{"id":"4a45efa0-bdd4-4dd4-bd01-c140bbf52488","arxiv_id":"1908.00874","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Direct 3D atom probe maps reveal that Si/SiGe buried interfaces have 0.2-0.3 nm roughness, 8-10 nm lateral correlation lengths, and lower growth temperature sharpens the interfaces.","lead":"This paper uses atom probe tomography to map the roughness of buried Si/SiGe interfaces atom by atom, finding roughness near 0.2 to 0.3 nm and lateral correlation lengths of 8 to 10 nm. A generalist might read it because these numbers have been treated as fitting parameters in device and laser models, and direct measurement can improve them.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 0.2–0.3 nm roughness values may be below the statistical and geometric resolution of the APT iso-surface pipeline; the reported H(τ) parameters are not separated from voxel counting noise or reconstruction artifacts.","rationale":"The reader's weakest assumption identifies the same load-bearing premise: the APT reconstruction, after thickness-only calibration, preserves the lateral and vertical geometry needed for sub-0.3 nm roughness and correlation-length measurements. I agree, and I sharpen the concern with a concrete numerical mechanism: the 1 nm³ voxel grid and finite Ge atom counts create a statistical noise floor in the iso-concentration surface height that is comparable to the reported σ values. This is not an internal inconsistency; the paper's own text acknowledges APT artifacts, but it only validates thicknesses against XTEM and only checks density correction against interface widths, not against σ or ξ. The optical modeling section is supportive but is a separate fitting argument and does not independently validate the 0.2–0.3 nm roughness scale; the SE-extracted interfacial widths are larger and carry different systematic errors. Because the central quantitative claims depend on the unvalidated sub-voxel height field, the conditional verdict is appropriate. No change to the reader's verdict is needed: the concern is real, but it is addressable with a synthetic reconstruction test, so CONDITIONAL remains the right call rather than outright rejection.","tokens_in":19782,"tokens_out":4197,"duration_ms":48325,"concrete_test":"Run a forward-model validation for S-16: simulate a Si/SiGe superlattice with prescribed interface roughness σ_true = 0.2 nm and ξ_true = 8 nm (same layer thicknesses and compositions), process it through a realistic APT evaporation/reconstruction model (e.g., TAPSim or an equivalent trajectory-aberration simulator), and then apply the manuscript's exact voxelization, 50% Ge iso-surface construction, and H(τ) fitting pipeline. If the recovered σ deviates from 0.2 nm by more than ~0.05 nm (25%) or ξ by more than ~1.5 nm, the reported values are dominated by reconstruction/voxelization artifacts rather than physical interface roughness. A faster complementary check on existing data: bootstrap-resample the Ge atoms in one S-16 dataset and recompute σ; if σ changes by more than ~0.05 nm, voxel counting noise alone can explain the result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline quantitative claim (σ ≈ 0.2–0.3 nm, ξ ≈ 8.1–10.1 nm) depends on the premise that Ge iso-concentration surfaces extracted from the APT reconstruction reproduce the true interface height field at sub-0.3 nm precision. That premise is not secured by the paper's calibration procedure. The text states that APT reconstruction parameters were iteratively adjusted only until reconstructed layer thicknesses matched XTEM within 5%, which constrains average vertical layer spacings but does not validate lateral geometry, local magnification, or the sub-voxel height fluctuations that define σ and ξ. The iso-surfaces are built from 1 nm³ voxels; with ~25 at.% Ge and typical APT detection efficiency, each voxel contains only tens of Ge atoms, so voxel-to-voxel counting noise (a few at.%) translates through a ~1 nm interfacial gradient into height fluctuations of order 0.1–0.2 nm—the same magnitude as the reported σ. The quoted uncertainties are standard deviations over interfaces and datasets, not fit errors or voxel-statistics errors, so they do not capture this floor. The authors explicitly acknowledge that APT reconstructions can be subject to artifacts, and the SI reports testing z-density correction only for interface widths, not for σ or ξ. Additionally, damaged top interfaces were excluded (top 3–4 for dataset 1; top 6 for dataset 2), so the surviving interface ensemble may be non-representative. Until the voxel-statistics floor and reconstruction fidelity are quantified, the 500–650 °C trend in σ and the correlation-length differences cannot be distinguished from measurement artifacts.