not the spectral energy resolution. At 173 K the optimized bias reduces the reported read noise by 0.08 e- but worsens the FWHM from 118.5 ± 1.7 eV to 120.8 ± 1.9 eV. While the FWHM change is modest and comparable to its uncertainty, the sign of the change is opposite to the noise trend, so the ADU-optimal point is not shown to deliver better spectral performance at the temperature used for the headline result. The authors should either report the Fe-55 FWHM at the optimized point as the primary metric, or justify why a 0.08 e- noise decrease is preferred over a 2.3 eV resolution degradation.","section":"§4, Table 3"}],"minor_comments":[{"comment":"Typo: 'Massachusets Institute of Technology' should be 'Massachusetts Institute of Technology.'","section":"Author affiliations"},{"comment":"Table headers contain a stray space: 'T emp (K)' should be 'Temp (K)'. Also, the notation for pixel rate is inconsistent: 'MPixels/s' and 'Mpixels/s' are both used.","section":"Tables 1-3"},{"comment":"The triangle plots would benefit from a color scale and a description of the contour color/white-space mapping. The text mentions 'gaps in the contours' where noise was unphysically small, but without axis labels and color bars it is difficult for the reader to compare the four temperatures.","section":"Figure 6 and Appendix A"},{"comment":"The phrase '∼20% recovery of the baseline and signal samples' is unclear: what is meant by 'recovery'—additional usable samples, improved settling, or something else? Please define.","section":"§3.1"},{"comment":"The abstract says 'simultaneous optimization of the output gate (OG), reset gate (RG), and reset drain (RD) biases,' but Section 4 scans four parameters (RGH, RGL, OG, RD). Since RG has high and low states, please clarify the terminology consistently.","section":"Abstract and §4"}],"recommendation":"major_revision","confidential_remarks":"This is a hardware-development paper for a well-defined mission. The central engineering measurements are plausible and internally consistent, and the heavy self-citation is appropriate for a device paper. The main obstacle is the missing gain calibration and the conflation of results at different speeds in the mission-readiness claim. These are fixable with additional measurements (gain calibration and a 5 Mpixels/s optimized spectrum) and a more careful wording of the requirements. I do not see a circularity or novelty problem; the bias-scan method is presented as a practical tuning tool, not as a new physical prediction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new things here are the integrated onboard SW/RG clock drivers and the four-parameter bias scan that steps over RGH, RGL, OG, and RD. Those are real engineering contributions, and the paper reports them clearly, with error bars and enough detail for someone in the field to reproduce the method on their own detector. The measured noise and FWHM numbers are internally consistent and plausible for this hardware; the 2.31 e- at 2 MPixels/s and the 5 MPixels/s capability, if both hold up, are directly relevant to AXIS-class focal planes. The team's self-citations are appropriate here, since they're building on their own prior work, and the paper is honest about what is new versus what was already reported.\n\nThe soft spots are real but not disqualifying. The central gap is the missing ADU-to-electron conversion. Section 4 assumes the relationship is \"well behaved\" across the bias grid, but never measures gain at each point. Table 3 shows the ADU-optimal bias at 173 K lowers noise from 2.39 to 2.31 e- while worsening FWHM from 118.5 to 120.8 eV. That is a small effect, and the FWHM change is within ~1.2 eV, so it isn't damning, but it does mean the optimization criterion is not validated against spectral resolution. If conversion gain varies across the bias grid, the ADU minimum could be picking a low-gain operating point. The fix is straightforward: report the Fe-55 peak centroid or a charge-injection gain measurement at the relevant bias points. I'd want that before accepting the headline number as a mission-ready claim.\n\nSecond, the Summary's language pairs the 2.31 e- noise with the 5 MPixels/s speed, but those numbers come from different operating points: at 5 MHz the best noise in Table 2 is 3.84 e-, and no optimized FWHM at 5 MPixels/s is presented. The paper doesn't hide this, but the summary glosses over it. A careful reader should not come away thinking the system does both simultaneously.\n\nThird, the leap from CCID-93 to CCID-100 is an extrapolation. The paper says this is the next step, so it's fine as a statement of intent, but the AXIS-readiness claim is based on small-format prototype metrics, not on the 16-channel full-scale device.