{"id":"0b13de19-ffce-4771-b0b5-df9a51511b9c","arxiv_id":"2411.09673","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"FANSIC, a 65 nm CMOS analog front-end ASIC with active summation, achieves roughly 3 ns output pulses, single-photon-level SNR, and about 5% post-calibration linearity for SiPM readout.","lead":"This paper reports FANSIC, a prototype 65 nm CMOS chip that reads out silicon photomultipliers for Cherenkov telescope cameras and shapes their signals into fast 3 ns pulses. The chip is designed to replace photomultiplier tubes in large-area detectors, potentially enabling more compact, lower-power cameras for gamma-ray astronomy.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 2.2's ENP<0.5 splitting argument ignores quadrature noise addition in the active summation; the reported Table 4 values project a 1-p.e. SNR near 3.5 for the summed four-input pixel, below the design requirement.","rationale":"The reader's weakest assumption correctly identifies the small-sensor-to-large-pixel extrapolation as the central gap. I go a step further: the extrapolation is not just unmeasured; the paper's own noise formulas and Table 4 make a quantitative prediction that the summed four-input configuration will fail the stated SNR requirement. Section 2.2's ENP < 0.5 is per-section before the active summation. At the summed output, a 1-p.e. event in one section sees the signal from that section plus quadrature noise from all four, so the 6.98 SNR measured with one small sensor projects to about 3.5. This is an internal consistency check, not an external consensus disagreement. The measurement remains credible for a single 3x3 mm2 sensor, and I do not see evidence of fabrication or carelessness in the acquired data. The Table 2 recovery-time discrepancy and Fig. 17/Table 4 uncertainty mismatch are real but secondary; they affect detailed reliability, not the central feasibility argument. A conditional verdict is appropriate, with the condition that the authors either provide a four-input summed measurement or a corrected noise budget showing SNR >= 5.","tokens_in":15549,"tokens_out":17153,"duration_ms":181422,"concrete_test":"Recompute the summed-pixel 1-p.e. SNR from the measured values in Table 4 as b / sqrt(4*sigma_e^2 + 4*sigma_s^2), using the target section capacitance scaling if the sections differ from the tested 3x3 mm2 sensor, and compare with the SNR >= 5 requirement. If the result is at least 5, the concern is refuted; if it is near 3.5, the Section 2.2 splitting rationale is invalid and the central claim is unsupported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 2.2 uses Eq. (2) to claim that splitting a roughly 1 cm2 pixel (Ceq = 650 pF) into four sections reduces the ENP below 0.5, and Section 3.2 then actively sums the four sections into one readout channel. The reduction applies only to a single section before summation. Because the summing stage is linear, the signal from a photoelectron located in one section has the same amplitude at the summed output as in the single-input measurement, but the noise of all four sections adds in quadrature. Using the measured single-channel parameters in Table 4 (gain b = 6.85 mV*ns, sigma_e = 0.813 mV*ns, sigma_s = 0.549 mV*ns), the 1-photoelectron SNR of the summed output is b / sqrt(4*sigma_e^2 + 4*sigma_s^2) = 6.85 / 1.96, which is approximately 3.5, below the SNR >= 5 requirement of Section 2.5. If the intended sections are larger than the tested 3x3 mm2 sensor, the per-photoelectron voltage signal is smaller and the projection worsens. The central large-pixel single-photon claim therefore rests not merely on an unmeasured configuration; the reported noise budget, once the active summation is included, does not close.