{"id":"b7efa4ad-a580-489c-a3fd-7cab4cf2345c","arxiv_id":"2608.06616","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":9,"one_line_summary":"The authors propose a graphene-screened MR-DWELL dosimeter in which the capacitance ratio eta suppresses radiation-induced resonance shifts by about threefold, with all headline numbers coming from an unvalidated analytical model.","lead":"This paper models a proposed radiation dosimeter that uses a graphene layer to shield its resonant tunneling structure from radiation-induced trapped charge. The model predicts high-dose tolerance and very fast response, but no device has been built or tested.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 1 MGy transport metrics (83% retention, PVR~10) come from unshown 'reduced transport functions' rather than the Landauer integral; the code even uses a different saturation law, so the central performance claim is not yet established.","rationale":"The reader's weakest assumption (fixed C_Q) is real and acknowledged by the authors, but it is a refinement concern: a self-consistent C_Q(µ) would shift η by tens of percent, not obviously invalidate the design. The more load-bearing gap is that the concrete headline numbers—83% peak-current retention, PVR~10, and the spectroscopic peak shifts—are produced by transport functions that are not shown in the paper and whose calibration is not specified. Because these numbers are the basis for 'preserving resonant transport up to 1 MGy,' the central claim is only as strong as that hidden calibration. Moreover, the manuscript contains a verifiable inconsistency between the claimed exponential 'physical chain' of Section 4.7 and the rational-law implementation in the provided Supplementary Software, making the provenance of the headline metrics unclear. This is a correctness risk independent of the electrostatic screening assumption. A direct Landauer integral is a simple, decisive check. I therefore keep the reader's CONDITIONAL verdict: the framework is promising and candid about its limitations, but the headline transport predictions should be verified from Eq. (7) before the 1 MGy claims are treated as established.","tokens_in":22139,"tokens_out":18874,"duration_ms":181837,"concrete_test":"Using the parameters of Table S1 and the exponential saturation of Eqs. (10), (19), and (S11)-(S12) with D0=500 Gy, ΔE1,max(η=2.09)=23 meV, Γ1(1 MGy)=4.4 meV, and the two-resonance Breit-Wigner transmission of Eq. (8), evaluate the Landauer integral of Eq. (7) over a bias range encompassing the first resonance and its valley at D=0 and D=1 MGy. Compute Ipeak(0), Ipeak(1 MGy), PVR(1 MGy)=Ipeak/Ivalley, and retention=Ipeak(1 MGy)/Ipeak(0), then compare with Table 3 and Fig. 5 (17% current drop, PVR~10). If the direct Landauer values differ by more than 10%, the reduced-function calibration is the source of the headline claims and must be corrected or the claims revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.1 states Eq. (7) is the physical transport basis, but Section 4.7 says the reported peak/valley currents and PVR are evaluated using 'calibrated reduced transport functions' PVR(ΔE) and Inorm(ΔE), calibrated against reference Landauer points; the manuscript does not display these functions or the calibration points. Table 3 and Fig. 5 headline values (17% current drop, PVR~10, ΔVpeak) therefore rest on an interpolation whose error at the large screened shift ΔE≈23 meV is unknown. The traceability gap is compounded by an internal inconsistency: the main text (Section 4.7, Eqs. (18)-(20)) declares the exponential saturation model to be the 'genuine physical chain,' with the rational D/(D+D0) used only for calibration, yet the provided Supplementary Software's electrostatics.py implements the rational saturation for ΔE(D,η) when evaluating PVR and Inorm. Thus the reported 1 MGy metrics are not demonstrably predictions of the claimed physical model. If the reduced functions deviate from direct Breit-Wigner/Landauer evaluation at ΔE≈23 meV, the quantitative claims of radiation tolerance (PVR~10, 83% retention) collapse, even if the electrostatic screening factor of Eq. (13) is correct.