{"id":"4af6dc3b-9438-4957-8a09-f0053e48e87d","arxiv_id":"1909.00712","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"HfO2 ALD nucleation on CVD graphene is mediated by adventitious carbon contamination and depends on graphene layer number via copper substrate assistance.","lead":"This paper images the earliest stage of atomic layer deposition of hafnium oxide on graphene and finds that nucleation happens on atmospheric carbon contamination rather than on clean graphene. It also shows that the copper catalyst below the graphene helps oxide growth only through one-atom-thick graphene, not through thicker graphene.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Carbon-seeding claim hinges on the unproven continuity of pre-ALD carbon under HfO2; the cited constant bare-graphene fraction does not actually rule out post-ALD carbon accumulation.","rationale":"The reader's conditional verdict is appropriate. The central novelty of the paper is the adventitious-carbon-mediated nucleation mechanism, and the other findings (graphene-layer-dependent Cu participation, step-edge nucleation, e-beam-induced crystallization to cubic/tetragonal HfO2) are well supported by the microscopy data. The weakest point is indeed the pre-ALD presence and sub-HfO2 location of the carbon, as the reader identified. I agree with that identification and add that the paper's quantitative argument from the roughly constant θ_graphene does not discriminate between the two scenarios: because carbon on top of HfO2 clusters would not be counted in the contamination fraction, the conservation relation θ_contamination + θ_oxide = 1 - θ_graphene is consistent with HfO2 nucleating on initially clean graphene followed by carbon decoration. The proposed controlled-exposure experiment would settle causality directly. If the test fails, the headline claim should be weakened from a seeding mechanism to a spatial correlation between HfO2 and carbon. Until such a test is run, the conditional verdict remains the right call.","tokens_in":18662,"tokens_out":6320,"duration_ms":71603,"concrete_test":"Use otherwise identical CVD graphene/Cu samples and vary only the pre-ALD carbon exposure: (a) about 1 day of ambient exposure as in the paper, (b) about 7 days of ambient exposure, and (c) in-situ vacuum or H2 annealing at roughly 300 °C immediately before ALD to desorb adventitious carbon. Then run the same 16-cycle substrate-assisted ALD and quantify HfO2 areal coverage and cluster density by SEM or STEM. If pre-existing carbon seeds nucleation, coverage should increase from (c) to (a) to (b); if coverage is statistically unchanged, the 'Carbon attracts HfO2' scenario is not supported and the observed correlation is consistent with post-deposition carbon accumulation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim (scenario ii, 'Carbon attracts HfO2') requires adventitious carbon to be present on graphene before ALD and to lie underneath the HfO2 deposits. The only direct evidence is post-ALD HAADF-STEM contrast: HfO2 clusters appear surrounded by, and possibly placed on, carbon-like regions, and the authors state that 'a plausible assumption' of a continuous carbon layer underneath HfO2 is based on 'visual spatial continuity' (Results, after Fig. 4b). This assumption is load-bearing. The quantitative argument against scenario (i) ('HfO2 attracts carbon') is not decisive: θ_graphene measures only atomically clean graphene areas. Carbon that accumulates on top of HfO2 clusters or at their edges during post-ALD ambient storage, wet transfer, or STEM imaging would not register as bare graphene, so θ_graphene can remain roughly constant while θ_oxide changes; the relation θ_contamination + θ_oxide = 1 - θ_graphene is compatible with HfO2 nucleating on clean graphene and carbon later decorating the remaining surface. The 6-cycle vs 16-cycle comparison uses different sample regions and has no error bars. The non-ALD control shows a similar bare-graphene fraction, but it is not the same sample region and cannot establish that carbon was present at nucleation sites pre-ALD. The authors' own caveat that e-beam exposure can induce contamination dewetting and attraction further weakens confidence that the observed spatial correlation is entirely pre-existing. Thus the causal direction of the carbon-HfO2 correlation remains unproven.