\n\nOverall, this is a useful, clearly written instrumentation paper. The metrology gap is fixable in revision, and the engineering results are worth refereeing. I'd send it to peer review with a request for gain calibration and a more careful summary that distinguishes speed and noise achievements at different operating points. It's a paper for the X-ray detector community, not a broad audience, but it delivers what it promises and the method is reusable.","headline":"Solid hardware paper with a real but fixable metrology gap: the bias scan optimizes ADU noise without a gain calibration, so the headline 2.31 e- number and the AXIS-readiness claim need a supporting measurement before they're taken at face value.","tokens_in":11327,"tokens_out":1565,"would_cite":true,"duration_ms":18700,"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":"CCD and ASIC readout chain meets AXIS speed and noise needs","keywords":["X-ray detectors","CCD readout","low-noise electronics","ASIC","AXIS mission","bias optimization","clock drivers","energy resolution"],"falsifier":"Read the same detector at the optimized 173 K bias but at 5 MPixels/s, and measure the Fe-55 FWHM and electron noise with a calibrated conversion gain; if the 5 MPixels/s FWHM or read noise exceeds the AXIS requirement, the claim that the system achieves the required speed and noise together fails. Also, a photon-transfer curve at each of the four scanned bias points would show whether the minimum-ADU-noise bias is truly the minimum-electron-noise bias.","tokens_in":1736,"feed_emoji":"🛰️","tokens_out":2114,"duration_ms":100570,"temperature":0.7,"pith_summary":"Future X-ray observatories need cameras that read out large images quickly without adding electronic noise. This paper reports a prototype readout chain for such cameras: a small test CCD built with a single-layer-polysilicon process, a compact custom amplifier chip that reads many pixels in parallel, clock drivers mounted close to the detector, and an automated routine that scans four bias voltages to find the quietest operating point. With this chain the authors measure 2.31 electrons of read noise and a 121 eV energy resolution at the 5.9 keV iron line while running at 2 MPixels/s, and they demonstrate the same electronics at serial speeds up to 5 MPixels/s. They argue these results meet the requirements of the proposed AXIS mission and that the approach scales to the full 16-output CCID-100 detector.","feed_headline":"CCD+ASIC readout hits 5 MPixels/s at 2.31 e- noise","feed_subtitle":"A detector-and-chip chain now meets the speed and noise targets for the proposed AXIS X-ray satellite.","key_machinery":"The paper's central machinery is the integrated readout chain: (1) the MIT-LL CCID-93 CCD, whose single-layer polysilicon gate structure needs small clock swings and whose two-stage pJFET/nMOSFET output provides high conversion gain; (2) the Stanford MCRC V1.0 ASIC, an 8-channel analog readout chip with selectable voltage/current input, user-selectable gain of 8 or 16 V/V, and differential output buffering; (3) locally buffered, opto-isolated summing-well and reset-gate clock drivers with RC snubber termination, which sit close to the CCD and recover about 20% of the usable waveform samples; and (4) an automated four-dimensional scan over RGH, RGL, OG, and RD that uses the overscan-region st","core_discovery":"The central result is that the combination of an MIT-LL single-polysilicon CCID-93 CCD, the Stanford MCRC V1.0 readout ASIC, locally buffered fast summing-well and reset-gate clock drivers, and automated reset-gate/output-gate/reset-drain bias optimization delivers the speed and noise expected of a next-generation X-ray imager: serial readout up to 5 MPixels/s and, at 173 K and 2 MPixels/s, a read noise of 2.31 e- with 121 eV FWHM at the 5.9 keV Mn K-alpha line. The locally buffered clocks recover roughly 20% of the usable waveform samples and lower noise at all speeds, while the bias scan finds operating points that reduce noise by up to 18% at warmer temperatures (243 K) compared with defa","pith_inferences":["If the ADU-to-electron conversion gain is not constant across the scanned bias grid, the minimum-ADU operating point may not be the minimum-electron-noise point; a photon-transfer measurement at each grid point would test this directly.","The temperature-dependent optimal biases hint at a physical model for reset-gate threshold or reset-feedthrough changes that, once understood, could let future systems predict optimal biases from temperature instead of scanning.","The same automated scan could be used during ground calibration of a flight instrument to track long-term bias drift, or to re-derive biases if the focal plane temperature changes over a mission.","Because the scan is fully automated, it could in principle be extended to other detector types with similar output-stage bias nodes, not only the MIT-LL CCID series."],"forward_implications":["The same readout approach should scale to the full AXIS focal-plane detector, the 1440x1440, 16-output CCID-100, using two MCRC ASICs in parallel to reach 5-20 frames/s.","Because the optimal reset-gate voltages shift with temperature, each output channel of a multi-output detector must be biased individually at its operating temperature to keep noise low.","The automated bias scan can replace manual tuning for these detectors; a 400-point, four-parameter scan takes about 45 minutes, making per-channel optimization practical.","The roughly 20% recovery of usable waveform samples from the onboard clock drivers is what enables 5 MPixels/s serial transfer without the