{"id":"7902a985-848e-43a4-a346-c38eb6d2f709","arxiv_id":"1908.05878","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"The TOFPET2 ASIC consumes 3.6 to 7.2 mW per channel, and the paper fits a model predicting this power from three software settings, with the tradeoff that lower power reduces timing performance by 20 to 40 ps.","lead":"This paper measures how much electrical power the TOFPET2 ASIC, a chip that reads out PET scanner light detectors, consumes under different settings. It also builds a model to predict that power and shows the settings affect timing and energy resolution, which helps designers choose a low-power configuration for combined PET/MRI machines.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Published model parameters don't reproduce the model: Eq. (6) references undefined a3, and Table II's b1,b2 signs would reverse the fe_ib2 power drop, so the analytical model is not usable as printed.","rationale":"The reader's weakest assumption -- additivity of the three dP terms -- is a legitimate concern, and the paper's own Section V acknowledges cross-parameter experiments were not performed. However, the random-tuple validation (Fig. 6) provides at least some evidence for additivity across the sampled region. In contrast, the internal inconsistency between Eq. (6)-(7) and Table II is a definite, checkable defect: Eq. (6) uses a3 which does not appear in the table, and the positive b1,b2 values would make dP_fe_ib2 positive for fe_ib2 >20, contradicting the monotonic decrease in Fig. 5b and the minimum benchmark at high parameter values. A model that cannot reproduce its own training curves cannot support the claim that power consumption can be computed for arbitrary configurations. This does not invalidate the measured power range (3.6-7.2 mW/channel) or the performance tradeoffs, which are the core experimental contribution, so the CONDITIONAL verdict stands, but it should explicitly require reconciliation of the model equations and parameters.","tokens_in":18336,"tokens_out":17553,"duration_ms":152527,"concrete_test":"Implement Eqs. (4)-(7) exactly as printed, taking the parameter values from Table II. Compute P_ch for the three benchmark tuples (60,30,32), (59,0,0), (0,0,0) and for the single-parameter sweeps of Fig. 5. Compare with Table I and Fig. 5. If the computed values do not match the measured curves (they cannot, because a3 is missing and b1/b2 have the wrong sign), the model is not self-consistent. Then try the plausible correction (a0 for the first segment, a1/a2 for the second; negative b1/b2) and check whether it reproduces the data; if it does, the paper needs a corrigendum before the model can be used.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central modeling claim -- that P_ch can be computed via Eq. (4) using the fitted dP functions -- is not reproducible from the printed text. Section IV.B defines dP_fe_ib1 (Eq. 6) with coefficients a1, a2, a3, but Table II lists only a0, a1, a2; a3 is never specified. For dP_fe_ib2 (Eq. 7), Table II gives b1 = +2.74e-3 and b2 = +6.70e-3. Over fe_ib2 in [21,30] these positive coefficients make dP_fe_ib2 a large positive number (e.g., +2.67 mW at fe_ib2=30), whereas Fig. 5b shows power decreasing monotonically and the benchmark minimum occurs at fe_ib2=30. The same table lists a0, a1 negative, suggesting the equations' labels are scrambled; regardless, a reader cannot evaluate the model. If the intent was different breakpoints or a sign convention, that is not stated. Because the model is validated only by comparison with its own measurements, and the paper presents no alternative parameter set, the claim 'allowing to compute the power consumption prior to experiments' is unsupported until the equations and table are reconciled.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper quantifies the power consumption of the TOFPET2 ASIC as a function of three software-configurable parameters: fe_ib1 (input stage impedance), fe_ib2 (discriminator noise), and disc_sf_bias (discriminator noise slew rate). The authors use a benchtop evaluation kit, measure the 1.2 V supply current via a shunt resistor inserted in place of a fuse on the FEB/D board, and report values from 3.6 to 7.2 mW/channel depending on the configuration. They propose an additive model P_ch = P0 + dP_fe_ib1 + dP_fe_ib2 + dP_disc_sf_bias with piecewise linear/parabolic fits, determine the coefficients by least-squares fits to single-parameter scans, and validate the model on a set