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a graphene/h-BN interface on a multi-resonant double-barrier quantum-dot-in-a-well (MR-DWELL) resonant-tunneling structure, using the ratio η = C_Q/C_geo of graphene quantum capacitance to geometric capacitance as a screening parameter. A capacitive-divider model (Eqs. (1)–(6)) yields the screening factor S = η/(1+η), which reduces the effective radiation-induced resonance shift to α_eff = α/(1+η) (Eq. (13)). The paper claims that for η = 2.09 this suppresses the low-dose shift coefficient threefold, limits the resonance shift at 1 MGy to about 23 meV, retains about 83% of the peak current with a peak-to-valley ratio of about 10, and gives an intrinsic RC response of 0.28 ps. It further introduces d²I/dV² spectroscopy as a readout, a three-population trapped-charge model for FLASH dose-rate discrimination, and design regimes in η. All results are analytical/numerical; no experimental data are presented.","tokens_in":22525,"tokens_out":3540,"duration_ms":34947,"significance":"The electrostatic screening concept is physically plausible and the capacitive-divider algebra is internally consistent. If the quantitative transport metrics were established, the framework would offer a simple design rule (η) for radiation-hard resonant-tunneling dosimeters and a plausible route to FLASH dose-rate discrimination. The paper has strengths worth acknowledging: it explicitly labels its transport functions as 'calibrated reduced models' rather than hiding them, provides Python code for reproducibility, clearly separates steady-state from transient trap populations, and marks the synthetic PCA/LDA classification as purely illustrative. However, the headline quantitative claims — 83% peak-current retention, PVR ≈ 10, and the associated 1 MGy metrics — currently rest on undocumented interpolation functions and on an internal inconsistency in the saturation law used to produce them. The central screening relation (Eq. (13)) is sound, but the transport-level predictions are not yet traceable to the stated Landauer basis.","major_comments":[{"comment":"The reported peak-current retention and PVR values are not obtainable from the manuscript's stated transport model. Section 3.1 presents Eq. (7) as the physical basis, but then states that 'the reported peak and valley currents are evaluated using calibrated reduced transport functions' without displaying PVR(ΔE) or Inorm(ΔE) or their calibration points. Section 4.7 confirms that these functions are 'not direct Landauer integrals.' Table 3 and Fig. 5 therefore rest on an interpolation whose error at the operating point ΔE ≈ 23 meV is unknown. The authors should either display the reduced functions and calibration points, or evaluate the headline metrics directly from Eqs. (7)–(8), so that the 1 MGy claims become reproducible predictions.","section":"§3.1 and §4.7"},{"comment":"There is a direct inconsistency between the claimed physical saturation model and the software used to generate the reported metrics. The main text (Eqs. (18)–(20)) declares the exponential law ΔE(D,η) = ΔE_max(η)[1−exp(−D/D0)] to be the 'genuine physical chain,' with the rational form D/(D+D0) used only for internal calibration. Yet the provided electrostatics.py implements the rational saturation law ΔE(D,η) = ΔE_max(η)·D/(D+D0) for the evaluation of PVR and Inorm. Since the 83% retention and PVR≈10 values are outputs of those routines, the headline transport claims are not demonstrably predictions of the exponential model presented as physical. The authors must either align the software with the declared model or show that the two saturation laws yield the same transport metrics.","section":"§4.7 vs. Supplementary Software (electrostatics.py)"},{"comment":"The claimed 'threefold suppression' and '23 meV maximum shift' are constructed from assumed inputs rather than derived predictions. Eq. (13) gives α_eff = α/(1+η) using α = 0.14 meV/mGy from the earlier work, and Eq. (20) gives ΔE_max(η) = ΔE_max(0)/(1+η) using ΔE_max(0) = 70 meV. Substituting η = 2.09 returns 23 meV by construction. This is not a problem per se, but the abstract and Section 1 present these as outcomes of the design framework. The manuscript should explicitly distinguish calibrated assumptions (α, ΔE_max(0), D0) from framework predictions, and should quantify how the headline numbers change under the acknowledged uncertainty in these inputs (e.g., α = 0.14 ± 0.01 meV/mGy).","section":"Eqs. (13), (20), and §3.2"},{"comment":"The fixed-C_Q assumption is load-bearing for the 1 MGy claims and is only acknowledged as a limitation. Equations (12)–(13) take C_Q constant at E_F ≈ 0.3 eV, but Section 3.2 states that the full gate-voltage dependence is reserved for future work, and Section 11 admits that trapped charge would modulate the graphene carrier density and couple C_Q(μ) to trap kinetics. If irradiation shifts the graphene Fermi level, the screening factor S in Eq. (6) changes, and the factor-three suppression no longer follows from Eq. (13). Since the central quantitative