{"id":"c5a6b6e4-8b2c-4a06-b571-7b9059705ed4","arxiv_id":"2607.20040","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"MgO magnetic tunnel junctions show simultaneous ~45% tunnel magnetoresistance and ~100x non-volatile resistive switching, with reversible electrical on/off control of the magnetic response.","lead":"Researchers made a single magnetic tunnel junction that works as both a magnetic field sensor and a reprogrammable resistor. The device keeps its magnetic sensing and can switch its resistance electrically, opening a route to reconfigurable spintronic circuits.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reversibility claim rests on unquantified recovery of the MgO barrier; the paper itself acknowledges incomplete filament reset.","rationale":"The reader's weakest_assumption identifies exactly the same load-bearing premise: that reset fully reconstructs the original MgO barrier. My reading of the full text confirms this is the central risk. The paper provides a single R(H) recovery curve and one Simmons fit, while explicitly acknowledging in Fig. 2a that the filament is not always 100% recovered. The missing data are paired TMR-recovery statistics over many cycles. Because the reader's CONDITIONAL verdict already captures this uncertainty with appropriate confidence, my independent assessment does not change the verdict. I do not see a more serious internal inconsistency or a fatal error; the electrical and magnetic measurements are internally coherent and the parallel-channel model is plausible. The main remaining issue is statistical support for reversibility, not a demonstrated physical flaw.","tokens_in":15350,"tokens_out":4186,"duration_ms":39857,"concrete_test":"Perform a paired endurance test on a single 55-μm device: repeat set/reset for at least 200 cycles, and after every reset measure a full R(H) loop at the same low bias as in Fig. 1b. Plot extracted TMR and HRS zero-field resistance versus cycle number, and compare both to the pre-forming values. If TMR or HRS resistance drifts monotonically, or if the spread exceeds the pre-forming measurement uncertainty (e.g., >5% TMR or >10% resistance), the 'fully reversible without degradation' claim is falsified. As a second check, fit each post-reset HRS I-V with the same Simmons model (Eq. 1) and verify that the extracted barrier thickness and height remain within the reported ±0.2 nm and ±0.01 eV of the pre-forming fit, not just once but across cycles.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central differentiator of this work is the claim that memristive switching can reversibly and completely suppress and recover the spintronic (TMR) functionality. The load-bearing premise is that the reset process fully reconstructs the original MgO tunnel barrier, so that the recovered TMR reflects a regenerated coherent tunneling channel and not a residual non-magnetic shunt. The evidence, however, is limited to a single R(H) cycle after reset (Fig. 3b) and a Simmons fit of one HRS I-V curve (Fig. 2c). The paper's own Fig. 2a caption states that the wide HRS dispersion is 'expected since the filament is not always 100% recovered in the reset process' — an explicit admission that the barrier is not always fully reconstructed. No endurance statistics link TMR recovery to cycling: the paper says R(H) measurements were performed 'for several cycles' without degradation, but no quantitative data, number of cycles, or error bars are provided. If the reset only partially dissolves the filament or leaves a permanent modified region, the recovered TMR may be degraded in ways not captured by a single representative trace, and the central claim of reversible, non-degrading multifunctionality fails after repeated cycling. Thus the weakest assumption is not merely hypothetical; it is contradicted by the observed HRS variability, and the claim of full barrier recovery is unsupported by the presented statistics.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports MgO-based magnetic tunnel junctions that simultaneously exhibit a linear, low-hysteresis TMR response (~45%) suitable for magnetic field sensing and a non-volatile, bipolar memristive resistance-switching effect (HRS/LRS ratio up to ~100). The authors show that the TMR is suppressed when the device is set to the LRS and recovered after reset to the HRS, and that intermediate resistance states can be programmed quasi-analogously. They also show area-dependent switching, attribute the LRS to a filamentary conduction path, and demonstrate that inserting an ultrathin Ta layer in the MgO barrier lowers the forming energy by ~20% at the cost of reducing TMR to 25%. The central claim is that memristive switching can reversibly and completely suppress and recover the spintronic functionality without degradation of either property.","tokens_in":15574,"tokens_out":3621,"duration_ms":40743,"significance":"If the reversibility claim holds, this is a significant advance: it would provide a single device that can be electrically reconfigured between a sensitive magnetic sensor and a high-resistance