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents FANSIC, a 65 nm CMOS ASIC for the readout of silicon photomultipliers, with an active summation stage to combine signals from four SiPM sections into one pixel channel. The design targets large-area SiPM pixels for Cherenkov telescope cameras. The authors describe the design requirements, circuit architecture, and a characterization performed with a single 3x3 mm2 Hamamatsu S13360 SiPM. The measured output pulse has a FWHM of 2.82 ns, the MPE fit yields a 1-p.e. SNR of 6.98, and the transfer characteristic is calibrated with a sixth-degree polynomial giving residuals within 5%. The paper claims compatibility with input capacitance up to 1 nF and single-photoelectron resolution up to 30 p.e., supported by simulations and extrapolation rather than by measurements in the large-pixel configuration.","tokens_in":15805,"tokens_out":9264,"duration_ms":91473,"significance":"The measured single-channel performance is credible and the characterization methodology (MPE fit, ND-filter transmittance calibration, automated acquisition) is careful and reproducible. If the results hold, the ASIC demonstrates a fast, low-power frontend for small SiPM pixels. However, the central claim of large-area SiPM readout is not supported by the present evidence: the ENP reduction argument in Section 2.2 ignores the noise added by the active summation, the 1 nF input-capacitance regime was never exercised in the laboratory, and the linearity metric is defined relative to a high-order polynomial rather than to a linear response. These issues are load-bearing for the advertised application, so the paper in its current form overstates its conclusions.","major_comments":[{"comment":"The ENP reduction argument for pixel splitting is invalid after the active summation. Equation (2) gives ENP≈1 for Ceq=650 pF; splitting the pixel into four sections reduces each section's ENP to about 0.5. However, Section 3.2 sums the four sections into a single output, so the noise of all four sections and their preamplifiers adds in quadrature while the signal from a photoelectron appears in only one section. Using the measured single-section noise values from Table 4 (sigma_e = 0.813 mV·ns, sigma_s = 0.549 mV·ns, gain b = 6.853 mV·ns), the 1-p.e. SNR of the summed output is 6.853/sqrt(4·(0.813^2 + 0.549^2)) ≈ 3.5, which is below the SNR ≥ 5 requirement stated in Section 2.5. The paper does not discuss this noise addition in the summation stage, so the design rationale for large-pixel readout is not established.","section":"Section 2.2, Eq. (2), Section 3.2, Table 4"},{"comment":"The central claim of readout of large SiPMs with input capacitance up to 1 nF and pixel areas around 1 cm² is not validated by the reported measurements. Section 4 characterizes FANSIC only with a single 3x3 mm² S13360 SiPM connected to one input; no measurement is presented with four sections summed, with all four inputs connected, or with an input capacitance near 1 nF. The compatibility with 1 nF is supported only by the analytical estimate in Eq. (2) and by circuit simulations. Consequently, the conclusions in Section 5 that the ASIC achieves 3 ns FWHM and single-photoelectron resolution up to 30 p.e. for the target large-pixel application are extrapolations, not demonstrated experimental results.","section":"Section 4, Section 5"},{"comment":"The claim of 'post-calibration non-linearity around 5%' is established by fitting a sixth-degree polynomial to the measured transfer function and reporting the residuals relative to that polynomial. This metric measures the goodness of fit of a flexible calibration curve, not the linearity of the detector response. A sixth-order polynomial can absorb arbitrary nonlinearity, so residuals within 5% are not evidence that the frontend meets the 5% linearity requirement of Section 2.7. The paper itself reports a quadratic dependence of pulse area on photoelectron number at low intensities; without defining a linear-calibration procedure (e.g., a single gain factor), the linearity claim is misleading.","section":"Section 4.2, Fig. 19"}],"minor_comments":[{"comment":"The abstract states a pulse duration of 3 ns, while Table 2 lists a measured FWHM of 2.82 ns; the minor discrepancy should be reconciled or explained.","section":"Abstract and Table 2"},{"comment":"The ENP discussion should explicitly state whether the quoted value is the per-section ENP before summation; the current text leaves this ambiguous.","section":"Section 2.2"},{"comment":"Equation (4) appears to have an incorrect integrand: the Gaussian tail probability should contain exp(-x²/(2σ²)), not exp(-x²/σ²). The resulting 0.62% is consistent with 2.5σ, so the numerical value is correct, but the expression as written is not.","section":"Section 2.5, Eq. (4)"},{"comment":"The claim of 'FWHM of 3 ns across a dynamic range of 1-250 p.e.' would be better supported by a dedicated plot of FWHM versus photoelectron number; the current figure combines several quantities and the text description is qualitative.","section":"Section 4, Fig. 16"},{"comment":"References [11] and [20] appear to be the same publication (Aguilar et al., NIM A 830, 2016) and should be merged or cross-referenced consistently.