of random parameter tuples, reporting a regression slope of 1.034 ± 0.018. They also evaluate coincidence resolution time, energy resolution, filtered count rate, satellite peak fraction, and dark-count trigger level at minimum, default, and maximum power settings, and compare the ASIC with other PET front-end ASICs.","tokens_in":18637,"tokens_out":6796,"duration_ms":61438,"significance":"Provided the analytic model is made reproducible, the paper is a useful engineering contribution: it establishes a tunable power-performance operating range for the TOFPET2 ASIC in a PET/MR context, and the measurement method (current sensing on the LDO input, exclusion of the FPGA line, stability checks over count rate and overvoltage) is carefully described and internally consistent. The comparison table of ASIC power consumptions and the explicit discussion of what the measured power does not include (ASIC-FPGA communication) are valuable. The paper does not claim device-level derivation; it is an empirical model, and the main limitation is that P0 must be measured for each setup and that cross-parameter interactions are not experimentally probed.","major_comments":[{"comment":"Equation (6) defines dP_fe_ib1 using coefficient a3 for the parabolic branch (41 ≤ fe_ib1 ≤ 60), but Table II lists only a0, a1, and a2 and does not contain a3. Table II additionally lists a0, which does not appear in Eq. (6) or in Eq. (7). As printed, the model cannot be evaluated for the upper fe_ib1 range, which includes the default (fe_ib1=59) and minimum (fe_ib1=60) benchmark configurations. The authors should correct the equation/table labels and provide the complete set of coefficients.","section":"IV.B, Eq. (6), Table II"},{"comment":"With the printed values b1 = +2.74×10−3 mW/channel and b2 = +6.70×10−3 mW/channel, dP_fe_ib2 is positive and increasing over fe_ib2 ∈ [21,30]; for example, it is approximately +1.35 mW/channel at fe_ib2=30. This contradicts Fig. 5b, which shows the power consumption decreasing monotonically in fe_ib2, and Table I, which places the minimum power consumption at fe_ib2=30. The equation signs or the table entries are therefore inconsistent, and the model predictions for half of the fe_ib2 range are wrong as printed. This must be reconciled and the corrected model re-validated on the random tuples.","section":"IV.B, Eq. (7), Table II"},{"comment":"The claim that the model 'can be used to compute the power consumption prior to experiments' is stronger than what the current evidence supports, because P0 must be measured experimentally for each setup and because the additive superposition of the three single-parameter effects has not been tested for interactions. The random-tuple validation in Fig. 6 is a useful check, but it uses the same setup and the same measurement method as the fits, and the tuple coverage is not reported. The authors should either extend the validation to include interaction-sensitive points (e.g., high fe_ib1 with non-zero fe_ib2 or disc_sf_bias) or explicitly restrict the claim to differential predictions once P0 is known, in the abstract as well as in Section V.","section":"IV.B, V, and Fig. 6"}],"minor_comments":[{"comment":"Section III.A states that fe_ib1 was scanned from 0 to 60, yet Section IV.A discusses a 'systematic increase for fe_ib1 > 60' and Fig. 5a marks this with a black circle. Please clarify the actual scan range and what the black circle indicates.","section":"III.A and IV.A"},{"comment":"The caption says 'KETEK PM3325', but the detector is referred to as 'KETEK PA3325' elsewhere in the text; this should be corrected.","section":"Fig. 4 caption"},{"comment":"The saturation formula is typeset ambiguously; please define the variables e and s and give the intended mathematical expression for the logarithmic term explicitly.","section":"Eq. (2)"},{"comment":"The 'linearity of 1.034 ± 0.018' is the slope of the measured-versus-computed regression, not a coefficient of determination; the wording should be changed to 'regression slope' for clarity.","section":"IV.B"},{"comment":"Table II would be more self-contained if the breakpoints of the piecewise functions (0–40/41–60 for fe_ib1 and 0–20/21–30 for fe_ib2) were included in the caption or as separate columns.","section":"Table II"}],"recommendation":"major_revision","confidential_remarks":"The equation/table inconsistency is the main obstacle; it is likely a typographical/labeling issue that can be fixed, but the corrected model must reproduce the random-tuple