{"id":"cf17fd78-ee84-4feb-b37a-b030a8d1705e","arxiv_id":"2505.24468","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Reactive MD simulations of bilayer TaOx/HfO2 ReRAMs show forming begins with electric-field-driven Ta/Hf cation migration and anodic Ta depletion, while Joule heating accelerates filament growth through thermally activated oxygen-vacancy generation near the cathode.","lead":"This paper uses atomic-scale computer simulations to watch what happens inside a TaOx/HfO2 memory device when voltage and heat are applied during the first 'forming' step. It finds that electric fields mostly move metal atoms near one electrode and create a shielding oxide layer, while heat mainly creates missing-oxygen spots that make the conductive filament grow faster.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Frozen Ta2O5 electrodes preclude anodic oxygen exchange with TiN; since the authors' ref. 11 documents ion exchange across TiN/TaOx interfaces, the observed minimal O response may be a boundary-condition artifact.","rationale":"The reader identifies the frozen-electrode/no-oxygen-exchange boundary condition as the weakest assumption; my reading reaches the same conclusion. The paper's strongest claim attributes the observed anodic Ta-depletion and minimal oxygen response to intrinsic electrostatics, but the simulation geometry removes the most plausible sink for oxygen at the anode. The authors explicitly cite Ma et al. (ref. 11) for two-stage forming in TaOx-based devices, and that same work demonstrates interfacial ion exchange across TiN/TaOx during electroformation, which the frozen Ta2O5 electrode cannot reproduce. If TiN absorbs oxygen, the anodic 'oxygen-rich' region would be smaller, and the 'screening layer' would be partly a modeling artifact. This concern is load-bearing because the decoupling of field-driven cation motion from thermally driven vacancy generation relies on oxygen being nearly immobile under 1.2 V. Secondary weaknesses—only two voltages for the threshold claim, no replicate runs, and the inferred rather than directly counted vacancy generation—are real but do not independently undermine the central mechanism; they would merely add noise to quantitative claims. The authors' own timescale limitation is acknowledged and does not change the conditional verdict. The suggested concrete test directly targets the assumption by allowing a dynamic TiN electrode and checking whether O uptake changes the anodic profiles. If the test shows negligible O exchange, the central claim is preserved; if not, the proposed mechanism would need to be reformulated. Therefore, the reader's conditional verdict remains appropriate, and no verdict change is needed.","tokens_in":18003,"tokens_out":3530,"duration_ms":43773,"concrete_test":"Run the same 1.2 V, 300 K protocol with a thin (0.5–1 nm) dynamic TiN electrode between the dielectric and a frozen outer layer, using a force field that includes Ti–N–O interactions (or a TaOx/TiN interface parametrization), and compare O concentration profiles and the anodic Ta/O composition against the frozen-Ta2O5 case. If O migrates into the TiN layer and the anodic O-rich layer intensity drops, the frozen-electrode condition is the cause and the central claim needs qualifiers; if O uptake is negligible within 500 ps, the assumption is adequate.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central mechanistic claim—that Ta cations migrate away from the anode while O anions barely move, creating a Ta-depleted/O-rich screening layer—depends critically on the frozen-electrode, no-oxygen-exchange boundary condition. In 'Device Layer Structure and Modeling Approach', the authors freeze both Ta2O5 electrode regions, justified by the assertion that TiN and conductive TaOx have low oxygen affinity and do not exchange O with the dielectric. Yet ref. 11 (Ma et al., ACS Appl. Mater. Interfaces 2020) is titled 'Exchange of Ions across the TiN/TaOx Interface during Electroformation' and documents TiN taking up oxygen during forming; this paper cites it for other purposes but does not reconcile it with the frozen boundary. In the simulation, oxygen anions near the anode cannot leave the functional layer, so the minimal O displacement and the 'oxygen-rich' anodic region could be enforced by the boundary condition rather than by intrinsic electrostatics. If real TiN electrodes act as an oxygen sink, the net anodic response