{"id":"1f05479d-ab37-433e-a78f-2289a1e92ecf","arxiv_id":"1908.02139","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Dwell times in superparamagnetic CoFeB/MgO/CoFeB junctions are explained by entropic pathways that yield a 17 nm activation volume, and switching rates can be tuned by combined magnetic and electric fields.","lead":"This paper explains why tiny magnetic tunnel junctions switch randomly much faster than textbook models predict: many microscopic switching paths act together, leaving a small 17 nm active region. It also shows that combining a magnetic field and a bias voltage can shift the switching-rate curve, a property usable in random-number and brain-inspired computing.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed 17 nm activation radius rests on an untested N! pathway-count ansatz; a diameter-scaling measurement would settle whether this entropy-based estimate is real or coincidental.","rationale":"The reader's weakest assumption (the w=N! model) is the correct point of vulnerability. I have sharpened it by noting that the two-level telegraph data favor a nucleation-count N rather than a permutation count N!, and by proposing a diameter-dependence experiment that would directly test the model. The energy-based route to rA from ΔE/K is more robust, but the paper's entropic reconciliation of the prefactor and the claimed mutual confirmation depend on the N! ansatz. The paper is a short experimental letter with no error bars on the Arrhenius intercepts, so CONDITIONAL is appropriate. My proposed test would either validate or falsify the central claim; until it is run, the concern prevents full acceptance but does not require rejection.","tokens_in":7541,"tokens_out":16899,"duration_ms":195223,"concrete_test":"Measure dwell-time Arrhenius intercepts on identical MTJs with at least three different electrode diameters, e.g., 80, 140, and 200 nm, while keeping the stack and processing otherwise identical. Under the paper's w≈N! model with a fixed intrinsic activation radius rA≈17 nm, N = (D/2 / 17 nm)^2, so ln w≈ln(N!) should rise from about 35 at 140 nm diameter to about 80 at 200 nm and fall to about 17 at 80 nm. Under the physically more natural single-nucleation count w≈N, ln w should change only by ln(area ratio), i.e., by about 1–2 across this range. The observed scaling of the intercept with diameter therefore distinguishes N! from N and settles whether the entropy-derived 17 nm radius is real or an artifact of the ansatz.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's entropic route to the activation radius (p.3) assumes w≈N! for N sub-volumes, then uses ln w≈35 to get N≈17 and rA = 70 nm/√N ≈ 17 nm. This is explicitly an assumption, and no independent evidence supports the N! counting. The observation (Fig. 1a) is a clean two-level telegraph, which is naturally described by a single nucleation event followed by rapid wall propagation; for such a process the number of equivalent paths should scale as N, not N!, and N! would require all sub-volumes to switch in every possible order even though only two resistance levels are observed. If the correct count is w≈N, then ln w≈35 would require N≈10^15 sub-volumes, incompatible with a 140-nm electrode; the entropic explanation of the fast dwell times would collapse, and the 'remarkable' agreement between the entropy-derived radius and the energy-derived radius from ΔE/K would be a coincidence rather than confirmation. Since the central claim is that entropic effects produce the small activation volume, this untested combinatorial ansatz is the most load-bearing link in the argument.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript reports time-resolved measurements of thermally driven switching in perpendicular CoFeB/MgO/CoFeB magnetic tunnel junctions with 140-nm-diameter pillars, a 1.1-nm free CoFeB electrode and MgO barriers of 1.2, 1.4 and 1.6 nm. The devices show a clean two-level telegraph with exponentially distributed dwell times. From the temperature and magnetic-field dependence of the mean dwell times the authors extract, for each barrier thickness, an activation energy ΔE and a Zeeman-energy product V·M_S. Because ΔE/V_E corresponds to an apparent anisotropy 12.7 to 24.3 times smaller than the quasistatically measured effective anisotropy K ≈ 330 kJ/m³, they infer a magnetic activation volume V_A = ΔE/K with radius r_A = r_E/√(K/K*) = (17±4) nm; recomputing M_S with V_A brings the saturation magnetization into the literature range. The Arrhenius intercept yields ln w ≈ 35 (w ≈ 10^15), and assuming w ≈ N! pathways among N switching sub-volumes gives N ≈ 17 and an equivalent radius of about 17 nm, which the authors call a remarkable agreement between the energetic and thermodynamic routes. Bias-voltage sweeps at the three barrier thicknesses are decomposed into a spin-torque contribution that decays exponentially with barrier thickness and a voltage-induced anisotropy contribution, and the switching-rate tuning curves obtained by sweeping field at fixed bias and bias at fixed field are Gaussian, matching a requirement for neural-like population coding.","tokens_in":7812,"tokens_out":27080,"duration_ms":244469,"significance":"If the central claims hold, the paper is a useful contribution to the sp-MTJ literature: it provides quantitative evidence that the thermally active volume in these nominally single-domain 140-nm electrodes is far smaller than the electrode (about 17 nm radius), that the independently measured quasistatic anisotropy combined with that volume reconciles the barrier heights, and that the resulting saturation magnetization agrees with literature values. The clean telegraph data, the verified exponential dwell-time statistics, the use of an independent anisotropy measurement as a cross-check, and the demonstration of Gaussian tuning curves shifted by bias voltage are genuine strengths; the paper also candidly admits that the anisotropy/spin-torque separation is not reliable. The specifically entropic result is, however, conditional: the values w ≈ 10^15, N ≈ 17 and the claimed corroboration of the activation volume rest on an unverified w ≈ N! pathway-counting model. As it stands, the paper establishes the reduced activation volume through the energetic analysis but does not establish the entropic mechanism announced in the abstract.","major_comments":[{"comment":"This step is load-bearing for the abstract's claim that including entropic effects leads to a magnetic activation volume much smaller than the electrode. The entropy-derived radius r_A ≈ 17 nm is obtained exclusively from the assumption w ≈ N!, i.e., that the magnetization can switch through any ordering of N sub-volumes; no microscopic justification is given, and the paper flags the step only with 'If we assume'. The assumption is in tension with the data in Fig. 1a, which show switching between exactly two resistance states: a process of a single nucleation event followed by rapid wall propagation would naturally have w ∝ N (the choice of nucleation site), and with w ∝ N the value ln w = 35 would force N ≈ 10^15 sub-volumes and a sub-atomic activation radius, destroying the entropic explanation. Because N is fixed by the same Arrhenius intercept that the model is meant to explain, the 'remarkable' agreement of the two radii tests only the assumed functional form. I ask the authors to justify N! microscopically, test it by measuring ln w versus pillar diameter (for w ≈ N! a doubling of diameter changes ln w by roughly a factor of four plus a large additive term, whereas for w ∝ N it changes by only ln 4), or reframe the entropic estimate as a hypothesis rather than as confirmation.","section":"Entropic estimate of activation volume (p. 3)"},{"comment":"The primary extracted quantities — ΔE, V·M_S, K*, M_S and r_A — are reported without error bars; the (17±4) nm uncertainty is only the spread across three barrier thicknesses, not a propagated fit error, and no fit ranges, numbers of points or weighting are given. The value ln w ≈ 35 is the intercept of the same Arrhenius regression and depends fully on the assumed attempt time τ0 = 10^-11 s (Ref. [16]): a factor-of-ten change in τ0 shifts ln w by 2.3 and N by about two to three sub-volumes, and a temperature dependence of K or τ0 over the measured range would bias the slope and intercept simultaneously. Since the paper's central message is an agreement of 'almost the same radii', the absence of error propagation makes that agreement impossible to evaluate; please provide uncertainties derived from the fits and a sensitivity statement for τ0.","section":"Table