validation before publication. If corrected, the paper would be a solid experimental characterization suitable for IEEE TRPMS. I would not require a full interaction study for acceptance, but the authors should temper the 'prior to experiments' claim accordingly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Useful paper with a fixable but real problem. The TOFPET2 power characterization (3.6–7.2 mW/channel across the three software settings) looks careful and credible. The performance tradeoffs—CRT, energy resolution, and trigger-level shifts between power settings—are exactly what a PET/MR system designer needs when deciding whether this ASIC fits. The power measurement method via a shunt on the 1.2-V line is described clearly, and the stability checks over count rate and overvoltage are reassuring.\n\nWhat is actually new is the complete mapping of power consumption as a function of all three configurable parameters (fe_ib1, fe_ib2, disc_sf_bias), plus the additive empirical model. The random-tuple validation showing a slope of 1.034 ± 0.018 is a nice check and suggests the model code works. But the printed model is not usable. Eq. (6) references an undefined a3, and the parameters in Table II for fe_ib2 have the wrong sign: b1 and b2 are positive, which would make dP_fe_ib2 increase with fe_ib2, while Fig. 5b shows power decreasing monotonically. The listed a1 is also about 20× too small to reproduce the plotted fe_ib1 drop. Either the equations or the table are scrambled, and a reader cannot evaluate the model without guessing.\n\nOther soft spots are minor in comparison. The model still requires a measured per-setup P0, so the claim that it computes power “prior to experiments” is overstated—it predicts the change from a baseline you must determine yourself. The additivity assumption is acknowledged, not tested, though the random-tuple check covers part of the space. The ASIC-to-ASIC comparison in Table III uses different crystal heights, so the CRT ranking is suggestive rather than decisive.\n\nThe stress-test note is right, and it does land: the analytical model is a load-bearing part of the paper’s contribution, and as printed it is not reproducible. The measurements themselves appear solid.\n\nWho should read this: anyone selecting a TOFPET2 for a PET/MR insert, and groups working with PETsys evaluation kits. It deserves a serious referee, but only with a mandatory revision that reconciles Eqs. (4)–(7) and Table II, or removes the model. I would not cite the model until that is fixed, but I would cite the measured power benchmarks and the stability data.","headline":"Careful ASIC power measurements with a model section that doesn't reproduce as printed—fixable, but equations and Table II need reconciliation.","tokens_in":19165,"tokens_out":3280,"would_cite":true,"duration_ms":31423,"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":"The TOFPET2 ASIC's power draw can be predicted from three software settings, ranging from 3.6 to 7.2 mW per channel.","keywords":["TOFPET2 ASIC","power consumption","positron emission tomography","time-of-flight PET","SiPM readout","PET/MR compatibility","application-specific integrated circuits","analytical power model"],"falsifier":"A targeted grid of two-parameter combinations, e.g., holding fe_ib2=15 and disc_sf_bias=16 while varying fe_ib1, would settle the independence assumption: if measured $P_{ch} - P_0$ differs from the sum of the single-parameter deltas $dP_{feib1}(fe_{ib1}) + dP_{feib2}(15) + dP_{discsfbias}(16)$ by more than the stated ~8% statistical error in a systematic pattern, the additive model fails off the parameter axes.","tokens_in":18138,"feed_emoji":"🔋","tokens_out":13635,"duration_ms":116801,"temperature":0.7,"pith_summary":"The paper aims to show that the TOFPET2 readout chip, an application-specific integrated circuit that digitizes silicon-photomultiplier signals in PET detectors, consumes between 3.6 and 7.2 mW per channel depending on three software settings: input-stage impedance, discriminator noise, and discriminator-noise slew rate. It further claims that per-channel power for any configuration can be computed before any experiment by adding independently measured contributions of those three settings to a count-rate-dependent baseline $P_0$. The practical point is that PET electronics inside an MRI scanner must run on low power supplied by linear regulators, since long analog cables degrade timing signals; a chip whose power draw is both low and