{"id":"ced64bd7-35a2-4816-912e-b3f3e5368d1c","arxiv_id":"2502.02724","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The readout electronics and firmware for the upgraded Belle II KLM detector were commissioned, with gain calibration on over 18,000 SiPMs and measured hit-time resolutions below 8 nanoseconds.","lead":"This paper describes the custom electronics, firmware, and calibration routines built for the outermost particle detector layers of the Belle II experiment in Japan. It shows how 18,560 silicon photomultipliers were calibrated without calibration sources and how the readout handles trigger rates up to 30,000 events per second.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 30 kHz trigger-rate claim in Sec. 3.2 rests on an undisclosed simulation; without its testbench and validation, the central 'meets Belle II requirement' assertion is unverifiable.","rationale":"I agree with the reader that the paper is a credible description of a working system, but I identify the undisclosed 30 kHz simulation as the most load-bearing gap because it directly supports the 'meets the Belle II readout requirement' part of the central claim. The Tcollect simplification affects only the reported time resolutions, which are presented as performance results rather than as evidence of commissioning; even if Tcollect varies, the system still works and the calibration constants per strip would partially absorb a constant offset. In contrast, an unsupported rate-capability claim could invalidate the design's headline specification. The real-data point at 2.8 kHz shows no rate problem at that luminosity, but the 30 kHz requirement is an order of magnitude higher and the paper provides no intermediate data or simulation detail. The reader's rationale already noted this, so my agreement is partial. The verdict remains CONDITIONAL because the concern is a missing justification, not a demonstrated failure.","tokens_in":15938,"tokens_out":4844,"duration_ms":43790,"concrete_test":"Run the SCROD firmware in a hardware-in-the-loop test with a synthetic L1 trigger generator producing a Poisson 30 kHz stream with 200 ns minimum spacing, injecting realistic multi-channel hit patterns and worst-case pedestal-fetch patterns (e.g., trigger-bit pattern 0b11111). Measure the fraction of hits that fall back to simple mode (digitization skipped) and the queue occupancy over a 1-hour run. If the simple-mode fraction is non-negligible or the queue grows without bound, the 30 kHz claim fails; if it passes with margin, the simulation concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper asserts in Section 3.2 that 'Simulation tests verify that this scheme allows the SCROD firmware to keep pace with a 30 kHz L1 trigger rate (Poisson-distributed with a minimum of 200 ns between consecutive L1 triggers), the requirement for Belle II unified readout-system design.' No simulation parameters, firmware version, traffic model, or validation results are provided. The only real-data evidence is at 2.8 kHz with 0.4% digitization skips, an order of magnitude below the requirement. The firmware's own description in §3.3–3.4 identifies a potential bottleneck: pedestal SRAM access is sequential and multi-channel hits can make SRAM the bottleneck; §3.2 also has a 'simple mode' fallback that skips waveform digitization. Without knowing how the simulation models these effects, an independent reviewer cannot assess whether the 30 kHz capability is real or an artifact of optimistic assumptions. This is load-bearing because the central claim explicitly includes meeting the 30 kHz requirement.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports the design, firmware development, calibration, and commissioning of the readout electronics for the K_L^0 and muon (KLM) detector of Belle II. The system comprises a scintillator readout chain based on TARGETX waveform-sampling ASICs and SCROD FPGA boards, an RPC front-end with TDC timestamping, and a Data Concentrator that merges both streams. The paper also describes an in-situ calibration procedure that uses single-photon spectra recorded in firmware to homogenize the gain of more than 17,000 SiPMs without dedicated calibration sources, and it reports measured hit-time resolutions from early 2024 data: standard deviations of 7.8 ns for RPCs, 5.4 ns for barrel scintillators, and 4.7 ns for endcap scintillators. The central claim is that the full readout system was successfully designed, installed, and commissioned and that it meets the Belle II requirement of sustaining a 30 kHz L1 trigger rate.","tokens_in":16156,"tokens_out":6174,"duration_ms":52526,"significance":"If the claims hold, this is a valuable instrumentation paper for a major running experiment. It documents a complete, working