{"id":"af4e3f7c-4723-40c5-9132-570e6a9e570a","arxiv_id":"2506.01139","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Optimized VO2 oscillator circuits reach 167 MHz oscillation, and reset relaxation is identified as the limiting factor for faster operation.","lead":"VO2 oscillator circuits redesigned with a transmission-line layout and ultra-small junction devices oscillate at up to 167 MHz, more than ten times faster than previous VO2 oscillators. The paper also identifies the memristor's reset relaxation time under oscillator bias, rather than circuit parasitics, as the practical speed limit for this approach.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Speed-ceiling claim rests on transition times measured at 103 MHz, not at the 167 MHz record; the 167 MHz trace's own reset fall time could settle it.","rationale":"The paper's main experimental contribution, the 167 MHz passively loaded VO2 oscillator, is directly measured and credible; the scaling collapse, the transmission-line simulations, and the 100 GHz verification at 103 MHz all support the record and the 100 MHz-range operation. The contested part is the negative ceiling claim. I examined the voltage-clamp assumption flagged by the reader and found it physically reasonable for this circuit: in a quasi-static transmission-line oscillator the memristor voltage is highest just before set and falls as the resistance drops, so large overshoots are not expected for the RS values required to reset. The reader's concern about engineered overshoot is thus less severe than it first appears. However, the numerical ceiling depends on the actual reset time at the maximum frequency, which is not reported. The pulse experiments are explicitly acknowledged by the authors to be non-representative (oscillator set time 1.2 ns vs 5.4 ns in pulses), and the Simulink model is fitted to the record. The only direct oscillator-condition measurement is at 103 MHz. Therefore the single most load-bearing gap is the absence of rise and fall time data at 167 MHz. A straightforward analysis of the existing trace, or a repeat measurement on the record device with the 100 GHz setup, would settle whether the ceiling claim is safe. This does not change the reader's CONDITIONAL verdict: the record stands, but the ceiling claim should be conditional on that measurement. I agree only partially with the reader's weakest-assumption diagnosis because the voltage-clamp assumption is not the main risk; the evidential gap at the maximum frequency is.","tokens_in":22512,"tokens_out":15635,"duration_ms":168119,"concrete_test":"Re-analyze the existing 167 MHz time trace (Fig. 1d): fit each falling edge to extract the 10-90% reset fall time and each rising edge for the set rise time, at the same V0 and circuit settings. Alternatively, repeat the 100 GHz bandwidth measurement of Fig. 6 on the specific device and circuit that produced the 167 MHz record, at the same drive voltage, and report the 10-90% rise and fall times. If the reset fall time at the record frequency is less than or about 2 ns, the claim that 'an oscillation significantly faster than 167 MHz is not realistic' is weakened because a 200 MHz oscillation with a 5 ns period could accommodate faster transitions; if the fall time is greater than or about 3 ns, the ceiling claim is directly supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is not just the 167 MHz record but the physical ceiling: 'an oscillation significantly faster than the fastest 167 MHz oscillation achieved is not realistic.' The load-bearing evidence for that ceiling is the oscillator-condition reset time. The only direct oscillator-condition measurement (Fig. 6, 100 GHz setup) yields a 10-90% reset fall time of 3.4 ns, but that trace oscillates at 103 MHz, not at the 167 MHz record. The pulse experiments (Fig. 5) give much longer set/reset times (5.4 ns/9.6 ns and up to 54 ns), but the authors themselves note these are not representative of oscillator operation: the set time in the actual oscillator is 1.2 ns, versus 5.4 ns for near-threshold single pulses. Thus the pulse data overestimate the switching times and cannot fix the ceiling numerically. The Simulink model reproduces 173 MHz with hand-picked tau_set=600 ps and tau_reset=1.5 ns, but because these constants are chosen to match the observed frequency, the model is a fit, not an independent bound. What is missing is the transition-time measurement at the actual maximum frequency: the 167 MHz trace in Fig. 1d was recorded with a 10 GS/s oscilloscope, so its falling edges contain the relevant reset time, but the paper does not report it. If the reset fall time at 167 MHz is much shorter than 3.4 ns, the ceiling could be higher; if it is about 3 ns, the ceiling claim is supported. This gap is the most load-bearing because the voltage-clamp argument alone does not specify the numerical relaxation time at the record condition.