{"id":"6501bc76-6d94-403a-adb3-d56f8dc318f2","arxiv_id":"2502.01348","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Low-frequency 1/f noise and random telegraph noise in graphene devices with interdigitated electrodes scale with the inverse of the active area, suggesting a bulk origin.","lead":"Researchers fabricated nine graphene ribbon devices with interdigitated electrodes and measured their low-frequency electrical noise. They report that the noise level scales with the inverse of the active graphene area, which could help designers choose optimal geometries for graphene-based sensors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Area-scaling claim rests on a post-hoc 7-point fit with one device per geometry; excluding G15W50 and G25W100 without independent evidence controls the 1/A correlation.","rationale":"Read in good faith, the paper's contribution is an empirical scaling relation between IDE/GMR geometry and low-frequency noise. For that relation to support the stated conclusion, the 1/A scaling must be a property of the device set, not of a subset chosen after inspection. The reader's weakest_assumption identifies exactly this. I find the concern load-bearing because the entire mechanistic claim (area conductance fluctuations, not contacts) is inferred from the slope of Fig. 6. Without error bars and with one sample per geometry, the exclusion of two points cannot be distinguished from legitimate device variability. The paper does include useful raw observations (ohmic I-V, 1/f spectra, RTN histograms), but the headline design rule is not yet quantitatively secure. The proposed fix is additional devices per geometry; until then a conditional verdict is appropriate. No concern about author integrity or novelty is raised.","tokens_in":5265,"tokens_out":5833,"duration_ms":59188,"concrete_test":"Fabricate and measure at least three additional G15W50 and G25W100 devices, plus a repeat of G8W50, G15W200, and G25W200, using the same transfer and noise protocol; then fit log(SR/R2) versus log(1/A) with all devices included and report per-geometry means and spreads. If the two formerly excluded geometries fall within the scatter of the fitted line and the slope remains ≈1, the area-scaling claim holds; if they remain systematically above the line, the outlier designation is supported; if the new replicates scatter broadly, the original 9-point conclusion was not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim—'the higher the active graphene area, the higher the SNR' and that the 1/f noise originates from the whole graphene area—stands on the linear trend in Fig. 6, where SR/R2 is plotted against 1/[G×W×(Nfingers−1)]. That trend is fit to only seven of the nine devices; G15W50 and G25W100 are labeled outliers with the qualitative suggestion of grain boundaries, wrinkles, or GMR edges, but no independent measurement is offered. With exactly one device per geometry and no error bars, this is a post-hoc selection: if those two points are included, the correlation is visibly weaker and the supposed 1/A law is not established. Since Fig. 6 is the only evidence that the noise is area-dominated rather than contact-dominated, the central conclusion inherits this fragility. A direct replication with multiple nominally identical devices per geometry is the minimal check that would decide whether the outliers are real geometry-specific effects or ordinary device-to-device scatter.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports low-frequency noise measurements (1/f and random telegraph noise) on nine single-layer graphene micro-ribbon devices with interdigitated electrodes of varying gap (8, 15, 25 μm) and width (50, 100, 200 μm). The authors observe 1/f behavior in all devices and RTN in several, extract the equivalent resistance noise SR from the slope of sqrt(SI) versus applied voltage, and plot the normalized resistance noise SR/R^2 against 1/A, where A = G × W × (Nfingers − 1). A linear trend is reported for seven of the nine devices, with two devices (G15W50, G25W100) labeled outliers. The paper concludes that the higher the active graphene area the higher the SNR, and that the low-frequency noise is conductance noise originating from the active graphene area rather than the contact regions; it also claims that RTN shows the same area dependence.","tokens_in":5406,"tokens_out":3089,"duration_ms":28132,"significance":"If the area-scaling result is correct, the paper would provide a useful design rule for graphene sensors with interdigitated electrodes and evidence about the physical origin of low-frequency noise in this geometry. The experimental effort is systematic in covering a 3×3 grid of device geometries, and the noise extraction method (linear regression of sqrt(SI) versus V) is standard. However, the central empirical claim is statistically fragile: the area-scaling conclusion rests on a fit to seven of nine points, with exactly one device per geometry, no error bars, and no quantitative outlier analysis. Because the paper's main conclusion depends on this fit, the result needs additional support before it can be considered established.","major_comments":[{"comment":"The central claim that SR/R^2 scales with 1/A rests on a linear fit to seven of nine devices, with G15W50 and G25W100 excluded as outliers. The