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an aberration-corrected STEM study of the nucleation stage of HfO2 grown by substrate-assisted ALD directly on CVD graphene that remains on its Cu growth foil. The authors identify several factors governing nucleation: a graphene-layer-dependent catalytic participation of the Cu support, preferential nucleation at graphene layer steps, a Volmer–Weber growth mode with amorphous as-deposited clusters, e-beam-induced crystallization to metastable cubic/tetragonal HfO2, and no preferential nucleation at graphene grain boundaries. The central mechanistic claim is that atmospheric adventitious carbon contamination acts as an unintentional seeding layer: HfO2 nucleates preferentially on pre-existing carbon deposits rather than on clean graphene basal planes. This claim is supported by spatial correlation between HfO2 clusters and carbon-like contrast, by approximately constant bare-graphene areal fractions across regions with very different HfO2 coverages, and by a comparison of 16-cycle and 6-cycle ALD samples together with a non-ALD control.","tokens_in":18959,"tokens_out":4730,"duration_ms":49944,"significance":"If the carbon-seeding conclusion holds, it is practically important for ALD process design on 2D materials, since adventitious carbon is ubiquitous in non-UHV processing. The paper's strengths include the use of polymer-free sample preparation, atomic-resolution and element-specific STEM imaging, direct quantification of areal coverages, the observation of non-equilibrium HfO2 polymorphs, and several control observations (non-ALD sample, grain-boundary comparison, layer-step comparison). The manuscript contains no fitted parameters or circular derivations, and the imaging evidence for layer-dependent Cu assistance and step-edge nucleation is visually compelling. However, the central causal inference about pre-existing carbon is presently supported by a stated unproven assumption and by single-image statistics, so the significance of the paper depends on strengthening that link.","major_comments":[{"comment":"The conclusion that scenario (ii) 'Carbon attracts HfO2' is correct rests entirely on the assumption that 'the adventitious carbon adsorbates are a continuous layer underneath the HfO2 deposits', which the authors introduce as a plausible assumption based on visual spatial continuity. The measured quantity θ_graphene alone cannot distinguish this scenario from scenario (i) 'HfO2 attracts carbon', because carbon that accumulates on top of or at the edges of HfO2 clusters after ALD is not counted as bare graphene and therefore does not affect θ_graphene. The relation θ_contamination + θ_oxide = 1 − θ_graphene is thus compatible with both scenarios. A direct test of carbon pre-existence is needed, for example by characterizing the graphene surface before ALD on identically prepared samples, by performing ALD on samples with controlled levels of ambient exposure, or by atomic-scale elemental mapping of the carbon/HfO2 interface to show that the carbon layer lies beneath the HfO2 rather than on top of it.","section":"Results and Discussion, paragraph following Fig. 4d"},{"comment":"The areal coverage fractions are reported as single values with no error bars and no statement of how many independent images or sample regions were analyzed: θ_oxide,monolayer ≈ 54%, θ_oxide,bilayer ≈ 20%, θ_graphene,monolayer ≈ 17%, θ_graphene,bilayer ≈ 22%, θ_oxide,6 cycles ≈ 15%, θ_graphene,6 cycles ≈ 24%, and θ_graphene,no ALD ≈ 25–30%. The central argument that θ_graphene is 'roughly constant' despite strongly varying θ_oxide is the quantitative linchpin of the carbon-seeding conclusion, but differences of 5–8 percentage points between the quoted values are not distinguishable from measurement noise without sampling statistics. The 6-cycle versus 16-cycle comparison also uses different sample regions and lacks paired statistics, so it cannot robustly corroborate the constancy of θ_graphene.","section":"Results and Discussion, Fig. 4b–d and surrounding text"},{"comment":"The authors note that e-beam exposure can induce contamination dewetting and/or attraction of additional contamination, and they state that such changes were minimal for their imaging conditions, but no quantitative evidence is provided. Because the observed spatial correlation between HfO2 clusters and carbon-like contrast is the primary evidence for the seeding mechanism, beam-induced carbon redistribution during STEM imaging could in principle produce the same apparent correlation even if HfO2 had nucleated on clean graphene. The authors should report low-dose control images or otherwise quantify the stability of the carbon distribution under the exact scan conditions used for the correlation analysis.","section":"Results and Discussion, paragraph preceding Fig. 4"}],"minor_comments":[{"comment":"The phrase 'a 5 few layer graphene region' in the Fig. 4c description is unclear; it presumably means a few-layer graphene region of five layers, and this should be stated explicitly.","section":"Fig. 4c caption and text"},{"comment":"The