{"id":"c3853e0a-15cf-41d6-b446-b5aa3a921d81","arxiv_id":"2507.22677","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A dry layer-by-layer process stacks wafer-scale single-crystal graphene and hBN monolayers with near-unity yield, clean interfaces, and programmable twist angles.","lead":"This paper reports a dry, wafer-scale process for stacking single layers of graphene and hexagonal boron nitride into clean multilayer films with controlled twist angles. It matters because it could turn hand-made atomically thin stacks into something manufactured at scale.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'near-unity yield' claim is supported only by area-averaged optical absorption, which cannot distinguish per-layer transfer yield from full-stack completeness; sparse missing layers would be hidden in the average.","rationale":"The reader's weakest assumption focuses on clean, repeated mechanical peeling from Ge(110) without losing orientation or accumulating contamination; my concern is closely related but more precise: the 'near-unity yield' evidence is an area average, and averaging is mathematically insensitive to the difference between per-layer transfer yield and complete-stack yield. This is a statistical/evidential gap in the paper's own data, not merely a request for replication. It is load-bearing because the title and abstract claim 'near-unity yield' at wafer scale, and the only wafer-scale yield metric presented is optical absorption linearity. The concern does not rest on outside consensus or ad hominem; it is a request to quantify what the reported optical data actually constrain. I therefore keep the reader's CONDITIONAL verdict unchanged: the paper is plausible and internally coherent, but the yield claim should be re-derived from pixel-resolved statistics before the headline is accepted at face value.","tokens_in":22520,"tokens_out":9690,"duration_ms":134585,"concrete_test":"Re-analyze the raw optical transmission and hyperspectral data underlying fig. S7 and fig. 2d at pixel resolution: for each multilayer film, compute the fraction of pixels whose local absorption falls within a stated tolerance (for example, ±5%) of the ideal NL times AML, and report the full-stack yield as a map across the 2-inch sample. If the median full-stack yield is below 99% or is not reported, the 'near-unity yield' claim is not supported. A complementary check is to count devices in the tunnel-device array whose R0A lies within one order of magnitude of the ideal NL-dependent value; a large deviation fraction would also weaken the atomic-thickness-control claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central 'wafer-scale films with pristine interfaces and near-unity yield' claim is not established by the reported yield evidence. In the main text and fig. S7, the authors deduce the per-layer optical absorption AML from multilayer films and note that the deduced AML is consistent with single-layer reference data and similar for different NL. That comparison constrains the average number of layers per unit area (roughly Y times NL), not the fraction of the area in which all NL layers are actually present. If each layer transfers independently with success probability Y, the average optical absorption is linear in NL with slope Y times AML, even when Y is substantially below 1. For Y = 0.98, the average absorption is about 98% of the ideal value, indistinguishable from 'near-unity' within typical optical uncertainty, yet only 0.98^10 ≈ 82% of the area of a 10-layer film has a complete stack. For Y = 0.95, only about 60% of the area is complete. The same averaging affects the tunnel-device R0A data, which are measured on small, selected junction areas and cannot report wafer-scale stack completeness. Thus the paper's reported evidence conflates per-layer transfer yield with full-stack yield, and sparse local failures such as tears, wrinkles, or missing layers would be hidden in the optical averages. This is load-bearing because 'near-unity yield' is one of the two headline capabilities of PCA.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a dry, wafer-scale 'programmed crystal assembly' (PCA) process in which epitaxial monolayer graphene and monolayer hBN grown on Ge(110) are repeatedly exfoliated and stacked under van der Waals control. The authors claim atomically pristine interfaces, near-unity stacking yield, layer-resolved chemical and twist-angle control, and demonstrate these via graphene/hBN superlattices, twisted multilayer graphene with programmed band structures, and batch-fabricated tunnel junctions whose resistance is controlled by hBN barrier thickness with single-atom precision. The evidence base includes cross-sectional STEM/EELS, LEED, XRD, XPS, AFM, optical absorption and hyperspectral imaging, ARPES, and circular dichroism measurements.","tokens_in":22814,"tokens_out":3690,"duration_ms":47505,"significance":"If the central claims hold, PCA would be a substantial advance: it would make wafer-scale, atomically clean van der Waals heterostructures of graphene and hBN routinely available, enabling artificial structures that are currently limited to small exfoliated flakes. The paper deserves credit for the breadth and independence of its structural characterization: STEM/EELS directly images alternating C/N layers, the XRD coherence length approaches the full stack thickness, LEED confirms wafer-scale crystallographic alignment, and ARPES resolves mini-gaps in twisted trilayers. The main weaknesses are