manuscript gives no quantitative outlier criterion (e.g., residual threshold, studentized residual, robust regression) and no independent evidence, such as Raman mapping or AFM, that these two devices are structurally different. With one device per geometry and no measurement repeats, the two excluded points may simply be device-to-device scatter. The authors should either justify the exclusion statistically (for example, by showing that the remaining points remain significant under robust regression and that the excluded points fall outside a stated confidence band) or provide multiple nominally identical devices per geometry. Without this, the conclusion that the noise scales strictly with 1/A is not supported.","section":"Section III.B, Fig. 6"},{"comment":"The fit to the 1/A dependence is presented without fit statistics: no R^2, no confidence interval for the slope, and no number of degrees of freedom are reported. Additionally, the area definition A = G × W × (Nfingers − 1) is asserted without justification. It is not explained why the product of gap, width, and finger number is the correct 'active area' for conductance fluctuations, nor whether the graphene under the metal fingers contributes. To support the conclusion that the noise is area-dominated rather than contact-dominated, the authors should report the regression parameters and compare with an alternative model that includes a constant contact-noise contribution independent of A.","section":"Section III.B, Fig. 6, Eq. (implicit)"},{"comment":"The conclusion that RTN exhibits the same area dependence as 1/f noise is not supported by the presented data. Figure 8 shows only two devices, both with gap 8 μm and widths 50 and 100 μm; no RTN was observed for G8W200, and no RTN results are shown for gaps of 15 and 25 μm. Two points are insufficient to establish an area-scaling law for RTN amplitude. This claim should either be removed from the conclusions or explicitly qualified as preliminary.","section":"Section III.C, Fig. 8 and Section IV"},{"comment":"The sentence 'It can be concluded without doubt that the higher the active graphene area, the higher the SNR' overstates the evidence. The paper does not directly measure signal-to-noise ratio, but rather the current noise PSD normalized by resistance; a high-SNR claim requires a defined signal level. Furthermore, since the underlying correlation is weakened by the outlier issue, the phrase 'without doubt' is inappropriate. The conclusion should be reworded to reflect the statistical uncertainty.","section":"Section III.B"}],"minor_comments":[{"comment":"The sentence 'Finally, e-beam resist was removed with standard acetone/IPA cleaning steps using.' is incomplete; the trailing 'using' should be removed.","section":"Section II.A"},{"comment":"The heading 'RESULTS AND DISCUSSSION' contains a typo: 'DISCUSSSION' should be 'DISCUSSION'.","section":"Section III"},{"comment":"The phrase 'The same triplet of devices was studied in detail' is unclear because the previous sentence refers to the nine devices in Table I. Please specify which three devices are meant.","section":"Section II.B"},{"comment":"In the sentence 'The latter are called outliers...', the antecedent of 'the latter' is ambiguous; G15W50 and G25W100 are actually two devices, not 'the latter' of a list of two items. Please rewrite to name the devices explicitly.","section":"Section III.B"},{"comment":"The caption says 'Line indicates the linear fit to normal data.' Please define what 'normal data' means and mark the excluded points explicitly in the plot (e.g., open symbols) to make the outlier treatment transparent.","section":"Fig. 6 caption"}],"recommendation":"major_revision","confidential_remarks":"The manuscript presents a plausible empirical study, but the core area-scaling claim is contingent on post-hoc outlier exclusion without quantitative justification. The authors should be encouraged to strengthen the statistical analysis or temper the conclusions. The paper fits the journal's scope in applied physics and device characterization, but it does not yet meet the standard for acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper gives us a clean, compact dataset of low-frequency noise on graphene micro-ribbons with interdigitated electrodes: nine devices spanning gaps of 8-25 µm and widths of 50-200 µm, with 1/f spectra and some RTN traces. That is genuinely useful for sensor designers, because it is exactly the kind of geometry sweep people need when choosing IDE parameters. The measurement chain is sound: I-V confirms ohmic conduction, the PSD shows 1/f with gamma in 0.8-1.2, and extracting the equivalent resistance noise SR from the slope of sqrt(SI) versus bias is a standard method. The RTN observations add a small empirical detail: the normalized amplitude is larger for narrower ribbons, consistent with area-dependent trapping. Credit where due: the raw data are presented clearly, the paper does not pretend to discover a new noise mechanism, and the citation pattern is fine — Balandin and Nah are both there.