I and Arrhenius/field fits (pp. 2-3)"},{"comment":"The decomposition of the bias-voltage effect rests on only three barrier thicknesses, and the paper itself states that the anisotropy contribution is 'not very reliable'. The quoted parameters confirm the concern: the spin-torque decay length is 13±6/nm (about 46% relative error) and the anisotropy coefficient is β = (30±15) fJ/(V·m) (about 50% relative error). In addition, the extracted activation energies are non-monotonic in barrier thickness (2.5, 1.3 and 2.4 eV for 1.2, 1.4 and 1.6 nm), which is unexplained and weakens confidence in the thickness trends used for the separation. The abstract's claim that the data allow one to separate the impacts of Zeeman energy, spin-transfer torque and voltage-induced anisotropy is therefore stronger than the evidence supports; the voltage-anisotropy value should be presented as tentative.","section":"Spin-torque/anisotropy separation (Eq. (3), Fig. 3)"}],"minor_comments":[{"comment":"The Gaussian tuning curve as printed, ν = ν0·exp(½((C−C0)/σ)²), is inverted (it has a minimum at C0 and maxima at the wings); even in a letter, the missing minus sign will confuse readers, since Fig. 4 shows peaked curves. It should read ν = ν0·exp(−½((C−C0)/σ)²).","section":"Eq. (4)"},{"comment":"The abstract uses the spelling 'Zeman energy' twice; the standard spelling is 'Zeeman energy'.","section":"Abstract"},{"comment":"The in-paper title 'Tuning superparamagnetism in perpendicular magnetic tunnel junctions' (p. 1) differs from the submission title 'Superparamagnetic dwell times and tuning of switching rates in perpendicular CoFeB/MgO/CoFeB tunnel junctions'; the two should be made consistent.","section":"Title consistency"},{"comment":"Table I uses decimal commas (e.g., 9,28; 2,5) inconsistently with the main text, and the column header 'V_E/A M_S, Anm²' does not state the power-of-ten prefactor of the units.","section":"Table I"},{"comment":"The caption of Fig. 1b contains 'the mean well times ¯τP/AP'; 'well' should be 'dwell'.","section":"Fig. 1 caption"},{"comment":"The manuscript does not state how many junctions were measured for each barrier thickness, how many switching events were collected per dwell-time value, or the temperature range and number of points used in the Arrhenius fits; these details are needed to judge the fit quality.","section":"Measurement details (p. 2)"},{"comment":"The grain-size result from supplement appendix C is the only microstructural evidence supporting granularity and is used to compare with the 17 nm radius; a one-sentence summary of the measured grain size should appear in the main text.","section":"Supplement"},{"comment":"The statement that 'for thick barriers the change of the anisotropy can have larger impact' is not borne out by the data in Fig. 3b, where the spin-torque term (about 0.045 eV/V) still exceeds the anisotropy term (about 0.021 eV/V) at t_MgO = 1.6 nm; please clarify the intended claim.","section":"Bias-voltage discussion (p. 3)"},{"comment":"Reference [8] contains a typo ('EUropean patent submission') and should be formatted as a standard patent reference.","section":"Ref. [8]"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about the N! ansatz does land on reading the manuscript: the entropic volume and the claimed 'remarkable' agreement are only as good as the unverified w ≈ N! counting, and the paper itself flags the counting only as an assumption. The energetic route to the reduced activation volume is noticeably stronger and could stand alone, so the manuscript is defensible with a reframing plus error bars, and I do not see grounds for rejection. The submission appears to be a 2019 letter-format preprint; the editor may want to check that the in-paper title matches the submitted title and that the authors' patent reference [8] is complete and correct."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper tries to explain why dwell times in 140-nm perpendicular CoFeB/MgO/CoFeB tunnel junctions are orders of magnitude shorter than single-domain Arrhenius predicts. The suggestion is entropic: many switching pathways lower the free energy barrier. If right, this is a useful resolution of a known puzzle in superparamagnetic MTJs, and their way of separating spin-torque from voltage-induced anisotropy effects by comparing three MgO thicknesses is a genuinely nice idea. The tuning-curve demonstration—shifting the Gaussian switching-rate peak with the other control parameter—is clean and directly relevant for stochastic-computing hardware.