predictable simplifies future simultaneous PET/MR systems. The same settings also change timing and energy performance, with lower power costing 20 to 40 ps in coincidence resolution time.","feed_headline":"Predict TOFPET2 ASIC power from three settings: 3.6-7.2 mW/channel","feed_subtitle":"That lets PET/MR designers balance timing performance against heat before building hardware.","key_machinery":"The load-bearing machinery is the linear superposition model of Eq. (4), which splits per-channel power into a baseline $P_0$ and three deltas: $dP_{feib1}$, $dP_{feib2}$, and $dP_{discsfbias}$. The parameters are software settings: fe_ib1 adjusts the input-stage impedance from 11 to 32 ohm, fe_ib2 adjusts a preamplifier current that sets discriminator noise and signal amplification, and disc_sf_bias adjusts the biasing of discriminator signal buffers; increasing any of the three lowers the power draw. Each delta is a least-squares fit of a piecewise linear or parabolic curve to single-parameter scans in which the other two parameters are held at zero. The experimental basis is a shunt-resistor measurement on the 1.2-V low-dropout regulator line, which isolates ASIC operating power from FPGA communication power. The model's defining move is to assert that those single-axis curves continue to hold throughout the three-dimensional parameter space, so a designer can compute the power of an arbitrary configuration triple before running the system.","core_discovery":"The central claim is that the TOFPET2 ASIC's power consumption is a software-tunable quantity, ranging from 3.6 mW/channel at minimum to 7.2 mW/channel at maximum, and that this variation is described by the additive model $P_{ch} = P_0 + dP_{feib1} + dP_{feib2} + dP_{discsfbias}$ (Eq. 4). Each delta is a piecewise linear or parabolic function of one configuration parameter, fitted to single-parameter scans performed with the other two parameters at zero; $P_0$ is the measured y-intercept at all-zero settings and absorbs count-rate dependence. Validation on random parameter tuples gave a fitted slope of 1.034 ± 0.018 and a negligible intercept of about 0.1 mW/channel, which the authors take as confirmation that the model works for arbitrary configurations. The paper further reports that power is stable across overvoltages from 0.75 V to 7.75 V and across count rates from about 1 kcps to 100,000 kcps, and that reducing power from maximum to minimum worsens coincidence resolution time by 20 to 40 ps while moving the effective photo-electron trigger thresholds.","pith_inferences":["A consequence the authors leave implicit is that the transportability of the fitted single-parameter curves can be established by two-parameter scans; if those scans reproduce the additive predictions, the model becomes a general design rule rather than a validation-set fit.","The count-rate dependence sitting entirely in $P_0$ suggests a future circuit-level model that separates digital switching current from analog bias current could predict the baseline analytically and eliminate the one remaining experimental calibration step.","The shunt-resistor metrology used here could be reapplied to other ASICs with on-board low-dropout regulators, giving system builders a comparable measure of analog front-end power that excludes FPGA and communication loads.","For PET/MR integration, average power is only part of the story: if low-power settings also change the time structure of the current drawn on the 1.2-V line, they could affect electromagnetic interference inside the MR bore in ways the paper does not quantify."],"forward_implications":["A system designer can compute per-channel power draw for any triple of the three settings from the published fit parameters plus one measured baseline, without benchtop measurements per configuration.","Choosing the minimum-power configuration (3.6 mW/channel) sacrifices 20 to 40 ps of coincidence resolution time relative to maximum power, while energy resolution changes by less than 0.5% absolute.","Because photo-electron trigger levels shift with power setting, threshold calibration (e.g., vth_t1) must be redone after switching between minimum, default, and maximum power configurations.","Measured power is stable over count rates from roughly 1 kcps to 100,000 kcps and over overvoltage from 0.75 V to 7.75 V, so the benchmark numbers are representative for realistic operation rather than single-point artifacts.","Including the estimated 1.2 mW/channel for ASIC-FPGA communication, the