memristive state, opening a route to reprogrammable spintronic circuits and multifunctional sensor arrays. The paper's strengths include a clear device stack and fabrication description, a direct two-terminal pulsed measurement protocol, and a simple two-channel shunt model whose signature (linear TMR vs RxA relation in Fig. 3d) is explicitly verified. The area dependence of HRS/LRS (Fig. 4) is consistent with filamentary switching. However, the central reversibility claim rests largely on a single representative reset cycle and on the interpretation of a Simmons fit, and the paper lacks the statistical evidence needed to establish that full barrier recovery is repeatable. The result is credible but not yet convincingly supported as a general property of the devices.","major_comments":[{"comment":"The paper's key differentiator is that the TMR can be 'completely suppress[ed] and recover[ed]' after reset, but the evidence is one R(H) cycle in Fig. 3b and the statement that measurements 'have been performed for several cycles, showing systematic results without degradation.' No number of cycles, quantitative recovery values, or error bars are provided. Moreover, the Fig. 2a caption explicitly states that the HRS dispersion is 'expected since the filament is not always 100% recovered in the reset process.' This admission directly weakens the claim of full barrier recovery; at minimum, the manuscript must quantify the distribution of recovered TMR and resistance after many set/reset cycles and show that the recovery is not accompanied by a progressive degradation of the magnetic response.","section":"§2.2, Fig. 2a and Fig. 3b"},{"comment":"The Simmons fit is used as evidence that 'the original barrier is recovered in the reset process,' but the fit is performed on a single HRS I-V curve and has two free parameters (barrier thickness t and average barrier height φ/2). The extracted values may also be consistent with a partially dissolved filament that leaves a non-magnetic, tunneling-like conduction path in parallel with the spin-dependent channel. The linear TMR–RxA relation in Fig. 3d is indeed the signature of a parallel shunt, but it does not identify the physical nature of the shunt. To support the barrier-recovery claim, the authors should show fits to multiple HRS cycles, report fitting residuals and the range of the data used, and compare the HRS I-V with the pristine (pre-forming) I-V. Without this, the claim that the original coherent tunneling channel is regenerated is not established.","section":"§2.2, Eq. (1) and Fig. 2c"},{"comment":"The paper states that 'all studied areas show a similar RxA, TMR, and memristive properties,' but the detailed characterization is presented for a single 55 μm device, and many key plots (RHSL, retention, resistive-switching distributions, intermediate-state R(H) curves) lack error bars or device-to-device statistics. For a device-oriented claim of multifunctionality and reproducibility, the absence of raw data and statistical measures (e.g., device counts, standard deviations, box plots) makes it impossible to assess whether the observed coexistence is a robust property or a singular result. The authors should provide statistics across several devices and cycles, at least for the main claims of TMR value, HRS/LRS ratio, and TMR recovery.","section":"§2.2, Fig. 1d, Fig. 2a, Fig. 3, and Experimental Section"}],"minor_comments":[{"comment":"The symbol φ is used in the equation and later defined as ϕ/2 (average barrier height). Please clarify whether φ in Eq. (1) is the same as the average barrier height or the full barrier height; the current notation is confusing.","section":"Eq. (1)"},{"comment":"In stack 1, the target RxA is given as '2.4 MΩ μcm2'; this should presumably be μm², and the same unit is used as MΩ μm² earlier. Please make the notation consistent.","section":"Experimental Section"},{"comment":"The abbreviation for the resistance hysteresis switching loop is inconsistent: the text uses both 'RSHL' and 'RHSL' (Fig. 1d caption vs. §2.2). Please unify to one abbreviation.","section":"Throughout"},{"comment":"The text says the standard deviation of Vset and Vreset is 0.5 V, 'which is lower than the step utilized to measure the IV curve (0.1V).' Since 0.5 V > 0.1 V, this appears to be either a typo or a mistaken comparison; please verify the intended statement.","section":"Fig. 2a inset"},{"comment":"The '≈100pJ' energy value for the present work likely refers to the reset operation, but the text gives 5 and 100 pJ for set and reset, and 350 pJ for forming. Please clarify which operation is quoted in the table and keep the notation consistent.","section":"Table 1"},{"comment":"Reference [49] has formatting errors ('M,C.,' and 'PREPRINT 2023'); please correct the citation.","section":"Reference list"}],"recommendation":"major_revision","confidential_remarks":"This is a conditional paper: the central observation of coexisting TMR and resistive switching appears real and is nicely supported by the shunt-model consistency, but the manuscript's primary differentiating claim—full