conclusions depend on this constancy, the authors should either provide a self-consistent estimate of the C_Q variation under the relevant trapped-charge densities or explicitly restrict the 1 MGy predictions to the fixed-C_Q regime and state the condition under which they fail.","section":"§3.2 and §11"}],"minor_comments":[{"comment":"The abstract states 'preserving resonant transport up to 1 MGy with about 83 percent peak-current retention'; Section 4.4 reports a 'peak current reduction at 1 MGy: ∼17%' for the screened case. These two statements are consistent (83% retention = 17% drop), but the abstract does not mention that this value depends on the undocumented reduced transport functions; adding a qualifier such as 'within the calibrated analytical model' would improve accuracy.","section":"Abstract and §4.4"},{"comment":"The Supplementary Material text in Fig. S1 refers to 'Eq. (6) of the main manuscript' for the transmission calculation, but the transmission function is defined by Eq. (8) of the main text, while Eq. (6) is the screening factor. This is a typographical cross-reference error that should be corrected.","section":"§3.1 and Supplementary Material S6"},{"comment":"The sub-picosecond RC time is computed with R_ch ≈ 1.2 kΩ, which is described as an order-of-magnitude estimate; the paper notes this is a low-temperature reference value. This is acceptable for a design framework, but Table 1 should state that τ_int is directly proportional to R_ch and is therefore an estimate rather than a device-specific prediction.","section":"§3.4 and Table 1"},{"comment":"The robustness bounds for PVR and J_p are obtained using phenomenological scaling relations (S9)–(S10) with coefficients (0.8, 1.0) chosen as illustrative. The text does report the extremal bounds, but the reader cannot tell how much of the PVR variation (e.g., −37%/+58% for D_it) comes from the transport model versus these illustrative coefficients. A sentence stating that the PVR/J_p bounds are illustrative scaling outcomes rather than Landauer results would clarify the status of the robustness claims.","section":"§4.8 and Eqs. (S9)–(S10)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is better described as an applied device-engineering design study than as a mathematical-physics contribution; the core mathematics is a simple capacitance-divider relation. That said, the analytical framework is clearly presented and the code availability is a genuine asset. The main risk is not the screening algebra but the traceability of the headline transport metrics: the reduced transport functions are not shown, and the software's saturation law disagrees with the one declared in the main text. These issues are fixable within the scope of a revision, so I do not recommend rejection. I would also flag that the paper relies heavily on self-citations to companion preprints (Refs. [5], [17], [18]) for the central input α and for the NEGF–Poisson methodology; the editor may wish to confirm that those sources are publicly accessible and that the present claims do not depend on unpublished details."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core idea here is sound and honestly framed: put graphene's quantum capacitance in series with the h-BN spacer, define eta = CQ/Cgeo, and you get a capacitive voltage divider that suppresses radiation-induced resonance shifts by 1/(1+eta). That is standard electrostatics, but applying it to an MR-DWELL dosimeter and organizing the design space around eta is a legitimate contribution, and the paper does it cleanly. It also ships reproducible Python code, flags its own assumptions, and explicitly labels the FLASH dose-rate business as speculative. Credit where due: the electrostatic algebra is internally consistent, and the design-regime map is useful for people thinking about radiation-hard resonant-tunneling sensors.  The soft spot is load-bearing. The paper's abstract and summary advertise 83% peak-current retention and PVR about 10 at 1 MGy. Those numbers are not computed from the Landauer integral that Section 3.1 presents as the physical basis. They come from 'calibrated reduced transport functions' PVR(DeltaE) and Inorm(DeltaE), and those functions plus their calibration points are never shown. Worse, the Supplementary Software's electrostatics.py evaluates the shift with the rational saturation law D/(D+D0), while Section 4.7 says the exponential law is the 'genuine physical chain' and the rational form was only for calibration. So as shipped, the code uses the non-preferred model to generate the headline metrics. The stress-test is right: the 1 MGy transport claims are not demonstrably predictions of the claimed physical model. If the reduced functions deviate from direct Breit-Wigner/Landauer evaluation at the large screened shift DeltaE about 23 meV, the quantitative claims collapse, even though the screening factor itself is correct.  