that the 'near-unity yield' claim rests on area-averaged optical absorption rather than a direct measure of stack completeness, and that the tunneling-model agreement is weakened by a fitted barrier height reused as an input. These are load-bearing for the two headline capabilities, but they are addressable with additional analysis or reframing.","major_comments":[{"comment":"The 'near-unity yield' claim is not established by the reported optical absorption data. Absorption measured over a macroscopic area is an average over the illuminated region, so it constrains the mean number of layers per unit area, not the fraction of the area in which all NL layers are actually present. If each layer is transferred independently with success probability p, the average absorption is linear in NL with slope p times the monolayer absorption, even when p is substantially below 1; for p=0.98 and NL=10 the average absorption would be about 98% of ideal while only about 82% of the area contains a complete stack, and for p=0.95 only about 60% of the area is complete. The R0A data in Fig. 3b, moreover, are measured on small, selected junction areas and cannot report wafer-scale completeness. Because 'near-unity yield' is repeated in the abstract and conclusion, the authors should either provide spatially resolved layer-count statistics over millimeter or wafer scales, or revise the claim to 'high average layer coverage' and state the distinction explicitly.","section":"Main text, 'Wafer-scale PCA' (Fig. 2d; fig. S7)"},{"comment":"The tunneling model validation contains a circular element. The barrier height phi is fitted to the same NL-dependent R0A data that the model is then used to explain, and the phonon energy threshold of ±63 meV is introduced as an additional input when simulating dI/dV. The agreement between the simulated and measured dI/dV curves in Fig. 3d is therefore not an independent confirmation of pristine interfaces. The consistency of the fitted 3.0 eV with the cited ARPES value is suggestive, but consistency with an external value is not the same as using that external value as a fixed input. The authors should either fix phi from the ARPES value and then treat the R0A fit as a predictive test, or show explicitly that the dI/dV features (positions of VA and VD and the zero-bias dip) are insensitive to the fitted phi and to the phonon threshold within reasonable ranges.","section":"Supporting Text, 'Tunneling model' and 'Estimation of tunneling barrier height' (Eqs. 1-3)"},{"comment":"The claim that twist angles are controlled to within about ±1° is supported by a single TEM diffraction pattern of one CTG stack and by LEED patterns in fig. S6. The edge-based alignment method is plausible and the LEED data in fig. S6E are encouraging, but the reported angle accuracy should be supported by statistics over multiple samples and over larger areas. If the PCA method is to be sold as providing programmable band structures with reliable twist control, a quantitative distribution of achieved theta_i across multiple wafers and stacks should be reported.","section":"Main text and Fig. 1e, twist-angle control"}],"minor_comments":[{"comment":"The sentence 'J decreased significantly by ~1 % per layer as NL increased' appears to be a typo; the data in Fig. 3b show a decrease of roughly one order of magnitude per hBN layer, not 1%. Please correct this to avoid confusion.","section":"Main text, Fig. 3b"},{"comment":"The main text refers to 'a theoretical model20' for the dz vs. theta_i dependence, while the Fig. 2f caption cites reference 18 for the theoretical values. These citations are inconsistent and should be reconciled.","section":"Figure 2f caption vs. main text"},{"comment":"The statement that chiro-optical spectral weight increases with NL is shown only up to NL=6 (fig. S18), yet the text says this demonstrates properties 'in multilayer films with large NL.' Please qualify the claim to the measured range or provide data for larger NL.","section":"Supporting Text, 'Circular dichroism from multilayer CTG'"},{"comment":"The phrase 'Near-unity yield of stacking over a macroscopic area was also confirmed by optical absorption measurements' is misleading because, as discussed in the major comments, absorption is an areal average. At minimum, rephrase to 'consistent with a high average layer coverage' and cite the limitation.","section":"Main text, 'Wafer-scale PCA' (Fig. 2d)"},{"comment":"The phrase 'The current densities J normalized by the junction area' is redundant; current density is already normalized by area. Consider simplifying to 'The current densities J were measured...'.","section":"Methods, 'Tunneling current measurement'"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically rich and the structural evidence for clean interfaces is strong, but the headline 'near-unity yield' claim needs a spatially resolved yield metric, and the tunneling model should be reframed to avoid the appearance of circular validation. These are fixable within the manuscript's scope, so I recommend major revision rather than rejection. I would also suggest the editor ask the authors to clarify the statistical basis for the ±1° twist-angle claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. The paper reports something the 2D-materials community has been missing: a repeatable, all-dry, wafer-scale way to stack single-crystal graphene and monolayer hBN with controlled twist. The prior state of the art either used small exfoliated flakes or was limited to TMDs. The authors extend their earlier all-dry transfer work (ref 19) to repeated stacking and show a breadth of characterization that generally supports the clean-interface claim. Cross-sectional STEM/EELS shows alternating layers without contamination, XRD coherence length matches nearly the full film thickness, and ARPES on CTG/ATG shows twist-programmed band structure with mini-gaps. That is real, and it is the core of the paper.