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents a method to map the roughness and lateral correlation of buried Si/SiGe interfaces in sub-10 nm superlattices using atom probe tomography (APT). The authors extract iso-concentration surfaces at 50% of the mean Ge concentration from APT reconstructions, compute the height-height correlation function H(τ), and fit it to extract RMS roughness σ and horizontal correlation length ξ. They report σ ≈ 0.2–0.3 nm and ξ ≈ 8.1–10.1 nm, with the 500 °C sample showing roughly 30% smaller σ and 24% larger ξ than the 650 °C samples. They also fit 1D Ge concentration profiles with a sigmoid to obtain interface widths of ~0.7–1.0 nm, and they compare four spectroscopic ellipsometry (SE) models (no interface, two effective-medium approximations, and a graded interface model) to argue that interfacial broadening must be included to describe the optical response.","tokens_in":20100,"tokens_out":7027,"duration_ms":63896,"significance":"The work addresses a real gap: buried interface roughness and correlation lengths in technologically relevant Si/SiGe heterostructures are presently treated as fit parameters in device and optical models, and direct 3D measurements are scarce. The multi-technique approach (APT, EELS, XTEM, HRXRD, SE) and the inclusion of multiple APT datasets per sample are strengths. If the quantitative claims hold, the results would provide useful input for modeling intersubband devices and Si nanosheet transistors. However, the central quantitative claims about σ and ξ rest on assumptions about APT reconstruction fidelity that are not fully validated, and the optical-modeling conclusion is logically narrower than the abstract states. The paper also includes a genuinely useful comparison of SE models with interface layers.","major_comments":[{"comment":"The reported RMS roughness values (σ ≈ 0.2–0.3 nm) are comparable to the height uncertainty expected from voxel counting statistics: with 1 nm³ voxels, ~25 at.% Ge, and typical APT detection efficiency, each voxel contains only tens of Ge atoms, so the iso-concentration surface height carries an uncertainty of order 0.1–0.2 nm. The manuscript does not quantify this noise floor or separate it from the physical roughness; the quoted ±11–13% uncertainties are standard deviations across interfaces and datasets, not measurement errors. Without a noise-floor estimate (e.g., from a Monte Carlo simulation of voxelated data or a reconstructed known-flat interface), the claim that σ varies from 0.2 to 0.3 nm cannot be distinguished from a constant value near the method's resolution limit.","section":"Fig. 2(a), Eq. (1), and Section 'Extraction of RMS roughness and horizontal correlation length'"},{"comment":"The APT reconstruction is calibrated only by iteratively matching layer thicknesses to XTEM within 5%, which constrains average vertical layer spacings but does not validate lateral geometry, local magnification, or sub-voxel height fluctuations. The SI reports that z-density correction did not change interface widths, but no test is presented for σ or ξ. The manuscript should either provide an independent validation of the roughness measurement (e.g., comparison with an established technique on a test structure) or explicitly discuss the systematic uncertainty this calibration leaves in the reported σ and ξ.","section":"Experimental section, 'Atom probe tomography', and SI Section 1"},{"comment":"The temperature dependence of σ and ξ is inferred from comparing S-16 and S-12 (650 °C), S-6 (600 °C), and S-3 (500 °C), but these samples differ not only in growth temperature but also in period count and layer thickness (e.g., S-3 has ~6–7.3 nm layers while S-16 has ~1.3–2.2 nm layers). The observed increase in ξ and decrease in σ at lower temperature could be influenced by thickness-dependent reconstruction artifacts or by the different number of interfaces averaged. The manuscript should acknowledge this confounding and, if possible, include samples grown at the same temperature with different periods to isolate the temperature effect.","section":"Fig. 2(d) and Table 1"},{"comment":"The conclusion that 'observed atomic-level roughening at the interface must be accounted for' is not directly supported by the optical modeling, because the Mint_σ model parameterizes interfacial width (a graded composition profile) rather than the roughness parameters σ and ξ measured by APT. No optical model in the paper actually incorporates the APT-derived roughness or correlation length. The conclusion should be limited to interfacial broadening (width) until a model with explicit roughness parameters is tested.","section":"SE studies, Fig. 5 and Table 2"}],"minor_comments":[{"comment":"The text refers to 'interface number 24' while the inset in the caption mentions 'interface 9'; this inconsistency should be corrected.","section":"Fig. 2(a) caption and text"},{"comment":"The uncertainty intervals such as ±13% and ±11.5% should be explicitly defined as relative standard deviations of the mean across interfaces/datasets, not confidence intervals for the measured value.","section":"Abstract and Fig. 2(d)"},{"comment":"The relationship between the sigmoid parameter ℒ and the reported interface width is only stated in the SI as dint^(i) = 4ℒ; this definition should appear in the main text where Eq. (2) is introduced.","section":"Eq. (2) and SI Section 3.2"},{"comment":"Table 1 is difficult to parse: the columns for XTEM, HRXRD, and SE thicknesses are