{"id":"294cf76f-8899-4364-b443-a7d2aeaef133","arxiv_id":"2611.17065","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"FPGA anti-tamper sensors drift substantially with aging and temperature, and their detection and response mechanisms can be bypassed, so protection must be assessed over the full sensing-to-clearing path.","lead":"This paper argues that the anti-tamper sensors protecting FPGAs, such as voltage, clock, and temperature monitors, are themselves a fragile and often bypassable attack surface. It combines recent attack results with new measurements of sensor drift under accelerated aging and temperature changes to motivate hardened detection and response designs.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Aging measurements rest on a single DUT without error bars, and the link between measured drift and exploitable attack success is deferred, not demonstrated.","rationale":"The reader's weakest_assumption correctly identifies that the paper does not demonstrate a link between measured drift and exploitable attack success, and that the aging data lack replication. My stress-test analysis confirms this as the single most load-bearing concern. The paper is honest, well-scoped, and the limitations are stated, but the central security claim is broader than the new evidence alone supports. The concrete test I propose directly addresses both gaps: multi-DUT replication to assess representativeness, and a combined aging-attack experiment to establish whether drift actually enables bypass. The reader's verdict of CONDITIONAL is appropriate; I agree and recommend no change. The paper should be accepted conditional on these validations, which the authors themselves identify as future work. No theatrical language or ad hominem is needed; the concern is technical and concrete.","tokens_in":12590,"tokens_out":682,"duration_ms":9628,"concrete_test":"Replicate the accelerated aging experiment on at least three additional DUTs of the same model (Spartan-7 XC7S50) under identical conditions (85°C, 1.2 V core, 535 h). Report RO frequency, TDC tap count, and XADC temperature with per-device traces and error bars. If the drift magnitude and shape vary by more than ±20% across devices, the single-DUT characterization is insufficient to support lifetime-wide conclusions. Separately, design and execute a combined aging-plus-attack experiment: after 100 h of accelerated aging, attempt a voltage glitch attack on the aged device while monitoring whether the sensor's detection threshold has shifted enough to miss the disturbance. If the attack succeeds on the aged device but not on a fresh control, the drift-to-exploit link is established; if not, the security impact remains speculative.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central security conclusion—that sensor drift and environmental effects can open silent detection gaps and enable bypass—rests on two unvalidated premises. First, the aging characterization (Figs. 2, 3, 6) uses a single Spartan-7 XC7S50 board with no reported device-to-device variation or error bars. The authors acknowledge this implicitly by reporting only point measurements, but the claim that the observed drift is representative of the device population is unsupported. Second, and more critically, the paper explicitly defers the connection between measured drift and actual attack success: Section VII states 'Future evaluations should connect the observed sensor drift to attack success and response completion across devices and combined environmental conditions.' Without this link, the measurements establish that sensor baselines change, but not that these changes create exploitable security gaps. The XADC zero readback (Fig. 6) is a striking observation, yet the authors note it does not identify the failure mechanism [b131] and no alarm was observed. The paper's strongest claim—that sensors are a fragile attack surface—therefore remains a plausible hypothesis supported by qualitative evidence, not a demonstrated vulnerability. The reader's verdict correctly identifies this as the weakest assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper consolidates recent attack results on FPGA anti-tamper sensors and presents new experimental data on accelerated aging and temperature-dependent spatial variation of soft-IP sensors (RO, TDC) and hard-IP XADC. It argues that sensor reliability degrades over lifetime and under environmental variation, creating potential detection gaps that could be exploited. The paper proposes countermeasures including asynchronous sensing/response, health monitoring, and controlled recalibration. The central claim is that sensors themselves are a fragile attack surface and that both detection and response can be bypassed without invasive tampering.","tokens_in":12831,"tokens_out":2039,"duration_ms":136908,"significance":"If the claims hold, the paper provides a useful consolidation of attack vectors and highlights that sensor baseline drift and environmental sensitivity are underexplored in threat models. The aging measurements (RO -2.6%, TDC -44%, XADC zero readback) are novel and raise legitimate concerns about long-term reliability of sensor-based protection. The paper also usefully distinguishes detection failure from response failure. However, the significance is tempered by the acknowledged lack of direct demonstration that these drifts translate into exploitable attack success, and by the limited experimental scope (single DUT, no error bars). The paper is more a position paper with supporting measurements than a definitive