and repeatable recovery of the magnetic functionality after reset—is asserted rather than demonstrated with sufficient statistics. The authors should be asked to provide endurance data and quantitative recovery distributions, and to soften the claims if such data cannot be supplied. The Simmons-fit interpretation also needs stronger support; otherwise the alternative explanation of a residual non-magnetic conduction path remains open. The paper fits the scope of Advanced Functional Materials but is not ready in its current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: This is the first MTJ+memristor paper I've seen where the coexistence numbers actually look good — ~45% TMR and ~100x switching ratio in the same device — and the central observation appears real rather than a fitting artifact. The weak point is the 'fully reversible' TMR claim, which is supported by one recovery trace and sits in tension with the paper's own admission that the filament is not always 100% recovered.\n\nWhat's new: Prior work either killed TMR during electroforming (refs 30, 34, 46) or got marginal coexistence (ref 31: TMR ~10%, ratio ~3). Here they get both effects at levels that could be useful, plus quasi-analog intermediate states and 10 ns switching. The two-channel shunt model is simple and the linear TMR–RxA relation in Fig 3d fits it. The Simmons fit on the HRS gives barrier thickness 1.3 nm and height 1.07 eV, consistent with a decent MgO barrier. The 200-cycle RHSL and 10^4 s retention data show the memristive side is genuine. Area dependence of HRS vs LRS supports a single-filament picture.\n\nSoft spots, in order of size. First, the reversibility differentiator. The paper says 'when the device is returned to the HRS after the reset process the TMR is recovered to its original value with no apparent degradation.' That is shown for one R(H) cycle in Fig 3b, plus a text note that 'several cycles' behaved systematically without plotting them. Meanwhile, Fig 2a's own caption says the large HRS dispersion is 'expected since the filament is not always 100% recovered in the reset process.' That is an explicit admission that full barrier reconstruction is not routine. A single Simmons fit cannot rule out a small residual filament. So the claim that the spintronic functionality can be completely suppressed and recovered without degradation is stronger than the data. This is a moderate concern; it doesn't sink the paper, because there is at least one clean recovery and the HRS resistance after reset is near the pristine value. But the reconfigurability pitch needs endurance statistics where TMR is measured after every reset, not just one trace.\n\nSecond, the quantitative core rests on one 55 μm device. They say all areas behave similarly and show area trends, but there are no error bars on the main claimed numbers (45% TMR, ratio ~100). Data availability is 'upon reasonable request.' That is a reproducibility soft spot, not a red flag.\n\nThe stress-test note flags the same reversibility issue and is right. I would not call it a load-bearing flaw because the authors do disclose the incomplete-reset behavior and the demonstrated recovery is real, if thin.\n\nWho should read this: applied spintronics and memristor groups interested in co-integration. It is a useful citation for coexistence performance comparisons. I would send it to a serious referee, with the request to add cycling-dependent TMR data and raw error bars. As a peer reviewer, I would accept with revisions.","headline":"Genuine step beyond the prior MTJ+memristor work, with the reversibility claim as the main soft spot.","tokens_in":16149,"tokens_out":3718,"would_cite":true,"duration_ms":37358,"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":"MgO magnetic tunnel junctions work as both linear magnetic sensors and reversible non-volatile memristors.","keywords":["magnetic tunnel junction","tunnel magnetoresistance","memristor","resistive switching","MgO barrier","multifunctional device","spintronics","neuromorphic computing"],"falsifier":"Cycle the device for at least 10,000 set/reset operations and measure the tunneling magnetoresistance after each reset; if the TMR decays or the high-resistance state drifts downward, the reversibility claim fails. Additionally, cross-sectional transmission electron microscopy of a device in the reset state should show an intact MgO barrier with no residual filament connecting the electrodes.","tokens_in":15165,"feed_emoji":"🧲","tokens_out":2383,"duration_ms":28875,"temperature":0.7,"pith_summary":"The paper shows that a single MgO-based magnetic tunnel junction can simultaneously behave as a high-quality magnetic field sensor and as a non-volatile, quasi-analog memristor. The key advance is that the memristive state can be switched reversibly to completely suppress and then restore the magnetoresistance, so neither functionality degrades the other. If correct, this means a conventional MTJ stack can be reconfigured after fabrication, opening practical routes to reprogrammable spintronic circuits and neuromorphic hardware without