not clearly aligned, and the 'surface rms. roughness' values are interleaved with the thickness data; reformatting is needed.","section":"Table 1"},{"comment":"The statement that 'there is no Ge segregation' is stronger than the data support, since the rising and falling interface widths are compared at the resolution limit of the measurement; consider softening this claim.","section":"Section 'Interfacial width' (discussion of Ge segregation)"},{"comment":"The fitted Hurst parameter α is never reported, even though it is a free parameter in H_fit(τ) and influences the relationship between σ and ξ; reporting α for at least the representative interfaces would aid reproducibility.","section":"Fig. 2(a) and Eq. (1)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal, but the headline quantitative claims (σ, ξ, and their temperature dependence) rest on measurement fidelity that is not demonstrated to sub-0.3 nm precision. The issues are addressable in a revision through a noise-floor analysis, an independent validation of roughness, and a disentangling of temperature from layer-thickness effects. The SE part is a useful contribution but the conclusion about roughness should be narrowed to interfacial broadening, or the model should actually import the measured roughness parameters. I recommend major_revision rather than rejection because the central methodology is plausible and the required additional analyses are feasible within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's central claim is that APT-derived iso-concentration surfaces give the first direct measurements of lateral correlation lengths for buried Si/SiGe interfaces. I think that claim holds. The height-height correlation analysis applied to APT surfaces is genuinely new relative to the QCL literature, where ξ is a fitting parameter, and the cross-checks with EELS, XTEM, HRXRD, and AFM are well done. The qualitative conclusion that SE models need interfacial broadening is solid.\n\nThe soft spot is the measurement floor. The reported RMS roughness values (0.2–0.3 nm) are the same order of magnitude as the expected voxel-statistics noise: with 1 nm³ voxels, ~25 at.% Ge, and typical APT detection efficiency, each voxel contains tens of Ge atoms, and the counting noise (a few at.%) translates through the ~1 nm interface gradient into height fluctuations of order 0.1–0.2 nm. Calibrating the APT reconstruction so layer thicknesses match XTEM within 5% constrains average vertical spacings, not the lateral height field that defines σ and ξ. The quoted uncertainties are standard deviations across interfaces and datasets; they do not include this floor. The SI shows the z-density correction was assessed for interface widths, not for σ or ξ. This does not sink the method, but the paper needs to demonstrate the floor explicitly, for example by pushing a known interface height field through the voxelization and reconstruction pipeline, or by perturbing reconstruction parameters and showing σ and ξ are stable.\n\nTwo smaller concerns. The temperature trend rests on one sample per temperature, with different layer thicknesses and period counts, so the 500–650 °C comparison is not clean. And the SE modeling shows graded-interface models fit better, but ellipsometry is sensitive to composition broadening, not to the distinction between roughness and intermixing; the step from 'interfacial broadening' to 'atomic-level roughening' is an interpretation, not what SE directly measures.\n\nThe citation pattern looks appropriate. This is a useful paper for the Si/SiGe, QCL, and gate-all-around communities. It deserves a serious referee and a major-revision decision, not a desk reject. If the authors quantify the voxel-statistics floor and show reconstruction insensitivity, the quantitative claims will be much more trustworthy.","headline":"First direct 3D APT measurement of buried Si/SiGe interface correlation lengths is likely novel and worth engaging, but the reported sub-0.3 nm roughness sits close to the method's voxel-statistics floor and the temperature trend is confounded.","tokens_in":20679,"tokens_out":5287,"would_cite":true,"duration_ms":47627,"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":"Atom-probe maps of buried Si/SiGe interfaces show growth temperature sets their roughness and correlation length, and that optical models must include this broadening.","keywords":["silicon-germanium superlattices","atom probe tomography","interfacial roughness","interface correlation length","interfacial width","spectroscopic ellipsometry","buried interfaces","height-height correlation function"],"falsifier":"A reconstruction simulation or a cross-check with a reconstruction-independent probe would settle the claim: start from a model superlattice with prescribed $\\sigma$ and $\\xi$, simulate atom probe evaporation with different evaporation fields for Si and SiGe, reconstruct the data, apply the iso-concentration-surface analysis, and see whether the input values are recovered; alternatively, measure identical superlattices with grazing-incidence small-angle X-ray scattering or scanning tunneling