{"id":"14a77bd5-0e85-4499-8610-07bff8b60d97","arxiv_id":"2607.14691","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"ATLAS ITk strip-sensor production reached 91% acceptance overall; four rejected batches showed non-uniform p-stop doping.","lead":"This paper summarizes the quality-control program for the silicon strip sensors that will go into ATLAS's upgraded inner tracker, reporting a 2.8% rejection rate and a few systematic fabrication issues. It matters because the ITk detector's tracking performance depends on catching bad sensors before installation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"P-stop V_PT acceptance band (12–18 V) is not validated against post-irradiation inter-strip isolation; batch rejection criteria may not predict tracker performance.","rationale":"The reader identified the V_PT threshold as the weakest assumption; I agree. The paper's central quantitative claims are the acceptance and rejection rates, and the p-stop batch rejection is a significant component. Without validation linking V_PT to post-irradiation performance, the rejection criteria are unsupported. The concrete test would settle whether the 12 V lower bound is meaningful. The arithmetic inconsistency in Table 1 is real but does not alter the qualitative conclusion; it should be corrected but is not the primary scientific concern. No evidence of fraud or misrepresentation; the paper is a straightforward QC summary with a missing validation step. Therefore, the conditional verdict is appropriate.","tokens_in":7629,"tokens_out":8065,"duration_ms":72018,"concrete_test":"Irradiate two samples of ~20 sensors each — one with pre-irradiation V_PT of 10–14 V (near/below the 12 V limit) and one with V_PT of 18–25 V (well inside the band) — to 1.6×10^15 neq/cm^2 and 66 Mrad. After irradiation, measure inter-strip isolation (e.g., inter-strip resistance or post-irradiation PTP) and compare distributions. If the low-V_PT group does not show significantly worse isolation, the 12–18 V acceptance band is not predictive and the four batch rejections are unjustified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim — that QC achieved a 2.8% rejection rate and accepted 91% of production — hinges on the validity of the pass/fail criteria, especially the punch-through protection voltage (V_PT) acceptance band (12–18 V) used to reject four batches for non-uniform p-stop doping (§5, §6). The paper converts V_PT to a p-stop density of 2×10^12 cm^-2 using TCAD simulations [10] that are submitted for publication, but provides no direct evidence that this room-temperature threshold predicts inter-strip isolation after irradiation to 1.6×10^15 neq/cm^2 and 66 Mrad. If the threshold is too strict, the rejection rate is overestimated and good sensors are discarded; if too loose, accepted sensors may fail in operation. The paper also does not address the false-negative rate of the QA sampling (one test-structure wafer per batch), so the reported 91% acceptance could be optimistic. This is not an internal inconsistency, but a correctness risk in the interpretation of the QC results as a success. The arithmetic discrepancy in Table 1 (sum 21,722 vs. stated 21,751) is a separate data-integrity issue that should be corrected, but the V_PT validation is the load-bearing scientific concern.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This proceedings-style paper reports the status and results of the quality-control (QC) program for ATLAS18 ITk n+-in-p strip sensors produced by Hamamatsu. It describes the standardized testing chain (visual inspection, IV/CV, full/fast strip tests, long-term stability, metrology) applied across seven institutes, and summarizes production yields: by September 2025, 23,614 sensors had been delivered (98% of the required 24,010), 22,376 had been QC-tested, and 21,751 were accepted (claimed 91% of the required total). The paper attributes the low overall rejection rate of 2.8% to two QC-rejected batches (instability / non-recoverable IV breakdown) and four QA-rejected batches (non-uniform p-stop doping identified via punch-through-protection voltage measurements), with an additional 1.8% individual-sensor rejection rate. It also discusses static-charge recovery procedures and presents a case study of p-stop doping non-uniformity across wafers.","tokens_in":8004,"tokens_out":6363,"duration_ms":63972,"significance":"If the reported yields and rejection rates are accurate, this is an important large-scale production QC result for the HL-LHC upgrade: it demonstrates that a multi-site, standardized QC chain can process ~500 sensors/month, that most individual failures are recoverable static-charge effects, and that test-structure QA can catch a systematic p-stop processing