total 4.8 to 8.4 mW/channel remains similar to or below other PET readout ASICs such as Triroc (10 mW/channel), STiC3 (25 mW/channel), and PETA4 (<40 mW/channel)."],"supporting_citations":[{"why":"Defines the three software parameters and the manufacturer's 5-8 mW/channel power specification that anchors the measurements.","marker":"[35]"},{"why":"Details the preamplifier circuit and its 2.5 mW/channel consumption, grounding the model's current terms.","marker":"[47]"},{"why":"Provides the data-processing chain, TDC characterization, and satellite-peak analysis used to evaluate performance at each power setting.","marker":"[49]"},{"why":"Documents the evaluation hardware and the manufacturer power range that the measurements are checked against.","marker":"[44]"},{"why":"Relates input-stage impedance to timing behaviour, supporting the interpretation of the coincidence-resolution-time changes.","marker":"[33]"},{"why":"Reports 8-11 mW/channel for the prior TOFPET ASIC generation, serving as the comparison baseline for TOFPET2's lower power.","marker":"[34]"},{"why":"Provides a comparable 64-channel SiPM readout ASIC at 10 mW/channel in the power comparison table.","marker":"[22]"},{"why":"Provides the <40 mW/channel comparison point used to argue that TOFPET2's power draw is competitive.","marker":"[20]"}],"fun_headline_variants":["Tune TOFPET2 power: 3.6-7.2 mW/channel via 3 settings","Analytical model predicts TOFPET2 ASIC power draw","TOFPET2: software-tunable power from 3.6 to 7.2 mW/ch","Power model for TOFPET2: three knobs, 3.6-7.2 mW","How to set TOFPET2 power: model links 3 settings to mW"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that fe_ib1, fe_ib2, and disc_sf_bias change power consumption independently, so single-parameter curves measured with the other two at zero remain valid across all combinations; the paper states that cross-parameter interaction experiments were not performed.","fun_headline_variants_meta":{"raw":{"variants":["Tune TOFPET2 power: 3.6-7.2 mW/channel via 3 settings","Analytical model predicts TOFPET2 ASIC power draw","TOFPET2: software-tunable power from 3.6 to 7.2 mW/ch","Power model for TOFPET2: three knobs, 3.6-7.2 mW","How to set TOFPET2 power: model links 3 settings to mW"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000225,"raw_usage":{"total_tokens":1509,"prompt_tokens":1036,"completion_tokens":473,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":652,"completion_tokens_details":{"reasoning_tokens":349}},"tokens_in":652,"tokens_out":473,"duration_ms":4332,"temperature":1.0,"reasoning_tokens":349,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:01:36.123853+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A targeted grid of two-parameter combinations, e.g., holding fe_ib2=15 and disc_sf_bias=16 while varying fe_ib1, would settle the independence assumption: if measured $P_{ch} - P_0$ differs from the sum of the single-parameter deltas $dP_{feib1}(fe_{ib1}) + dP_{feib2}(15) + dP_{discsfbias}(16)$ by more than the stated ~8% statistical error in a systematic pattern, the additive model fails off the parameter axes.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the three software parameters and the manufacturer's 5-8 mW/channel power specification that anchors the measurements."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the data-processing chain, TDC characterization, and satellite-peak analysis used to evaluate performance at each power setting."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents the evaluation hardware and the manufacturer power range that the measurements are checked against."},{"cited_title":"EXPERIMENTAL RESULTS WITH TOFPET2 ASIC FOR TIME-OF-FLIGHT APPLICATIONS,","cited_arxiv_id":null,"evidence_quote":"Relates input-stage impedance to timing behaviour, supporting the interpretation of the coincidence-resolution-time changes."},{"cited_title":"1.2) , 1.2, PETsys Electronics S.A., PETsys Electronics SA, Taguspark - Lisboa Science and Technology Park, Edificio Tecnologia I, 26, 2740-257 PORTO SALVO, Portugal, Feb","cited_arxiv_id":null,"evidence_quote":"Reports 8-11 mW/channel for the prior TOFPET ASIC generation, serving as the comparison baseline for TOFPET2's lower power."},{"cited_title":"TRIROC, A VERSATILE 64-CHANNEL SIPM READOUT ASIC FOR TIME-OF-FLIGHT PET,","cited_arxiv_id":null,"evidence_quote":"Provides a comparable 64-channel SiPM readout ASIC at 10 mW/channel in the power comparison table."}],"review_version":1}