microscopy and compare the extracted roughness and correlation lengths with the atom probe values.","tokens_in":19555,"feed_emoji":"🔬","tokens_out":7089,"duration_ms":65171,"temperature":0.7,"pith_summary":"The paper sets out to measure, directly and in three dimensions, the roughness of interfaces buried inside sub-10 nm Si/SiGe superlattices, a quantity that shapes electronic and optical behavior but has usually been treated as a fitting parameter. Using atom probe tomography, the authors construct iso-concentration surfaces at each buried interface and analyze their height-height correlations. They report that the root-mean-square roughness is about 0.3 nm for growth at 650 °C and about 0.2 nm at 500 °C, with horizontal correlation lengths of about 8 nm and 10 nm respectively. They also find about 30% narrower interfacial widths at the lower growth temperature and attribute the difference to thermally activated atomic exchange at the growth surface. Finally, by comparing optical models with increasing interfacial complexity, they show that the atomic-level roughening must be included to reproduce spectroscopic ellipsometry data; models that omit the interface systematically overestimate layer thicknesses.","feed_headline":"Cooler growth sharpens buried Si/SiGe interfaces","feed_subtitle":"Atom-by-atom maps pin the roughness at 0.2–0.3 nm and the horizontal correlation length at 8–10 nm.","key_machinery":"The central object is the iso-concentration surface: a three-dimensional surface drawn at 50% of the mean Ge concentration within each SiGe layer, taken to represent each buried interface. Its height map $z(x,y)$ feeds the height-height correlation function $H(\\tau)=\\langle |z(x,y)-z(x',y')|^2\\rangle$, whose fit to $2\\sigma^2[1-\\exp(-(\\tau/\\xi)^{2\\alpha})]$ yields the RMS roughness $\\sigma$ and the horizontal correlation length $\\xi$. A second device is the sigmoid fit $c(x)=c_0+d_0/[1+\\exp(-(x_0\\pm x)/\\mathcal{L})]$ to one-dimensional concentration profiles, whose length parameter $\\mathcal{L}$ defines the interfacial width. The optical counterpart is a graded interfacial alloy model in which each interface is a graded layer with Si content following the same sigmoid function, discretized into 21 sublayers; this is the mechanism that connects the measured atomic-level broadening to the optical response.","core_discovery":"The central claim is that atom probe tomography, applied to Si/SiGe superlattices with individual layers 1.5–7.5 nm thick, can map the vertical height of every buried interface as a three-dimensional iso-concentration surface, and that from these surfaces the RMS roughness $\\sigma$ and horizontal correlation length $\\xi$ can be extracted quantitatively. For interfaces defined at 50% of the mean Ge content, the height-height correlation function $H(\\tau)$ follows the phenomenological form $H_{\\mathrm{fit}}(\\tau)=2\\sigma^2[1-\\exp(-(\\tau/\\xi)^{2\\alpha})]$, yielding mean $\\sigma = 0.3$ nm ($\\pm11.5\\%$) and $\\xi = 8.1$ nm ($\\pm5.8\\%$) for growth at 650 °C, and $\\sigma = 0.2$ nm ($\\pm13\\%$) and $\\xi = 10.1$ nm ($\\pm6.2\\%$) for growth at 500 °C. The authors further claim that the one-dimensional Ge concentration profiles across the same interfaces give average interfacial widths of about 1.0 nm at 650 °C and 0.7 nm at 500 °C, with no asymmetry between Si→SiGe and SiGe→Si transitions, indicating suppressed Ge segregation. On the optical side, they show that a graded-alloy interfacial model reproduces the ellipsometric spectra better than models that omit or homogenize the interfaces, and that omitting interfacial broadening systematically overestimates layer thicknesses.","pith_inferences":["If the reported roughness and correlation lengths are physical, interface-roughness scattering models for Si/SiGe quantum cascade and nanosheet devices can be evaluated with measured inputs rather than fitted ones; a direct test would be to compute intersubband scattering rates from these values and compare them with measured linewidths.","The same iso-concentration-surface methodology should transfer to other epitaxial pairs, such as Si/SiC, GeSn/SiGeSn, or III–V superlattices, but the reconstruction calibration would need to be revalidated because evaporation-field contrasts differ between materials.","The apparent link between smaller roughness, larger correlation length, and lower growth temperature suggests a growth-front smoothing mechanism: at lower temperature the surface exchange length increases, so the roughness undulates more gently; this could be tested by growing a temperature staircase within a single superlattice and mapping the interfaces one by one.","Because spectroscopic ellipsometry consistently overestimates interface width relative to atom probe and electron energy loss data by less than 1 nm, a combined workflow could serve as a calibration transfer for non-destructive inline metrology of buried-interface sharpness."],"forward_implications":["Interface roughness and correlation length can be obtained directly from atom probe data for buried interfaces, replacing the practice of treating the correlation