\n\nNow the soft spots. The headline claim — that SR/R2 scales like 1/A, and therefore noise comes from the whole graphene area — rests entirely on the linear fit in Fig. 6. That fit includes seven of the nine devices. G15W50 and G25W100 are declared outliers and dropped, with only qualitative mention of grain boundaries, wrinkles, or ribbon edges. There is no quantitative outlier criterion, no independent measurement to support those attributions, and each geometry is represented by a single device. With one device per geometry and no error bars, the fit cannot distinguish a true 1/A law from device-to-device scatter. The sentence \"It can be concluded without doubt\" is too strong for the evidence; if the two excluded points are included, the correlation is visibly weaker. The RTN section is also exploratory: a few devices, no statistics on capture/emission times, but that is fine as a preliminary observation.\n\nWhere does that leave us? The paper's value is empirical, not conceptual. It confirms that 1/f noise in these IDE devices tracks active area in the way Balandin and Nah already reported, and it extends the data to a useful parameter range. The sensor-design guideline — larger active area, higher SNR — is plausible and probably correct, but the paper does not yet establish it firmly.\n\nRecommendation: send it to peer review. A good referee can push for multiple nominally identical devices per geometry, for error bars, and for a defensible outlier policy. With those additions, the design rule would be credible. As it stands, it is a solid empirical report with an overconfident headline.","headline":"Useful empirical noise data on graphene IDE devices, but the headline 1/A scaling rests on a seven-point fit with two post-hoc outliers, so the design rule needs replication before it hardens into a claim.","tokens_in":6007,"tokens_out":2868,"would_cite":false,"duration_ms":27285,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["72.70.+m","73.63.-b"],"model":"deepseek-v4-flash","headline":"Single-layer graphene devices with interdigitated electrodes show 1/f and random telegraph noise whose normalized amplitude scales inversely with the active graphene area, implying the noise comes from conductance fluctuations in the…","keywords":["Graphene","Low Frequency Noise","Interdigitated Electrodes","Random Telegraph Noise","Contact Resistance","1/f noise","conductance fluctuations"],"falsifier":"Re-measure the two excluded devices on fresh chips and redo the $S_R/R^2$ versus $1/A$ fit with all nine points; if the linear trend disappears or the slope stops matching $1/A$, the central claim fails. A second check is to hold $G$ and $W$ fixed while varying $N_{\\mathrm{fingers}}$ so that $A$ changes by a controlled integer factor and confirm that the noise follows $1/A$ point-for-point, rather than depending on individual device variability.","tokens_in":5041,"feed_emoji":"📉","tokens_out":8724,"duration_ms":69270,"temperature":0.7,"pith_summary":"This paper tries to establish that the low-frequency noise of single-layer graphene micro-ribbon devices with interdigitated electrodes is set by the size of the active graphene area, not by the metal contacts. Across nine devices with three gap spacings and three ribbon widths, the normalized resistance noise $S_R/R^2$ falls linearly with $1/A$, where $A = G \\times W \\times (N_{\\mathrm{fingers}}-1)$, and the same geometry dependence appears in random telegraph noise amplitudes. If the area-scaling claim is right, sensor designers can improve signal-to-noise ratio by enlarging the active graphene area, and noise models for graphene sensors should treat the whole channel as the fluctuation source. The paper also reads the $1/f$ spectra and two-level RTN signals as evidence of carrier trapping at the graphene/SiO$_2$ interface.","feed_headline":"Graphene sensor noise falls as active area grows","feed_subtitle":"Normalized 1/f and telegraph noise track 1/area across nine devices, pointing to the graphene channel as the noise source.","key_machinery":"The load-bearing object is the active-area product $A = G \\times W \\times (N_{\\mathrm{fingers}}-1)$, with $G$ the gap between interdigitated fingers, $W$ the graphene micro-ribbon width, and $N_{\\mathrm{fingers}}$ the number of electrodes; the device is modeled as $(N_{\\mathrm{fingers}}-1)$ bottom-gate graphene FETs in series, which licenses the use of standard MOSFET low-frequency-noise formalism. The argument runs by extracting the equivalent resistance noise $S_R$ from the slope of $S_I^{1/2}$ versus voltage at 20 Hz and then fitting normalized noise $S_R/R^2$ versus $1/A$. The linearity of that fit is what converts individual spectra into a geometric scaling law.","core_discovery":"The central claim is that normalized low-frequency resistance noise in these IDE graphene devices obeys a $1/A$ area law: plotting $S_R/R^2$ against $1/[G W (N_{\\mathrm{fingers}}-1)]$ for the nine devices yields a linear correlation, and the paper states that the higher the active graphene area, the higher the SNR. Because the trend holds across different finger gaps and ribbon widths, the paper concludes that the noise is resistive conductance fluctuations distributed over the whole graphene area rather than noise generated at the graphene–aluminium contacts. Random telegraph noise observed in some devices, with two current levels whose histogram fits two Gaussians, is attributed to capture and emission of carriers by traps at the SLG/SiO$_2$ interface, and its amplitude shows the same increase for smaller ribbon