{"id":"7e055f2f-14b9-413b-b51c-dcd2107adf15","arxiv_id":"1908.02103","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":2,"one_line_summary":"The AGIPD detector installed at the European XFEL delivers the high-speed, wide-dynamic-range X-ray imaging that enabled the first megahertz serial crystallography.","lead":"AGIPD is a custom X-ray camera that records up to 352 images per 220-nanosecond pulse train at the European XFEL, combining single-photon sensitivity with a dynamic range of 10,000 photons. The paper reports the performance of the installed 1-megapixel camera and the detector systems now being built.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Calibration of medium/low gains via integration-time scan is indirect and unquantified; Sec. 10 admits gain uncertainty during readout, putting the dynamic-range claim at risk.","rationale":"The reader's weakest assumption identified the indirect electrical-stimulus calibration of medium and low gains as the key vulnerability. I agree that this is the core issue, but I sharpen it with the paper's own admission in Sec. 10 of 'increasing uncertainty of the gain determination during readout', which directly implicates the stability of that calibration across the 352 memory cells. The paper provides no systematic-uncertainty estimate for the headline performance numbers, so the margin by which the system meets the dynamic-range and gain-switching requirements cannot be assessed. This is a load-bearing concern because the central claim 'System fulfils all requirements' depends on the calibration to convert measured signals into photon numbers across all gains. The concern does not, however, warrant rejection: the high-gain calibration is direct, the system has been in user operation since September 2017, and a published megahertz crystallography experiment (Ref. 21) demonstrates real-world capability. A CONDITIONAL verdict remains appropriate, and the reader's verdict is therefore unchanged. The concrete test of comparing pulsed-capacitor versus current-source calibration would settle whether the indirect method has a systematic bias.","tokens_in":10691,"tokens_out":9399,"duration_ms":94131,"concrete_test":"Use the in-pixel pulsed capacitor (Sec. 3) to inject calibrated charge steps into each gain setting and directly measure the medium and low gain sensitivities relative to high gain; compare with the integration-time-scan results from Sec. 4. If the two methods disagree by more than the reported 0.44% nonlinearity, the indirect electrical-stimulus calibration has a systematic bias that affects the dynamic-range claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The Summary (Sec. 10) claims the SPB system meets all requirements: noise <310 e-, single-photon sensitivity, dynamic range ≥10^4 photons at 12 keV, and 4.5 MHz frame rate. The dynamic-range and gain-switching accuracy rest on the calibration in Sec. 4: high gain is calibrated with fluorescence photons, but medium and low gains are calibrated only indirectly by scanning preamplifier integration time with the in-pixel current source. This assumes the current source is stable and the preamplifier gain is independent of integration time; neither assumption is quantified. Section 10 itself admits a 'remaining shortcoming' of 'increasing uncertainty of the gain determination during readout' for the installed 1.1 ASIC, implying gain calibration drifts across the 352 memory cells. No systematic-uncertainty estimate is given for any headline number, so it is impossible to verify that the detector meets the requirements with margin.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes the AGIPD hybrid pixel detector developed for the European XFEL, covering the source requirements, the pixel ASIC architecture, the installed 1 Mpixel cameras at SPB and MID, the calibration procedure, first user experiments, and plans for the 4 Mpixel SFX system, the HIBEF system with high-Z sensors, and future faster detectors. The central claim is stated in Section 10: the AGIPD 1 Mpixel system installed at SPB fulfills all requirements, specifically noise below 310 e- (≤1.2 keV), single-photon sensitivity, a dynamic range of at least 10^4 photons at 12 keV, and a 4.5 MHz frame rate. This claim is supported by ASIC-level characterization measurements (Figures 4-6), by a hybrid calibration procedure using fluorescence photons and electrical stimuli, and by citation of successful first user experiments including MHz serial crystallography.","tokens_in":10845,"tokens_out":2232,"duration_ms":25688,"significance":"If the stated performance is correct, AGIPD is a unique instrument: it provides single-photon sensitivity at 6-12 keV together with a dynamic range of 10^4 photons and an in-burst framing rate of 4.5 MHz, which enabled the first MHz serial crystallography experiments at the European XFEL. The paper gives a useful system-level overview of a complex detector and reports direct, external-stimulus measurements of noise, transfer characteristics, and single-photon spectra rather than relying on simulation. The practical calibration method and the openly described remaining shortcomings are also valuable. The main