{"id":"8a5b4d4c-1232-4583-bd71-0ef570b50315","arxiv_id":"2607.16566","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A series-resistor-plus-LSMO two-terminal circuit amplifies small AC signals up to ~11.5x when biased into the negative differential resistance window of the metal-insulator transition.","lead":"This device paper shows that a simple two-terminal metal-insulator transition (MIT) resistor can amplify small voltage signals up to about 11.5 times when biased into its negative-differential-resistance regime. The approach offers a compact, axon-like building block for neuromorphic circuits and can amplify spikes from a VO2-based artificial neuron.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (3)'s gain prediction assumes the quasi-static I-V slope persists at 1 kHz; thermal/kinetic effects could invalidate this, and no frequency-dependent check is provided.","rationale":"The empirical phenomenon—voltage amplification in a divider with an NDR element—is physically plausible and directly evidenced by the time traces and the replayed neuristor spike experiment. So I do not challenge the existence of some gain. The load-bearing weakness is the model's claim to predict that gain from a quasi-static I-V characteristic. The entire derivation in Suppl. A is an identity for a pure resistor; the physics enters when ΔI/ΔV is identified with the static slope. Whether that identification holds at 1 kHz is an empirical question the paper leaves open. The agreement in Fig. 3b is at a single frequency and a single operating point; it cannot distinguish between the quasi-static slope being the correct dynamic conductance and the fortuitous matching of a smoothed slope. A frequency-resolved impedance measurement would settle it. The reader's weakest assumption is the same; I agree. The verdict remains CONDITIONAL pending this test.","tokens_in":11033,"tokens_out":13696,"duration_ms":148012,"concrete_test":"Measure the AC gain and the device's small-signal differential conductance at the same bias conditions (VDC = 4.59 V, AC amplitude = 40 mV, Rs = 400 Ω) as a function of frequency from 10 Hz to 100 kHz, using a lock-in amplifier for the impedance and an oscilloscope for the gain. Extract the quasi-static ΔI/ΔV from the I-V ramp at the corresponding DC bias. If the real part of the dynamic conductance at 1 kHz differs from the quasi-static slope by more than ~15%, or the gain deviates from Eq. (3) beyond this margin at any frequency, the quasi-static assumption is falsified and the model is not generally predictive.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central model, Eq. (3) (gain = 1 - Rs·ΔI/ΔV), is derived in Suppl. A under the small-signal assumption that the AC current and voltage are related by the differential conductance measured from a quasi-static DC I-V ramp (Fig. 3a, recorded with a Keithley 2450). However, the gain is measured at 1 kHz with a 40 mV AC excitation superimposed on a DC bias near the MIT switching threshold. At this operating point, the device is in the NDR region where the static I-V is both steep and nonlinear; the AC swing (40 mV, and in some panels up to 500 mV) can cross the switching threshold, entering a regime where the local slope is not uniquely defined due to hysteresis. More importantly, the MIT switching in LSMO involves an insulating phase barrier whose formation kinetics, along with Joule-heating-related thermal time constants, could make the dynamic differential conductance at 1 kHz differ from the quasi-static ΔI/ΔV. The paper provides no frequency-dependent gain data and no direct measurement of the small-signal dynamic impedance at the operating point. If the dynamic conductance is smaller, phase-shifted, or otherwise different, Eq. (3) will not predict the observed gain, undermining the claim that the gain can be predicted from easily measurable transport characteristics. This is the load-bearing point because the model is the article's proposed design rule for optimizing MIT-based amplifiers.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports AC signal amplification in a two-terminal La0.7Sr0.3MnO3 (LSMO) metal-insulator-transition (MIT) device connected in a voltage-divider circuit with a series resistor. When the device is DC-biased into the negative differential resistance (NDR) regime near the MIT switching threshold, a small 1-kHz AC excitation is amplified by up to about 11.5×. The authors show that the gain can be tuned by DC bias, AC amplitude, series resistance, and temperature, and they demonstrate amplification of a recorded VO2-neuristor spiking sequence. They propose a design rule, Eq. (3): AC gain = 1 - Rs·ΔI/ΔV, in which the gain is determined from the measured I-V characteristic, and they argue that NDR is the enabling mechanism. The paper claims this establishes a general framework for axon-like amplification in nonlinear electronic materials.","tokens_in":11222,"tokens_out":6332,"duration_ms":75970,"significance":"If the claims hold, this is a useful conceptual and practical result: it shows that a simple, compact two-terminal device based on a MIT material can provide local signal amplification in a neuromorphic circuit, without a conventional CMOS amplifier, and it identifies an easily measurable criterion (negative ΔI/ΔV) for above-unity gain. The demonstration of spike amplification using a real VO2 neuristor is a