{"id":"228a6295-d472-4309-89fa-5a807ec90c45","arxiv_id":"2506.16758","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A ferroic HfO2-ZrO2 gate bilayer in an N-polar GaN HEMT simultaneously increases on-current by about 3x and reduces gate leakage by over an order of magnitude, attributed to negative capacitance.","lead":"Researchers added an ultrathin hafnium-zirconium oxide layer to the gate of a gallium nitride transistor and measured both higher on-current and lower leakage, a combination usually at odds. The work points to a route for improving high-frequency and power transistors without the usual performance trade-off.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The NC gate-stack claim is not yet secured: no equal-thickness non-ferroelectric GaN control was fabricated, and the Schottky Cmax reference is truncated by the D>1 leakage window, so the reported capacitance and ON-current boosts may not require negative capacitance.","rationale":"The reader's weakest assumption identifies exactly the two evidential gaps that matter most: the absence of a non-ferroelectric gate-dielectric control on GaN and the leakage-truncated Schottky capacitance reference. My stress-test pass found no additional internal inconsistency that would move the verdict. The device-level comparison is carefully done: matched recess depth by AFM and TEM, comparable sheet resistance from TLM, multiple devices, and RSD-corrected intrinsic overdrive comparisons. Those strengths support the empirical observation of higher ON current and lower leakage in the HZO-gated device. However, the mechanistic conclusion that this improvement is caused by a stabilized negative capacitance in HZO is not established by the presented data. The only capacitance-boost control is on Si MOS capacitors (Fig. 4B), and the only direct structural evidence is mixed FE/AFE phases in HRTEM, which is necessary but not sufficient for stabilized negative capacitance. The paper itself concedes that no MIS-HEMT was experimentally investigated, and the Schottky C-V data are only accepted where D<1. Therefore, the central scientific claim should remain conditional: the empirical device result is plausible, but the negative-capacitance mechanism and the 'beyond Schottky limit' interpretation require the proposed control experiments to be confirmed.","tokens_in":19070,"tokens_out":5269,"duration_ms":53345,"concrete_test":"Fabricate a third HEMT on the same wafer with an equal-thickness (1.8 nm) ALD HfO2 gate dielectric and identical TiN/W metallization and HF clean, processed side-by-side with the HZO and Schottky devices. Measure 2 MHz C-V and ID vs Vov,i for all three. If the HfO2 control shows Cmax below the Schottky reference and lower ID at Vov,i = 1-2 V, while the HZO device exceeds both, the NC attribution is supported. Also re-measure the Schottky C-V with pulsed bias (or a lower-leakage Schottky variant) to test whether its D>1 truncated Cmax is the true Cg,Sch.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that Cg,eq of the NC-HEMT exceeds Cg,Sch because a stabilized negative capacitance in the HZO layer more than offsets the series positive capacitance of the added dielectric. Two linked gaps make this premise insecure. First, the Cg,Sch reference is measured only in the narrow D<1 window (approximately -1.5 to -0.5 V at 2 MHz; Fig. 3D). Schottky leakage makes C-V unreliable at higher forward bias, so the quoted Cmax for the control may not be the true maximum capacitance of the same heterostructure; the reported ~30% boost could be an artifact of comparing a measurable MIS-like C-V to a truncated Schottky C-V. Second, no GaN HEMT with an equal-thickness non-ferroelectric dielectric (e.g., 1.8 nm HfO2) was fabricated; the Fig. 1B caption explicitly states that the MIS-HEMT was only conceptual and not measured. The devices differ not only by HZO but also by the TiN capping layer and by the dilute-HF surface preparation before ALD, and the Schottky control shows hysteresis while the NC device is nearly hysteresis-free (fig. S9). Thus the higher capacitance and ~3x ON current at fixed intrinsic overdrive could in principle result from interface or stack changes rather than from a stabilized negative capacitance in HZO. The Si MOS capacitor comparison (Fig. 4B) and HRTEM mixed-phase assignment support plausibility but do not establish that this specific GaN gate stack operates in the negative-capacitance