risk is that the headline performance claim is not accompanied by quantitative uncertainty estimates on the reported values, and the medium/low-gain calibration is indirect.","major_comments":[{"comment":"The summary claim that the SPB system fulfills all requirements, in particular noise below 310 e- and dynamic range of 10^4 photons at 12 keV, is supported by measurements that appear to be performed on single ASICs or small systems rather than on the installed 1 Mpixel camera. Please state explicitly which measurements were made on the installed SPB system and which were made on ASIC-level test structures, and give uncertainties or at least confidence intervals for the headline values (for example, ENC ≈ 320 e- is reported without a statistical or systematic error). This distinction is load-bearing because the abstract and Section 10 attribute the performance to the installed system.","section":"Sections 3.1 and 10, Figures 4-6"},{"comment":"The calibration of the medium and low gain settings is indirect: their sensitivities are determined relative to the high gain by scanning the preamplifier integration time with the in-pixel constant current source, while only the high gain is calibrated with fluorescence photons. This assumes that the current source is stable and that the preamplifier gain does not depend on integration time; neither assumption is quantified. Since the dynamic-range and gain-switching accuracy claims depend directly on these relative sensitivities, please provide an estimate of the resulting systematic uncertainty or explain quantitatively why the lever-arm procedure is insensitive to these effects.","section":"Section 4"},{"comment":"The paper admits an 'increasing uncertainty of the gain determination during readout' as a remaining shortcoming of the installed AGIPD 1.1 ASIC, but does not quantify how this affects the calibrated images or the stated dynamic-range and gain-accuracy figures. Without a bound on this effect, the claim that the system fulfills all requirements, especially the dynamic-range requirement, is not fully verifiable. Please either quantify the gain uncertainty versus memory-cell index or temper the summary claim to reflect the known limitation.","section":"Section 10"}],"minor_comments":[{"comment":"There are several typographical errors, for example 'camara' should be 'camera' and 'loosing' should be 'losing'; please proofread the text.","section":"Section 5"},{"comment":"The words 'comissioning' and 'decribed' should be 'commissioning' and 'described'; also 'sheild' in Section 2 should be 'shield'.","section":"Sections 7.1 and 8"},{"comment":"The text says the memory stores 352 images, but the footnote states that the current firmware reads 300 separate amplitude and gain frames; please clarify whether the remaining 52 memory cells are unused or used for other purposes.","section":"Section 3, footnote 9"},{"comment":"Several references contain formatting artifacts (for example, missing spaces or misplaced punctuation); please normalize the bibliography to the journal style.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"This is a detector description paper rather than a new-physics result, and the heavy self-citation is normal for a large multi-institute detector collaboration. The main issue is that the central 'fulfills all requirements' statement would be strengthened by clearer separation of ASIC-level and system-level measurements and by quantitative systematic uncertainties, particularly for the indirect medium/low-gain calibration. In my view these requests are within the manuscript's scope and do not require new beam time, so major revision is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a system status report, not a new physics result. The core numbers – noise below ~310 e-, single-photon sensitivity, 10^4 photon dynamic range, 4.5 MHz operation – have appeared in earlier AGIPD papers from the same collaboration. What this paper adds is the installed-system perspective: how the 1-Mpixel camera is actually put together, the calibration flow, the new readout board for SFX/HIBEF, and the ecAGIPD electron-collecting variant. If you want to know what a deployed XFEL integrating detector looks like in practice, this is the most complete single reference I know of.\n\n\n\nThe central claim that the SPB system meets requirements is supported. Figures 4-6 show measurements with an IR laser, fluorescence lines, and a Petra III beam; the cited megahertz serial crystallography run (Wiedorn et al.) is genuine external validation that the camera works for science. The paper is also honest about its warts: connector reliability problems at SPB, the gain uncertainty during readout, and a new ASIC version to fix it. Citation pattern is normal for a multi-institute detector collaboration; I don't see circularity.