concrete step toward integrated artificial neurons and axons. The paper is clearly written, the circuit derivation in Supplementary A is elementary and correct, and the parameter sweeps in Fig. 2 give a useful map of the operating window. However, the quantitative claims are weakened by the absence of error bars or repeated measurements, by the ambiguous definition of gain for the strongly distorted output waveforms, and by the lack of any frequency-dependent check of the quasi-static I-V slope used in the model. These are load-bearing issues for the paper's central 'model predicts gain' claim and for the reproducibility of the ~11.5× figure.","major_comments":[{"comment":"Eq. (3) is derived as an algebraic identity from Kirchhoff's voltage law: gain = 1 - Rs·(ΔI_AC/ΔV_AC,in). As such, it is exact by construction for any circuit obeying that relation. The only predictive content comes from replacing ΔI_AC/ΔV_AC,in with the quasi-static ΔI/ΔV obtained from the I-V ramp in Fig. 3a. The manuscript provides no evidence that the dynamic small-signal conductance at 1 kHz equals the quasi-static slope. Phase-transition kinetics, Joule-heating thermal time constants, or capacitive/RC effects could make the dynamic differential conductance smaller, phase-shifted, or hysteretic, particularly when the 40-mV AC swing approaches the abrupt switching threshold. The paper should either (a) measure the gain at several frequencies (e.g., 100 Hz to 100 kHz) and show that the prediction holds, or (b) measure the small-signal dynamic impedance at the operating point. As writt","section":"Fig. 3a-b and Supplementary A"},{"comment":"The AC gain is defined as V_AC,out/V_AC,in, but the output waveform in the high-gain regime is visibly distorted, with abrupt jumps at the MIT switching threshold. No information is given about how the AC amplitude was extracted: peak-to-peak, RMS, fundamental Fourier component, or some other convention. For a strongly nonlinear waveform, these definitions give very different numbers, and the reported 11.5× may be dominated by harmonic content rather than by linear small-signal amplification. The authors should state the amplitude extraction method, report total harmonic distortion (THD) or the amplitude of the fundamental, and ideally separate 'small-signal gain at 1 kHz' from 'peak-to-peak amplification of a distorted waveform'. This is essential for evaluating the quantitative claim and for comparing with future work.","section":"Eq. (2) and Fig. 1d"},{"comment":"All quantitative claims (gain values, tuning curves, temperature map, spike amplification factor of ~5×) are presented without error bars, without the number of repeated measurements, and without specifying how many nominally identical devices were tested. The abstract calls the amplification 'robust,' but there is no statistical evidence. At minimum, the authors should show representative repeated traces or include error bars/standard deviations in Figs. 2b-d, and state the number of devices and measurement cycles. This is needed to assess device-to-device and cycle-to-cycle variability, especially because the MIT switching is stochastic and hysteretic.","section":"Figs. 2 and 4"},{"comment":"The derivation in Supplementary A is correct but is a rearrangement of the definitions, not a physical model. The paper should explicitly state that Eq. (3) is an exact circuit relation and that the model's usefulness depends entirely on the quasi-static/dynamic equivalence described in the first major comment. This is not an error, but framing it as 'we propose a model that predicts the gain' is misleading unless the dynamic equivalence is tested. Please adjust the wording and add the missing validation.","section":"Supplementary A, Eq. (3)"}],"minor_comments":[{"comment":"Equation (1) is confusing as written: V_in = V_FG + (50 Ω)·I. The text says the FG has a built-in 50-Ω series resistor, so the effective series resistance is Rs + 50 Ω. Please clarify the sign convention, the physical meaning of V_in, and why V_in differs from V_FG. As written, a reader cannot reproduce the calculation of the 'effective series resistance' or the gain without guessing.","section":"Eq. (1)"},{"comment":"The caption says 'I-V characteristic of an LSMO device in series with a 400 Ω resistor,' but the main text says 'of an LSMO device in series with a 400 Ω resistor.' Please clarify whether the voltage axis is the total applied voltage (across device plus series resistor) or the voltage across the LSMO device alone. This is important because Eq. (3) uses ΔI/ΔV, and the derivative should be taken with respect to the input voltage, not the device voltage.","section":"Fig. 3a"},{"comment":"The caption mentions 'peak differential conductance (orange)' but the axes and units are not visible. Please label the y-axis clearly (e.g., ΔI/ΔV in S or mS) and specify how the 'peak' was extracted from the noisy derivative.","section":"Fig. 3d"},{"comment":"There is a typo: 'NRD regime' should be 'NDR regime.'","section":"Discussion, last paragraph before Summary"},{"comment":"The inset claims the spike shape is 'well-preserved,' but the output shows visible differences in amplitude and shape. Please quantify the preservation (e.g., normalized cross-correlation, spike-width error) or soften the wording.","section":"Fig. 4b"},{"comment":"The paper does not state the measurement bandwidth, the oscilloscope input impedance, or whether parasitic capacitances were considered. Given the 1-kHz operating frequency, these are probably minor, but a brief note would improve reproducibility.