vulnerability demonstration.","major_comments":[{"comment":"The aging characterization is performed on a single DUT (Spartan-7 XC7S50) with no reported device-to-device variation or error bars. The authors acknowledge this implicitly by reporting only point measurements, but the claim that the observed drift is representative of the device population is unsupported. Without multiple samples or statistical bounds, the generalizability of the 2.6% RO drift and 44% TDC tap count reduction is unknown. This limits the strength of the conclusion that 'sensor reliability cannot be assumed constant over a device lifetime' (b4 context). The authors should either provide multiple-device data, discuss known device-to-device variation from literature, or explicitly state this as a single-sample observation and temper the claim accordingly.","section":"IV.A.1, Figs. 2-3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper explicitly defers the connection between measured drift and actual attack success: 'Future evaluations should connect the observed sensor drift to attack success and response completion across devices and combined environmental conditions.' Without this link, the measurements establish that sensor baselines change, but not that these changes create exploitable security gaps. The paper's strongest claim—that sensors are a fragile attack surface—therefore remains a plausible hypothesis supported by qualitative evidence, not a demonstrated vulnerability. The authors should either provide at least one concrete attack scenario where the measured drift would enable a bypass, or clearly state that the paper's contribution is the characterization of drift, not the demonstration of exploitability.","section":"VII, Conclusion","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The XADC zero readback observation is striking, but the authors note that 'constant chamber temperature does not establish constant die temperature' and 'without an independent thermal reference, the measurements cannot distinguish sensor drift from changes in self-heating.' Furthermore, they state that 'continued soft-IP operation does not rule out an XADC-specific conversion, clocking, reference, or readout failure' and that 'no alarm was observed.' The failure mechanism is unidentified. While the observation is valuable as a data point, it does not support a conclusion about sensor security without understanding whether the failure is a one-time hardware defect, a systematic issue, or a recoverable condition. The paper should discuss whether the zero readback is reproducible or could be triggered by an attacker.","section":"IV.B.1, Fig. 6","authors_acknowledge":"Continued soft- IP operation does not rule out an XADC-specific conver- sion, clocking, reference, or readout failure.","authors_acknowledge_p":0.9},{"comment":"The temperature-dependent TDC measurements show spatial variation and ITD effects, but the experimental setup uses a different board (NewAE CW305 with XC7A35T) than the aging experiment (Digilent Arty S7-50 with XC7S50). The authors do not discuss whether the observed spatial variation is device-specific or generalizable. Additionally, the temperature range (0°C to 80°C) is within commercial specifications, so the observed variation may be normal operation rather than a security vulnerability. The paper should clarify how these measurements inform threat models, given that sensors are typically calibrated for the expected operating range.","section":"IV.A.2, Fig. 5","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proposed countermeasures (asynchronous sensing/response, health monitoring, bounded recalibration, placement characterization) are reasonable but are presented at a high level without implementation details or validation. The paper acknowledges that 'these recommendations require validation of the complete sensing-to-clearing path under combined attacks and environmental variation.' As such, the countermeasure section reads more as a list of desiderata than a concrete contribution. The authors should either provide more specific design guidance (e.g., circuit-level recommendations) or clearly state that these are directions for future work.","section":"V, Countermeasures","authors_acknowledge":"The effectiveness of this approach depends on two distinct requirements, namely the sensing mechanism must identify the disturbance, and the response must complete before the adversary achieves the security objective.","authors_acknowledge_p":0.64}],"minor_comments":[{"comment":"The abstract states 'we show that both the detection and response mechanisms of state-of-the-art sensors can be bypassed without invasive tampering.' This claim is supported by citing prior work (Chypnosis, Chypothermia, GlitchSnipe) rather than new experiments. The phrasing could be clarified to indicate that the paper consolidates and extends prior results, rather than claiming new demonstrations of bypass.","section":"Abstract","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The introduction mentions 'accelerated aging and temperature variation affect sensor outputs' but does not preview the key quantitative findings (2.6% RO drift, 44% TDC drift, XADC zero readback). Adding a brief summary of the main experimental results would help readers assess the paper's contribution early.","section":"I, Introduction","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The aging experiment uses 20% overvoltage (1.2V vs 1.0V nominal) and 85°C temperature. The authors should discuss whether these stress conditions are