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The single-channel measurements are sound and the characterization methodology is a strength. The main concern is that the paper's advertised large-area SiPM capability is not supported by the data: the noise of the active summation is not accounted for, which is a fundamental issue in the design rationale, and the 1 nF configuration was never tested. A major revision that either supplies measurements of the summed four-input configuration or substantially tempers the claims about large-pixel readout would be necessary before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take on 2411.09673. The chip is real, and the single-channel measurements look careful. The 2.82 ns FWHM, the SNR of 6.98 at 1 p.e., and the ~5% post-calibration linearity are believable for a 3x3 mm2 S13360; the MPE fit and ND-filter calibration are appropriate. That part earns its keep.\n\nThe soft spot is load-bearing. Section 2.2 argues that splitting a 650 pF pixel into four sections reduces the ENP below 0.5. That is true per section before summation. But Section 3.2 then actively sums the four sections, and the summing stage is linear: a photoelectron in one section gives the same output amplitude as in the single-section case, while the noise of all four sections adds in quadrature. Using the paper's own Table 4 values (b = 6.85 mV·ns, σe = 0.813, σs = 0.549), the 1-p.e. SNR at the summed output is 6.85 / sqrt(4·(0.813² + 0.549²)) ≈ 3.5. That is below the SNR ≥ 5 requirement from Section 2.5. And the intended sections are larger than the tested 3x3 mm2, so the real condition is worse. The splitting argument doesn't close the noise budget for the pixel it claims to serve.\n\nOther issues are smaller. The abstract's claim of 'pulse FWHM of 3 ns across a dynamic range of 1–250 p.e.' overstates Fig. 16, where FWHM grows with intensity. Table 2 shows a 0.1% recovery time that differs by an order of magnitude between simulation (808 ns) and measurement (33 ns) without discussion. And the fit uncertainty for σe in Fig. 17's legend (0.6031) disagrees with Table 4 (0.03). These are fixable with text corrections.\n\nWho gets value from this? Detector front-end people in gamma-ray astronomy will find the single-channel characterization a useful data point. But as a demonstration of large-area SiPM readout, the paper overreaches.\n\nMy recommendation: send it to peer review, but expecting major revision. The authors should either measure a real four-section summed pixel or rework the noise analysis to show how the summed output meets the stated SNR requirement. Without that, the architecture is an interesting prototype, not a demonstrated solution.","headline":"The single-channel measurements are credible, but the quadrature-noise analysis shows the summed four-input pixel falls below the paper's own SNR requirement, so the large-SiPM claim does not close.","tokens_in":16419,"tokens_out":8015,"would_cite":false,"duration_ms":71961,"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":"FANSIC, a 65 nm CMOS analog front-end, claims to read out large silicon photomultipliers with 3 ns pulses, single-photon resolution up to 30 p.e., and 23 mW per pixel by actively summing four pixel sub-sections.","keywords":["FANSIC","ASIC","silicon photomultiplier","active summation","Cherenkov telescope camera","single-photoelectron resolution","pulse shaping","gamma-ray astronomy"],"falsifier":"Take a FANSIC channel, connect four SiPM sections (or an equivalent network) whose total anode capacitance is near 650 pF–1 nF to the four inputs, and measure the summed output: the claim fails if the FWHM exceeds about 5 ns, if the single-photon SNR