The fixed-CQ assumption is also acknowledged by the authors, so I treat that as a stated limitation rather than a hidden flaw - but it does mean the factor-three suppression is conditional on CQ staying at its EF about 0.3 eV value.  Who gets value from this? Groups designing or evaluating radiation-hard resonant-tunneling detectors, and people who want a compact design framework for graphene/h-BN electrostatic screening. It is not an experimental validation, and the authors do not claim it is.  Recommendation: send it to peer review. The conceptual framework is worth refereeing with code and an honest limitations section. But the referee must require the reduced transport functions to be displayed, the calibration points specified, and the exponential-versus-rational inconsistency between text and code resolved before any quantitative radiation-tolerance claim is accepted. As it stands, the paper should be treated as a design framework with promising electrostatics and unproven transport metrics.","headline":"The capacitive-divider screening idea is real and clearly presented, but the headline MGy transport numbers rest on reduced functions that are never shown and on code that contradicts the main text's saturation law.","tokens_in":753,"tokens_out":2215,"would_cite":false,"duration_ms":36800,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82D37"],"pacs":[],"model":"deepseek-v4-flash","headline":"Tuning the ratio of graphene's quantum capacitance to the h-BN spacer's capacitance cuts radiation-induced resonance shifts threefold and preserves resonant transport to 1 MGy in a quantum-dot tunneling dosimeter.","keywords":["MR-DWELL","resonant tunneling","electrostatic interface engineering","quantum capacitance","radiation dosimetry","radiation hardness","second-derivative spectroscopy","FLASH dose-rate regime"],"falsifier":"The paper's own validation roadmap doubles as the falsifier: build two otherwise identical MR-DWELL stacks, one with a metal top contact and one with a graphene/h-BN contact, and measure the $\\mathrm{d}^2I/\\mathrm{d}V^2$ peak shift versus accumulated dose for h-BN thicknesses of 0.5, 1, 2, and 3 nm. The model predicts the low-dose shift coefficient falls as $\\alpha_{\\mathrm{eff}} = 0.14/(1+\\eta)$ meV/mGy and the 1 MGy shift as $70/(1+\\eta)$ meV; a clear deviation from these curves, or a dose-dependent motion of the graphene Fermi level comparable to 0.3 eV, would falsify the fixed-$C_Q$ screening picture.","tokens_in":21919,"feed_emoji":"☢️","tokens_out":19748,"duration_ms":145951,"temperature":0.7,"pith_summary":"This paper proposes a design framework for a radiation-tolerant dosimeter built on a resonant-tunneling heterostructure (an MR-DWELL: a multi-resonant double-barrier quantum-dot-in-a-well stack) topped with a graphene contact separated by an h-BN spacer. The central idea is that graphene's finite density of states gives it a quantum capacitance, and the ratio of that quantum capacitance to the geometric capacitance of the spacer, $\\eta = C_Q/C_{\\mathrm{geo}}$, sets a screening factor $S = \\eta/(1+\\eta)$ that controls how much radiation-induced trapped charge perturbs the resonant levels. The model predicts that choosing $\\eta \\approx 2$ (about a 1 nm h-BN spacer) cuts the effective resonance-shift coefficient threefold, keeps roughly 83% of the peak current with a peak-to-valley ratio near 10 after 1 MGy, and holds the intrinsic RC response near 0.28 ps. The paper further argues that second-derivative spectroscopy, $\\mathrm{d}^2I/\\mathrm{d}V^2$, offers a practical lock-in readout of the resonance shifts, and that a three-population trapped-charge model opens a path to FLASH dose-rate discrimination even though steady-state dosimetry saturates near 500 Gy. A sympathetic reader would care because the framework turns a known failure mode of resonant-tunneling dosimeters — uncontrolled interface charge growth — into a tunable design variable.","feed_headline":"Graphene screening triples a tunneling dosimeter's radiation tolerance","feed_subtitle":"A graphene/h-BN gate screens trapped charge, keeping resonant transport alive past one megagray.","key_machinery":"The load-bearing object is the capacitance ratio $\\eta = C_Q/C_{\\mathrm{geo}}$ together with the screening factor $S = \\eta/(1+\\eta)$ it generates. The