\n\nThe central quantitative claim, though, is shakier than the tone suggests. The entropy factor ln(w)≈35 is extracted from the intercept of the same Arrhenius plot that gives the energy barrier, and the conversion to a 17-nm activation radius assumes w≈N! for N sub-volumes. That is an explicit ansatz with no independent support. A telegraph signal between two resistance levels, as in Fig. 1a, is naturally described by one nucleation event followed by fast wall propagation; the number of equivalent paths should then scale as N, not N!. If w≈N, ln(w)≈35 would require ~10^15 sub-volumes, which is incompatible with a 140-nm electrode, and the entropic explanation collapses. The agreement between the entropy-derived radius and the energy-derived one is therefore not yet confirmation—it may be coincidence, because both derive from the same linear fit and the same ΔE.\n\nThe paper is also sloppy about error bars. Table I lists no uncertainties, and the voltage-separation fit gives one parameter with ±6/nm on an exponent, so the STT/anisotropy split is indicative, not precise. The claim that the tuning curves 'match exactly the requirements' for neural-like computing is overstated; they match the shape.\n\nWhat is solid: the data are real, the two-state statistics are clean, the field/voltage dependence follows the expected exponential form, and the barrier-thickness method is worth pursuing. The proposed neurons-in-population-code application is plausible, not proven.\n\nI would send this to a serious referee, but I would tell the referee to focus on the N! assumption and to ask for either a direct test—e.g., diameter scaling of the dwell-time statistics—or a much more cautious presentation of the entropy-derived radius. As is, the paper is a good concept paper with one load-bearing link missing.\n\nRegards.","headline":"Entropic explanation of fast MTJ switching is plausible but rests on an untested N! pathway-count ansatz; the tuning-curve demonstration is the solid part.","tokens_in":8299,"tokens_out":2376,"would_cite":false,"duration_ms":25819,"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":"The paper claims that thermally driven switching in perpendicular CoFeB/MgO/CoFeB junctions is governed by a ~17 nm activation volume, not the full 70 nm electrode, and that entropic pathway counting resolves the fast-dwell-time puzzle.","keywords":["superparamagnetic tunnel junctions","magnetic tunnel junctions","perpendicular magnetic anisotropy","Arrhenius dwell times","magnetic activation volume","spin-transfer torque","voltage-controlled magnetic anisotropy","stochastic neural networks"],"falsifier":"Measure dwell times on electrodes with different diameters (e.g., 100 nm, 140 nm, 200 nm) at fixed thickness and temperature; if the entropic model is right, the inferred activation radius stays near 17 nm regardless of electrode size, whereas if nucleation is tied to grain size or the full electrode, the radius would scale with electrode size.","tokens_in":7339,"feed_emoji":"🧲","tokens_out":9302,"duration_ms":81187,"temperature":0.7,"pith_summary":"This paper studies superparamagnetic magnetic tunnel junctions whose free CoFeB electrode switches thermally on millisecond timescales at temperatures near 50 °C. The measured dwell times follow an Arrhenius law but are orders of magnitude shorter than a single-domain model would allow. The authors argue that including entropic effects – a large number w of possible switching pathways – removes this discrepancy, reducing the effective activation volume to a region of about 17 nm radius. They then use MgO barrier thickness as a control to separate spin-transfer torque from voltage-induced anisotropy changes, and show that switching-rate tuning curves follow Gaussians whose peak shifts with the other input. If correct, this makes such junctions usable as tunable stochastic units for true random number generation and population-coding neural networks.","feed_headline":"Switching in sp-MTJs traces to a 17-nm activation volume","feed_subtitle":"Entropic pathways reconcile millisecond dwell times with measured barriers, enabling field-tunable stochastic devices.","key_machinery":"The central object is the entropic multiplicity $w$ of switching pathways in the free-energy barrier. The paper writes the mean dwell time as $\\bar{\\tau} = (\\tau_0/w) \\exp(\\Delta E/k_B T)$, so that a large $w$ shortens the effective prefactor without changing the barrier. To estimate $w$, it assumes switching can nucleate in any of $N$ interchangeable sub-volumes, giving $w \\approx N!