{"id":"d9ad90c5-d3a9-4fb6-8d71-452c7351d345","arxiv_id":"2502.09090","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Ferroelectric switching in sputtered Al0.92Sc0.08N/GaN appears to initiate at the GaN interface, based on post-switching STEM images.","lead":"This paper uses atomic-resolution electron microscopy to show that ferroelectric switching in a sputtered AlScN film on GaN starts at the GaN interface, leaving a sharp polarization boundary there and small unswitched regions near the top platinum electrode. The work matters because knowing where and how switching begins in AlScN/GaN stacks is key to designing non-volatile memories and high-power transistors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim of interface-nucleated switching rests on ex situ remanent domain images; the observed planar boundary at GaN could be a pinned final state from top-down or bulk nucleation.","rationale":"The reader's weakest-assumption analysis is accurate and identifies the decisive gap. I agree that the central load-bearing claim is the interface-nucleation pathway, and that ex situ remanent-state images are insufficient to pin it down. The text actually strengthens the concern: the 'partial' switching state is explicitly described as already majority-switched, so it is a late-stage snapshot rather than an image of the first nuclei. The complementary evidence (EELS, EDS, lattice-spacing maps, MOCVD comparison) supports the quality of the structural observations and the existence of the planar head-to-head interface, but not the temporal order of nucleation. The proposed in situ biasing experiment directly tests the missing temporal link; an ex situ single-pulse amplitude series on identical capacitors would be a practical fallback. The paper is transparent about its limitations and explicitly requests theoretical validation, so the reader's CONDITIONAL verdict remains appropriate and does not need to be changed.","tokens_in":18808,"tokens_out":12000,"duration_ms":136166,"concrete_test":"Perform in situ biasing TEM on a freshly prepared cross-section lamella from the same Pt/Al0.92Sc0.08N/n-GaN stack: mount it on an electrical-biasing holder, apply a slowly increasing voltage below and through the coercive threshold, and record serial ABF/HAADF-STEM images of the same field of view. If the first N-polar contrast appears at the GaN interface and grows upward, the central claim is confirmed. If the first contrast appears at the Pt electrode or in the bulk, and the planar GaN-interface wall appears only later, the conclusion should be downgraded from interface-nucleated to interface-pinned final state. Repeat on two or more lamellae to exclude FIB-damage artifacts.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central mechanistic claim (Secs. II B and II C, Fig. 7b) is that M-to-N switching 'is initiated' at the GaN interface, with the first AlScN monolayer inverting first. The support is ex situ ABF/HAADF-STEM imaging of remanent states. A horizontal head-to-head polarization discontinuity at the N-polar AlScN / M-polar GaN interface is observed after switching, together with residual M-polar domains at the Pt electrode. But the head-to-head discontinuity is simply the interface between N-polar AlScN and M-polar GaN; it appears in the final state regardless of where switching nucleated. The remanent images cannot exclude the scenario in which domains nucleate at the Pt electrode (or just under a non-switching electrode-adjacent layer), propagate downward, and are pinned at the GaN interface. The 'partially switched' sample is not an early stage: Sec. II B states that 'the majority of the polar volume is already switched' under sub-saturating fields, so the snapshot is late-stage. The N-at-bottom/M-at-top pattern is consistent with bottom-up growth, but also with a downward front that has already reached the interface and left pinned M material at the top. The authors' own call for theoretical validation (end of Sec. II C) acknowledges the missing link. Without observing the first nuclei, the interface-initiated pathway is an interpretation, not an established mechanism.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports the sputter epitaxy of nearly lattice-matched Al0.92Sc0.08N films on GaN and characterizes their ferroelectric domain structure before and after ex situ switching using HRSTEM, ABF/HAADF-STEM, DPC-STEM, EELS, and EDS. The central claim is that M-to-N ferroelectric polarization inversion is initiated at the Al0.92Sc0.08N/GaN interface, forming an atomically sharp planar head-to-head inversion domain boundary at that interface, while residual M-polar domains remain pinned at the Pt electrode. The authors also propose an atomic model of the interfacial inversion boundary, involving a basal-plane stacking fault and octahedrally coordinated metal cations at the interface, and compare the observed switching pathway with that in MOCVD-grown Al0.85Sc0.15N films.","tokens_in":19065,"tokens_out":3875,"duration_ms":41462,"significance":"If