regime. The paper explicitly frames the result as 'otherwise not attainable with conventional dielectrics'; without a conventional-dielectric GaN control this load-bearing contrast is unverified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an N-polar AlGaN/GaN HEMT whose gate stack incorporates an ultrathin (~1.8 nm) ALD-grown HfO2-ZrO2 ferroelectric/antiferroelectric bilayer (HZO) capped with TiN. Compared with a Schottky-gated control fabricated on the same heterostructure, the HZO device exhibits simultaneous improvements: roughly 3x higher ON current at fixed intrinsic overdrive, lower gate leakage by more than an order of magnitude, and a gate capacitance about 30% larger than the Schottky device's. The authors attribute the capacitance and current enhancements to negative capacitance (NC) in the mixed-phase HZO layer, which they argue more than compensates for the series positive capacitance of the added dielectric. Supporting evidence includes TEM phase identification (orthorhombic ferroelectric plus tetragonal antiferroelectric), C-V data on Si MOS capacitors with the same HZO, and careful structural/electrical equivalence checks (AFM, TEM, TLM, four-probe measurements).","tokens_in":19397,"tokens_out":5388,"duration_ms":56865,"significance":"If the NC interpretation is correct, the paper overturns the classical series-capacitance penalty that has constrained gate dielectric integration in 2DEG-based HEMTs and would be of substantial scientific and technological interest. The empirical data collection is careful: the AFM and TEM confirm identical recess depth and epitaxial layer thicknesses, TLM shows similar sheet resistance, and the four-probe and RSD corrections give confidence in the intrinsic current comparison. The paper also transparently states that the MIS-HEMT structure was only conceptual. However, the load-bearing mechanistic claim that the HZO acts as a negative capacitance on GaN is not directly established by the present experiments; the missing non-ferroelectric control on GaN and the truncated Schottky C-V reference leave alternative explanations open. The empirical result is likely publishable, but the strong NC interpretation needs additional support or substantial reframing.","major_comments":[{"comment":"The Schottky C-V curve is reliable only in the narrow dissipation-factor window (D<1), which the caption indicates spans roughly -1.5 V to -0.5 V. At more forward biases the Schottky gate leaks, so the reported maximum capacitance for the control may not be the true maximum capacitance of the bare heterostructure. Because the central quantitative claim is that the NC device exceeds Cg,Sch by ~30%, the control's Cmax must be obtained without leakage truncation, for example by pulsed C-V, by a lower-leakage Schottky barrier metal, or by a MIS reference with a non-ferroelectric dielectric.","section":"Fig. 3D, 'Electrical Measurements' section"},{"comment":"The only dielectric control measurements are on Si MOS capacitors, not on GaN HEMTs. The HZO on the GaN HEMT is separated from the 2DEG by a 4 nm GaN cap and 2.6 nm AlGaN barrier; the series capacitance and interface conditions are therefore different from the Si/SiO2 case. The Si result is plausibility evidence, not proof that the same HZO layer operates in the NC regime in the GaN stack. A GaN HEMT with an equivalent-thickness non-ferroelectric gate dielectric (e.g., 1.8 nm HfO2 without ZrO2) should be fabricated and measured to show that the capacitance and current boosts are specific to the ferroic HZO layer.","section":"'Mechanism of the capacitance changes and ON current increase' and Fig. 4B"},{"comment":"The paper uses the capacitance increase as evidence for negative capacitance and then explains the capacitance increase by negative capacitance, which is an interpretive loop. The Si MOS comparison does not break this loop for the GaN stack because it does not demonstrate negative capacitance on the GaN heterostructure. Direct evidence of negative capacitance in the GaN gate stack (e.g., transient C-V or local polarization imaging) or a falsifiable control experiment (a non-ferroelectric HfO2 gate showing reduced capacitance and current) is needed to support the central claim.","section":"'Mechanism of the capacitance changes and ON current increase'"},{"comment":"The Schottky HEMT shows hysteresis in its bidirectional transfer sweeps while the NC HEMT is nearly hysteresis-free. This