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports VO2 relaxation oscillators with a record 167 MHz oscillation frequency, more than an order of magnitude above the previous 9 MHz state of the art. The improvement is attributed to three factors: nanoscale devices with a confined active region and minimized stray capacitance, a transmission-line-based circuit layout with an in-line series resistor and matched 50-ohm readout, and the identification of internal VO2 relaxation times as the practical speed limit. The authors support these claims with LTspice transmission-line simulations, an analytic voltage build-up formula, measurements of oscillation frequency versus drive voltage and memristor-to-resistor distance, pulsed switching experiments, and Matlab Simulink models with finite set/reset time constants. They conclude that under oscillator bias the reset relaxation time is the key limiting factor and that oscillations significantly faster than 167 MHz are not realistic.","tokens_in":22873,"tokens_out":3961,"duration_ms":41394,"significance":"If the central claim holds, this is a substantial advance for VO2-based oscillatory neural networks: it moves individual passive VO2 oscillators from the few-MHz range into the 100 MHz range and identifies a concrete physical bottleneck, namely the reset relaxation time under oscillator conditions rather than circuit capacitance or wiring alone. The paper's strengths include direct time-domain observation of the 167 MHz oscillation, an independent 100 GHz-bandwidth measurement showing 1.2 ns rise and 3.4 ns fall times at 103 MHz, the collapse of the measured frequency data under v/d normalization, and an analytic voltage build-up formula that is tested against experiment. The authors are also honest about the limits of the simplest model: the scaling fails above about 75 MHz, and the 167 MHz record is stated to rely on a fine interplay of device parameters that is not reproduced in every circuit.","major_comments":[{"comment":"The speed-ceiling claim—that an oscillation 'significantly faster than the fastest 167 MHz oscillation achieved is not realistic'—rests on a 3.4 ns reset fall time measured on a 103 MHz oscillation, not on the 167 MHz record trace. The 167 MHz trace in Fig. 1d was recorded at 10 GS/s and contains reset edges; reporting their 10–90% fall time (and the corresponding set rise time) would directly test whether the ceiling is supported at the actual maximum frequency. If those edges are much faster than 3.4 ns, the ceiling would need to be revised upward; if they are approximately 3 ns, the claim is strongly supported. This is the most load-bearing missing datum in the paper.","section":"Investigation of the internal relaxation time-scales (Fig. 6)"},{"comment":"The Simulink model reproduces 173 MHz using tau_set = 600 ps and tau_reset = 1.5 ns, but these constants are not independently measured; they are selected so that the simulated frequency matches the experimentally observed 167 MHz. The model is therefore a consistency check, not an independent bound on the maximum oscillation frequency. To make the model load-bearing for the ceiling claim, the authors should extract tau_set and tau_reset from the 100 GHz oscillator trace, or from voltage-resolved measurements during oscillation, rather than from frequency matching alone.","section":"Supporting Information S3"},{"comment":"The premise that oscillator-condition voltages are clamped between Vset and Vreset is stated rather than verified: in the in-line circuit only the current is measured, and the authors themselves allow that 'minor overshoots or undershoots are possible.' Since the pulse experiments show that switching times depend strongly on overvoltage and undervoltage, the claim that oscillator set/reset times are order(s) of magnitude longer than the 15 ps/600 ps pulsed values would be considerably stronger with a direct high-bandwidth measurement of Vbias, or an indirect reconstruction from the current trace, during oscillation.","section":"Investigation of the internal relaxation time-scales, first paragraph"}],"minor_comments":[{"comment":"The 167 MHz oscillation is difficult to read from the compressed multi-period trace; a zoomed single-period view with marked rise and fall times would help