{"id":"653c9a52-5c52-4951-ade4-1c42610a689a","arxiv_id":"2505.06547","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The oscillation-frequency tunability (nonlinearity) of three-terminal magnetic tunnel junction oscillators can be reduced to zero by a 1.1 nm free layer and varied electrically by spin-orbit current.","lead":"This paper measures how the current tunability of the oscillation frequency in nanoscale magnetic tunnel junction oscillators depends on magnetic field direction, free-layer thickness, and additional spin-orbit currents. It shows the tunability can be engineered to zero and modulated electrically, which matters for stable microwave generators and compact neuromorphic devices.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Zero-nonlinearity claim depends on excluding curved high-current points attributed to Joule heating; if that curvature is intrinsic nonlinearity, N=0 is an artifact of the fit window.","rationale":"The reader identified the same load-bearing assumption: the high-current data points in the 1.1 nm sample are excluded as Joule heating, and the zero-nonlinearity claim depends on that exclusion. My reading of the manuscript confirms this is the weakest link. The paper itself states that the observed frequency-current relation is quadratic rather than linear, which is exactly the kind of curvature expected from either heating or from higher-order terms in the nonlinear auto-oscillator model. Without an independent thermal calibration or a full-range fit that includes both the theoretical nonlinearity and a thermal term, the near-zero slope extracted from a restricted window cannot be distinguished from a small nonlinearity that is masked by the fit choice. The absence of error bars on the extracted slopes in Figure 3(k) further weakens the claim. The SOT-modulation section has its own issue—the model in Eq. (5) uses a parameter η fitted to the same data—but that is secondary because the headline result is the zero nonlinearity at 1.1 nm. Since the reader already marked the verdict CONDITIONAL with these concerns, my assessment does not change the verdict; it should remain conditional pending the full-range analysis or equivalent thermal calibration.","tokens_in":9759,"tokens_out":3066,"duration_ms":37193,"concrete_test":"Re-fit the full 1.1 nm data in Figure 3(h)-(j) to f(J) = f0 + (N/2π)·(ζ−1)/(ζ+Q) + βJ², where β is a thermal coefficient independently estimated from the 1.3 nm control sample or from the known Joule power and thermal resistance, and N, Q, f0 are free parameters. If the 95% confidence interval for N excludes zero, the zero-nonlinearity conclusion is an artifact of excluding the high-current points; if it includes zero, the exclusion is justified and the claim stands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in the abstract and Section 3.2—that nonlinearity reaches zero in 1.1 nm CoFeB and frequency is unaffected by drive current—rests on the data in Figure 3(e)-(j). The authors explicitly state that the frequency versus J_STT curves in the 1.1 nm sample are quadratic, not linear as predicted by Eq. (1), and that the high-current points (red) were excluded 'to eliminate the influence of heating.' The near-zero slope is then extracted from the remaining low-current points (yellow). This makes the conclusion sensitive to the fit window: the same curvature that motivates the heating correction also determines which points define the zero slope. If the curvature is partly intrinsic—for example, the full Eq. (2) with finite Q and supercriticality not much larger than Q, or a current-dependent M_eff that is itself part of the nonlinear response—then the low-current slope is not N but d f/d J evaluated at a particular operating point, and the claim of drive-current-independent frequency is unsupported. The paper provides no error bars on the slopes in Figure 3(k), so 'extremely low N' could be consistent with a small but nonzero N. Section 3.2 attributes the zero to M_eff ≈ 0 from high H_k in the thinner film, but no independent magnetization or anisotropy measurement is presented to confirm that the linear regime was actually sampled.