representative of realistic deployment scenarios or are intended to accelerate aging. The relationship between accelerated aging time and equivalent real-time aging is not provided, making it difficult to interpret the 535-hour result in terms of device lifetime.","section":"IV.A.1, Figs. 2-3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The TDC tap count drops from 58.0 to 32.4 taps (44% reduction). The authors attribute this to aging slowing signal propagation, but do not discuss whether the TDC remains functional or whether the reduced tap count still allows meaningful measurements. A 44% reduction could push the sensor outside its calibrated operating range; the paper should comment on whether the sensor would still detect anomalies.","section":"IV.A.1, Fig. 3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The XADC temperature readings show a 6.0°C increase over the first 14 hours, but the authors note this could be self-heating rather than sensor drift. The paper should consider whether the initial increase is consistent with expected self-heating from the heavy workload (four large ROs) and whether the subsequent zero readback could be related to thermal stress on the XADC reference circuitry.","section":"IV.B.1, Fig. 6","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The color scale in Figure 5 is not clearly labeled in the caption. The caption states 'Red and blue indicate increases and decreases relative to the reference measurement; white indicates approximately zero change.' However, the scale is labeled 'Mean TDC (Operating Temperature - Room Temperature)' which is in units of taps. The figure would benefit from a color bar with numerical values.","section":"IV.A.2, Fig. 5","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Figure 1 shows the sensor architecture but the labels are small and some abbreviations (PFD, CP, LP, VCO) are not defined in the caption. The figure is referenced in the text but the caption should be self-contained.","section":"II.B.2, Fig. 1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Several references are to arXiv preprints or works in progress (e.g., [4], [5], [13]). While this is acceptable for a timely paper, the authors should ensure that the cited results are publicly available and that the paper does not rely on unpublished results for its main claims. The paper's own measurements are independent, so this is a minor concern.","section":"References","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses the term 'anti-tamper' throughout, which is a common spelling. However, the more standard term in the literature is 'anti-tamper' or 'anti-tampering'. Consistency is fine, but the authors may want to check journal style.","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Colleague,\n\nThis is a well-written special-session paper that consolidates recent FPGA anti-tamper sensor bypass results (much of it from the same WPI group) and contributes new measurements of accelerated aging and temperature-dependent spatial variation in soft-IP sensors. The new empirical content is real: the RO (-2.6%) and TDC (-44%) drift under 535 hours of accelerated aging on a single Spartan-7, and the temperature-dependent spatial variation across 79 TDC instances on an Artix-7. The XADC zero readback after 36 hours is a striking observation. The paper is honest about its limitations—it explicitly states that constant chamber temperature does not establish constant die temperature, that drift cannot be distinguished from self-heating without an independent thermal reference, and that the XADC failure mechanism is unidentified. The authors also defer the connection between measured drift and actual attack success to future work.\n\nThe soft spots are real but not fatal for the venue. First, the aging characterization rests on a single DUT with no reported device-to-device variation or error bars. The authors acknowledge this implicitly by reporting only point measurements, but the claim that the observed drift is representative is unsupported. Second, the central security conclusion—that sensor drift and environmental effects can open silent detection gaps and enable bypass—rests on the unvalidated premise that the measured baseline changes translate into exploitable attack success. The authors explicitly defer this link in Section VII. Third, several of the cited attack results (Chypnosis, Chypothermia, GlitchSnipe) are from the same group, so the 'bypass without invasive tampering' claim leans heavily on in-press or co-authored work. That is not a flaw per se, but it means the paper's strongest claim is more of a survey-plus-hypothesis than a demonstrated vulnerability.\n\nWho this is for: researchers working on FPGA security, especially those designing or evaluating anti-tamper sensors. The paper is a useful consolidation and the aging data is a genuine contribution, even if preliminary. It deserves a serious referee—the measurements are real, the writing is clear, and the limitations are stated. I would accept peer review and recommend publication with a request for error bars and a clearer separation of the new evidence from the survey material.\n\nBest,\n[You]","headline":"Honest consolidation with new aging data, but the security claim is broader than the evidence supports.","tokens_in":13468,"tokens_out":554,"duration_ms":329210},"pith_extraction":null,"created_at":"2026-11-13T10:20:36.679339+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":null,"supporting_citations":[],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}