exotic materials.","feed_headline":"Magnetic sensor and analog memory share one MgO junction","feed_subtitle":"A single device keeps 45% tunneling magnetoresistance while switching resistance by 100x, enabling reprogrammable spintronics.","key_machinery":"The central mechanism is a parallel two-conduction-channel model: one channel is the coherent spin-polarized tunneling through the MgO barrier that produces the tunnel magnetoresistance; the other is a voltage-tunable conductive filament (oxygen vacancies) that shunts the barrier and carries no spin polarization. The variable parallel resistance accounts for the memristive switching, the suppression of TMR in the low-resistance state, and the linear TMR-versus-resistance-area relationship, and its full dissolution in the reset step restores the pristine tunneling properties.","core_discovery":"The paper demonstrates that a 55 μm CoFeB/MgO/CoFeB magnetic tunnel junction exhibits a tunneling magnetoresistance of about 45% in the high-resistance state and a bipolar resistive switching ratio of about 100 between the high- and low-resistance states, at room temperature. The low-resistance state shorts out the tunneling magnetoresistance, and the reset process restores the original barrier, recovering the TMR to its pre-forming value. The resistance can be set to intermediate levels by partial reset pulses, with a linear correlation between the TMR and the resistance–area product, explained by a two-channel model: a spin-dependent tunneling path in parallel with a variable, spin-indepen","pith_inferences":["If the filament is indeed oxygen-vacancy based, then controlling oxygen stoichiometry or using reactive electrode materials could further tune forming voltage, endurance, and retention in these MTJs.","The reversibility claim rests on the reset pulse fully dissolving the filament; an endurance test over thousands of cycles with TMR recovery measured after each reset would be a natural stress test of this assumption.","The two-channel model is generic: any defect or parasitic conduction path in an MTJ could be exploited to program TMR, which might extend to STT-MRAM cells as multi-level analog synapses.","The Ta doping trade-off suggests that other dopants or barrier engineering might achieve even lower switching energy while preserving a larger fraction of the TMR."],"forward_implications":["A single MTJ device can be used as both a magnetic field sensor and a non-volatile memory element, enabling reconfigurable sensor bridges that can be switched into gradiometer mode after fabrication.","Because the memristive state can fully suppress and then recover the TMR, spintronic circuits can have their functionality reprogrammed post-fabrication by a simple pulsing protocol.","The nanosecond write pulses and picojoule-level energy consumption put these devices in the range needed for neuromorphic synapses and crossbar arrays.","The observed area scaling suggests that smaller junctions will have larger resistance ratios, so the multifunctional behavior should improve when scaled toward nanometer dimensions.","Ta doping of the MgO barrier reduces the forming energy by about 20%, providing a concrete materials route to low-power operation, albeit with reduced TMR."],"fun_headline_variants":["MTJ double duty: magnetoresistance and memristive switching","One MgO junction enables magnetic sensing and analog memory","Coexisting TMR and memristance in a single MTJ device","MgO barrier gives linear TMR and 100x resistance states","Reversible spintronic gating via memristive states in MgO"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The reset process completely and repeatedly removes the conductive filament, so the recovered tunneling magnetoresistance comes from a reconstructed MgO barrier rather than from a new non-magnetic conduction path.","fun_headline_variants_meta":{"raw":{"variants":["MTJ double duty: magnetoresistance and memristive switching","One MgO junction enables magnetic sensing and analog memory","Coexisting TMR and memristance in a single MTJ device","MgO barrier gives linear TMR and 100x resistance states","Reversible spintronic gating via memristive states in MgO"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000417,"raw_usage":{"total_tokens":2008,"prompt_tokens":788,"completion_tokens":1220,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":532,"completion_tokens_details":{"reasoning_tokens":1129}},"tokens_in":532,"tokens_out":1220,"duration_ms":11044,"temperature":1.0,"reasoning_tokens":1129,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T10:57:40.819600+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Cycle the device for at least 10,000 set/reset operations and measure the tunneling magnetoresistance after each reset; if the TMR decays or the high-resistance state drifts downward, the reversibility claim fails. Additionally, cross-sectional transmission electron microscopy of a device in the reset state should show an intact MgO barrier with no residual filament connecting the electrodes.","supporting_citations":[],"review_version":1}