difference indicates different interface trap densities or slow trap responses between the two gate stacks. Because the stacks differ not only by the HZO layer but also by the dilute-HF surface preparation and the TiN capping layer, the higher capacitance and current could in principle result from improved interface quality rather than from negative capacitance. This confound should be explicitly discussed or controlled.","section":"Supplementary fig. S9"}],"minor_comments":[{"comment":"The caption correctly states that the MIS-HEMT was 'not part of the experimental investigation,' but the main text and abstract do not contain this qualification; please move it into the main text to avoid misleading readers.","section":"Fig. 1B caption"},{"comment":"'N Polar' should be hyphenated as 'N-polar' for consistency with the rest of the manuscript.","section":"Page 5, 'Electrical Measurements' paragraph"},{"comment":"The caption describes the unreliable regions as 'dotted gray,' but the figure appears to use dashed lines; please make the figure and caption consistent.","section":"Fig. 3D"},{"comment":"The term 'ferroic' is used without definition; on first use, please define it as encompassing both ferroelectric and antiferroelectric order.","section":"Page 1, 'negative capacitance (NC) effect' paragraph"},{"comment":"The sentence 'the bilayer provided a similar increase in capacitance as seen before' is vague; please specify the exact comparator (SiO2 alone, HfO2 control, or both) and quantitative magnitude of the boost.","section":"Page 8, 'Mechanism' section, Fig. 4B description"}],"recommendation":"major_revision","confidential_remarks":"The empirical result (higher ON current plus lower gate leakage in the HZO-gated HEMT) appears solid and internally consistent; I am confident in the device equivalence checks. However, the paper's central mechanistic claim of negative capacitance is not yet supported on GaN. The missing experiments are a non-ferroelectric HfO2 control on the same GaN HEMT and a non-truncated estimate of the Schottky Cmax. These are within the scope of the work and should be required before publication. I would not reject, but the strong NC language in the title and abstract needs to be matched by evidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: careful device paper. Ultrathin HZO bilayer on an N-polar GaN HEMT gives higher on-current and lower gate leakage than a Schottky control. The empirical result looks solid and is worth taking seriously. The mechanistic claim—that the gain comes from stabilized negative capacitance in the HZO—is plausible but not proven.\n\nThe new bit is the specific combination: mixed-phase HZO on N-polar AlGaN/GaN with simultaneous current increase and leakage reduction. The concept isn't new—prior NC-HEMT work used BaTiO3 on Ga-polar AlGaN/GaN (ref 25)—but this material/platform combination is. The device work is genuinely careful: both samples etched together, AFM and TEM confirm identical cap/barrier/channel thicknesses, TLM shows nearly identical sheet resistance, and the electrical dataset is internally consistent, including higher ID at fixed intrinsic overdrive and lower leakage across multiple devices. The nearly hysteresis-free NC device versus the hysteretic Schottky control is a nice observation, though it cuts both ways.\n\nThe stress-test is right about the soft spots. There is no GaN control with a non-ferroelectric dielectric of equal thickness; the MIS-HEMT in Fig 1B is explicitly conceptual, not measured. The Schottky Cmax is captured only in the D<1 window (roughly -1.5 to -0.5 V), so the reported ~30 percent capacitance boost may be inflated by comparing a measurable MIS-like C-V to a truncated Schottky C-V. The devices also differ in more than the dielectric: the NC stack has the TiN cap and a dilute-HF clean before ALD, and the hysteresis difference indicates the interfaces are not the same. So the higher capacitance and roughly 3x current could in principle come from interface or stack changes rather than from a stabilized negative capacitance in the HZO. The Si MOS comparison (Fig 4B) and HRTEM mixed-phase assignment support plausibility but do not transfer automatically to this specific GaN stack. And the paper's framing—'otherwise not attainable with conventional dielectrics'—is exactly the contrast that was not measured.