readout chain for two detector technologies with a single Data Concentrator, and it demonstrates a practical method for calibrating tens of thousands of SiPMs without calibration sources. The reported in-situ gain homogenization and the measured timing resolutions are concrete performance results that will be useful to the Belle II collaboration and to future detector projects. The paper's strengths include the explicit use of real commissioning data, the statement of measured digitization-skip rates, and the discussion of known limitations such as preamplifier saturation. However, two load-bearing quantitative claims—the 30 kHz trigger-rate capability and the timing-resolution numbers—rest on assumptions that are not fully substantiated in the manuscript.","major_comments":[{"comment":"The sentence in §3.2 that 'Simulation tests verify that this scheme allows the SCROD firmware to keep pace with a 30 kHz L1 trigger rate (Poisson-distributed with a minimum of 200 ns between consecutive L1 triggers), the requirement for Belle II unified readout-system design' is the only support for the conclusion that the system operates 'at trigger rates up to 30 kHz.' The simulation is not described: no firmware version, no traffic model, no treatment of the SRAM bottleneck identified in §3.4, and no validation against the measured 2.8 kHz data with 0.4% digitization skips. Because the SRAM access is explicitly stated to become the bottleneck for multi-channel hits, and because the 'simple mode' in §3.2 skips waveform digitization entirely, it is essential to know how the simulation modeled these effects. Please provide a fuller description of the simulation (or a reference to a public note), or substantiate the 30 kHz capability with a bench or beam test, or soften the claim to 'designed for 30 kHz' with the simulation presented as indicative. As written, this load-bearing claim cannot be independently assessed.","section":"3.2, 3.3, 3.4, 8"},{"comment":"In the definition of t0, the term Tcollect is described as 'small and currently treated as a constant,' but no evidence is given that its variation with pulse height (i.e., number of fired SiPM pixels) is negligible. If Tcollect varies with pulse amplitude, the per-strip calibration constants (Tcable) fitted from the same data could absorb part of this variation, and the quoted standard deviations (7.8 ns for RPCs, 5.4 ns for barrel scintillators, 4.7 ns for endcap scintillators) could underestimate the true timing resolution. Please either demonstrate that Tcollect variation is small by, for example, comparing t0 in bins of pulse height, or explicitly state in the conclusions that the quoted resolutions assume a constant Tcollect and may be optimistic. This is load-bearing for the quantitative performance results.","section":"7, Eq. (1)"}],"minor_comments":[{"comment":"The sentence 'The t0 resolutions (FWHMs) for RPCs, barrel scintillators, and endcap scintillators are 7.8 ns (14.0 ns), 5.4 ns (5.6 ns), and 4.7 ns (3.8 ns), respectively' is confusing because the numbers in parentheses are FWHMs but the phrase 'resolutions (FWHMs)' implies the first numbers are also FWHMs. Suggest rewording to 'The standard deviations (FWHMs) are 7.8 ns (14.0 ns), 5.4 ns (5.6 ns), and 4.7 ns (3.8 ns), respectively.'","section":"7"},{"comment":"The text states 'The mean gain slope is 15 ADC counts / PE / V,' while the example fit in Fig. 18 yields a slope of 14.01 ADC counts / PE / V. Please clarify whether 15 is a rounded average over all channels or a different quantity, and consider quoting the distribution mean with uncertainty.","section":"6.4"},{"comment":"The phrase 'Poisson-distributed with a minimum of 200 ns between consecutive L1 triggers' is ambiguous because a true Poisson process has no minimum inter-arrival time. Presumably the simulated trigger stream is Poisson subject to a 200 ns dead-time or spacing constraint; please specify the model precisely.","section":"3.2"},{"comment":"The sentence 'According to the SiPM vendor, at a 70 V bias, the frequency of SiPM pulses larger than 1.2 pixels is 75 kHz, and 750 kHz for 0.5 pixels, respectively' is awkward; rephrasing as two separate statements would improve clarity.","section":"6.3"},{"comment":"The t0 definition is presented as a displayed equation but is not numbered. In the journal version, numbered equations would facilitate reference.","section":"7"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid commissioning report, and the central engineering claims appear credible. The main concern is the unsubstantiated 30 kHz simulation claim, which is repeated in the conclusion; the authors may have internal notes with the required detail, but they are not in the manuscript. The timing-resolution caveat is also worth