the interfacial-initiation mechanism is correct, it would establish a distinctly different ferroelectric switching pathway in sputtered AlScN compared with MOCVD-grown material, with direct implications for interface engineering of ferroelectric nitride devices. The manuscript provides a valuable atomic-scale dataset of polar domains and domain boundaries in Al0.92Sc0.08N/GaN heterostructures, and the imaging and spectroscopic analyses are careful and well documented. The authors are also transparent about the tentative nature of their structural model and the need for further theoretical and experimental validation. The main weakness is that the load-bearing claim of interface-nucleated switching is inferred from ex situ remanent images, which cannot distinguish nucleation from pinning.","major_comments":[{"comment":"The central claim that M-to-N polarization inversion is 'initiated' at the GaN interface is not supported by the ex situ images presented. The observed planar head-to-head boundary at the GaN interface is the final remanent state, and it would appear regardless of where the domains actually nucleated: if domains nucleate at the Pt electrode or in the bulk and are then pinned at the GaN interface, the same post-switching domain pattern would result. The 'partially switched' sample in Fig. 3e is described as having 'the majority of the polar volume already switched' under a sub-saturating field, so it is a late-stage snapshot, not an early nucleation stage. To substantiate the initiation claim, the authors need time-resolved or in situ observations, or a series of pulse-width/field-amplitude experiments that capture the early stages of switching. As written, the conclusion in the abstract and in Sec. II D that the film 'is evidenced to be initiated at the GaN interface' overstates what the data can establish. A defensible statement would be that the final state is consistent with interface-initiated switching, but not uniquely so.","section":"Sec. II B and II C, Figs. 3 and 7b"},{"comment":"The quantitative lattice-spacing values central to the structural model are reported without uncertainties. The values Δd(0001) = -2.3%, +26%, +11.5%, -4.6%, and -3.8% are each given as single numbers, with no error bars, number of measurements, or statistical analysis. The proposed identification of an extra anion layer α′ at the interface rests on the observed +26% increase in metal-metal monolayer separation, yet the precision of this measurement is not documented. The authors should either provide uncertainties and measurement statistics, or explicitly state that the atomic model is one of several possibilities consistent with the images. Without this, the atomic-scale structural claims are not fully quantitative.","section":"Sec. II C, Fig. 4"},{"comment":"The proposed atomic model of the interfacial inversion domain boundary assumes a chemically sharp AlScN/GaN interface, but the EDS data in Fig. 5a show elevated O and Si signals at the interface. The authors acknowledge that oxygen can play a role in forming inversion domain boundaries in AlN-based films and state that 'it can not be ruled out that it plays a role.' Because the model's local coordination (e.g., the distorted octahedra at the interface) depends on the exact atomic species and occupancy at the interfacial layers, the presence of interfacial O or Si could alter the proposed structure. The manuscript should address whether the observed lattice spacings and contrast are compatible with alternative models involving O or Si incorporation, or explicitly limit the model to the analyzed impurity-free regions.","section":"Sec. II C, Fig. 4g,h and Fig. 5a"},{"comment":"The comparison between the PVD (Al0.92Sc0.08N, 110 nm) and MOCVD (Al0.85Sc0.15N, 230 nm) films is confounded by simultaneous differences in composition, thickness, and growth method. The claim that the switching pathway is fundamentally different for sputtered films is based on this single comparison, so the difference could be due to Sc content, strain state, thickness, or defect density rather than deposition technique per se. While the authors discuss potential explanations, they do not control for these variables. The statement that 'sputtered Al0.92Sc0.08N films feature locally tail-to-tail domain walls, while MOCVD-grown Al0.85Sc0.15N films only feature head-to-head domain walls' should be presented as a property of the specific samples compared, not as a general growth-method distinction.","section":"Sec. II D, Fig. 7"}],"minor_comments":[{"comment":"The text states that 'HRSTEM micrographs of the interface prior to ferroelectric switching are presented in Figure 4a and 4c,' but Figure 4c shows a switched interface, not the as-grown one; the reference should be to Figure 4a and 4b.","section":"Sec. II C, paragraph 2"},{"comment":"The phrase 'is evidenced to be initiated' is too strong given the ex situ nature of the data; recommend 'is consistent with initiation' or 'suggests initiation.'","section":"Sec. II