falls below 5, or if the post-calibration non-linearity exceeds 5% within 1–250 p.e.","tokens_in":15249,"feed_emoji":"🔭","tokens_out":8647,"duration_ms":74170,"temperature":0.7,"pith_summary":"FANSIC is a 65 nm CMOS analog front-end ASIC built to make large-area silicon photomultipliers practical for the cameras of imaging atmospheric Cherenkov telescopes, where photomultiplier tubes are still standard. The central claim is that splitting a large pixel into four sections and actively summing their signals on-chip overcomes the noise and slow-tail penalties of large SiPM capacitance, keeping output pulses at about 3 ns full width at half maximum while preserving single-photon resolution. Measured with a 3x3 mm2 S13360 SiPM, the prototype delivers a 3 ns FWHM across 1–250 photoelectrons, single-photon resolution up to about 30 p.e., post-calibration non-linearity around 5%, and 23 mW per pixel. If the design-level claim of input-capacitance compatibility up to 1 nF holds, this would let Cherenkov telescopes replace PMTs with SiPMs over square-meter camera surfaces at lower power and finer granularity.","feed_headline":"CMOS chip keeps large SiPM pulses at 3 ns","feed_subtitle":"Single-photon resolution to 30 p.e. at 23 mW per pixel would let Cherenkov telescopes swap PMTs for silicon.","key_machinery":"The load-bearing element is the active summation stage: four pre-amplifiers convert each SiPM-section voltage pulse into currents through series resistors into the virtual ground of an inverting summing amplifier, where Kirchhoff's current law adds them before a feedback resistor sets the overall gain. Around this, a configurable band-pass transfer function keeps only the fast component of the SiPM pulse, with the high-pass edge set by internal AC-coupling capacitors and the low-pass edge by amplifier bandwidth, replacing the ~1 µs recharge tail with a brief undershoot and preventing night-sky pile-up. The four-way pixel split is what keeps the equivalent noise charge below half a photoelectron, based on the SiPM electrical model with Ceq = 650 pF and gain $10^{4}$.","core_discovery":"On its own terms, the paper reports that a prototype ASIC with four parallel voltage-mode pre-amplifiers feeding an active summation stage can combine a pixel's sub-sections into one readout channel without losing the fast signal. With a single 3x3 mm2 S13360 SiPM and a pulsed 375 nm laser, the measured output pulse has a FWHM of about 2.8 ns, the electronics SNR at one photoelectron is 8.4, the combined sensor-plus-electronics SNR is 7.0, and the multi-photoelectron histogram resolves individual photoelectron peaks up to about 30 p.e. The pulse integral stays within 5% of a calibrated polynomial up to about 300 p.e. The slow ~1 µs SiPM recharge tail is filtered into a short undershoot, preventing baseline pile-up from night-sky background rates of 300 MHz to 1 GHz. The paper concludes that these results satisfy the requirements for Cherenkov cameras and that the architecture is compatible with pixels up to about 1 cm2 and input capacitances up to 1 nF.","pith_inferences":["If the simulated 1 nF input-capacitance regime holds in hardware, the same active-summation layout could extend to pixel areas beyond 1 cm2 by adding more sub-sections, since the noise penalty scales roughly as the square root of the summed capacitance.","The band-pass strategy of keeping only the fast decay component and discarding the slow tail is not SiPM-specific: any large-capacitance photodetector with widely separated time constants could use the same filter topology.","The measured gain varies linearly with photoelectron number because the amplifier transconductance is quadratic in voltage; an on-chip linearization or per-channel correction would reduce the calibration effort the camera currently needs."],"forward_implications":["A camera front-end could sustain 10 MHz trigger rates with a lower threshold: extending the pulse FWHM from 3 ns to 5 ns would raise the threshold by about 40%.","Single-photon resolution up to about 30 p.e. lets the trigger detect faint 20 GeV Cherenkov events with a false-positive probability around 0.6% at a half-photoelectron threshold.","Power of 23 mW per pixel makes a multi-thousand-pixel