issue. The paper's strengths include the unusually large dataset, the explicit pass/fail specifications, the independent HPK-vs-ATLAS comparison for full-depletion voltage, and the reproducible standardized procedures. However, the central yield claims contain an internal arithmetic inconsistency, and the interpretation of the p-stop rejection criterion as a performance indicator is not established in this paper. These issues must be fixed before the quantitative conclusions can be relied upon.","major_comments":[{"comment":"The 'Accepted' column entries sum to 21,722, not the stated total 21,751. With the stated QC-tested total of 22,376, the column-sum accepted value gives a rejected fraction of 654/22,376 = 2.92%, not 2.8%, and an acceptance fraction of 21,722/24,010 = 90.5%, not 91%. Please correct Table 1 and recompute all derived percentages in the Abstract, Section 5, and the Conclusion.","section":"Table 1, p. 5"},{"comment":"The sentence 'Six batches were rejected ... These account for 2.8% of the total tested sensors' is internally inconsistent with the stated batch size of 30–50 sensors: six full batches would account for at most 300 of 22,376 tested sensors (~1.3%), not 2.8%. The 2.8% must be a combination of batch-level and individual-sensor rejections; the text should state this decomposition explicitly, since the current wording makes the claim appear arithmetically impossible.","section":"§5, p. 5"},{"comment":"The four p-stop batch rejections and the claimed 'clear correspondence' between QA test-structure results and main-sensor punch-through behavior are judged against a V_PT acceptance band of 12–18 V and an associated p-stop density of 2×10^12 cm^-2 obtained from TCAD simulations in reference [10], which is submitted for publication. No evidence is given that this room-temperature threshold predicts inter-strip isolation after irradiation to 1.6×10^15 neq/cm^2 and 66 Mrad. This is a correctness risk for interpreting the rejection rate as a quality metric. Please either provide a reference to prior validation of this band, add a short argument linking V_PT to inter-strip isolation, or explicitly state that the acceptance numbers are relative to ATLAS production specifications and not to post-irradiation performance.","section":"§5–§6"},{"comment":"The batch-rejection criterion in footnote 1 (reject a batch when ≥4 sensors fail the same QC test) is introduced without statistical justification. Given that batch subsets are only 2–5% of the batch, the false-positive and false-negative properties of this threshold are unknown, and the QA sampling is one wafer per batch with eight peripheral PTP measurements. The paper's conclusion that QA failures are 'reliable indicators' of main-sensor problems is based on selected batches shown in Figs. 7–8, not on a sensitivity study. Please add a quantitative discussion of sampling sensitivity or temper the wording accordingly.","section":"§6, footnote 1"}],"minor_comments":[{"comment":"Percentages are reported without statistical uncertainties. For small numbers (e.g., 12% for R0, 10% for R2), the corresponding counts should be quoted so the reader can judge the significance; this is especially relevant because the rejection rates are dominated by batch-level rejections.","section":"Table 1 and §3"},{"comment":"The phrase 'Over 91% of the production, totaling over 590 batches, were tested and accepted as-is' is grammatically ambiguous: 'production' refers to sensors, but 'totaling' refers to batches. Please rephrase to state the number of sensors and the number of batches separately.","section":"Abstract"},{"comment":"The conclusion refers to 'nearly 21,800 strip sensors'; the Table 1 total (21,751, or 21,722 per the column sum) does not support '21,800' as a precise value. Use the corrected number from Table 1.","section":"Conclusion"},{"comment":"Reference [4] (Ullán et al., NIM A 981, 164521) lists a URL pointing to the JINST article for reference [3]. The URL should be corrected to the NIM A DOI or removed.","section":"References"},{"comment":"The captions say 'For better visibility, only every fifth batch is mentioned in the caption.' It is unclear which batches are displayed in the figure versus named in the caption; please clarify.","section":"Fig. 7"}],"recommendation":"major_revision","confidential_remarks":"I share the reader's conditional verdict. The dataset is valuable and the QC framework is described clearly, but the exact yield claim is compromised by the Table 1 sum discrepancy and by the ambiguous 'six batches account for 2.8%' sentence. The p-stop threshold issue is a legitimate correctness risk, although it may be addressable by referencing existing ATLAS specifications rather than by adding new radiation data. These are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth your time if you care about how the ITk sensor production actually went. The genuinely new material is the full-statistics yield report, the static-charge recovery observations with the one-year persistence claim, and the lateral p-stop gradient on full-size sensors. The p-stop study is the most valuable part: it shows a clear spatial gradient in punch-through voltage consistent with reduced p-stop doping and demonstrates a correlation between test-structure QA failures and main-sensor performance. The paper is clearly written and the HPK-vs-ATLAS V_fd comparison is a nice inter-laboratory check.