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a systematic experimental study of nonlinearity (the current-induced frequency tunability) in three-terminal magnetic tunnel junction spin-torque nano-oscillators (MTJ-STNOs). The authors vary the external magnetic field magnitude and direction, the CoFeB free-layer thickness (1.3 nm vs. 1.1 nm), and the combined application of spin-transfer torque (STT) and spin-orbit torque (SOT) currents. They find that the field magnitude only weakly affects the nonlinearity in 1.3 nm CoFeB, while the field angle can continuously tune it from positive to negative values. For 1.1 nm CoFeB they report an extremely low nonlinearity, which they interpret as N approaching zero, making the oscillation frequency nearly independent of drive current. They also demonstrate that SOT currents can modulate the frequency-current slope and propose a modified auto-oscillator model in which the SOT current acts as an additive effective STT current, with a conversion factor eta extracted from the data.","tokens_in":10104,"tokens_out":6675,"duration_ms":63380,"significance":"If the central claims hold, the paper offers two practically useful routes for controlling STNO nonlinearity: a thickness-based approach that can suppress nonlinearity without requiring a specific magnetic field, and an electrical approach that can tune nonlinearity in situ. The field-angle sign change of N is a clean demonstration of the Slavin-Tiberkevich theory in three-terminal devices. The proposed modified model extends an established framework to combined STT+SOT driving and could be a useful design tool. However, the significance is currently tempered by two issues: the zero-nonlinearity claim rests on a selective fit window after excluding high-current data points, and the model validation is based on a parameter extracted from the same data to which the model is then compared. These issues limit the strength of the conclusions until they are addressed.","major_comments":[{"comment":"The conclusion that N reaches zero for the 1.1 nm CoFeB sample rests entirely on a linear fit to the low-current (yellow) data points after excluding the high-current (red) points on the grounds of Joule heating. Because the full f(J_STT) dependence is quadratic, the low-current slope is a local differential quantity and cannot be identified with the nonlinearity N defined in Eq. (2) unless the heating interpretation is independently verified. The paper provides no independent measurement of Joule heating (e.g., resistance or temperature change as a function of current) and no independent determination of M_eff (e.g., FMR or magnetometry). In addition, Fig. 3(k) shows no error bars on the extracted slopes, and the reproducibility across devices of the same thickness is not demonstrated. I therefore request: (i) a fit of the full frequency-current curve to a model that explicitly includes both the intrinsic nonlinearity and a heating term, (ii) error bars on all reported slopes, and (iii) an explicit statement of the number of devices measured and the device-to-device variation. Without these, the N=0 claim could be an artifact of the chosen fit window.","section":"Section 3.2, Fig. 3(h)-(j)"},{"comment":"The refined model in Eq. (5) introduces a free parameter eta that is extracted by fitting the same frequency-current curves to which the model is subsequently compared. The inset of Fig. 4(e) shows that eta is estimated from a linear fit of Delta J vs. J_SOT, and the solid lines in Fig. 4(e) are then described as 'theoretical calculation'. This is a fit, not an independent validation of the model. To make the model credible, the authors should either determine eta from an independent measurement (e.g., harmonic Hall or spin-torque ferromagnetic resonance) or demonstrate predictive power by fixing eta on one dataset and predicting another. They should also state explicitly which parameters (N, Q, J_th,STT, omega_FMR) are fixed and which are fitted when generating the solid lines in Fig. 4(e) and (k).","section":"Section 3.3, Eq. (5), Fig. 4(e)"},{"comment":"The relation between the measured slope d omega/dJ_STT and the nonlinearity N uses the prefactor (1+Q)/(zeta+Q)^2 from Eq. (2). The authors assume that this prefactor is approximately constant to justify the proportionality, but they do not report values of Q or zeta for any of the measurements. Without these parameters, one cannot assess how much of the observed variation (or lack thereof) in Fig. 2(e) and Fig. 3(k) is due to a change in the prefactor rather than a change in N. I recommend reporting a table of the fitted parameters (omega_FMR, J_th, Q, and the extracted N) for each condition, together with confidence intervals.","section":"Section 3.1, Eq. (2), Fig. 2(e)"}],"minor_comments":[{"comment":"The sentence 'With such property, the output power of the STNO remains unaffected by the driving current' is physically incorrect: even when N=0, the oscillation power