$; with $\\ln w \\approx 35$ this yields $N \\approx 17$ and an activation radius $r_A = r_E/\\sqrt{K/K^*} \\approx 17\\,\\text{nm}$. This entropic prefactor is what reconciles the measured Arrhenius slopes with the fast observed switching.","core_discovery":"The central claim is that the inconsistency between measured superparamagnetic dwell times and the single-domain thermal-activation barrier disappears once the entropy $S = k_B \\ln w$ of switching pathways is added to the free energy. With the free energy $F = E - T S$, the mean dwell time becomes $\\bar{\\tau}_{P/AP} = (\\tau_0/w) \\exp(\\Delta E_{P/AP}/k_B T)$. Fitting the field and temperature dependence gives an activation energy $\\Delta E$ and an apparent anisotropy $K^* = \\Delta E/V_E$ that is 13 to 24 times smaller than the measured anisotropy $K$. Interpreting the ratio as a reduced magnetic activation volume yields $r_A = r_E/\\sqrt{K/K^*} = (17\\pm4)\\,\\text{nm}$, and the same volume makes the extracted saturation magnetization fall in the reported 500 kA/m to 1 MA/m range. The paper also reports that the same 17 nm radius follows from counting $w \\approx N! \\approx 10^{15}$ nucleation pathways, with $N \\approx 17$. For the bias-voltage response, the spin-torque contribution scales exponentially with MgO thickness while the anisotropy contribution scales as $1/t_{\\text{MgO}}$, allowing the two to be separated; the resulting switching-rate tuning curves are Gaussian in both magnetic and electric fields.","pith_inferences":["Going beyond the paper, if the $N!$ pathway count is taken literally, the inferred 17 nm region is comparable to the domain-wall width in similar CoFeB films, suggesting reversal may nucleate and expand from a small reversed domain; time-resolved magnetic imaging of individual switching events could test this directly.","A further implication is that the pathway-counting assumption is the softest link; an independent measurement of the attempt prefactor $\\tau_0/w$ from ferromagnetic resonance or from the temperature dependence of the dwell-time distribution would constrain $w$ without relying on combinatorial arguments.","Another testable extension is to vary the electrode diameter at fixed thickness: the entropic model predicts the activation radius should remain near 17 nm, while a model that ties sub-volumes to grain size would predict a different scaling.","Finally, the demonstrated Gaussian, shiftable tuning curves suggest that populations of such junctions could implement basis-function coding; a concrete next step is checking whether the peak-shift behaviour survives when many junctions are averaged in parallel."],"forward_implications":["Dwell-time statistics of superparamagnetic MTJs can be described by the Arrhenius law once the entropic prefactor is included, with an activation volume of roughly 17 nm radius rather than the full electrode.","The bias-voltage effect separates by MgO thickness: spin-transfer torque dominates thin barriers and voltage-induced anisotropy change becomes relatively stronger for thick barriers.","Switching rate as a function of magnetic field or voltage is a Gaussian whose peak position can be shifted by the other input, satisfying the tuning-curve requirements of population-coding neural networks.","Quasistatic coercivity measurements become time-scale dependent: if the measurement time exceeds the mean dwell time, the apparent coercive field drops to zero even though the intrinsic anisotropy remains.","Using the reduced activation volume instead of the full electrode volume restores physically plausible saturation magnetization values for these CoFeB films."],"supporting_citations":[{"why":"Provides the exponential thermal-activation expression for switching rates that the paper fits to dwell times.","marker":"[13]"},{"why":"Establishes