would include O loss, not just Ta depletion, and the inferred field-shielding picture would need revision. Because the whole VCM/TCM decoupling argument rests on O ions being essentially immobile, this is the most load-bearing assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents reactive molecular dynamics simulations of a Ta2O5/HfO2 bilayer ReRAM stack under applied bias, using an extended CTIP+EChemDID method implemented in LAMMPS. The model is built from STEM/EDS data of real devices, with amorphous Ta2O5 and HfO2 functional layers between frozen Ta2O5 electrodes. Under 1.2 V at 300 K for 500 ps, the authors observe downward drift of Ta and Hf cations, with small upward displacement of O, leading to a Ta-depleted O-rich anodic layer; conductive metal atoms cluster near the cathode and nucleate a filament. At 900 and 1300 K, filament size increases, attributed to thermally activated generation and agglomeration of oxygen vacancies near the filament edge rather than enhanced vertical field-driven migration. At 0.6 V no filament nucleates even at 1300 K, leading to the claim of a threshold voltage for vacancy clustering. The paper concludes that forming in these bilayer devices decouples electric-field-driven cation redistribution from thermally driven vacancy generation, reconciling VCM and TCM pictures.","tokens_in":18219,"tokens_out":6227,"duration_ms":72816,"significance":"The central qualitative mechanism is novel and plausible: it provides a concrete atomistic rationale for why Ta cations are mobile during forming despite their low diffusion constant, and it offers a way to reconcile conflicting VCM/TCM observations in TaOx/HfO2 devices. The paper's strengths include the experimentally informed stack geometry, the use of a literature-optimized variable-charge potential with validation against cohesive energies and radial distribution functions, the explicit EChemDID implementation and promised code release, and the falsifiable predictions (threshold voltage, temperature-dependent forming). If the results are confirmed by ensemble-replicated and boundary-condition-robust simulations, they would be a significant contribution to the atomistic understanding of electroforming in bilayer ReRAMs.","major_comments":[{"comment":"The frozen-electrode, no-oxygen-exchange boundary condition is load-bearing for the paper's central observation that oxygen ions respond only minimally under bias. The text asserts (p. 8) that TiN and conductive TaOx have low oxygen affinities and do not exchange O, but the cited ref. 11 is titled 'Exchange of Ions across the TiN/TaOx Interface during Electroformation' and documents TiN taking up oxygen during forming. Because both Ta2O5 electrode regions are frozen, oxygen anions in the model cannot leave the functional dielectric; the anodic O-rich zone and the minimal O displacement could therefore be enforced by the boundary condition rather than by intrinsic electrostatics. The authors should either relax this assumption (e.g., allow O exchange at the anode, or use an oxygen reservoir) or provide a quantitative justification, such as O-affinity data or a control simulation, before the VCM/TCM decoupling claim can be considered robust.","section":"Results, 'Device Layer Structure and Modeling Approach' (also Methodology, 'External Electrochemical Potential…"},{"comment":"All composition profiles, displacement profiles, and charge distributions are derived from a single 500 ps MD trajectory per condition. Figure 7(d) reports error bars from temporal averaging over the final 100 ps of one run, not run-to-run variability. Molecular dynamics is stochastic, and the displacement hierarchy Ta > Hf > O, the anodic depletion profile, and the threshold behavior are quantitative claims that require at least several independent initializations (e.g., different random seeds or melt-quench replicas) with reported means and standard deviations. Without such replicates, the central quantitative conclusions are not statistically supported.","section":"Results, 'Atomistic Response to 1.2 V...' and 'Atomistic response to 0.6 V...' (Figs. 4-10)"},{"comment":"The claimed threshold voltage for filament nucleation is inferred from exactly two bias conditions: 0.6 V (no filament) and 1.2 V (filament), each from one trajectory. A threshold cannot be located from two points, and no error bar is attached to the 'no nucleation' outcome. A proper