the reader verify the claimed frequency.","section":"Fig. 1d"},{"comment":"The typeset form of Eq. (3) appears to have an unbalanced parenthesis in the argument of the logarithm; please check the mathematical expression.","section":"Eq. 3"},{"comment":"The notation tA1, tO1, tP1 and related labels is introduced in the text but not all of these times are marked in the figure panels; adding direct labels to the arrows would improve readability.","section":"Fig. 5"},{"comment":"The estimate '150 µm·18/2 + 150 µm = 1.5 mm' for the effective memristor-to-resistor distance is not self-evident; a sentence explaining the factor 18/2 would make the calculation transparent.","section":"Methods, meander resistor fabrication"}],"recommendation":"major_revision","confidential_remarks":"This is a strong experimental paper with a defensible central result. The main risk is that the speed-ceiling statement goes beyond the evidence: the numerical ceiling is supported by a 103 MHz trace and by fitted Simulink parameters rather than by transition-time data taken at the 167 MHz record itself. If the authors can provide an analysis of the 167 MHz reset edge or an independent extraction of the relaxation constants, I would be comfortable recommending acceptance; otherwise the ceiling claim should be softened to 'under the conditions realized here.'"},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know up front: the 167 MHz record is real, directly measured, and the best part of the paper. The softer spot is the universal ceiling claim attached to it. The paper concludes that oscillation significantly faster than 167 MHz is not realistic, but the load-bearing transition times come from a 103 MHz trace, not from the record trace. That gap is genuine, and it is a gap in the ceiling claim, not in the record.\n\nWhat is new: a passive VO2 relaxation oscillator at 167 MHz, over an order of magnitude above the previous 9 MHz best; the in-line transmission-line geometry that enables it; the f·d/v scaling collapse across different memristor-to-resistor distances and its breakdown above 75 MHz; and the identification of oscillator-condition reset relaxation as the practical speed limiter. The 100 GHz real-time traces (1.2 ns rise, 3.4 ns fall) are strong evidence for the relaxation-limited picture, and the d = 1.5 mm failure with recovery via a 1 pF capacitor is a nice confirmation of their stabilization argument. The authors are honest with the reader: they flag device-to-device variation and full relaxation dynamics as out of scope, report the scaling breakdown instead of hiding it, and let Eq. 3 be tested against experiment, failing honestly at high frequencies.\n\nSoft spots, in proportion. The Simulink model uses hand-picked tau_set = 600 ps and tau_reset = 1.5 ns to reproduce the 167 MHz trace; that makes it a fit, not an independent bound. The pulse experiments give far longer times (up to 54 ns), but the authors correctly note those overestimate oscillator-condition switching. So the only direct oscillator-condition numbers are from the 103 MHz trace. The stress-test suggestion is exactly right: the 167 MHz trace was recorded at 10 GS/s, and its falling edges contain the reset time at the record condition. Report that number. If it is around 3 ns, the ceiling claim holds up; if it is sub-nanosecond, the ceiling is higher. The voltage-clamp argument itself is explicit and reasonable, so the missing measurement, not the assumption, is what needs attention. Also minor: no error bars or statistics on the frequencies, and no released simulation code or data.\n\nThis paper is for anyone working on VO2 or Mott-oxide oscillators, neuromorphic hardware, or the speed limits of relaxation oscillators. It deserves a serious referee: the record and the scaling data merit publication, and the ceiling claim is worth arguing about, but it needs either softening or direct support at the record frequency.