text states the HAADF intensity 'scales linearly with specimen thickness' and also 'provides materials contrast which is dependent on Z with Z~1.64'. The notation Z~1.64 is ambiguous; it should be written as Z^1.64 to indicate the power-law dependence.","section":"Results and Discussion, paragraph following Fig. 2d"},{"comment":"The phase assignment to cubic and/or tetragonal HfO2 is made on a qualitative presence/absence basis. The authors should clarify whether the data can distinguish between cubic and tetragonal, and if not, state that the two phases are indistinguishable at the available resolution.","section":"SAED phase identification, Fig. 3d and Supporting Fig. S3"},{"comment":"The polymer-free direct transfer is cited to reference 40 (Regan et al.), but it is worth stating explicitly that no polymer support layer is used in that method, since the absence of polymer residues is a key premise of the paper.","section":"Methods, transfer procedure"},{"comment":"Several minor grammatical issues appear (e.g., 'the salient features observed in image labelled' in the Fig. 2d caption, and 'under conditions' instead of 'under our conditions' in the Conclusions). A careful proofreading pass is recommended.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The central claim is novel and potentially important, but it is currently under-supported by the evidence presented. The authors need either a direct observation of pre-ALD carbon at nucleation sites or a control experiment that manipulates carbon exposure before ALD. The rest of the paper's observations—layer-dependent Cu assistance, step-edge nucleation, e-beam crystallization, and grain-boundary comparison—are solid and could stand as a useful contribution even if the carbon-seeding claim were softened to a correlation rather than a causal mechanism. I would not reject, but the carbon-seeding conclusion must be either strengthened or appropriately reframed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis is a paper you should know about if you care about ALD on 2D materials. The headline: it provides atomic-scale images of HfO2 nucleation on polymer-free CVD graphene, with what looks like clear spatial correlation between HfO2 clusters and adventitious carbon. The authors argue that the carbon acts as an unintentional seeding layer. That is the interesting claim, and it is probably right, but the paper does not actually prove it.\n\nWhat is genuinely new: they do substrate-assisted ALD directly on graphene/Cu, avoiding polymer transfer residues, and then image the nucleation stage with HAADF-STEM. The layer-dependent coverage (monolayer >> bilayer/few-layer) is a clean demonstration of the Cu substrate contribution. The step-edge nucleation and the e-beam-induced crystallization to cubic/tetragonal HfO2 are nice additions, and the SAED phase assignment is reasonable, if qualitative.\n\nThe soft spot is the causal direction. The evidence is post-ALD spatial correlation. To conclude \"carbon attracts HfO2\" you need carbon present before ALD and underneath the clusters. The authors rely on visual continuity and on the claim that the bare-graphene fraction stays roughly constant while HfO2 coverage changes. But that constant-fraction argument is weaker than it looks: θ_graphene only counts atomically clean graphene; carbon that lands on top of HfO2 clusters or at their edges after ALD would not register as bare graphene. So the data are compatible with HfO2 nucleating on clean graphene and carbon decorating it later. The non-ALD control shows similar bare-graphene fractions, but on a different sample region, so it cannot establish pre-existence at the nucleation sites. The authors do flag the e-beam artifact concern themselves, but that doesn't close the gap.\n\nThe quantification is also thin: coverage fractions come from single STEM images, without error bars or any statement about how many regions/samples were analyzed. That is a minor problem for the layer-dependence claim but not a fatal one; the trend is consistent across SEM and TEM data.