D"},{"comment":"The abstract's statement that 'already the first atomic layer of Al1-xScxN changes its polarization' implies a direct observation of the first monolayer flipping, whereas the evidence shows the first monolayer in the final state has N-polarity; the wording should be softened.","section":"Abstract"},{"comment":"The text describes the fringed layer at the Pt interface as 'extending about 10 nm,' while the figure label reads '>10 nm'; these should be harmonized.","section":"Sec. II B, Fig. 3e"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a high-quality experimental dataset and is within the scope of the journal. The main issue is that the central mechanistic claim is overinterpreted relative to the ex situ evidence; this is fixable by reframing the conclusions and adding missing uncertainty quantification or further experiments. I do not see a fundamental circularity or internal inconsistency; the authors are appropriately cautious in some places but not consistently so. The comparison with MOCVD material is useful but currently confounded by different compositions and thicknesses."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nWhat you should know: this is a careful, well-executed STEM study of ferroelectric domain structures in sputtered Al0.92Sc0.08N/GaN, and the main new observation—a planar head-to-head inversion boundary at the GaN interface after switching, with residual M-polar domains at the top Pt electrode—is real and worth knowing. The interpretation that this means switching initiates at the interface is plausible but not proven, because the evidence is ex situ remanent images.\n\nThe paper does several things well. The sputter epitaxy quality is documented with XRD/RSM and SAED, the polarity assignment is clearly done with ABF/HAADF and Atomap, and the comparison to MOCVD-grown Al0.85Sc0.15N is a genuine contrast: MOCVD films show inclined head-to-head domain walls, while these sputtered films show a flat boundary at the GaN interface and curved tail-to-tail walls. The atomic structure models for the interface are explicitly tentative, and the authors call for theoretical validation; that is honest.\n\nThe soft spot is the central claim. Post-switching images show a final domain pattern. A horizontal head-to-head discontinuity at the AlScN/GaN interface is exactly what you would expect from any switching path, because GaN is M-polar and the switched AlScN is N-polar. The residual M-polar layer at the top electrode and the vertical M-polar domains in the partially switched sample could equally come from top-down nucleation that got pinned near the interface, rather than interface-initiated switching. The 'partially switched' sample is not an early nucleation stage—the majority of the film is already switched, so you are looking at a late-stage snapshot. The paper's own language ('suggests', 'could be inferred') shows this. To nail the mechanism you would need in situ biasing or at least a genuinely early-stage partial-switch series.\n\nMinor issues: the Δd(0001) values (+26%, +11.5%) are reported without uncertainties, which is a problem for quantitative claims about the interface reconstruction. The EELS interpretation is speculative, and the authors admit it. The selection of 'undistorted' interface regions could bias the structural analysis.\n\nOverall: the microscopy is credible, the new domain geometry is a solid contribution, and the mechanistic narrative should be softened until direct evidence appears. A serious referee should engage with it—there is enough here for a good paper after revision, but the load-bearing claim needs reframing or new evidence.","headline":"Solid microscopy, plausible mechanism; the interface-nucleated switching claim is an interpretation, not a demonstrated result.","tokens_in":19612,"tokens_out":3477,"would_cite":true,"duration_ms":31217,"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":"Ferroelectric switching in sputtered AlScN/GaN nucleates at the GaN interface, forming an atomically sharp domain boundary.","keywords":["AlScN","ferroelectric switching","GaN heterostructure","sputter epitaxy","inversion domain boundary","scanning transmission electron microscopy","wurtzite ferroelectrics","polarization reversal"],"falsifier":"An in situ biasing experiment inside the TEM, recording domain nucleation during a voltage ramp, would settle the claim directly: N-polar domains should first appear at the AlScN/GaN interface if the paper is right, and at the Pt electrode if it is wrong. A complementary check is a sub-coercive pulse series on pristine capacitors: the first switched regions should appear at the GaN interface with increasing pulse width or amplitude if interface-initiated nucleation dominates.","tokens_in":18623,"feed_emoji":"🔬","tokens_out":11020,"duration_ms":96227,"temperature":0.7,"pith_summary":"This paper uses atomic-resolution electron microscopy to determine where and how ferroelectric polarization reverses in a sputter-epitaxial Al0.92Sc0.08N