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The core observation — above-unity AC gain in a voltage divider made of a series resistor and an MIT device biased into NDR — appears real and is worth publishing after the authors address the frequency-response validation, gain-definition ambiguity, and reproducibility statistics. These are fixable with additional experiments and revision, so I recommend major revision rather than rejection. The authors should also be encouraged to present Eq. (3) as an exact circuit identity and to state clearly that the predictive claim rests on the quasi-static/dynamic conductance equivalence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a solid, readable experimental paper. The core result—AC voltage gain up to ~11.5x in a voltage divider built from a linear resistor and an LSMO two-terminal device biased into its NDR regime—is new for a conventional MIT material and clearly supported by the I-V and gain traces. The spike-amplification demo is a bonus, even if it uses a recorded VO2 spike train rather than a live coupled neuristor.\n\nWhat the paper does well: the circuit physics is transparent, the tuning of gain via bias, AC amplitude, series resistance, and temperature is mapped, and the link between gain and negative dI/dV is directly demonstrated. The comparison to Brown et al.'s ~1.6x in LaCoO3 is honest and puts the improvement in context.\n\nThe soft spots are about the model and the evidence. Eq. (3) is an algebraic identity—it's just Kirchhoff's voltage law divided by dV_in, with dI/dV taken from the measured I-V curve. Calling it a 'model that predicts gain' overstates what it does; it's a compact re-description of the measurement. That is still useful as a design rule—it tells you to look for a sharp NDR with minimal hysteresis—but it is not an independent prediction from material properties. I don't see that as a fatal flaw, just a wording problem.\n\nMore substantial: there are no error bars or statistics anywhere. The 11.5x comes from a single trace at one bias point with a 40 mV excitation and the output is visibly distorted. The paper acknowledges the distortion, but I would want at least a few repeated sweeps to know how stable those numbers are. Also, the reported gain is purely voltage gain; there is no power or energy analysis, which matters when you pitch this for neuromorphic hardware. Operation is at 80 K; the room-temperature claim rests on prior work, not this paper.\n\nThe stress-test worry about quasi-static dI/dV not holding at 1 kHz is not borne out by the paper's own Fig. 3b, where the gain calculated from the quasi-static I-V tracks the measured 1 kHz gain quite well. That doesn't prove the dynamic conductance stays identical at all frequencies, and a frequency sweep would be cheap to do and would settle it, but the evidence in the paper suggests the assumption is reasonable at the operating point tested.\n\nBottom line: this deserves a serious referee. The core observation is likely correct and the simplicity of the device is a genuine plus. I'd send it to review with a request for error bars, a frequency check, and a softer claim about what Eq. (3) actually is.","headline":"A clean, useful demonstration of NDR-based voltage gain in a two-terminal LSMO device, with a model that is more a rearrangement of the measured I-V than a true prediction, and a need for error bars and a frequency check before trusting the 11.5x.","tokens_in":11887,"tokens_out":3073,"would_cite":true,"duration_ms":35176,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A simple voltage divider made of one linear resistor and one metal-insulator device amplifies small AC signals by up to ~11.5x when the device is biased into its negative differential resistance regime.","keywords":["metal-insulator transition","negative differential resistance","signal amplification","axon emulation","neuromorphic computing","volatile resistive switching","LSMO","voltage divider"],"falsifier":"Measure the AC gain at several frequencies (e.g., 100 Hz to 1 MHz) under the same DC bias and series resistance, and compare with Eq. (3) computed from the quasi-static I-V curve; if the gain changes substantially with frequency or the device cannot repeatedly return to the same NDR branch each cycle, the quasi-static model is invalid.","tokens_in":10790,"feed_emoji":"⚡","tokens_out":3457,"duration_ms":37052,"temperature":0.7,"pith_summary":"The paper claims that a two-terminal device made of a metal-insulator transition material, when biased at the verge of its phase transition, acts as a high-gain amplifier of small AC signals, achieving up to ~11.5x amplification. The mechanism is that the device's negative differential resistance reverses the sign of the current response, so the AC voltage across the device grows rather than shrinks when the input rises. The paper derives a simple predictive rule: the AC gain equals one minus the series resistance times the differential conductance of the device, meaning anything with a steep negative dI/dV can amplify. It demonstrates amplification of realistic neuron-like spikes from a VO2 oscillator, pointing to compact axon-like amplifiers for