\n\nBottom line: the paper is a solid empirical contribution with an honest, well-hedged interpretation. The carbon-seeding mechanism is plausible and consistent with prior XPS work, but it is not proven. This deserves a serious referee, and the authors should be asked to either soften the causal claim to \"spatial correlation and a plausible mechanism\" or provide pre-ALD evidence (for example, before/after imaging or in-situ ALD in the STEM). I would take it with a grain of salt for the mechanism, but it is a useful paper for the ALD community.","headline":"Strong microscopy and a plausible, well-hedged carbon-seeding story, but the causal claim rests on an unproven assumption; deserves review and a careful revision.","tokens_in":19495,"tokens_out":2553,"would_cite":true,"duration_ms":26464,"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":"Atomic-scale imaging shows hafnium oxide nucleates on adventitious carbon contamination from ambient air, which acts as an unintentional seeding layer.","keywords":["atomic layer deposition","graphene","hafnium oxide","nucleation","adventitious carbon","scanning transmission electron microscopy","substrate-assisted ALD","high-k dielectric"],"falsifier":"Image the identical graphene region before and after ALD, or after in-situ removal of the adventitious carbon: if HfO$_2$ appears on regions that were atomically clean before deposition, or if carbon is absent underneath freshly nucleated clusters, the \"carbon attracts HfO$_2$\" scenario is wrong; a companion test is an UHV-clean, ambient-free ALD run, which should strongly suppress carbon-correlated nucleation.","tokens_in":18492,"feed_emoji":"🔬","tokens_out":11605,"duration_ms":93221,"temperature":0.7,"pith_summary":"Using atomically resolved electron microscopy, this paper tries to establish why hafnium oxide nucleates where it does when it is grown by atomic layer deposition (ALD) on graphene that has not been transferred off its copper growth catalyst. The central claim is that under realistic, non-ultra-high-vacuum conditions the oxide does not nucleate on clean graphene basal planes: it starts on adventitious carbon contamination picked up from ambient air, which therefore acts as an unintentional seeding layer. The paper also argues that the copper support participates catalytically in a graphene-layer-dependent way, with monolayer graphene acquiring far more HfO$_2$ than bilayer or few-layer regions, and that graphene layer steps act as chemically reactive nucleation lines. If this is right, ambient-air exposure history is a first-order control variable in ALD coating of two-dimensional materials, alongside precursor chemistry and temperature.","feed_headline":"Airborne carbon, not clean graphene, seeds hafnium oxide growth","feed_subtitle":"Atomic-scale imaging shows hafnium oxide nucleates on ambient carbon contamination, so air exposure steers ALD coating.","key_machinery":"The decisive object is the atomic-number-contrast HAADF-STEM image: carbon (Z=6) is dark, hafnium (Z=72) is bright, and the graphene lattice can be resolved, so each nucleation site can be classified as sitting on bare graphene, on adventitious carbon, or on the HfO$_2$ clusters themselves. The analytical step that carries the argument is the areal-fraction comparison encapsulated in $\\theta_{\\mathrm{contamination}} + \\theta_{\\mathrm{oxide}} = 1 - \\theta_{\\mathrm{graphene}}$, together with the observation that $\\theta_{\\mathrm{graphene}}$ stays nearly constant while $\\theta_{\\mathrm{oxide}}$ changes strongly. That combination rules out the competing \"HfO$_2$ attracts carbon\" scenario and identifies the carbon as the pre-existing seed.","core_discovery":"The paper reports direct atomic-scale HAADF-STEM evidence that ALD HfO$_2$ on chemical-vapor-deposited graphene nucleates preferentially on pre-existing adventitious carbon rather than on atomically clean graphene. The authors explicitly weigh two scenarios, \"HfO$_2$ attracts carbon\" versus \"carbon attracts HfO$_2$\", and conclude for the latter from the visual spatial continuity of a carbon network underneath the oxide and from measured areal fractions: the bare-graphene fraction stays roughly constant (about 17–24%) across regions that differ strongly in HfO$_2$ coverage and across ALD runs with 6 versus 16 cycles, whereas the oxide coverage changes from about 54% on monolayer graphene to about 20% on bilayer graphene. They attribute the layer-number dependence to the copper catalyst's electronic influence decaying through additional graphene layers. The as-deposited HfO$_2$ is amorphous and, under the electron beam, crystallizes to metastable cubic/tetragonal polymorphs rather than the equilibrium monoclinic phase, and no epitaxial relationship with the graphene lattice appears.","pith_inferences":["Editorial inference: the carbon-seeding mechanism is likely to transfer to other ALD oxide/2D-material combinations where ambient exposure is not controlled, since the paper's cited XPS work already connects it to graphite and MoS$_2$; a controlled comparative study would make the generality quantitative.","Editorial inference: if contamination is the true seed, deliberately depositing a uniform carbonaceous or oxygen-functional adlayer before ALD could convert a source of run-to-run irreproducibility into a reproducible nucleation-density knob.","Editorial inference: the cleanest