\n\nBottom line: treat the empirical device result as provisional but real, and the NC mechanism as a hypothesis. The fix is straightforward: a GaN MIS-HEMT with an equal-thickness non-ferroelectric dielectric (e.g., HfO2), and a C-V measurement not truncated by leakage. This paper deserves serious peer review; a referee should ask for that control before accepting the mechanism, not before accepting the data. I'd bring it to a GaN or NC reading group.","headline":"Careful GaN device data with a plausible but unproven negative-capacitance mechanism; the empirical result deserves scrutiny, and a referee should request a non-ferroelectric control before accepting the mechanism.","tokens_in":20029,"tokens_out":2547,"would_cite":true,"duration_ms":23294,"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":"Adding a 1.8 nm ferroic HfO2-ZrO2 bilayer to a GaN HEMT gate raises capacitance above the Schottky value, tripling ON current and cutting gate leakage by over an order of magnitude.","keywords":["negative capacitance","GaN high-electron-mobility transistor","ferroelectric-antiferroelectric HfO2-ZrO2","Schottky gate","gate leakage","two-dimensional electron gas","N-polar AlGaN/GaN","capacitance boost"],"falsifier":"A GaN HEMT with an equal-thickness non-ferroelectric HfO2 gate dielectric would settle it: if that device also shows a maximum capacitance above the Schottky value, the negative-capacitance explanation is not required.","tokens_in":18875,"feed_emoji":"⚡","tokens_out":8939,"duration_ms":87088,"temperature":0.7,"pith_summary":"This paper reports GaN high-electron-mobility transistors whose gate stack replaces the usual direct metal-semiconductor Schottky contact with an ultrathin mixed-phase ferroelectric/antiferroelectric HfO2-ZrO2 bilayer. The authors claim that this bilayer supplies a negative capacitance in series with the heterostructure's own capacitance, so the total gate capacitance exceeds the Schottky-gate value $C_{g,\\mathrm{Sch}}$ instead of being reduced as a conventional dielectric would. On their N-polar AlGaN/GaN devices this produces a roughly threefold higher ON current at fixed intrinsic overdrive together with more than an order of magnitude lower gate leakage. If correct, this breaks the long-standing trade-off in HEMTs between maximizing channel charge and suppressing gate leakage.","feed_headline":"Ferroic gate lifts GaN transistor past Schottky limits","feed_subtitle":"A 1.8 nm HfO2-ZrO2 bilayer lifts gate capacitance above the Schottky ceiling while cutting leakage tenfold.","key_machinery":"The central object is the ferroic HfO2-ZrO2 bilayer, ~1.8 nm thick (13 ALD cycles of ZrO2 followed by 5 cycles of HfO2), acting as a negative-capacitance element in series with the intrinsic Schottky gate capacitance $C_{g,\\mathrm{Sch}}$. Negative capacitance is the regime where the ferroelectric's voltage decreases as charge increases, so in series it amplifies rather than dilutes the total capacitance: $C_{g,\\mathrm{eq}} = (1/C_{\\mathrm{NC}} + 1/C_{g,\\mathrm{Sch}})^{-1} > C_{g,\\mathrm{Sch}}$. The argument carries on the stabilization of this otherwise unstable state by the mixed ferroelectric/antiferroelectric phase coexistence, the in-situ TiN capping layer, and the resulting depolarization fields that flatten the energy landscape.","core_discovery":"The paper's central claim is that a ferroic negative-capacitance dielectric can push a GaN HEMT past the classical Schottky-gate limit: the measured maximum gate capacitance of the HZO-gated device is about 30 percent larger than that of an otherwise identical Schottky-gated HEMT, so $C_{g,\\mathrm{eq}} > C_{g,\\mathrm{Sch}}$. Because charge and drive current scale with gate capacitance, the NC device delivers nearly three times the ON current at the same intrinsic overdrive while the additional physical thickness and wide bandgap of the HZO layer reduce gate leakage by more than an order of magnitude. The paper attributes the capacitance boost to a stabilized negative capacitance in the ~1.8 nm HfO2-ZrO2 bilayer, inferred from the capacitance increase, from mixed ferroelectric orthorhombic and antiferroelectric tetragonal phases seen in TEM, and from matching capacitance-boost behavior on silicon control stacks.","pith_inferences":["A control device with an equal-thickness non-ferroelectric HfO2 gate would isolate whether the boost comes from the ferroic negative capacitance rather than from the bilayer's high permittivity or a measurement artifact.","Because the HZO stack was