geometric capacitance is $C_{\\mathrm{geo}} = \\varepsilon_0\\varepsilon_r a_{\\mathrm{emit}}/t_d$, with $t_d$ the h-BN spacer thickness, and the graphene quantum capacitance is $C_Q = (2e^2k_BT/\\pi(\\hbar v_F)^2)\\ln\\left[2\\cosh(\\mu/2k_BT)\\right]$, fixed in this work at the $E_F \\approx 0.3$ eV operating point where the areal value is about 7.2 $\\mu$F/cm$^2$. The identity $\\alpha_{\\mathrm{eff}} = \\alpha_i(1-S) = \\alpha_i/(1+\\eta)$ turns the capacitance ratio into a direct prediction for the dose response, which is then fed through first-order trap-filling kinetics $Q_{\\mathrm{trap}}(D) = Q_{\\max}[1-\\exp(-D/D_0)]$ with $D_0 = 500$ Gy, and through a Breit–Wigner transmission and Landauer current model with two resonances ($E_{10} = 82$ meV, $\\Gamma_{10} = 4$ meV; $E_{20} = 126$ meV, $\\Gamma_{20} = 7$ meV). A separate three-population trap kinetics (deep, slow, and fast traps) with critical dose rate $\\dot{D}_{\\mathrm{crit}} = D_{\\mathrm{fast}}/\\tau_{\\mathrm{fast}}$ carries the FLASH dose-rate discrimination hypothesis.","core_discovery":"On the paper's own terms, the discovery is that radiation hardness in a resonant-tunneling dosimeter can be engineered electrostatically rather than metallurgically. Graphene is not a perfect shield: its finite density of states yields a quantum capacitance $C_Q$, so when radiation traps charge at the interface, the induced potential partitions across the series combination of $C_Q$ and the geometric capacitance $C_{\\mathrm{geo}}$ of the h-BN spacer. Only the fraction $1-S = 1/(1+\\eta)$ of the bare electrostatic perturbation reaches the resonant levels, so the effective low-dose shift coefficient obeys $\\alpha_{\\mathrm{eff}} = \\alpha_i/(1+\\eta)$ and the asymptotic high-dose shift is suppressed by the same factor. With $\\eta = 2.09$ the unscreened 70 meV asymptotic shift falls to about 23 meV, the peak-to-valley ratio stays near 10, and the device retains about 83% of its peak current at 1 MGy, where the unscreened reference is already strongly degraded. The model further claims a sub-picosecond intrinsic RC time constant near 0.28 ps, a packaging-limited system response near 20 ps, and a design map in which h-BN thickness selects between high-sensitivity, balanced, and maximum-hardness operating regimes.","pith_inferences":["If the screening picture survives experimental test, the same $\\eta = C_Q/C_{\\mathrm{geo}}$ parameter should transfer to any resonant-tunneling sensor whose degradation channel is electrostatic drift of level positions — the framework is about stabilizing levels, not specifically about radiation.","The paper fixes $C_Q$ at one Fermi level; a gate-tuned graphene contact would in principle make the screening factor adjustable during operation, letting a single device trade sensitivity for hardness mid-measurement — an extension the authors explicitly reserve for future work.","The fast-trap kinetics imply a concrete untested prediction: a pulse at $10^4$ Gy/s should produce a transient $\\mathrm{d}^2I/\\mathrm{d}V^2$ peak shift that is absent at 0.1 Gy/s even at equal accumulated dose, because the fast-trap population saturates only in the high-rate regime.","The radiation-type descriptor $R = \\beta/\\alpha$ is demonstrated only on synthetic data; the natural next step is to calibrate $\\alpha$ and $\\beta$ for two well-characterized radiation fields and check whether their ratio is genuinely species-dependent in a real device."],"forward_implications":["At $\\eta = 2.09$ the model predicts a resonance shift near 23 meV, roughly 83% peak-current retention, and a peak-to-valley ratio near 10 after 1 MGy, moving resonant-tunneling dosimetry from the hundreds-of-gray range into the megagray range.","The design rules assign each h-BN thickness to an application class: $\\eta < 1.5$ for high sensitivity, $1.5 \\lesssim \\eta \\lesssim 4$ for balanced FLASH-compatible operation, and $\\eta > 4$ for maximum hardness at the cost of sensitivity.","Because the intrinsic RC time is near 0.28 ps, the speed ceiling shifts to interconnects and front-end electronics, so the estimated 20 ps system response is set by packaging rather than by the tunneling structure itself.","The $\\mathrm{d}^2I/\\mathrm{d}V^2$ readout, which is measurable with standard lock-in techniques, is predicted to track the radiation-induced resonance shifts while remaining well resolved up to 1 kGy in the model.","With $D_{\\mathrm{fast}} \\approx 50$ Gy and $\\tau_{\\mathrm{fast}} \\approx 10$ ms, the fast-trap kinetics give a critical dose rate near $5\\times10^3$ Gy/s, spanning $5\\times10^2$–$5\\times10^4$ Gy/s