addressing explicitly. The scope fits NIM-A well. The unusual disclaimer in the acknowledgments is not a scientific issue. I recommend major revision with the expectation that the two load-bearing points can be resolved by added detail or by appropriately softening the claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Sergei,\n\nHere's my read on arXiv:2502.02724. It's a commissioning report for the Belle II KLM readout electronics, and it does what that genre should: it describes the hardware, firmware, and calibration procedures in enough detail that someone could reproduce the system. The genuinely new piece is the firmware-resident single-photon spectrum histogram, which lets them calibrate 18,560 SiPMs without calibration sources. That is a real trick, and the resource-saving FIFO-concatenation for pedestal averaging is a nice engineering detail. The paper is honest about its own limitations: preamplifiers saturate, about 1,000 channels don't converge in the gain fit, and the time calibration treats Tcollect as a constant.\n\nThe soft spots are real but not disqualifying. The 30 kHz L1 rate claim in Sec. 3.2 rests on an undisclosed simulation. The paper says 'Simulation tests verify...' with no parameters, traffic model, or validation. That's a gap. The measured 2.8 kHz with 0.4% digitization skips is good real evidence, but it's an order of magnitude below the design rate, and the firmware's own description shows a plausible SRAM bottleneck for multi-channel hits. A referee should ask for the simulation setup or at least a clear statement that the 30 kHz figure is a design simulation, not a measured capability. The time-resolution numbers (7.8, 5.4, 4.7 ns) are reported without uncertainties, and the Tcollect-constant assumption is stated but not tested. These are the kind of things that are easy to fix in revision.\n\nThe citation pattern looks fine—TARGET references are appropriate, and the 30 kHz requirement is tied to the Belle II unified readout paper. I don't see any self-citation inflation or missing credit.\n\nBottom line: this is a solid detector paper for NIM-A. It deserves a serious referee. The central claim—system designed, installed, commissioned, working at current luminosity—is supported. The 30 kHz claim needs either more detail or a softer wording. I'd send it to review, and ask the authors to pin down the simulation and add uncertainties to the time resolutions.\n\nI'd probably cite this the next time I write about Belle II instrumentation. Reading group? Maybe, if we're doing detector electronics; it's a bit niche otherwise.\n\n— [Your name]","headline":"A solid, workmanlike commissioning report for the Belle II KLM readout; the firmware-resident SPS method is the real novelty, and the main soft spot—the unverifiable 30 kHz simulation claim—is worth a referee's attention but not a rejection.","tokens_in":17079,"tokens_out":2547,"would_cite":true,"duration_ms":23851,"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":"Belle II's outer detector readout, built around a 1 GHz waveform-sampling ASIC, was installed and commissioned to meet the 30 kHz trigger rate.","keywords":["Belle II","KLM detector","readout electronics","TARGETX ASIC","SCROD firmware","SiPM calibration","single-photon spectra","resistive plate chamber"],"falsifier":"Measure leading-edge arrival time as a function of pulse amplitude on a single scintillator strip using the TARGETX waveforms: if the time of the leading edge at a fixed threshold shifts by more than about 1 ns between single-photoelectron and saturated multi-photoelectron pulses on the same channel, the constant-$T_\\mathrm{collect}$ assumption fails and the reported resolution would change after an amplitude-dependent correction. A simpler version: compute the $t_0$ residual distribution before and after adding a pulse-height-dependent term to the timing model; if the width shrinks, the constant assumption was the limiting factor.","tokens_in":15735,"feed_emoji":"⚛️","tokens_out":6598,"duration_ms":55285,"temperature":0.7,"pith_summary":"This paper reports the design, installation, and commissioning of the complete electronic readout for the K-long and muon (KLM) detector at Belle II, the outermost particle detector of the experiment. The authors' claim is that a single readout chain built around the TARGETX waveform-sampling ASIC, SCROD FPGA control boards, and a Data Concentrator can merge hits from two different detector technologies—scintillator strips read by silicon photomultipliers and legacy resistive-plate chambers—while keeping up with Belle II's 30 kHz L1 trigger requirement. They support this with simulation showing the firmware sustains that rate and with physics data at a 2.8 kHz trigger rate where only 0.4% of hits skip waveform