threshold characterization would require scanning several voltages (for example, 0.7-1.1 V) at multiple temperatures and reporting nucleation probability or average filament size, ideally with ensemble statistics. This is load-bearing because the voltage-gating of vacancy clustering is one of the paper's main conclusions.","section":"Results, 'Atomistic response to 0.6 V...' and 'Discussion and Conclusions'"},{"comment":"The mechanism attributing the cation-dominated displacement to larger per-atom charges on metal ions than on oxygen relies on the CTIP charge equilibration model with imposed charge bounds. The reported mean charges are not validated against independent electronic-structure data, and the charge bounds could artificially reduce the oxygen partial charges and thus the electrostatic force on O. The authors should compare the CTIP charges with DFT-derived Bader charges or perform a sensitivity test varying the charge bounds, since this explanation is central to the proposed field-driven cation migration.","section":"Results, 'Atomistic Response to 1.2 V...' (Figs. 6(c-d) and p. 15)"}],"minor_comments":[{"comment":"The caption says '(a) and (b) show the atomic snapshots' and then '(b) and (d) plot the profiles'; the correct references should be (a)/(c) for snapshots and (b)/(d) for profiles.","section":"Figure 9 caption"},{"comment":"The effective diffusivity k is never given a numerical value; for reproducibility, report k (and its units) or state how it was set relative to the chosen iteration count.","section":"Methodology, Eq. (4)"},{"comment":"The phrase 'above the 50 Åz position' should read 'above z = 50 Å' or 'above the 50 Å z-position'.","section":"Results, p. 13"},{"comment":"The sentence 'can aide future all-atom reactive MD simulations' should be 'can aid future all-atom reactive MD simulations'.","section":"Discussion and Conclusions, p. 24"},{"comment":"The GitHub repository is mentioned but no URL is given; include a link or a DOI in the final version for reproducibility.","section":"Supporting Information section"},{"comment":"The y-axis label 'Vertical displacement' should specify the sign convention (positive upward vs downward) so that the 'downward' direction claims can be interpreted unambiguously.","section":"Figure 5(a)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a computational materials science journal and the implementation appears careful, but the central mechanistic claims need additional support through ensemble-replicated simulations, a revisited boundary-condition treatment, and a more thorough threshold-voltage scan. I would be willing to review a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look. This is the first atomistic MD study of forming in a bilayer TaOx/HfO2 ReRAM stack, and it lays out a concrete mechanism: under a 1.2 V bias, Ta cations migrate away from the anode, creating a Ta-depleted, oxygen-rich screening layer, while oxygen vacancies cluster near the cathode and nucleate a filament. Joule heating then accelerates growth by generating vacancies near the filament edge, not by pushing ions along the field. The displacement hierarchy Ta > Hf >> O and the threshold between 0.6 and 1.2 V are new and experimentally testable.\n\nThe paper is careful about its method. The CTIP + EChemDID implementation is a genuine extension, the amorphous structures are validated against known density and RDFs, and the authors openly state the timescale limitation. They also bring their own experimental forming-voltage-versus-temperature data in the SI, which supports the thermal part of the story. The separation of field-driven cation rearrangement from thermally driven vacancy generation is a useful synthesis, and it gives device engineers a rationale for compliance-current effects.\n\nThe soft spot that matters most is the frozen-electrode boundary condition. The authors freeze the Ta2O5 electrode regions and assume TiN and conductive TaOx have low oxygen affinity, so they do not exchange oxygen with the functional layers. But ref. 11, which they cite for other purposes, documents TiN taking up oxygen during electroforming of TaOx devices. With no TiN in the model and no pathway for oxygen to leave the stack, the small oxygen displacement and the anodic O-rich layer could be partly enforced by the boundary condition rather than by intrinsic electrostatics. If real electrodes act as an oxygen sink, the anodic response would include O loss, not just Ta depletion, and the field-shielding argument would need revision. This does not destroy the central claim, but it is the load-bearing assumption.