camera feasible on a 1.2 V supply.","Post-calibration linearity of about 5% up to 300 p.e. fits the camera-level error budget that assigns the front-end a 5% share of the 8% pixel-response requirement.","The output stage drives both single-ended and pseudo-differential 50 Ω loads, so the same chip can feed commercial or custom 1 GHz ADCs without extra components."],"supporting_citations":[{"why":"Defines the upgraded camera requirements and the trigger simulation used to set the 3 ns FWHM target and threshold budget.","marker":"[1]"},{"why":"Supplies the Cherenkov flash duration and spectral assumptions, the dynamic-range argument, and the 1 GHz digitization concept behind the pulse-shaping choices.","marker":"[4]"},{"why":"Prior active-summation ASIC for SiPM arrays, cited as the precedent for the summation architecture.","marker":"[15]"},{"why":"Provides the SiPM electrical model and naming convention used in Eqs. (1)-(3) for equivalent capacitance, noise charge, and time constants.","marker":"[16]"},{"why":"Supports the capacitance-scaling and noise analysis that motivates dividing the pixel into four sections.","marker":"[19]"},{"why":"Supplies the generalized Poisson probability density function (equation 3.2) used to fit the multi-photoelectron histograms and extract gain, noise, and cross-talk.","marker":"[22]"}],"fun_headline_variants":["FANSIC chip reads large SiPMs with 2.8 ns pulses","Fast ASIC merges SiPM segments for 30 p.e. clarity","Cherenkov-ready chip resolves 30 photoelectrons per pulse","Active summation in 65 nm CMOS speeds up large SiPM readout","SiPM readout chip sharpens pulses to 3 ns for Cherenkov"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central claim that FANSIC reads out large ~1 cm2 SiPM pixels rests on the assumption that the performance measured with a single 3x3 mm2 S13360 SiPM transfers to four-input active summation with combined input capacitance up to 1 nF, a regime the paper evaluates with the SiPM electrical model and circuit simulations rather than laboratory measurements.","fun_headline_variants_meta":{"raw":{"variants":["FANSIC chip reads large SiPMs with 2.8 ns pulses","Fast ASIC merges SiPM segments for 30 p.e. clarity","Cherenkov-ready chip resolves 30 photoelectrons per pulse","Active summation in 65 nm CMOS speeds up large SiPM readout","SiPM readout chip sharpens pulses to 3 ns for Cherenkov"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000414,"raw_usage":{"total_tokens":2151,"prompt_tokens":971,"completion_tokens":1180,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":587,"completion_tokens_details":{"reasoning_tokens":1080}},"tokens_in":587,"tokens_out":1180,"duration_ms":10820,"temperature":1.0,"reasoning_tokens":1080,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T20:25:11.832827+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a FANSIC channel, connect four SiPM sections (or an equivalent network) whose total anode capacitance is near 650 pF–1 nF to the four inputs, and measure the summed output: the claim fails if the FWHM exceeds about 5 ns, if the single-photon SNR falls below 5, or if the post-calibration non-linearity exceeds 5% within 1–250 p.e.","supporting_citations":[{"cited_title":"Heller, et al., The next generation cameras for the Large-Sized Tele- scopes of the Cherenkov Telescope Array Observatory, PoS ICRC2023 (2023) 740","cited_arxiv_id":null,"evidence_quote":"Defines the upgraded camera requirements and the trigger simulation used to set the 3 ns FWHM target and threshold budget."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the SiPM electrical model and naming convention used in Eqs. (1)-(3) for equivalent capacitance, noise charge, and time constants."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supports the capacitance-scaling and noise analysis that motivates dividing the pixel into four sections."},{"cited_title":"Large scale characterization and calibration strategy of a SiPM-based camera for gamma-ray astronomy","cited_arxiv_id":"2008.04716","evidence_quote":"Supplies the generalized Poisson probability density function (equation 3.2) used to fit the multi-photoelectron histograms and extract gain, noise, and cross-talk."}],"review_version":1}