digitization. They also report that the gain of more than 17,000 SiPMs was equalized using in-firmware single-photon spectra, and that calibrated hit-time resolutions are 7.8 ns for RPCs, 5.4 ns for barrel scintillators, and 4.7 ns for endcap scintillators. A sympathetic reader would care because these numbers determine how well Belle II can identify muons and K-longs in the high-background environment at design luminosity.","feed_headline":"Belle II outer detector readout hits 30 kHz and 5 ns timing","feed_subtitle":"Custom ASIC and FPGA chain times hits to 4.7–7.8 ns and calibrates 17,000 SiPMs without calibration sources.","key_machinery":"The load-bearing object is the readout chain rather than any single chip: the TARGETX ASIC (a 16-channel, 1 GHz waveform-sampling chip with 214 storage cells per channel and a per-channel trigger comparator), the SCROD FPGA board that controls up to ten such ASICs, handles L0 self-triggers and L1 lookback, and performs feature extraction, and the Data Concentrator that merges scintillator and RPC hit packets. The feature-extraction stage uses constant-value discrimination with linear interpolation to define leading-edge time because the preamplifiers saturate on large pulses, so constant-fraction discrimination is impossible. A second essential mechanism is the in-firmware single-photon spectrum: dark-rate pulses are histogrammed inside FPGA block RAM, each spectrum is fit to a sum of Gaussians spaced by the gain, and the gain-versus-bias linear fit is used to trim each SiPM's bias in 20 mV steps. Together these mechanisms carry the paper's two claims: the system keeps pace with 30 kHz triggers, and the SiPMs are uniformly calibrated without calibration sources.","core_discovery":"The central discovery is that the full KLM readout system works as designed: it was installed and commissioned, and it satisfies the Belle II unified-readout requirement of sustaining a 30 kHz L1 trigger rate. Waveform digitization of silicon-photomultiplier signals is performed by the TARGETX ASIC, which samples at 1 GHz into a 16.384 µs analog buffer per channel; the SCROD FPGA timestamps self-triggers, masks regions of interest on L1, digitizes, subtracts per-cell pedestals, and extracts leading-edge time and pulse height. The system merges these scintillator data with timestamps from RPC front-end boards in a Data Concentrator. The authors further report that all 18,560 installed SiPMs were gain-calibrated without any calibration light source by recording dark-rate single-photon spectra inside the firmware; more than 17,000 channels were then set to a common gain. Per-strip cable-delay calibration in the Belle II conditions database yields the reported hit-time resolutions, with the paper noting that the charge-collection time in the SiPM is currently treated as a small constant.","pith_inferences":["The dark-rate single-photon-spectrum calibration, done entirely in firmware without calibration sources, is a transferable recipe for large SiPM arrays where LED or laser calibration would be impractical.","If the charge-collection time $T_\\mathrm{collect}$ varies with pulse amplitude more than the paper assumes, the timing constants would need an amplitude-dependent term; this can be tested by comparing leading-edge times for single- versus multi-photoelectron pulses on the same strip.","The fixed-latency lookback scheme with a digitization queue and a fallback 'simple mode' (timestamp-only, zero pulse height) is a general pattern for front-end readout at high-rate colliders, since it bounds dead time under bursty triggers.","The paper's anticipated improvement toward about 1 ns hit-time resolution, if realized by better feature extraction, would directly strengthen background rejection and low-momentum muon identification at design luminosity."],"forward_implications":["The SCROD firmware can sustain Belle II's design L1 trigger rate of 30 kHz (Poisson-distributed, 200 ns minimum spacing), so the waveform-readout architecture is not a bottleneck at design luminosity.","At the early-2024 luminosity of $1.9\\times10^{34}\\,\\mathrm{cm^{-2}s^{-1}}$ and a 2.8 kHz L1 rate, only 0.4% of hits skip waveform digitization, meaning multi-channel ambiguity is resolved for essentially all events.","Waveform digitization disambiguates multi-channel hits on a single TARGETX: without it, group hits degrade strip resolution from 4 cm to 16 cm; with it, normal strip-level resolution is restored.","Per-strip cable-delay calibration gives hit-time resolutions of 7.8 ns (RPC), 5.4 ns (barrel scintillator), and 4.7 ns (endcap scintillator), which are the input to track and K-long cluster reconstruction.","The gain of more than 17,000 SiPMs was set to a common value (target 30 ADC counts per