\n\nOther issues are smaller. The composition and displacement profiles come from single 500 ps trajectories with no error bars for most quantities. The threshold voltage is inferred from only two bias points. The vacancy-generation mechanism is inferred from enhanced lateral displacement rather than directly observed. The code is promised but not yet released, so independent verification is limited. These are common MD limitations, and the authors acknowledge the timescale gap.\n\nOverall, this is a serious mechanistic hypothesis, not an established result. It deserves careful peer review. The referee should press for a treatment of electrode oxygen exchange, replicate runs, and code or data release. I would not cite it as fact yet, but I would bring it to a group discussion.","headline":"First atomistic MD of forming in bilayer TaOx/HfO2 gives a plausible VCM/TCM reconciliation, but the frozen-electrode assumption may predetermine the oxygen immobility it claims to explain.","tokens_in":18796,"tokens_out":4234,"would_cite":false,"duration_ms":46813,"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":"In bilayer TaOx/HfO2 ReRAM, forming begins with voltage-driven cation migration, and Joule heating accelerates it by creating oxygen vacancies near the filament edge.","keywords":["ReRAM","electroforming","TaOx/HfO2 bilayer","reactive molecular dynamics","charge transfer ionic potential","EChemDID","oxygen vacancy filament","Joule heating"],"falsifier":"Track oxygen and metal atom positions during early-stage forming in a real TiN/TaOx/HfO2 device held below 1.2 V—using elemental mapping in a transmission electron microscope or isotopic oxygen tracers. If oxygen ions migrate substantially toward the anode before tantalum depletion appears, or if a filament nucleates at 0.6 V while the device is held at 1300 K, the paper's cation-first, no-oxygen-exchange picture would be contradicted.","tokens_in":17790,"feed_emoji":"⚡","tokens_out":7968,"duration_ms":100269,"temperature":0.7,"pith_summary":"The paper tries to establish a division of labor in the forming of bilayer TaOx/HfO2 ReRAM devices: applied voltage rearranges metal cations, while heat supplies the defects that grow the filament. Reactive molecular dynamics at 1.2 V show tantalum ions moving away from the anode most strongly, hafnium ions next, and oxygen ions barely at all; the result is a tantalum-depleted, oxygen-rich screening layer at the anode and a cluster of oxygen vacancies at the cathode where the filament nucleates. Raising the temperature to 900–1300 K accelerates filament growth not by speeding vertical ion drift but by thermally generating oxygen vacancies near the filament edge, where they are stabilized by the filament acting as a virtual cathode. A 0.6 V simulation nucleates no filament even at 1300 K, implying a voltage threshold for vacancy clustering. If correct, the picture reconciles the valence-change and thermochemical descriptions for bilayer devices and explains why hotter devices form at lower voltages.","feed_headline":"Metal ions, not oxygen, start the ReRAM filament","feed_subtitle":"Voltage moves tantalum and hafnium out of the anode region; heat then builds the filament by generating oxygen vacancies.","key_machinery":"The argument runs on a reactive molecular-dynamics scheme in which the charge transfer ionic potential (CTIP)—a hybrid potential treating metallic and ionic bonding with dynamic atomic charges—is coupled to the electrochemical dynamics with implicit degrees of freedom (EChemDID) method, which propagates the electrode potential through conductive clusters in the dielectric. Under applied bias, electrode electronegativities are shifted, atomic charges are re-equilibrated, and a diffusion equation carries the local potential through connected metallic atoms so the growing filament behaves as a virtual cathode. Conductive metal atoms are identified by oxygen coordination of 5 or below, and the filament is defined by cluster analysis with a 3.9 Å cutoff. This machinery is what lets the paper separate field-driven vertical displacement from thermally activated lateral motion and track filament size at 300, 900, and 1300 K.","core_discovery":"The central claim is that electroforming