optimized for silicon transistors, re-tuning the ZrO2:HfO2 cycle ratio, thickness, and capping stress for the GaN cap could push the capacitance and current gains beyond the roughly threefold value reported here.","The same series-capacitance reasoning would apply to any heterostructure 2DEG transistor, but the size of the gain will depend on how the negative-capacitance layer couples to the specific cap and barrier stack, which this paper does not demonstrate."],"forward_implications":["HEMT gate stacks would no longer face the classical choice between high channel charge via a Schottky gate and low leakage via a conventional dielectric: the ferroic layer gives both at once.","The ~30 percent capacitance rise over $C_{g,\\mathrm{Sch}}$ means more 2DEG charge is induced for the same gate voltage, which is why the ON current nearly triples at fixed intrinsic overdrive.","Gate leakage falls by more than an order of magnitude because the HZO adds physical thickness and a wider bandgap without the usual series-capacitance penalty.","The concept transfers to other 2DEG transistors, including Ga-polar GaN and III-V HEMTs, and stronger nitride ferroelectrics could enlarge the effect."],"supporting_citations":[{"why":"Supplies the ultrathin ferroic HfO2-ZrO2 gate-stack demonstration on silicon whose capacitance boost this work reproduces on GaN.","marker":"(26)"},{"why":"Provides the SiO2-only control curve and the silicon-based capacitance-boost baseline used to establish the HZO negative-capacitance behavior.","marker":"(29)"},{"why":"Proposes that a ferroelectric negative capacitance in series with a dielectric raises total capacitance; this is the mechanism the paper invokes.","marker":"(31)"},{"why":"Provides experimental evidence of ferroelectric negative capacitance in nanoscale heterostructures.","marker":"(32)"},{"why":"Shows spatially resolved steady-state negative capacitance, supporting stabilization of the NC state.","marker":"(33)"},{"why":"Reviews the theory of ferroelectric negative capacitance used to interpret the capacitance boost.","marker":"(34)"},{"why":"Demonstrates negative capacitance in ferroelectric superlattices, supporting the mixed-phase stabilization picture.","marker":"(35)"},{"why":"Demonstrates negative capacitance in a ferroelectric capacitor, the basis for the series ferroelectric-dielectric argument.","marker":"(41)"},{"why":"Reveals the double-well energy landscape in a ferroelectric layer, grounding the stabilization mechanism.","marker":"(42)"},{"why":"Prior demonstration of negative-capacitance HEMTs on AlGaN/GaN with a different ferroelectric, the direct precedent this work extends.","marker":"(25)"}],"fun_headline_variants":["Negative capacitance breaks Schottky ceiling in GaN HEMTs","NC gate triples GaN current while cutting leakage 10x","Ferroic HZO layer outdoes Schottky gate in GaN transistors","GaN HEMT surpasses Schottky limits with negative-cap gate","More current, less leakage: NC beats Schottky in GaN HEMTs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on the assumption that the thin HZO layer is working in a stabilized negative-capacitance regime; the paper infers this from the capacitance increase and from similar silicon stacks, but it does not directly measure a negative capacitance or include a GaN control with a non-ferroelectric dielectric of the same thickness.","fun_headline_variants_meta":{"raw":{"variants":["Negative capacitance breaks Schottky ceiling in GaN HEMTs","NC gate triples GaN current while cutting leakage 10x","Ferroic HZO layer outdoes Schottky gate in GaN transistors","GaN HEMT surpasses Schottky limits with negative-cap gate","More current, less leakage: NC beats Schottky in GaN HEMTs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000813,"raw_usage":{"total_tokens":3553,"prompt_tokens":919,"completion_tokens":2634,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":535,"completion_tokens_details":{"reasoning_tokens":2534}},"tokens_in":535,"tokens_out":2634,"duration_ms":20139,"temperature":1.0,"reasoning_tokens":2534,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:19:00.769836+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A GaN HEMT with an equal-thickness non-ferroelectric HfO2 gate dielectric would settle it: if that device also shows a maximum capacitance above the Schottky value, the negative-capacitance explanation is not required.","supporting_citations":[],"review_version":2}