P=(zeta-1)/(zeta+Q) depends on the current. The intended statement is presumably that the frequency is unaffected; please correct this.","section":"Section 3.2, paragraph after Fig. 3(k)"},{"comment":"The extracted slopes are presented without error bars or uncertainty estimates. Given that the conclusions hinge on the magnitude and trend of these slopes, error estimates (from the linear fits) should be added.","section":"Figures 3(k), 4(f), and 4(l)"},{"comment":"The parameter Q is defined after Eq. (1) but not redefined when Eq. (5) is introduced later; please keep the notation consistent and restate the definition for readability.","section":"Section 3.3, Eq. (5)"},{"comment":"The text contains many garbled symbols (e.g., 'd' rendered as '𝑑𝑑', and inconsistent use of 'N' for the nonlinearity). This appears to be a LaTeX rendering issue; the typeset manuscript should be checked carefully so that equations and symbols are legible.","section":"Throughout"},{"comment":"The statement that the asymmetric impact of positive and negative SOT currents 'largely rules out the influence of thermal effects' is too strong. Joule heating is symmetric in current, but the asymmetry could also arise from the nonlinear dependence of the denominator in Eq. (5). Please either show a quantitative estimate of the thermal contribution or soften this claim.","section":"Section 3.3, discussion of thermal effects"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a relevant and timely topic, and the experimental data set is substantial. However, the main claims--zero nonlinearity at 1.1 nm and the success of the modified model--are currently supported only by fits that involve selective data exclusion or a parameter extracted from the same data. I would recommend the editors ask for the additional analyses and independent checks described in the major comments before considering the paper for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's my read of arXiv:2505.06547. The genuinely new thing is experimental: a three-terminal MTJ STNO with a 1.1 nm CoFeB free layer shows an almost flat frequency vs STT-current slope in the low-current window, and the slope of the 1.3 nm device changes systematically with an applied SOT current. The field-angle dependence (N going from positive to negative as theta increases) is textbook-like and well presented. Those data are worth a look.\n\nThe paper is less solid at its headline claim. The N=0 result for 1.1 nm is obtained by excluding the high-current points from the f vs J curves because they curve downwards and the authors attribute that to Joule heating. The problem is that the same curvature, if it has any intrinsic component, means the low-current slope is not N but df/dJ at one operating point. Since Eq. (2) already gives a current-dependent factor (1+Q)/(zeta+Q)^2, the curvature could be partly from finite supercriticality rather than temperature. There are no error bars on the extracted slopes and no independent measurement of M_eff or H_k to back the claim that M_eff is near zero. One device per thickness does not help.\n\nThe SOT modulation section is more robust. The data in Fig. 4 show a clear, monotonic change in df/dJ_STT as J_SOT is swept, and the asymmetric response between positive and negative currents makes a pure heating story unlikely. The model addition, eq (5), is a minor extension of Slavin-Tiberkevich, and yes, eta is extracted by fitting the same family of curves that the theory then reproduces; that is not fully circular, since the Delta-J vs J_SOT relation in the inset is an independent linear relation, but it does mean the \"theoretical calculation\" is not a prediction.\n\nBottom line: the paper deserves a serious referee. It reports new experiments and a plausible mechanism for electrical tuning of nonlinearity, and the zero-nonlinearity claim is testable. What it needs before publication is error bars, more statistics, and either a thermal calibration or a fit over the full current range that lets heating and intrinsic nonlinearity compete. If the claims are softened to \"near-zero N in the low-current regime,\" the core content stands.