\n\nThe weak spot is the 'near-unity yield' claim. The evidence is optical absorption scaling with layer number, which only constrains the average number of layers per area. As the stress-test note correctly points out, a per-layer success rate of 98% would be nearly indistinguishable in that measurement, yet only 82% of a 10-layer stack would be complete. The paper conflates per-layer yield with full-stack yield. This is load-bearing because 'near-unity yield' is a headline capability. The tunnel-device data are measured on small selected junctions, so they don't fix it. The authors need direct spatial statistics—Raman or AFM mapping over a meaningful area, or quantitative optical imaging that can count missing layers.\n\nThe tunneling validation also has a circular element: the barrier height is fit to the layer-dependent R0A and then used in the dI/dV simulation. That doesn't sink the paper—the fitted 3.0 eV agrees with an independent ARPES value for hBN on graphite—but the agreement should be presented as consistency, not as confirmation.\n\nOverall, this is a serious experimental paper with a plausible central capability and strong multi-technique support for interface quality. The yield claim is overreaching relative to the evidence. I would send it to peer review and ask for direct yield statistics and a more measured claim.","headline":"A genuinely useful wafer-scale dry-stacking capability for graphene/hBN with strong interface evidence, but the 'near-unity yield' claim is not supported by the area-averaged optics.","tokens_in":23390,"tokens_out":2358,"would_cite":true,"duration_ms":27764,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Programmed dry stacking builds wafer-scale films of 2D crystals with pristine interfaces.","keywords":["graphene","hexagonal boron nitride","van der Waals heterostructures","two-dimensional materials","wafer-scale assembly","twisted multilayer graphene","tunnel junctions","programmed crystal assembly"],"falsifier":"A decisive observation would be to cut cross-sections at several random positions of a wafer-scale PCA stack and use atomic-resolution EELS or EDS to search for a continuous contamination layer, such as amorphous hydrocarbon, at any graphene/hBN or graphene/graphene interface.","tokens_in":22357,"feed_emoji":"🔬","tokens_out":12354,"duration_ms":122401,"temperature":0.7,"pith_summary":"The paper reports programmed crystal assembly (PCA), an all-dry method for stacking graphene and monolayer hexagonal boron nitride into wafer-scale films with atomically clean interfaces. The central claim is that epitaxial films grown on Ge(110) can be peeled and re-stacked repeatedly, van der Waals forces doing the work, to build multilayers whose layer composition and twist angle are set in advance. Because the interfaces stay pristine, the resulting films behave like single crystals: hBN tunnel barriers show thickness-controlled resistance over orders of magnitude, and twisted multilayer graphene shows the mini-gaps and symmetry-broken bands predicted for chiral and achiral stacks. If the method works as claimed, clean, wafer-scale van der Waals heterostructures become a routine platform for electronics and for testing correlated and topological states.","feed_headline":"Wafer-scale 2D crystal stacks now assemble with near-unity yield","feed_subtitle":"Dry van der Waals stacking keeps graphene/hBN interfaces atomically clean and twist angles precise.","key_machinery":"The central object is the one-atom-thick crystal itself, grown epitaxially on Ge(110) and used as a reusable assembly unit. The mechanical trick is that the interaction energy between the Ge(110) substrate and the as-grown film is lower than the van der Waals binding between layered materials, so a film capped with a soft gold layer can be peeled off dry, aligned with straight crystalline edges to set the twist angle, stacked, and the cycle repeated. The gold support makes conformal contact and is removed only after the stack is complete, keeping polymers and etchants away from the interfaces. The straight edges of the Ge(110) wafer serve as a global angular reference, so the twist angle at each interface is set by optical alignment rather than by post-hoc characterization.","core_discovery":"The central discovery is that atomically clean, wafer-scale assembly of graphene and monolayer hBN is achievable by exploiting the weak van der Waals interaction between the as-grown films and their Ge(110) growth substrates. Because the film-substrate bond is weaker than the interlayer bond within layered crystals, a film can be peeled off dry with a gold support, placed on another as-grown film, and the cycle repeated, with no polymer or etchant touching the interfaces. The paper demonstrates the result in two working systems: vertical graphene/hBN superlattices with alternating chemical composition, and twisted multilayer graphene in which each interface is rotated by a programmed angle, including chiral twisted graphite (constant rotation sense) and achiral twisted graphite (alternating sign). The evidence