film grown on GaN. The authors show that, after an applied voltage switches the film from its as-grown metal-polar state to nitrogen-polar, the inversion begins at the AlScN/GaN interface: even the first AlScN monolayer flips, producing a planar head-to-head inversion domain boundary that is atomically sharp. A thin metal-polar layer remains at the top Pt electrode. The paper argues that this interface-initiated switching pathway is specific to sputtered films and differs from the top-initiated, inclined-domain-wall pathway seen in MOCVD-grown Al0.85Sc0.15N, and that knowing the location and atomic structure of these boundaries is key for designing ferroelectric nitride devices.","feed_headline":"AlScN flips at the GaN interface, not the top electrode","feed_subtitle":"Atomic-resolution images show polarization inversion starts at the interface, opposite to MOCVD-grown films.","key_machinery":"The load-bearing object is the atomically sharp planar head-to-head inversion domain boundary that forms at the AlScN/GaN interface after switching. It is identified by atomic-resolution ABF- and HAADF-STEM imaging of the polar stacking sequence, which shows the first AlScN monolayer flipped from M- to N-polarity, and by quantitative monolayer-spacing maps that reveal a local +11.5% to +26% increase in the metal-metal layer distance at the reconstructed boundary. A structural model attributes the boundary to a basal-plane stacking fault that inserts a cubic block and an additional anion layer, giving the interfacial metal atoms a distorted octahedral coordination; STEM-EELS of the Sc-L2,3 and N-K edges and EDS of the interface provide supporting chemical information. This boundary is the evidence that switching initiates at the GaN interface, and its sharp planar geometry is what distinguishes the sputtered-film pathway from the inclined wedge-shaped domain walls of MOCVD films.","core_discovery":"The central claim is that in nearly lattice-matched, sputter-epitaxial Al0.92Sc0.08N/GaN, ferroelectric M-to-N polarization inversion is initiated at the GaN interface. High-resolution STEM images of partially and fully switched capacitors show the first AlScN monolayer at the interface has changed to N-polarity, creating an atomically sharp planar head-to-head inversion domain boundary between the N-polar AlScN and the M-polar GaN; atomic modeling and electron energy-loss spectroscopy suggest the boundary is a reconstructed interface containing a basal-plane stacking fault, a cubic stacking block, and distorted octahedrally coordinated metal atoms. In the same films, small M-polar domains remain pinned at the Pt electrode interface, even after nominally saturating fields. By contrast, MOCVD-grown Al0.85Sc0.15N films switch by vertical growth of wedge-shaped domains that stop tens of nanometers above the GaN interface, so the paper concludes that the switching pathway depends on the deposition route and the associated interfacial strain and defect structure.","pith_inferences":["The ex situ images cannot by themselves separate nucleation from pinning: the planar boundary at GaN could alternatively form if domains nucleate at the Pt electrode and then stall at the interface. An in situ switching experiment that observes where domains first appear would resolve this.","If the GaN interface is indeed the nucleation site, then deliberately introducing interfacial strain or oxygen (already detected at the interface by EDS) might be a route to lower switching voltages or to localize switching in selected device regions - an extension the paper does not explicitly propose.","The EELS fine-structure change across the interface region (from roughly 5-10 nm wide in the as-grown film to about 2 nm after switching) could serve as a local fingerprint of whether a given interface has been switched, offering a way to map switching completeness in devices without electrical access.","The proposed octahedral interfacial layer resembles the local coordination in rock-salt-type AlScN; if correct, the switched interface may be viewed as a thin buried layer of a different structural motif, which would affect how the polarization discontinuity screens and how carriers scatter at the interface."],"forward_implications":["Interface-initiated switching means the GaN/AlScN interface, not the metal electrode, controls where polarization reversal begins, so engineering that interface (strain, roughness, chemistry) should directly tune switching behavior.","The persistent thin M-polar layer at the Pt electrode even after full switching provides a built-in seed for the reverse N-to-M switch, which should influence imprint, retention, and coercive-field asymmetry in devices.","The observed atomically sharp head-to-head boundary is exactly the configuration that theory predicts to strongly enhance interface sheet charge in AlScN/GaN, supporting the feasibility of ferroelectric HEMT concepts based on polarization-controlled two-dimensional electron gases.","Because the switched interface includes a