\n\nI wouldn't cite it in my own work as it stands, but I'd bring it to a reading group to debate the fit-window issue.","headline":"Solid experiments on SOT tuning of nonlinearity in three-terminal MTJ STNOs, but the headline zero-nonlinearity claim rests on a fit window that excludes the curved high-current data.","tokens_in":10590,"tokens_out":2588,"would_cite":false,"duration_ms":28053,"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":"This paper reports that the nonlinearity of three-terminal magnetic tunnel junction spin torque nano-oscillators can be tuned by magnetic field angle and free-layer thickness, reaching zero at 1.1 nm CoFeB, and can also be modulated…","keywords":["spin torque nano-oscillator","magnetic tunnel junction","nonlinear auto-oscillator theory","spin-transfer torque","spin-orbit torque","zero nonlinearity","CoFeB free layer thickness","current-tunable frequency"],"falsifier":"Run the same frequency-versus-current sweep in a 1.1 nm CoFeB device using short current pulses (or a low-temperature stage) so that Joule heating is suppressed, and compare the full-curve slope with the dc measurement: the zero-nonlinearity claim predicts the slope stays at zero over the whole range, while the heating-artifact alternative predicts a nonzero slope once the high-current points are included. As a complementary check, heat the sample externally and see whether the excluded curvature is reproduced as a purely thermal effect.","tokens_in":9564,"feed_emoji":"🧲","tokens_out":10170,"duration_ms":96241,"temperature":0.7,"pith_summary":"Spin torque nano-oscillators emit microwaves whose frequency normally shifts when the drive current changes; the size of that shift is the nonlinearity. This paper tries to show that in three-terminal magnetic tunnel junctions the nonlinearity is not a fixed property but can be engineered: it changes from positive to negative as the external magnetic field tilts out of the film plane, falls to zero when the CoFeB free layer is thinned to 1.1 nm, and can be continuously shifted by sending current through either the junction or the spin-orbit channel. To explain the combined current response, the paper extends the standard nonlinear auto-oscillator model with an effective spin-orbit-torque current term. If these claims hold, oscillator designers gain a material and electrical route to a current-independent frequency, which would improve microwave spectral purity, and a way to set the current-frequency response of individual devices for neuromorphic circuits.","feed_headline":"1.1 nm free layer cancels oscillator nonlinearity","feed_subtitle":"At that thickness, output frequency no longer follows drive current, while STT and SOT currents still tune nonlinearity.","key_machinery":"The load-bearing object is the nonlinear auto-oscillator relation $f(J)=f_{\\mathrm{FMR}}+\\frac{N}{2\\pi}P$, with normalized power $P=(\\zeta-1)/(\\zeta+Q)$ and supercriticality $\\zeta=J/J_{\\mathrm{th}}$, in which the frequency shift is proportional to the nonlinearity $N$ and the nonlinear damping $Q$ controls how power grows above threshold. Experimentally, the slope of the measured $f$ versus current-density curve is treated as a direct readout of $N$. The paper's modification adds the spin-orbit-torque current into $\\zeta$ as an effective current contribution $\\eta J_{\\mathrm{SOT}}$, so STT and SOT act as one combined drive; this is the mechanism that turns raw power spectral density measurements into the claims about thickness-zeroed and electrically tunable nonlinearity.","core_discovery":"The central claim is that the nonlinearity $N$, read from the slope $\\mathrm{d}f/\\mathrm{d}J$ of frequency versus current density, can be tuned over a wide range and even zeroed by choices that are built into the device. In a 1.3 nm CoFeB free layer, rotating the applied field angle $\\theta$ from $0^\\circ$ to $60^\\circ$ changes $N$ from positive to negative while keeping the field magnitude fixed, and the magnitude of an in-plane field leaves $N$ nearly unchanged. In a 1.1 nm CoFeB layer, the perpendicular magnetic anisotropy nearly cancels the demagnetizing field, making $M_{\\mathrm{eff}}\\approx 0$, so the frequency current slope becomes effectively zero and the oscillation frequency stops following the drive current. The paper also claims that both spin-transfer-torque and spin-orbit-torque currents can modulate $N$ in either direction, and it accounts for this with the refined relation $f(J_{\\mathrm{STT}}) = f_{\\mathrm{FMR}} + \\frac{N}{2\\pi}\\left(1-\\frac{1+Q}{(J_{\\mathrm{STT}}+\\eta J_{\\mathrm{SOT}})/J_{\\mathrm{th,STT}}+Q}\\right)$, where the spin-orbit current acts as an equivalent shift $\\Delta J=\\eta J_{\\mathrm{SOT}}$ of the effective drive; the calculated curves agree with the measured spectra.","pith_inferences":["The same compensation of $M_{\\mathrm{eff}}$ that zeroes $N$ at 1.1 nm thickness could in principle be reached by voltage-controlled magnetic anisotropy or mechanical strain, giving