widths as the $1/f$ noise.","pith_inferences":["The $1/A$ law implies a resolution-versus-noise tradeoff for graphene sensors: shrinking the sensing area to improve spatial localization raises normalized noise, so applications must balance SNR against footprint; the slope of the reported fit gives the quantitative exchange rate.","If the outlier explanation is correct, wafer-scale defect maps such as Raman D-band intensity or grain-boundary density should predict which identical-geometry devices depart from the area law; this is a testable prediction the paper does not make.","The voltage-independence of $\\Delta I/I_H$ suggests each RTN event is dominated by a single effective trap; combining the measured $\\Delta I/I_H$ with capture and emission time statistics could convert the area scaling into an interface trap density estimate.","Because the devices are modeled as series-connected GFETs, the same data imply that the gate-referred voltage noise spectral density should also scale with $1/A$; re-analyzing already-published IDE graphene noise data with this normalization would provide an independent check."],"forward_implications":["Sensor design can use active area as a tunable knob: making $A$ larger lowers normalized $1/f$ noise and raises SNR, all else equal.","Noise characterization of such devices should normalize by $1/A$ so results from different IDE layouts are directly comparable.","RTN amplitude follows the same geometry trend, so interface traps at the SLG/SiO$_2$ boundary are part of the same area-dependent noise budget.","The series-GFET model implies that adding fingers (increasing $N_{\\mathrm{fingers}}$) should reduce normalized noise in proportion to the added area, a prediction testable without changing material growth."],"supporting_citations":[{"why":"Prior study of low-frequency noise in graphene sensors with different IDE geometries; the comparison baseline this work extends.","marker":"[4]"},{"why":"Establishes the $1/f$ noise behavior in graphene devices that the measured spectra are checked against.","marker":"[5]"},{"why":"Supplies the interpretation of $1/f$ noise as a probe of defects, scattering, and screening in two-dimensional materials.","marker":"[8]"},{"why":"Provides the grain-boundary, wrinkle, and ribbon-edge mechanisms invoked to explain the two outlier devices.","marker":"[9]"},{"why":"Attributes graphene/SiO$_2$ interface interaction to device properties, grounding the RTN trap assignment at that interface.","marker":"[10]"}],"fun_headline_variants":["Graphene noise scales with area, not electrode contacts","Larger graphene area reduces 1/f and telegraph noise","Noise in graphene sensors stems from channel, not contacts","Area law for graphene noise: bigger area, better SNR","Graphene 1/f noise: area-dependent, contact-independent"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The area-scaling law depends on treating two of the nine devices (G15W50 and G25W100) as outliers and removing them from the linear fit; if they are included, the claimed $1/A$ correlation is significantly weakened, and the paper offers no independent evidence that grain boundaries, wrinkles, or ribbon edges caused their deviation.","fun_headline_variants_meta":{"raw":{"variants":["Graphene noise scales with area, not electrode contacts","Larger graphene area reduces 1/f and telegraph noise","Noise in graphene sensors stems from channel, not contacts","Area law for graphene noise: bigger area, better SNR","Graphene 1/f noise: area-dependent, contact-independent"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000201,"raw_usage":{"total_tokens":1314,"prompt_tokens":819,"completion_tokens":495,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":435,"completion_tokens_details":{"reasoning_tokens":412}},"tokens_in":435,"tokens_out":495,"duration_ms":4788,"temperature":1.0,"reasoning_tokens":412,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T15:34:13.869073+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-measure the two excluded devices on fresh chips and redo the $S_R/R^2$ versus $1/A$ fit with all nine points; if the linear trend disappears or the slope stops matching $1/A$, the central claim fails. A second check is to hold $G$ and $W$ fixed while varying $N_{\\mathrm{fingers}}$ so that $A$ changes by a controlled integer factor and confirm that the noise follows $1/A$ point-for-point, rather than depending on individual device variability.","supporting_citations":[{"cited_title":"Electrical and Low Frequency Noise Characterization of Graphene Chemical Sensor Devices Having Different Geometries,","cited_arxiv_id":null,"evidence_quote":"Prior study of low-frequency noise in graphene sensors with different IDE geometries; the comparison baseline this work extends."},{"cited_title":"1 / f noise in van der Waals materials and hybrids,","cited_arxiv_id":null,"evidence_quote":"Supplies the interpretation of $1/f$ noise as a probe of defects, scattering, and screening in two-dimensional materials."},{"cited_title":"Impacts of graphene/SiO2 interaction on FET mobility and Raman spectra in mechanically exfoliated graphene films,","cited_arxiv_id":null,"evidence_quote":"Attributes graphene/SiO$_2$ interface interaction to device properties, grounding the RTN trap assignment at that interface."}],"review_version":1}