\n\nSoft spots: Table 1 does not add up — the row sum is 21,722, not the stated 21,751, so the rejection rate is 2.9% rather than 2.8%. The footnote saying accepted numbers include batch rejections is confusing and needs rewording. The one-year persistence of charge-up recovery is asserted without supporting data. The V_PT acceptance band (12–18 V) is taken as a given; the paper does not show that it predicts post-irradiation inter-strip isolation. That may be fine for a QC report, but the authors should at least cite where that validation is documented. Also, the \"over 91%\" phrasing is optimistic given 21,751/24,010 is 90.6%, and there are no error bars. QA sampling is one test-structure wafer per batch and the false-negative rate is not quantified; the correlation study mitigates this but is not a complete answer.\n\nNone of this is fatal. The qualitative picture stands. This is a conference proceedings, not a groundbreaking technique, but it is a solid, honest production report that deserves referee time. Fix the arithmetic and footnote; the rest is fine.","headline":"A solid, honest production QC summary with a genuinely useful p-stop case study; the central yield numbers have a fixable arithmetic inconsistency and the V_PT spec validation is left to prior work.","tokens_in":708,"tokens_out":2367,"would_cite":true,"duration_ms":61671,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["29.40.Gx","29.40.Wk"],"model":"deepseek-v4-flash","headline":"This paper reports that ATLAS's ITk strip-sensor quality-control program accepted 91% of required production (21,751 sensors), rejected six of 642 batches (2.8%), and identified non-uniform p-stop doping — detected via punch-through voltage","keywords":["quality control","silicon strip sensors","ATLAS ITk","HL-LHC","p-stop doping","punch-through voltage","sensor production","static charge"],"falsifier":"Irradiate sensors from a rejected low-V_PT wafer and from an accepted high-V_PT wafer to a fluence of 1.6×10^15 n_eq/cm^2, then measure inter-strip resistance and strip noise as a function of position. If the low-V_PT side shows equal or better isolation than the high-V_PT side, the 12 V rejection threshold is unnecessarily strict; if sensors passing at high V_PT degrade, the threshold is too loose. This can be done with test structures already present on half-moons.","tokens_in":7557,"feed_emoji":"🔬","tokens_out":5294,"duration_ms":46840,"temperature":0.7,"pith_summary":"The paper reports that the multi-year quality-control program for the ATLAS ITk silicon strip sensors, run across seven laboratories at roughly 500 sensors per month, has accepted 91% of the required production: 21,751 sensors, with an overall rejection rate of 2.8% (six rejected batches out of 642). It identifies non-uniform p-stop doping as the main systematic failure mode, discovered through punch-through protection voltage (V_PT) measurements that show a lateral gradient across affected wafers, falling below the 12 V acceptance limit on one side. A sympathetic reader would care because this is evidence that the detector's sensor supply is on track for the HL-LHC, and because the paper shows how a specific fabrication-process issue can be caught and isolated. It also shows that static charge accumulated during shipping is reversible through UV irradiation, ionizing air, or baking, which recovered a substantial fraction of initial electrical failures.","feed_headline":"ATLAS18 QC accepts 91% of sensors; six batches rejected","feed_subtitle":"Non-uniform p-stop doping, the top systematic failure, was caught by punch-through voltage checks before installation.","key_machinery":"The punch-through protection (PTP) voltage measurement is the key mechanism: it records the threshold voltage at which parasitic punch-through current flows between neighbouring strip implants, and this voltage acts as an indirect measure of p-stop doping density, with the 12 V lower limit calibrated to 2×10^12 cm^-2 via TCAD simulation. The measurement on full-size sensors from affected batches, taken at multiple