length as a free fitting parameter in scattering and transport models.","Growth temperature controls interface morphology: lowering the growth temperature from 650 °C to 500 °C reduces the RMS roughness by about 30% and the interfacial width by about 30%, while increasing the horizontal correlation length by about 24%.","The absence of a difference between Si→SiGe and SiGe→Si interface widths indicates that Ge segregation is suppressed in these reduced-pressure chemical vapor deposition superlattices, so interface broadening is set by kinetically limited surface atomic exchange rather than by segregation.","Optical modeling of Si/SiGe heterostructures needs an interface layer with graded composition; the graded sigmoid model gives the lowest layer-thickness errors, while effective-medium approximations of the interface are inadequate.","Because the method works for layers as thin as 1.3 nm of Si and 2.2 nm of SiGe, it can be applied to gate-all-around nanosheet superlattices and other sub-10 nm multilayer devices where interface scattering limits performance."],"supporting_citations":[{"why":"Supplies the justification for analyzing only atoms within a 30 nm central cylinder, namely that evaporation is most uniform from the center of an atom probe tip.","marker":"[13]"},{"why":"Provides the height-height correlation function formalism and the fitting function $H_{\\mathrm{fit}}(\\tau)=2\\sigma^2[1-\\exp(-(\\tau/\\xi)^{2\\alpha})]$ from which $\\sigma$ and $\\xi$ are extracted.","marker":"[14]"},{"why":"Corroborates the same height-height correlation analysis used to extract roughness and correlation length from height maps.","marker":"[15]"},{"why":"Supplies the sigmoid function applied to one-dimensional Ge concentration profiles to define interfacial width from the fit parameter $\\mathcal{L}$.","marker":"[17]"},{"why":"Bases the Bruggeman effective-medium approximation used in the EMA interface optical models against which the graded model is compared.","marker":"[24]"},{"why":"Provides the optical constants of pseudomorphic SiGe layers used as inputs to the spectroscopic ellipsometry models.","marker":"[29]"}],"fun_headline_variants":["3D atom maps reveal Si/SiGe interface roughness down to 0.2 nm","Cooler growth sharpens Si/SiGe interfaces, atom probe confirms","Atom probe maps 3D roughness and correlation length in superlattices","Lower growth temperature improves Si/SiGe interface abruptness","Atomic roughness in Si/SiGe layers alters optical response, study finds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the atom probe reconstruction, after iterative calibration so that its layer thicknesses match transmission electron microscopy within 5%, preserves true vertical heights and horizontal separations at the buried interfaces to better than about 0.2 nm; if reconstruction artifacts shift heights at this scale, the reported roughness and correlation lengths would be measurement artifacts rather than physical interface properties.","fun_headline_variants_meta":{"raw":{"variants":["3D atom maps reveal Si/SiGe interface roughness down to 0.2 nm","Cooler growth sharpens Si/SiGe interfaces, atom probe confirms","Atom probe maps 3D roughness and correlation length in superlattices","Lower growth temperature improves Si/SiGe interface abruptness","Atomic roughness in Si/SiGe layers alters optical response, study finds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000531,"raw_usage":{"total_tokens":2672,"prompt_tokens":1176,"completion_tokens":1496,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":792,"completion_tokens_details":{"reasoning_tokens":1399}},"tokens_in":792,"tokens_out":1496,"duration_ms":11217,"temperature":1.0,"reasoning_tokens":1399,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:29:20.415233+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A reconstruction simulation or a cross-check with a reconstruction-independent probe would settle the claim: start from a model superlattice with prescribed $\\sigma$ and $\\xi$, simulate atom probe evaporation with different evaporation fields for Si and SiGe, reconstruct the data, apply the iso-concentration-surface analysis, and see whether the input values are recovered; alternatively, measure identical superlattices with grazing-incidence small-angle X-ray scattering or scanning tunneling microscopy and compare the extracted roughness and correlation lengths with the atom probe values.","supporting_citations":[{"cited_title":"Aspnes, A","cited_arxiv_id":null,"evidence_quote":"Supplies the justification for analyzing only atoms within a 30 nm central cylinder, namely that evaporation is most uniform from the center of an atom probe tip."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the height-height correlation function formalism and the fitting function $H_{\\mathrm{fit}}(\\tau)=2\\sigma^2[1-\\exp(-(\\tau/\\xi)^{2\\alpha})]$ from which $\\sigma$ and $\\xi$ are extracted."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Corroborates the same height-height correlation analysis used to extract roughness and correlation length from height maps."}],"review_version":1}