as the time constant varies from 1 to 100 ms — a window overlapping the FLASH regime — so transient dose-rate discrimination is claimed to be possible even though steady-state dosimetry saturates at 500 "],"supporting_citations":[{"why":"Source of the unscreened MR-DWELL baseline: it supplies the low-dose shift coefficient $\\alpha_1 = 0.14$ meV/mGy and the current sensitivity 0.3 $\\mu$A/mGy that the graphene architecture must improve on.","marker":"[5]"},{"why":"Supplies the measured graphene quantum-capacitance value, about 7.2 $\\mu$F/cm$^2$ at $E_F \\approx 0.3$ eV, which anchors the screening factor numerically.","marker":"[25]"},{"why":"Demonstrates second-harmonic $\\mathrm{d}^2I/\\mathrm{d}V^2$ spectroscopy of localized states in graphene/h-BN stacks, motivating the proposed readout scheme.","marker":"[16]"},{"why":"Source of the Landauer transport formalism used to convert the screened transmission $T(E)$ into the tunnelling current.","marker":"[22]"},{"why":"Provides the DWELL resonance energy $E_{10} = 82$ meV used as a fixed input parameter of the two-resonance model.","marker":"[23]"},{"why":"Supplies DWELL resonant-tunnelling measurements consistent with the linewidths $\\Gamma_{10} = 4$ meV and $\\Gamma_{20} = 7$ meV taken as model inputs.","marker":"[24]"},{"why":"Establishes the h-BN dielectric platform whose thickness $t_d = 0.5$–3 nm and permittivity $\\varepsilon_r = 3.9$ set the geometric capacitance and hence $\\eta$.","marker":"[19]"},{"why":"Describes the NEGF–Poisson computational architecture whose convergence criteria underlie the extraction of the unscreened shift coefficient.","marker":"[17]"}],"fun_headline_variants":["Graphene screen cuts dose shift 3x in tunneling dosimeter","Electrostatic design keeps dosimeter working past 1 megagray","Tunneling dosimeter gains 3x radiation hardness via graphene gate","h-BN spacer tunes graphene shield for radiation-tolerant tunneling sensor"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that radiation-induced trapped charge behaves as a static, uniformly coupled sheet whose electrostatic effect splits cleanly between the graphene quantum capacitance, held fixed at its $E_F \\approx 0.3$ eV value, and the geometric capacitance; if irradiation moves the graphene Fermi level or the charge couples unevenly to the resonant states, the identity $\\alpha_{\\mathrm{eff}} = \\alpha_i/(1+\\eta)$ fails and the threefold suppression and all 1 MGy metrics collapse.","fun_headline_variants_meta":{"raw":{"variants":["Graphene screen cuts dose shift 3x in tunneling dosimeter","Electrostatic design keeps dosimeter working past 1 megagray","Tunneling dosimeter gains 3x radiation hardness via graphene gate","h-BN spacer tunes graphene shield for radiation-tolerant tunneling sensor"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000518,"raw_usage":{"total_tokens":2510,"prompt_tokens":947,"completion_tokens":1563,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":563,"completion_tokens_details":{"reasoning_tokens":1486}},"tokens_in":563,"tokens_out":1563,"duration_ms":10723,"temperature":1.0,"reasoning_tokens":1486,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T04:10:35.016314+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The paper's own validation roadmap doubles as the falsifier: build two otherwise identical MR-DWELL stacks, one with a metal top contact and one with a graphene/h-BN contact, and measure the $\\mathrm{d}^2I/\\mathrm{d}V^2$ peak shift versus accumulated dose for h-BN thicknesses of 0.5, 1, 2, and 3 nm. The model predicts the low-dose shift coefficient falls as $\\alpha_{\\mathrm{eff}} = 0.14/(1+\\eta)$ meV/mGy and the 1 MGy shift as $70/(1+\\eta)$ meV; a clear deviation from these curves, or a dose-dependent motion of the graphene Fermi level comparable to 0.3 eV, would falsify the fixed-$C_Q$ screening picture.","supporting_citations":[{"cited_title":"Dosymova, M.V","cited_arxiv_id":null,"evidence_quote":"Source of the unscreened MR-DWELL baseline: it supplies the low-dose shift coefficient $\\alpha_1 = 0.14$ meV/mGy and the current sensitivity 0.3 $\\mu$A/mGy that the graphene architecture must improve on."},{"cited_title":"Narihiro, G","cited_arxiv_id":null,"evidence_quote":"Supplies DWELL resonant-tunnelling measurements consistent with the linewidths $\\Gamma_{10} = 4$ meV and $\\Gamma_{20} = 7$ meV taken as model inputs."},{"cited_title":"Dosymova, M.V","cited_arxiv_id":null,"evidence_quote":"Describes the NEGF–Poisson computational architecture whose convergence criteria underlie the extraction of the unscreened shift coefficient."}],"review_version":1}