\n\n\n\nThe weak spot is exactly where the reader put it. The medium and low gain calibrations are derived from an integration-time scan with the in-pixel current source, assuming the source is stable and the preamplifier gain is independent of integration time. The paper acknowledges the indirectness but gives no systematic uncertainty for any headline number. And it is not purely cosmetic: Section 10 admits the installed 1.1 ASIC has \"increasing uncertainty of the gain determination during readout,\" and Section 4 says remaining calibration uncertainty drove the AGIPD 1.2 redesign. So the dynamic-range claim has a real, unquantified component. The stress-test note says this \"puts the dynamic-range claim at risk\" – I would soften that. The detector clearly works well enough for the first user experiments; what is at risk is only the margin by which it meets the 10^4-photon specification, not the claim that it operates at that scale.\n\n\n\nMinor: the abstract says 6.5 MHz, the summary says 4.5 MHz, and the paper never clearly states which limit applies to the installed system. That should be cleaned up.\n\n\n\nWho is this for: detector developers at XFELs and users who want a system-level understanding of AGIPD. It deserves a serious referee and would come back with minor revisions, mainly asking for uncertainty estimates on the headline numbers and a clarified speed claim.","headline":"A candid, useful system status report: the installed AGIPD cameras work and the headline numbers hold up, but the dynamic-range claim rests on an indirect calibration with an unquantified systematic uncertainty.","tokens_in":11519,"tokens_out":3584,"would_cite":true,"duration_ms":36337,"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":"AGIPD, a hybrid pixel detector with per-pixel adaptive gain, claims to meet an X-ray free-electron laser's extreme requirements: single-photon sensitivity, a dynamic range of $10^4$ photons per pixel, and a 4.5 MHz frame rate.","keywords":["X-ray detector","hybrid pixel detector","AGIPD","adaptive gain","single photon sensitivity","MHz frame rate","serial femtosecond crystallography","free-electron laser"],"falsifier":"Compare the detector's response to a directly calibrated X-ray beam at a synchrotron, using attenuators and Poisson statistics to cover the medium- and low-gain ranges; if the reconstructed photon counts disagree with the electrical-stimulus calibration by more than the stated 0.44% nonlinearity, the indirect calibration premise is wrong.","tokens_in":10510,"feed_emoji":"📸","tokens_out":8798,"duration_ms":81844,"temperature":0.7,"pith_summary":"The paper claims that AGIPD (Adaptive Gain Integrating Pixel Detector), a hybrid pixel detector in which each pixel's amplifier has three automatically selected gains, meets the requirements of the X-ray free-electron laser: noise below 310 electrons (about 1.2 keV), single-photon sensitivity at energies above about 6 keV, a dynamic range of at least $10^4$ photons per pixel at 12.4 keV, and a frame rate of 4.5 MHz. A one-megapixel AGIPD camera has been in user operation at a laser endstation since September 2017 and produced the first megahertz serial crystallography data. This matters because the facility's pulses arrive in 2700-pulse trains spaced by 220 ns, so a detector must store many frames on the pixel and read them out between trains. The paper also describes calibration, a new readout board, and an electron-collecting variant for higher energies.","feed_headline":"Megapixel X-ray camera hits 4.5 MHz with photon-level sensitivity","feed_subtitle":"One frame can register a single photon and a 10,000-photon flash in the same pixel.","key_machinery":"The load-bearing mechanism is the AGIPD readout ASIC: a 64x64 array of 200 µm pixels, each with a charge-sensitive preamplifier with three switchable gains, a discriminator that selects the gain when the preamplifier output crosses a threshold, a correlated double sampling stage, a 352-image analogue memory, and a second buffer for readout. An in-pixel constant current source and pulsed capacitor provide electrical calibration stimuli. This lets every pixel choose its own sensitivity for every pulse, so a single frame can contain both single photons and intense Bragg spots; the 99.4 ms gap between pulse trains is then used to digitise the stored frames.","core_discovery":"The central result is that a single pixel can be simultaneously sensitive to a single 6 keV photon and to a $10^4$-photon flash by switching its amplifier gain in three steps under discriminator control. Each 200 µm pixel contains a charge-sensitive preamplifier whose feedback capacitance is switched from 60 fF to 3 pF and 10 pF; a correlated double sampling stage removes reset noise, and the resulting amplitude plus gain bit is stored in a 352-frame analogue memory. A one-megapixel system built from these ASICs reaches a reported equivalent noise charge below 310 electrons, better than 0.44% nonlinearity up to $5\\times10^3$ 12.4 keV photons, and has been used at 1.1 MHz to record diffraction patterns from protein microcrystals.","pith_inferences":["Because the medium- and low-gain calibration is indirect, the quoted dynamic-range accuracy likely carries a systematic uncertainty that is not reflected in the calibration constants; a direct X-ray cross-check at a synchrotron would settle this.","The adaptive-gain on-pixel memory architecture is not tied to this laser's specific 220 ns spacing; it could be adapted to other high-repetition-rate sources, such as electron