in a pristine bilayer TaOx/HfO2 stack under 1.2 V is initiated by electric-field-driven migration of metal cations—Ta more than Hf—away from the positive electrode, not by oxygen anion migration. This creates a Ta-depleted, oxygen-rich region at the anode that partially screens the bulk dielectric from the field, while oxygen vacancies accumulate at the cathode and cluster into the seed of the conductive filament. Joule heating then acts by increasing the generation rate of oxygen-vacancy defects in the layer near the filament tip; these defects aggregate onto the filament because the filament, as a virtual cathode, stabilizes them. The paper further claims a threshold voltage between 0.6 and 1.2 V below which no filament nucleates even at 1300 K, so temperature accelerates forming only once voltage has enabled clustering. Together these claims assign the electric field the role of cation rearrangement and nucleation, and temperature the role of vacancy generation and growth.","pith_inferences":["The cation-first response implies a design lever the paper does not explore: forming voltage could be tuned by doping or interface modifications that change Ta/Hf mobility or cation charge, rather than by engineering oxygen transport.","Because the central picture depends on oxygen-inert electrodes, a natural test is to rerun the same stack with electrodes that can absorb oxygen; if oxygen migration becomes substantial, the no-exchange boundary condition is responsible for part of the result.","If the threshold behavior transfers to single-layer TaOx and HfO2 devices, it would unify their apparently conflicting VCM/TCM observations; the authors gesture at this, but the extrapolation is ours.","A compact circuit model could encode a temperature-activated vacancy-generation rate localized at the filament tip, which would make electroforming-voltage predictions depend on thermal resistance and ambient temperature rather than on vacancy drift alone."],"forward_implications":["Limiting Joule heating—through a compliance current or a series resistor—should yield a smaller filament and lower post-forming conductance, because temperature's main role is defect generation rather than field-driven drift.","The anodic Ta-depleted, oxygen-rich layer shields the bulk from the field, which explains why forming in these devices requires a high applied voltage (experimentally above 4 V) rather than continuous anodic electroreduction.","Raising the device's ambient temperature lowers the required forming voltage; the authors report observing a linear correlation in similar bilayer devices.","Below the nucleation threshold, extra heat alone cannot start the filament; the applied voltage must first overcome vacancy–vacancy repulsion to make clusters stick.","The combined field-plus-thermal mechanism reconciles valence-change and thermochemical behaviors observed in TaOx and HfO2 ReRAMs by giving each model a different stage of forming."],"supporting_citations":[{"why":"Supplies the experimental bilayer TaOx/HfO2 device stack, STEM/EDS layer structure, and the high forming-voltage reference that the simulation model is built to explain.","marker":"[15]"},{"why":"Provides the optimized variable-charge interatomic potential parameters for the Hf/Nb/Ta/Ti/Zr/O system used in the CTIP implementation.","marker":"[46]"},{"why":"Introduces the modified charge-transfer embedded-atom method formalism that CTIP extends to metal/metal-oxide redox systems.","marker":"[37]"},{"why":"Defines the charge transfer ionic–embedded atom method potential used for the non-electrostatic interactions.","marker":"[38]"},{"why":"Introduces the voltage equilibration method that propagates the applied bias through metallic clusters in reactive molecular dynamics.","marker":"[39]"},{"why":"Supplies the oxygen-coordination criterion and cluster-analysis cutoff used to identify conductive atoms and the filament, and the HfO2 full-cycle MD methodology being adapted.","marker":"[10]"},{"why":"Reports experimental TaOx forming via thermal-gradient-induced cation accumulation, the single-layer counterpart that motivates the cation-motion picture.","marker":"[9]"},{"why":"Documents TiN/TaOx ion exchange and two-stage thermal/compositional runaway during electroforming, used for high-temperature values and as the experimental boundary-condition context.","marker":"[11]"},{"why":"Shows that electric field alone leaves