\n\nRecommendation: send it to peer review, with a request for transition-time analysis of the 167 MHz trace and more statistics.","headline":"Credible 167 MHz record for passive VO2 oscillators, but the paired speed-ceiling claim lacks transition-time data at the record frequency — a fixable gap, not a fatal flaw.","tokens_in":23445,"tokens_out":4405,"would_cite":true,"duration_ms":40611,"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":"The paper demonstrates a record 167 MHz oscillation in passive VO2 relaxation oscillators and argues that the reset relaxation time under oscillator bias sets a practical speed ceiling near this value.","keywords":["vanadium dioxide","Mott memristor","relaxation oscillator","resistive switching","oscillating neural network","transmission line circuit","ultrafast oscillation"],"falsifier":"Feed an oscillating VO2 circuit with a drive waveform that adds a short overvoltage at each set phase and record the oscillation frequency and the device voltage with a probe of at least 100 GHz bandwidth; a sustained oscillation well above 167 MHz, or set and reset edges remaining sub-nanosecond under oscillation bias, would falsify the claimed ceiling.","tokens_in":22318,"feed_emoji":"⚡","tokens_out":4948,"duration_ms":46770,"temperature":0.7,"pith_summary":"The paper claims that nanoscale VO2 memristors can be made to oscillate at 167 MHz, more than an order of magnitude beyond the previous 9 MHz record, by shrinking the switching region, cutting stray capacitance, and arranging the oscillator as a transmission-line circuit with the series resistor close to the device. It further claims that this is close to the practical ceiling: under oscillator bias the device voltage is clamped between Vset and Vreset, so the ultrafast 15 ps set and 600 ps reset seen with optimized single pulses cannot be reached. Instead, set and reset take nanoseconds, with reset relaxation the limiting factor, and the fastest oscillation is set by matching the circuit-induced voltage build-up delay to this internal relaxation time. If true, VO2-based oscillating neural networks can operate in the 100 MHz range with simple passive circuits, but GHz operation is not realistic in this design.","feed_headline":"Record 167 MHz oscillation in a VO2 oscillator circuit","feed_subtitle":"Tiny switching spots and a transmission-line layout beat the 9 MHz record; reset relaxation sets the ceiling.","key_machinery":"The central object is the nanoscale VO2 Mott memristor with a V-shaped electrode that confines switching to a roughly 30 nm spot, combined with an in-line oscillator circuit in which the memristor and series resistor are connected by a short transmission line of length d and the transmitted current is measured through a matched 50 Ω load. The argument is carried by an analytic voltage build-up time $\\tau_{0\\to V_{\\mathrm{set}}} = \\frac{d}{v}\\left[\\frac{\\ln\\left(1 - V_{\\mathrm{set}}\\frac{1-R_M R_S}{2V_0 T_S R_M}\\right)}{\\ln(R_M R_S)} - 1\\right]$ that counts how many back-and-forth reflections between the resistor and the memristor are needed to reach Vset, and by pulsed resistance-relaxation measurements that show how set and reset times lengthen under oscillator-like conditions. LTspice hysteresis-switch and Simulink relaxation models reproduce the observed frequencies and the failure of ultra-short-distance oscillators without a parallel capacitor.","core_discovery":"The paper demonstrates a record 167 MHz self-oscillation in a passive VO2 relaxation oscillator and argues that this sits near a fundamental practical ceiling for this class of circuits. With the switching confined to an ultrasmall, roughly 30 nm wide active region and the circuit rebuilt as a transmission-line arrangement where only the distance d between the series resistor and the memristor limits the signal build-up, the authors show that oscillation frequencies above 100 MHz are achievable and that the normalized frequency f·d/v collapses onto a single curve across distances. The limiting step is identified by pulsed experiments that mimic oscillator bias: because the device voltage is clamped near Vset and Vreset during oscillation, set and reset times stretch from the 15 ps and 600 ps of optimized single pulses to nanoseconds, with reset relaxation consistently the slower process. The conclusion is that an oscillation significantly faster than the demonstrated 167 MHz is not realistic for this passive-circuit design.","pith_inferences":["If overshoot beyond Vset could be engineered deliberately during oscillation, for instance with pulsed drive or a nonlinear load, set times might approach the 15 ps single-pulse value, potentially lifting the ceiling toward GHz.","The paper's conclusion concerns purely passive series-resistor circuits; active or hybrid circuits, such as the transistor-assisted TaOx benchmark the authors compare against, are a separate route that may bypass the reset-limited ceiling.","The normalized f·d/v scaling suggests a testable design rule: choose d so that the voltage build-up delay roughly matches the internal relaxation time, and the optimum should shift with temperature or device stoichiometry that changes Vset and Vreset.","For oscillating neural networks, the coupling network itself must