stacking fault and octahedral coordination, the atomic structure after switching is a metastable reconstruction; the paper notes that back-switching may leave a comparably complex interface rather than restoring the as-grown structure.","The comparison with MOCVD films implies that the growth route, through its effect on interfacial coherence and strain, determines the switching mechanism, so reports of switching behavior from different deposition methods should not be assumed to transfer from one film to another."],"supporting_citations":[{"why":"Supplies the theoretical prediction that a sharp head-to-head domain wall at the AlScN/GaN interface would greatly enhance interface sheet charge, motivating why the observed boundary matters.","marker":"[26]"},{"why":"Provides the MOCVD-grown Al0.85Sc0.15N reference sample and the STEM methodology for determining film polarity and domain structure.","marker":"[33]"},{"why":"Reports the electric-field-induced domain structures in MOCVD AlScN that this paper directly compares against, establishing the top-initiated wedge-domain pathway.","marker":"[34]"},{"why":"Provides the sputter epitaxy growth details and the observation of strain-relaxation-driven polarity flips that frame the as-grown M-polar state.","marker":"[35]"},{"why":"Documents persistent M-polar domains near electrode interfaces in AlScN, supporting the interpretation of the pinned residual M-polar layer at the Pt electrode.","marker":"[37]"},{"why":"Shows that local lattice strain can initiate ferroelectric switching in (Al,Sc)N layered structures, evidence for the strain-based mechanism proposed for interface-nucleated switching.","marker":"[53]"}],"fun_headline_variants":["AlScN polarization reversal initiates at GaN, not electrode","Sputtered AlScN switches at GaN interface, defying MOCVD route","First AlScN layer at GaN flips polarity in ferroelectric switch","Interfacial layer, not top contact, triggers AlScN polarization flip"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conclusion that switching is initiated at the GaN interface rests on interpreting the post-switching domain pattern as a memory of where inversion began; the same pattern would result if domains nucleated at the top electrode and were then blocked or slowed at the GaN interface.","fun_headline_variants_meta":{"raw":{"variants":["AlScN polarization reversal initiates at GaN, not electrode","Sputtered AlScN switches at GaN interface, defying MOCVD route","First AlScN layer at GaN flips polarity in ferroelectric switch","Interfacial layer, not top contact, triggers AlScN polarization flip"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000289,"raw_usage":{"total_tokens":1731,"prompt_tokens":1022,"completion_tokens":709,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":638,"completion_tokens_details":{"reasoning_tokens":625}},"tokens_in":638,"tokens_out":709,"duration_ms":6932,"temperature":1.0,"reasoning_tokens":625,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T22:39:28.634192+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An in situ biasing experiment inside the TEM, recording domain nucleation during a voltage ramp, would settle the claim directly: N-polar domains should first appear at the AlScN/GaN interface if the paper is right, and at the Pt electrode if it is wrong. A complementary check is a sub-coercive pulse series on pristine capacitors: the first switched regions should appear at the GaN interface with increasing pulse width or amplitude if interface-initiated nucleation dominates.","supporting_citations":[{"cited_title":"Li , author X","cited_arxiv_id":null,"evidence_quote":"Supplies the theoretical prediction that a sharp head-to-head domain wall at the AlScN/GaN interface would greatly enhance interface sheet charge, motivating why the observed boundary matters."},{"cited_title":"Yassine , author A","cited_arxiv_id":null,"evidence_quote":"Provides the MOCVD-grown Al0.85Sc0.15N reference sample and the STEM methodology for determining film polarity and domain structure."},{"cited_title":"van Deurzen , author T.-S","cited_arxiv_id":null,"evidence_quote":"Reports the electric-field-induced domain structures in MOCVD AlScN that this paper directly compares against, establishing the top-initiated wedge-domain pathway."},{"cited_title":"Casamento , author C","cited_arxiv_id":null,"evidence_quote":"Provides the sputter epitaxy growth details and the observation of strain-relaxation-driven polarity flips that frame the as-grown M-polar state."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents persistent M-polar domains near electrode interfaces in AlScN, supporting the interpretation of the pinned residual M-polar layer at the Pt electrode."},{"cited_title":"Stolyarchuk , author T","cited_arxiv_id":null,"evidence_quote":"Shows that local lattice strain can initiate ferroelectric switching in (Al,Sc)N layered structures, evidence for the strain-based mechanism proposed for interface-nucleated switching."}],"review_version":1}