a testable route to zero nonlinearity that the paper does not pursue.","A pulsed-current or cryogenic repetition of the 1.1 nm measurement would decide whether zero-$N$ survives outside the small-current fitting window; the paper's own data leave that question open.","If nonlinearity can be tuned electrically during operation, a single oscillator could be reconfigured in situ between a stable microwave generator and a nonlinear neuromorphic element, a capability the paper hints at but does not demonstrate.","The asymmetry between positive and negative SOT currents in the data suggests the electrical effect is torque-like rather than thermal; a direct test would compare nonlinearity modulation under positive and negative currents of equal magnitude at identical dissipated power."],"forward_implications":["At 1.1 nm CoFeB, the oscillation frequency should no longer shift with drive current, so output frequency becomes stable against current noise and easier to use in high-quality microwave generation.","Because the field angle can switch $N$ between positive and negative in 1.3 nm CoFeB, the same device can be configured for either sign of current-frequency tunability without changing the field magnitude.","The refined model with the $\\eta J_{\\mathrm{SOT}}$ term predicts the frequency response for arbitrary STT/SOT current combinations, allowing a three-terminal oscillator to be designed with a specified nonlinearity.","At the zero-nonlinearity point, output linewidth is expected to narrow, giving cleaner microwave spectra.","Electrical tuning of nonlinearity by STT and SOT currents provides a compact bias knob for reconfigurable STNO-based synapses and neurons."],"supporting_citations":[{"why":"Supplies the nonlinear auto-oscillator formalism, Eq. (1), that defines the relation between frequency shift and normalized power.","marker":"[30]"},{"why":"Provides the fuller theory of current-driven auto-oscillators from which $N$ is extracted as proportional to $\\mathrm{d}f/\\mathrm{d}J$.","marker":"[32]"},{"why":"Establishes the earlier route of tuning $N$ through interfacial $M_{\\mathrm{eff}}$ modification, the benchmark for the thickness-based zero-$N$ approach.","marker":"[36]"},{"why":"Previous study of the same three-terminal devices showing SOT-assisted STT oscillation, used to interpret the threshold-current changes.","marker":"[28]"},{"why":"Documents current-induced Joule heating changes of $H_k$ and $M_s$, the basis for excluding high-current data points in the 1.1 nm sample.","marker":"[40]"},{"why":"Bounds the field-like-torque effective field at about 3 mT, used to rule out that mechanism for the electrical modulation of $N$.","marker":"[41]"}],"fun_headline_variants":["1.1 nm free layer eliminates oscillator nonlinearity","Zero nonlinearity via 1.1 nm CoFeB thickness","Spin currents electrically tune nonlinearity","Field angle flips nonlinearity sign in MTJ oscillators"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The zero-nonlinearity claim rests on excluding the high-current data points in the 1.1 nm sample and treating their curved frequency response as pure Joule heating; if that curvature is intrinsic nonlinearity rather than heating, the conclusion that $N=0$ is an artifact of the fitting window.","fun_headline_variants_meta":{"raw":{"variants":["1.1 nm free layer eliminates oscillator nonlinearity","Zero nonlinearity via 1.1 nm CoFeB thickness","Spin currents electrically tune nonlinearity","Field angle flips nonlinearity sign in MTJ oscillators"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001064,"raw_usage":{"total_tokens":4567,"prompt_tokens":1158,"completion_tokens":3409,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":774,"completion_tokens_details":{"reasoning_tokens":3345}},"tokens_in":774,"tokens_out":3409,"duration_ms":26006,"temperature":1.0,"reasoning_tokens":3345,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:39:26.600960+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same frequency-versus-current sweep in a 1.1 nm CoFeB device using short current pulses (or a low-temperature stage) so that Joule heating is suppressed, and compare the full-curve slope with the dc measurement: the zero-nonlinearity claim predicts the slope stays at zero over the whole range, while the heating-artifact alternative predicts a nonzero slope once the high-current points are included. As a complementary check, heat the sample externally and see whether the excluded curvature is reproduced as a purely thermal effect.","supporting_citations":[],"review_version":1}