ringing penalty seen with external drivers.","Higher frame rates at low noise would reduce pile-up from bright sources and suppress particle background for faint diffuse sources, the two scientific drivers for AXIS."],"supporting_citations":[{"why":"supplies the earlier fast/low-noise readout design whose 4 MHz serial-speed limit the onboard summing-well and reset-gate clocks overcome.","marker":"[8]"},{"why":"provides the design and fabrication description of the MIT-LL single-polysilicon CCID-93 CCD used in the measurements.","marker":"[10]"},{"why":"characterizes charge-transfer behavior in the single-polysilicon gate structure, supporting the detector's fast transfer claims.","marker":"[12]"},{"why":"describes the AXIS high-speed camera and its 16-output CCID-100 detector, the mission target the paper claims its results meet.","marker":"[18]"},{"why":"reports the initial MCRC V1 readout ASIC architecture that provides the eight parallel analog readout channels.","marker":"[19]"},{"why":"demonstrates the MCRC ASIC's 5 Mpixel/s-per-channel capability that the combined system builds on.","marker":"[20]"},{"why":"reports the latest MCRC ASIC performance, the readout baseline for the spectra and noise values reported here.","marker":"[21]"},{"why":"describes the X-ray beamline test chamber used for the Fe-55 spectrum and noise measurements.","marker":"[22]"}],"fun_headline_variants":["5 MPix/s CCD+ASIC readout at 2.31 e- noise","CCD+ASIC chain: 5 MPix/s, 2.31 e- noise","Future X-ray CCD: 5 MPix/s, 2.31 e- noise","AXIS-ready CCD+ASIC: 5 MPix/s, 2.31 e-"],"cache_read_input_tokens":12928,"weakest_assumption_plain":"The quoted noise numbers rest on the assumption that the overscan-region scatter in raw digital units tracks true electron read noise at every bias setting, with no direct conversion-gain calibration, so the bias that minimizes ADU noise is assumed to be the best operating point for spectroscopy.","fun_headline_variants_meta":{"raw":{"variants":["5 MPix/s CCD+ASIC readout at 2.31 e- noise","CCD+ASIC chain: 5 MPix/s, 2.31 e- noise","Future X-ray CCD: 5 MPix/s, 2.31 e- noise","AXIS-ready CCD+ASIC: 5 MPix/s, 2.31 e-"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000355,"raw_usage":{"total_tokens":1780,"prompt_tokens":773,"completion_tokens":1007,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":517,"completion_tokens_details":{"reasoning_tokens":908}},"tokens_in":517,"tokens_out":1007,"duration_ms":8898,"temperature":1.0,"reasoning_tokens":908,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T18:43:59.374800+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Read the same detector at the optimized 173 K bias but at 5 MPixels/s, and measure the Fe-55 FWHM and electron noise with a calibrated conversion gain; if the 5 MPixels/s FWHM or read noise exceeds the AXIS requirement, the claim that the system achieves the required speed and noise together fails. Also, a photon-transfer curve at each of the four scanned bias points would show whether the minimum-ADU-noise bias is truly the minimum-electron-noise bias.","supporting_citations":[{"cited_title":"Development and characterization of a fast and low noise readout for the next generation x-ray charge-coupled devices,","cited_arxiv_id":null,"evidence_quote":"supplies the earlier fast/low-noise readout design whose 4 MHz serial-speed limit the onboard summing-well and reset-gate clocks overcome."},{"cited_title":"Toward fast, low-noise charge-coupled devices for Lynx,","cited_arxiv_id":null,"evidence_quote":"provides the design and fabrication description of the MIT-LL single-polysilicon CCID-93 CCD used in the measurements."},{"cited_title":"Charge transfer effects in a CCD with a single polysilicon gate structure,","cited_arxiv_id":null,"evidence_quote":"characterizes charge-transfer behavior in the single-polysilicon gate structure, supporting the detector's fast transfer claims."},{"cited_title":"The high-speed x-ray camera on AXIS,","cited_arxiv_id":null,"evidence_quote":"describes the AXIS high-speed camera and its 16-output CCID-100 detector, the mission target the paper claims its results meet."},{"cited_title":"MCRC V1: development of integrated readout electronics for next generation x-ray CCD detectors for future satellite observatories,","cited_arxiv_id":null,"evidence_quote":"reports the initial MCRC V1 readout ASIC architecture that provides the eight parallel analog readout channels."},{"cited_title":"X-ray speed reading with the MCRC: a low noise CCD readout ASIC enabling readout speeds of 5 Mpixel/s/channel,","cited_arxiv_id":null,"evidence_quote":"demonstrates the MCRC ASIC's 5 Mpixel/s-per-channel capability that the combined system builds on."},{"cited_title":"X-ray speed reading with the MCRC: prototype success and next generation upgrades,","cited_arxiv_id":null,"evidence_quote":"reports the latest MCRC ASIC performance, the readout baseline for the spectra and noise values reported here."},{"cited_title":"The XOC x-ray beamline: probing colder, quieter, and softer,","cited_arxiv_id":null,"evidence_quote":"describes the X-ray beamline test chamber used for the Fe-55 spectrum and noise measurements."}],"review_version":1}