photoelectron), making trigger thresholds and pulse-height comparisons uniform across the detector."],"supporting_citations":[{"why":"Belle II technical design report that defines the KLM detector layout, the steel flux return, and the interaction-length context for the muon and K-long system.","marker":"[1]"},{"why":"Describes the scintillator strips, wavelength-shifting fibers, and SiPMs that the new readout electronics serve.","marker":"[4]"},{"why":"Documents the Belle II data acquisition system that the KLM Data Concentrator interfaces to via the Belle2Link protocol.","marker":"[5]"},{"why":"Introduces the TARGET digitizer chip family from which the TARGETX ASIC used here descends.","marker":"[6]"},{"why":"Sets the 30 kHz L1 trigger rate and unified readout-system design requirement that the SCROD firmware is claimed to meet.","marker":"[10]"},{"why":"Describes the automated production test system for the 20,000-channel readout and supplies initial TARGETX register tuning procedures.","marker":"[11]"},{"why":"Documents the Belle II conditions database that stores the per-strip timing calibration constants and SiPM bias settings.","marker":"[12]"},{"why":"Describes the Belle II core software framework in which the timing calibration constants are applied during hit reconstruction.","marker":"[13]"}],"fun_headline_variants":["Belle II KLM readout sustains 30 kHz trigger rate","KLM readout: 18,560 SiPMs calibrated without light source","KLM electronics pass 30 kHz with in-firmware SiPM calibration","Belle II outer detector readout hits 30 kHz, self-calibrates"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The timing calibration treats the SiPM charge-collection time $T_\\mathrm{collect}$ as a small constant independent of pulse height and the number of pixels fired; if it actually varies with pulse amplitude, the reported 4.7–7.8 ns hit-time resolutions and the per-strip calibration constants would be systematically biased.","fun_headline_variants_meta":{"raw":{"variants":["Belle II KLM readout sustains 30 kHz trigger rate","KLM readout: 18,560 SiPMs calibrated without light source","KLM electronics pass 30 kHz with in-firmware SiPM calibration","Belle II outer detector readout hits 30 kHz, self-calibrates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000486,"raw_usage":{"total_tokens":2366,"prompt_tokens":886,"completion_tokens":1480,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":502,"completion_tokens_details":{"reasoning_tokens":1396}},"tokens_in":502,"tokens_out":1480,"duration_ms":9722,"temperature":1.0,"reasoning_tokens":1396,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T11:20:50.081119+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure leading-edge arrival time as a function of pulse amplitude on a single scintillator strip using the TARGETX waveforms: if the time of the leading edge at a fixed threshold shifts by more than about 1 ns between single-photoelectron and saturated multi-photoelectron pulses on the same channel, the constant-$T_\\mathrm{collect}$ assumption fails and the reported resolution would change after an amplitude-dependent correction. A simpler version: compute the $t_0$ residual distribution before and after adding a pulse-height-dependent term to the timing model; if the width shrinks, the constant assumption was the limiting factor.","supporting_citations":[{"cited_title":"Yamada, R","cited_arxiv_id":null,"evidence_quote":"Documents the Belle II data acquisition system that the KLM Data Concentrator interfaces to via the Belle2Link protocol."},{"cited_title":"Bechtol et al., TARGET: A multi-channel digitizer chip for very-high-energy gamma-ray telescopes, Astroparticle Physics 36 (1) (2012) 156–165","cited_arxiv_id":null,"evidence_quote":"Introduces the TARGET digitizer chip family from which the TARGETX ASIC used here descends."},{"cited_title":"Nakao et al., Performance of the unified readout system of Belle II, IEEE Transactions on Nuclear Science 68 (8) (2021) 1826–1832","cited_arxiv_id":null,"evidence_quote":"Sets the 30 kHz L1 trigger rate and unified readout-system design requirement that the SCROD firmware is claimed to meet."},{"cited_title":"Edralin, Design and performance of an automated pro- duction test system for a 20,000 channel single-photon, sub- nanosecond electronic readout for a large area muon detector, Ph.D","cited_arxiv_id":null,"evidence_quote":"Describes the automated production test system for the 20,000-channel readout and supplies initial TARGETX register tuning procedures."},{"cited_title":"Ritter et al., Belle II conditions database, J","cited_arxiv_id":null,"evidence_quote":"Documents the Belle II conditions database that stores the per-strip timing calibration constants and SiPM bias settings."}],"review_version":1}