neuromorphic circuits. A sympathetic reader would care because this offers a scalable alternative to CMOS amplifiers for large neural networks.","feed_headline":"Amplify signals 11.5x with one resistor and a phase-change device","feed_subtitle":"A two-terminal metal-insulator amplifier, biased at its transition, predicts gain from the I-V curve and could shrink neuromorphic hardware.","key_machinery":"The central object is the negative differential resistance (NDR) region of the metal-insulator transition device's I-V characteristic, accessed by DC-biasing at the onset of MIT switching. NDR is the regime where current decreases as voltage increases (dI/dV < 0). The load-bearing identity is AC gain = 1 - Rs·(dI/dV): because dI/dV is negative in the NDR region, the subtracted term flips sign, so an increase in AC voltage reduces the current, lowering the voltage drop across the series resistor and increasing the voltage across the device itself, which is the output. The quasi-static differential conductance dI/dV, measured from a slow voltage ramp, serves as the input to the model.","core_discovery":"Biasing an LSMO device into the negative differential resistance (NDR) region of its I-V curve yields AC gain up to ~11.5x in a voltage-divider circuit, and the gain is quantitatively captured by AC gain = 1 - Rs·(dI/dV), where dI/dV is taken from the quasi-static I-V characteristic. The gain is largest when the I-V curve shows abrupt switching with minimal hysteresis, and it can be controlled by DC bias, AC amplitude, series resistance, and temperature. The device also amplifies spiking sequences from a real VO2 neuristor with a factor of ~5x while preserving spike shape, demonstrating practical axon-like functionality.","pith_inferences":["If the quasi-static dI/dV model holds, the gain should be approximately frequency-independent up to the speed of the phase transition; a testable prediction is that gain will roll off when the AC period approaches the switching time constant.","Because larger AC excitations reduce gain, the circuit has an inherent nonlinear compression or limiting behavior that could be exploited for signal normalization in spiking networks.","The principle may extend to other first-order phase-transition or volatile resistive switching systems beyond LSMO, provided the switching is repeatable and the NDR branch can be held stably for long times.","The DC-bias-controlled attenuation/amplification transition could be used as an analog gain clamp or automatic gain control element in neuromorphic or conventional analog circuits, going beyond axon emulation."],"forward_implications":["The same amplifier principle should apply to any material with a steep, dynamically accessible NDR; N-type NDR materials can be used directly, while S-type NDR materials like VO2, NbO2, and SmNiO3 could work if the circuit is modified for current-controlled conditions.","Varying the DC bias switches the circuit between attenuation (PDR) and amplification (NDR), enabling a single device to act as both an axon-like amplifier and a tunable gain/synaptic-weight element.","Amplifying spikes from a VO2-based neuristor shows that MIT-based neurons and axon-like amplifiers can be integrated directly in the same materials platform.","Nonvolatile memristors with NDR are unsuitable because their NDR is not dynamically accessible to small AC signals; volatile switching is essential for the amplification mechanism.","Optimizing gain requires maximizing the NDR while minimizing I-V hysteresis, giving a concrete materials-selection rule for future axon-like amplifiers."],"fun_headline_variants":["11.5x gain from a two-terminal phase-change device","Axon-like amplifier: 11.5x gain in a single resistor circuit","Phase-transition device amplifies neural spikes 11.5x","Metal-insulator amplifier: gain predicted from I-V curve","Simple two-terminal device boosts signals 11.5x"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The model assumes the device's small-signal response at the operating frequency is fully described by the quasi-static I-V slope dI/dV; if phase-transition kinetics, thermal time constants, or capacitive effects alter the dynamic conductance, the predicted gain and the proposed amplification mechanism will not hold.","fun_headline_variants_meta":{"raw":{"variants":["11.5x gain from a two-terminal phase-change device","Axon-like amplifier: 11.5x gain in a single resistor circuit","Phase-transition device amplifies neural spikes 11.5x","Metal-insulator amplifier: gain predicted from I-V curve","Simple two-terminal device boosts signals 11.5x"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000727,"raw_usage":{"total_tokens":3089,"prompt_tokens":737,"completion_tokens":2352,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":2263}},"tokens_in":481,"tokens_out":2352,"duration_ms":16559,"temperature":1.0,"reasoning_tokens":2263,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T20:34:09.432527+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the AC gain at several frequencies (e.g., 100 Hz to 1 MHz) under the same DC bias and series resistance, and compare with Eq. (3) computed from the quasi-static I-V curve; if the gain changes substantially with frequency or the device cannot repeatedly return to the same NDR branch each cycle, the quasi-static model is invalid.","supporting_citations":[],"review_version":1}