includes cross-sectional STEM and EELS showing pristine interfaces and consistent interlayer spacings of $\\d_z = 3.37$ Å for graphene/hBN and $3.41$ Å for twisted graphene, XRD showing coherent diffraction over millimeter scales, tunnel devices whose resistance scales exponentially with hBN layer number, and ARPES showing angle-dependent mini-gaps and hybridized parabolic bands.","pith_inferences":["If the central claim holds, the usual size-versus-yield trade-off of exfoliation-based assembly disappears, so experiments that need many clean interfaces at once, such as magic-angle hierarchies in N-layer twisted graphene, could be run on centimeter-scale samples instead of micrometer flakes.","The edge-alignment trick sets twist angles from the Ge(110) crystal edges; a natural extension is to test whether the same geometry can lock angles after stacking through moiré reconstruction, which would matter for devices needing angles near the magic angle.","The pristine-interface claim rests on spectroscopy and imaging that show no contaminants; an independent check would be to compare transport or optical linewidths of PCA-made stacks with those of exfoliated pick-up stacks of the same geometry.","Because the process uses a gold support and thermal release tape, it could plausibly be automated and integrated into a fabrication line, a path the paper does not itself detail."],"forward_implications":["hBN tunnel barriers made by PCA show zero-bias resistance-area products that scale exponentially with the number of monolayers, meaning the barrier thickness is controlled one atom at a time across a centimeter-scale array.","Chiral twisted graphite with a constant rotation sense shows equally spaced hybridized Dirac bands with mini-gaps of about 200 meV, while alternating-twist graphite shows a larger mini-gap of about 300 meV from collective hybridization across interfaces.","The measured interlayer spacing of twisted graphite rises by about 2% from 3.34 Å at zero twist to a plateau at twist angles above 3°, matching a relaxation model, which means the assembled stacks take on the equilibrium structure of clean interfaces.","The assembly unit extends beyond graphene and hBN: a graphene/MoS2 superlattice was formed with near-unity yield, so metal, semiconductor, and insulator layers can be assembled in programmed sequences.","Optical circular dichroism in chiral twisted graphite strengthens with the number of layers, indicating that multi-interface properties scale predictably when the interfaces are pristine."],"supporting_citations":[{"why":"Supplies wafer-scale single-crystal graphene growth on Ge(110), the reusable assembly unit.","marker":"[14]"},{"why":"Supplies aligned monolayer hBN growth on germanium, giving the second assembly unit.","marker":"[18]"},{"why":"Establishes the weak film-substrate interaction and the all-dry peeling principle used in PCA.","marker":"[19]"},{"why":"Provides the structural-reconstruction picture for small twist angles that explains the interlayer spacing trend.","marker":"[5]"},{"why":"Provides the theoretical relaxation model for moiré patterns that the measured $\\d_z$ versus twist angle is compared to.","marker":"[20]"},{"why":"Supplies the tunneling model and vertical-junction geometry used to analyze hBN barrier transport.","marker":"[21]"},{"why":"Supplies the hBN band-structure offset used as a consistency check for the fitted tunnel barrier height.","marker":"[22]"},{"why":"Provides theoretical predictions for twisted graphene multilayers with many layers that PCA aims to realize.","marker":"[28]"},{"why":"Provides theoretical predictions for chiral and achiral twisted graphite band topology that the measured bands are read against.","marker":"[29]"}],"fun_headline_variants":["Wafer-scale 2D stacks now assemble atomically clean","Dry van der Waals stacking builds pristine 2D films","Programmed stacking gives wafer-scale pristine 2D crystals","Atomic precision twist-stacking of 2D films at wafer scale"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that graphene and hBN monolayers grown on Ge(110) can be peeled off and re-stacked many times in dry conditions without losing their single-crystal orientation or picking up contamination at the new interfaces.","fun_headline_variants_meta":{"raw":{"variants":["Wafer-scale 2D stacks now assemble atomically clean","Dry van der Waals stacking builds pristine 2D films","Programmed stacking gives wafer-scale pristine 2D crystals","Atomic precision twist-stacking of 2D films at wafer scale"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000712,"raw_usage":{"total_tokens":3199,"prompt_tokens":937,"completion_tokens":2262,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":553,"completion_tokens_details":{"reasoning_tokens":2191}},"tokens_in":553,"tokens_out":2262,"duration_ms":17346,"temperature":1.0,"reasoning_tokens":2191,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T11:24:16.581180+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive observation would be to cut cross-sections at several random positions of a wafer-scale PCA stack and use atomic-resolution EELS or EDS to search for a continuous contamination layer, such as amorphous hydrocarbon, at any graphene/hBN or graphene/graphene interface.","supporting_citations":[{"cited_title":"H.; Hovden, R.; Tsen, A","cited_arxiv_id":null,"evidence_quote":"Provides the structural-reconstruction picture for small twist angles that explains the interlayer spacing trend."}],"review_version":1}