defect-generation barriers high in HfO2 while injected electrons lower them, supporting the thermally and electron-assisted vacancy generation near the filament.","marker":"[49]"}],"fun_headline_variants":["Metal cations, not oxygen, start ReRAM filaments","Electroforming begins with tantalum drift, not oxygen","Voltage moves Ta and Hf; heat grows the ReRAM filament","Cation migration nucleates filament, heat drives its growth","Ta ions seed the filament; oxygen vacancies build it by heat"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the frozen electrodes, which cannot absorb or release oxygen, faithfully represent real TiN and conductive TaOx contacts during the early stage of forming; if those contacts exchange oxygen with the dielectric under a 1.2 V bias, the simulated suppression of oxygen motion could be an artifact.","fun_headline_variants_meta":{"raw":{"variants":["Metal cations, not oxygen, start ReRAM filaments","Electroforming begins with tantalum drift, not oxygen","Voltage moves Ta and Hf; heat grows the ReRAM filament","Cation migration nucleates filament, heat drives its growth","Ta ions seed the filament; oxygen vacancies build it by heat"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1472,"prompt_tokens":1007,"completion_tokens":465,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":623,"completion_tokens_details":{"reasoning_tokens":381}},"tokens_in":623,"tokens_out":465,"duration_ms":6291,"temperature":1.0,"reasoning_tokens":381,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:22:23.949303+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Track oxygen and metal atom positions during early-stage forming in a real TiN/TaOx/HfO2 device held below 1.2 V—using elemental mapping in a transmission electron microscope or isotopic oxygen tracers. If oxygen ions migrate substantially toward the anode before tantalum depletion appears, or if a filament nucleates at 0.6 V while the device is held at 1300 K, the paper's cation-first, no-oxygen-exchange picture would be contradicted.","supporting_citations":[{"cited_title":"Advanced electronic ma- terials 2022, 8, 2200448","cited_arxiv_id":null,"evidence_quote":"Supplies the experimental bilayer TaOx/HfO2 device stack, STEM/EDS layer structure, and the high forming-voltage reference that the simulation model is built to explain."},{"cited_title":"Developing a variable charge potential for Hf/Nb/Ta/Ti/Zr/O system via machine learning global optimization","cited_arxiv_id":null,"evidence_quote":"Provides the optimized variable-charge interatomic potential parameters for the Hf/Nb/Ta/Ti/Zr/O system used in the CTIP implementation."},{"cited_title":"Modified charge transfer–embedded atom method potential for metal/metal oxide systems","cited_arxiv_id":null,"evidence_quote":"Introduces the modified charge-transfer embedded-atom method formalism that CTIP extends to metal/metal-oxide redox systems."},{"cited_title":"W.; Wadley, H","cited_arxiv_id":null,"evidence_quote":"Defines the charge transfer ionic–embedded atom method potential used for the non-electrostatic interactions."},{"cited_title":"Voltage equilibration for reactive atomistic simulations of electrochemical processes","cited_arxiv_id":null,"evidence_quote":"Introduces the voltage equilibration method that propagates the applied bias through metallic clusters in reactive molecular dynamics."},{"cited_title":"L.; Islam, M","cited_arxiv_id":null,"evidence_quote":"Supplies the oxygen-coordination criterion and cluster-analysis cutoff used to identify conductive atoms and the filament, and the HfO2 full-cycle MD methodology being adapted."},{"cited_title":"A.; Cullen, D","cited_arxiv_id":null,"evidence_quote":"Reports experimental TaOx forming via thermal-gradient-induced cation accumulation, the single-layer counterpart that motivates the cation-motion picture."},{"cited_title":"A.; Goodwill, J","cited_arxiv_id":null,"evidence_quote":"Documents TiN/TaOx ion exchange and two-stage thermal/compositional runaway during electroforming, used for high-temperature values and as the experimental boundary-condition context."},{"cited_title":"W.; Cottom, J.; Larcher, L.; Shluger, A","cited_arxiv_id":null,"evidence_quote":"Shows that electric field alone leaves defect-generation barriers high in HfO2 while injected electrons lower them, supporting the thermally and electron-assisted vacancy generation near the filament."}],"review_version":1}