positions, reveals the spatial doping gradient that the standard QA sample (eight test-structure measurements on one wafer per batch) can miss. The QC workflow itself — IV/CV, full and fast strip tests, long-term stability checks, visual inspection and metrology — provides the stat","core_discovery":"The central discovery is a production-scale demonstration that a distributed QC program can hold a 2.8% rejection rate across roughly 24,000 silicon strip sensors, and the isolation of a specific, systematic defect: a lateral gradient in p-stop doping across the wafer, seen in a monotonic decrease of punch-through voltage V_PT from left to right, dipping below the 12 V specification (corresponding to a p-stop density of 2×10^12 cm^-2 from TCAD simulations). Four of six rejected batches failed because of this non-uniformity, while two failed due to non-recoverable IV breakdown and electrical instability. The paper also establishes a clear correspondence between QA test-structure results and t","pith_inferences":["If the V_PT-to-p-stop-density calibration is transferable, then the 12–18 V acceptance band could also be used as a spatially resolved quality metric in future sensor runs, beyond the six affected batches.","The acceptance thresholds are based on room-temperature measurements; a direct irradiation study comparing inter-strip isolation for low-V_PT vs high-V_PT sides of the same wafer would test whether the rejection band is the right cutoff for post-irradiation performance.","Since the paper found no CV failures without accompanying IV failures, it may be possible to simplify QC by dropping standalone CV testing on the subset that already passes IV, reducing throughput time — a cost-saving extension the authors do not propose."],"forward_implications":["If the reported yield holds through the end of production, ATLAS will have roughly 21,800 accepted sensors, sufficient for the ITk strip tracker's 17,888 installed sensors with a replacement margin.","The demonstrated correlation between QA test-structure results and full-size sensor behaviour means batch-level QA can continue to serve as a reliable release gate, saving full-sensor testing cost.","The recovery procedures (UV-A/UV-C irradiation, ionizing air blower, high-temperature baking) convert a substantial share of initial electrical failures into accepted sensors, and the effect persists over at least one year.","The p-stop gradient points to a furnace-dependent process non-uniformity; tracking V_PT distributions against furnace equipment (A vs B) provides an in-production early-warning signature."],"fun_headline_variants":["91% of ATLAS18 sensors pass; p-stop gradient flags 4 batches","Punch-through voltage test isolates p-stop doping gradient in ATLAS18","ATLAS18 QC: 24k sensors, 2.8% reject, p-stop non-uniformity top cause","Systematic p-stop defect caught by PT voltage in ATLAS18 production","90%+ acceptance in ATLAS18 strip sensors; four batches fail p-stop test"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the pass/fail thresholds — especially the V_PT acceptance band of 12–18 V and its mapping to a p-stop density of 2×10^12 cm^-2 from TCAD simulations — reliably predict inter-strip isolation after the sensors receive the HL-LHC radiation dose; the paper offers no direct measurement connecting room-temperature V_PT to post-irradiation performance.","fun_headline_variants_meta":{"raw":{"variants":["91% of ATLAS18 sensors pass; p-stop gradient flags 4 batches","Punch-through voltage test isolates p-stop doping gradient in ATLAS18","ATLAS18 QC: 24k sensors, 2.8% reject, p-stop non-uniformity top cause","Systematic p-stop defect caught by PT voltage in ATLAS18 production","90%+ acceptance in ATLAS18 strip sensors; four batches fail p-stop test"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000285,"raw_usage":{"total_tokens":1586,"prompt_tokens":888,"completion_tokens":698,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":632,"completion_tokens_details":{"reasoning_tokens":601}},"tokens_in":632,"tokens_out":698,"duration_ms":6566,"temperature":1.0,"reasoning_tokens":601,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T01:19:41.711965+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Irradiate sensors from a rejected low-V_PT wafer and from an accepted high-V_PT wafer to a fluence of 1.6×10^15 n_eq/cm^2, then measure inter-strip resistance and strip noise as a function of position. If the low-V_PT side shows equal or better isolation than the high-V_PT side, the 12 V rejection threshold is unnecessarily strict; if sensors passing at high V_PT degrade, the threshold is too loose. This can be done with test structures already present on half-moons.","supporting_citations":[],"review_version":1}