the single-domain thermal-fluctuation model whose predicted dwell times are orders of magnitude too long here.","marker":"[14]"},{"why":"Source of the entropy term S = k_B ln w and of the demonstration that topological magnetic objects can have huge pathway counts.","marker":"[15]"},{"why":"Reports superparamagnetic behavior in ultrathin MgO/CoFeB/Ta structures and supports the granularity hypothesis used to interpret the activation volume.","marker":"[7]"},{"why":"Introduces the magnetic activation volume concept used to convert ΔE/K into a reduced volume.","marker":"[19]"},{"why":"Provides the typical domain-wall width in CoFeB films with perpendicular anisotropy, used to validate the 17 nm activation radius.","marker":"[20]"},{"why":"Gives the voltage-induced anisotropy change coefficient for MgO/CoFeB/Ta, used to compare the extracted β value.","marker":"[24]"},{"why":"Defines the tuning-curve requirements for population coding that the measured Gaussian switching-rate curves are claimed to satisfy.","marker":"[27]"}],"fun_headline_variants":["Entropy of pathways explains superparamagnetic switching rates","17-nm activation volume resolves MTJ dwell time paradox","Field-tunable stochastic switching from entropic paths in MTJs","17-nm volume gives tunable stochastic bits for neural nets"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The 17 nm activation radius rests on the assumption that switching can nucleate in any of N sub-volumes and that the number of distinct pathways is N!, so a different relationship between pathway count and sub-volume count would change the inferred radius and could erase the agreement with the energy-based estimate.","fun_headline_variants_meta":{"raw":{"variants":["Entropy of pathways explains superparamagnetic switching rates","17-nm activation volume resolves MTJ dwell time paradox","Field-tunable stochastic switching from entropic paths in MTJs","17-nm volume gives tunable stochastic bits for neural nets"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000244,"raw_usage":{"total_tokens":1537,"prompt_tokens":958,"completion_tokens":579,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":574,"completion_tokens_details":{"reasoning_tokens":510}},"tokens_in":574,"tokens_out":579,"duration_ms":5878,"temperature":1.0,"reasoning_tokens":510,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:53:16.312013+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure dwell times on electrodes with different diameters (e.g., 100 nm, 140 nm, 200 nm) at fixed thickness and temperature; if the entropic model is right, the inferred activation radius stays near 17 nm regardless of electrode size, whereas if nucleation is tied to grain size or the full electrode, the radius would scale with electrode size.","supporting_citations":[{"cited_title":"Petracic","cited_arxiv_id":null,"evidence_quote":"Provides the exponential thermal-activation expression for switching rates that the paper fits to dwell times."},{"cited_title":"Thermal fluctuations of a single-domain particle","cited_arxiv_id":null,"evidence_quote":"Establishes the single-domain thermal-fluctuation model whose predicted dwell times are orders of magnitude too long here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the entropy term S = k_B ln w and of the demonstration that topological magnetic objects can have huge pathway counts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Reports superparamagnetic behavior in ultrathin MgO/CoFeB/Ta structures and supports the granularity hypothesis used to interpret the activation volume."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the magnetic activation volume concept used to convert ΔE/K into a reduced volume."},{"cited_title":"Yamanouchi, A","cited_arxiv_id":null,"evidence_quote":"Provides the typical domain-wall width in CoFeB films with perpendicular anisotropy, used to validate the 17 nm activation radius."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the voltage-induced anisotropy change coefficient for MgO/CoFeB/Ta, used to compare the extracted β value."},{"cited_title":"Salinas and L","cited_arxiv_id":null,"evidence_quote":"Defines the tuning-curve requirements for population coding that the measured Gaussian switching-rate curves are claimed to satisfy."}],"review_version":1}