diffraction or plasma X-ray sources, whenever frames arrive in bursts.","The power-scaling argument implies that below about 100 µm pixel pitch, continuous 100 kHz operation becomes thermally impractical before bandwidth does; future designs may need to trade pixel size for speed or use per-pixel digitisation to cut analog readout."],"forward_implications":["A 1-megapixel AGIPD camera can record a full diffraction pattern at the facility's burst rate, making megahertz serial femtosecond crystallography a practical experiment.","At the installed system, users can collect up to 352 images per pulse train with per-pixel dynamic range spanning single photons to $10^4$ photons at 12.4 keV.","The new readout board eliminates most connectors and power supply cabling, so future 4-megapixel systems can operate each module standalone with only an optical fibre and a single power supply.","For the high-energy beamline, an electron-collecting ASIC with high-Z sensors is designed to detect photons up to about 25 keV, where silicon becomes transparent.","At future continuous-wave operation around 100 kHz, the current detector's readout bandwidth and power dissipation would be limiting; the paper identifies in-pixel digitisation as the necessary next step."],"supporting_citations":[{"why":"Defines the detector requirements (frame timing, dynamic range, noise) that AGIPD must meet.","marker":"[3]"},{"why":"Describes the assembled AGIPD detector system and its overall architecture.","marker":"[7]"},{"why":"Reports the noise and performance characterisation of the AGIPD ASIC.","marker":"[10]"},{"why":"Provides the full-scale chip characterisation including transfer curves and gain switching.","marker":"[15]"},{"why":"Demonstrates single-photon sensitivity of the AGIPD pixel.","marker":"[17]"},{"why":"Shows the first megahertz serial crystallography experiment performed with the installed detector.","marker":"[21]"},{"why":"Supplies the sensor design and radiation-hardness measures underlying the detector's performance.","marker":"[11]"},{"why":"Introduces the calibration circuitry concept used for the gain calibration.","marker":"[22]"}],"fun_headline_variants":["X-ray camera captures single photons at megahertz rates","From one photon to 10,000 in a single pixel","AGIPD camera: 6.5 MHz burst imaging with single-photon sensitivity","1-megapixel X-ray detector with 352-frame analog memory","Pixel-level gain switching enables wide X-ray dynamic range"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported dynamic range and gain-switching accuracy depend on the stability of the in-pixel current source and on the preamplifier gain being independent of integration time, because the medium- and low-gain calibrations are derived from an electrical scan rather than direct photon exposure.","fun_headline_variants_meta":{"raw":{"variants":["X-ray camera captures single photons at megahertz rates","From one photon to 10,000 in a single pixel","AGIPD camera: 6.5 MHz burst imaging with single-photon sensitivity","1-megapixel X-ray detector with 352-frame analog memory","Pixel-level gain switching enables wide X-ray dynamic range"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001076,"raw_usage":{"total_tokens":4570,"prompt_tokens":1078,"completion_tokens":3492,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":694,"completion_tokens_details":{"reasoning_tokens":3403}},"tokens_in":694,"tokens_out":3492,"duration_ms":27951,"temperature":1.0,"reasoning_tokens":3403,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:53:25.122420+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the detector's response to a directly calibrated X-ray beam at a synchrotron, using attenuators and Poisson statistics to cover the medium- and low-gain ranges; if the reconstructed photon counts disagree with the electrical-stimulus calibration by more than the stated 0.44% nonlinearity, the indirect calibration premise is wrong.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the detector requirements (frame timing, dynamic range, noise) that AGIPD must meet."},{"cited_title":"Allahgholi et al., ”The Adaptive Gain Integrating Pixel Detector AGIPD”, JINST 11, C02066 (2015), DOI:10.1088/1748-0221/11/02/C02066","cited_arxiv_id":null,"evidence_quote":"Reports the noise and performance characterisation of the AGIPD ASIC."},{"cited_title":"Mezza et al., ”Characterization of AGIPD 1.0: the full scale chip”, Nucl","cited_arxiv_id":null,"evidence_quote":"Provides the full-scale chip characterisation including transfer curves and gain switching."},{"cited_title":"Becker et al.: ”The single photon sensi- tivity of the Adaptive Gain Integrating Pixel Detector”, Nucl","cited_arxiv_id":null,"evidence_quote":"Demonstrates single-photon sensitivity of the AGIPD pixel."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows the first megahertz serial crystallography experiment performed with the installed detector."},{"cited_title":"Schwandt et al, ”Design of the AGIPD sensor for the European XFEL”, Nucl","cited_arxiv_id":null,"evidence_quote":"Supplies the sensor design and radiation-hardness measures underlying the detector's performance."}],"review_version":1}