falsification the authors could not perform in situ is a same-region before/after comparison; with a heating stage or in-situ cleaning in the microscope one could strip the adventitious carbon, deposit ALD cycles, and check whether HfO$_2$ then nucleates elsewhere or not at all.","Editorial inference: the electron-beam crystallization to cubic/tetragonal HfO$_2$ may not reproduce furnace annealing, so the polymorph claim for device processing should be checked with conventional thermal anneals before relying on the high-k phases."],"forward_implications":["Ambient air exposure before or during ALD should be treated as a deliberate process parameter, because it controls where and how densely HfO$_2$ nucleates.","Repeatedly interrupting ALD with air exposure should incrementally contaminate still-clean graphene and thereby homogenize subsequent oxide coating, matching the earlier Al$_2$O$_3$ result cited in the paper.","On Cu-supported graphene, monolayer regions will coat more completely than bilayer or few-layer regions, so the areal uniformity of the graphene layer number sets an upper bound on oxide film uniformity.","Graphene layer steps and folds will collect thicker oxide and may act as leakage or contact points in devices, since they are preferential nucleation lines.","Thin ALD HfO$_2$ films on graphene start amorphous and, when crystallized, favor metastable cubic/tetragonal phases, which is relevant because those phases have higher dielectric constants than monoclinic HfO$_2$."],"supporting_citations":[{"why":"Supplies the substrate-assisted ALD process parameters and prior calibration of thickness versus cycle number.","marker":"[28]"},{"why":"Proposes the substrate-assisted ALD mechanism in which the Cu catalyst participates electronically through the graphene, which the paper tests layer by layer.","marker":"[34]"},{"why":"Documents adventitious carbon as ubiquitous on non-UHV graphene and identifiable in TEM, grounding the contamination identification.","marker":"[42]"},{"why":"Shows evaporated metals sit preferentially on carbon adsorbates on graphene, the closest prior observation of the carbon-seeding pattern.","marker":"[43]"},{"why":"Provides prior XPS evidence that adventitious carbon on graphite changes Al2O3 ALD composition, integral support for contamination-mediated nucleation.","marker":"[45]"},{"why":"Extends the graphite XPS study to trimethylaluminum/ozone chemistry, reinforcing that precursor adsorption is governed by carbonaceous surface condition.","marker":"[46]"},{"why":"Shows HfO2 ALD on MoS2 is affected by contamination, indicating the carbon-seeding effect is not unique to graphene.","marker":"[47]"},{"why":"Reports that interrupting Al2O3 ALD on graphene with air exposure improves film homogeneity, which the paper interprets as independent support for incremental carbon seeding.","marker":"[68]"}],"fun_headline_variants":["Hafnium oxide nucleates on airborne carbon, not pristine graphene","Airborne carbon seeds HfO2 on graphene, not clean lattice","Nucleation on graphene: carbon contamination does the seeding","Copper catalyst and adventitious carbon direct ALD on graphene"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The carbon-seeding conclusion assumes the carbon seen under and around the HfO$_2$ was already on the graphene before deposition and formed a continuous layer underneath the oxide, rather than accumulating at HfO$_2$ sites during or after ALD or being redistributed by the electron beam.","fun_headline_variants_meta":{"raw":{"variants":["Hafnium oxide nucleates on airborne carbon, not pristine graphene","Airborne carbon seeds HfO2 on graphene, not clean lattice","Nucleation on graphene: carbon contamination does the seeding","Copper catalyst and adventitious carbon direct ALD on graphene"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000561,"raw_usage":{"total_tokens":2748,"prompt_tokens":1114,"completion_tokens":1634,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":730,"completion_tokens_details":{"reasoning_tokens":1560}},"tokens_in":730,"tokens_out":1634,"duration_ms":83786,"temperature":1.0,"reasoning_tokens":1560,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:37:50.669334+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Image the identical graphene region before and after ALD, or after in-situ removal of the adventitious carbon: if HfO$_2$ appears on regions that were atomically clean before deposition, or if carbon is absent underneath freshly nucleated clusters, the \"carbon attracts HfO$_2$\" scenario is wrong; a companion test is an UHV-clean, ambient-free ALD run, which should strongly suppress carbon-correlated nucleation.","supporting_citations":[],"review_version":1}