be transmission-line compatible, otherwise the interconnect reflections that were removed inside the oscillator will reappear at the network level."],"forward_implications":["Oscillation frequency becomes tunable by the memristor-to-resistor distance d and the parallel capacitance C, with the normalized product f·d/v collapsing to a single curve across geometries.","The previous 9 MHz ceiling for VO2 oscillators is not fundamental; optimized devices routinely reach 75–100 MHz and have reached 167 MHz.","Since reset relaxation under oscillator bias is the slow step, circuit designs that shorten or assist the reset transition while preserving oscillation would be the route to higher frequencies.","Integrated on-chip resistors with very short d fail to oscillate without a parallel capacitor, because the voltage passes through the switching window too quickly for the memristor to complete set and reset.","VO2-based oscillating neural networks stand to operate at roughly 100 MHz, about an order of magnitude faster than previously demonstrated, which would speed up and make more energy-efficient the computations they perform."],"supporting_citations":[{"why":"Provides the prior record of 9 MHz VO2 oscillation that the present 167 MHz result exceeds.","marker":"[26]"},{"why":"Supplies the 15 ps set and 600 ps reset single-pulse benchmarks, the 2 fF stray capacitance estimate, and the 100 GHz bandwidth setup used for the fast oscillation measurement.","marker":"[31]"},{"why":"Gives theoretical frequency limits for crossbar VO2 devices that the confined-geometry devices are said to surpass.","marker":"[30]"},{"why":"Demonstrates 250 MHz TaOx oscillation using an active transistor, the passive-circuit comparison point.","marker":"[28]"},{"why":"Motivates why higher oscillation frequency improves the speed and energy efficiency of oscillatory neural networks.","marker":"[25]"},{"why":"Establishes the V-shaped confined active region that is the basis of the ultrafast device geometry.","marker":"[29]"}],"fun_headline_variants":["VO2 oscillators break 100 MHz, hit 167 MHz record","167 MHz VO2 oscillator: 18x faster via circuit redesign","Tiny switching spot enables 167 MHz VO2 oscillators","VO2 oscillator circuit hits record 167 MHz"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"During oscillation the voltage on the VO2 element is assumed to stay clamped between Vset and Vreset, so switching never sees the large overvoltages that produce 15 ps set and 600 ps reset times in optimized single pulses; if that clamp fails or is engineered away, the speed ceiling would move.","fun_headline_variants_meta":{"raw":{"variants":["VO2 oscillators break 100 MHz, hit 167 MHz record","167 MHz VO2 oscillator: 18x faster via circuit redesign","Tiny switching spot enables 167 MHz VO2 oscillators","VO2 oscillator circuit hits record 167 MHz"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001064,"raw_usage":{"total_tokens":4443,"prompt_tokens":913,"completion_tokens":3530,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":529,"completion_tokens_details":{"reasoning_tokens":3459}},"tokens_in":529,"tokens_out":3530,"duration_ms":25568,"temperature":1.0,"reasoning_tokens":3459,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:51:20.604772+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Feed an oscillating VO2 circuit with a drive waveform that adds a short overvoltage at each set phase and record the oscillation frequency and the device voltage with a probe of at least 100 GHz bandwidth; a sustained oscillation well above 167 MHz, or set and reset edges remaining sub-nanosecond under oscillation bias, would falsify the claimed ceiling.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the prior record of 9 MHz VO2 oscillation that the present 167 MHz result exceeds."},{"cited_title":"A Micram DAC10004 100GSa/s DAC unit together with a Centellax UA0L65VM broadband amplifier served as the driving unit","cited_arxiv_id":null,"evidence_quote":"Supplies the 15 ps set and 600 ps reset single-pulse benchmarks, the 2 fF stray capacitance estimate, and the 100 GHz bandwidth setup used for the fast oscillation measurement."},{"cited_title":"Carapezzi, A","cited_arxiv_id":null,"evidence_quote":"Gives theoretical frequency limits for crossbar VO2 devices that the confined-geometry devices are said to surpass."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Demonstrates 250 MHz TaOx oscillation using an active transistor, the passive-circuit comparison point."},{"cited_title":"Delacour, S","cited_arxiv_id":null,"evidence_quote":"Motivates why higher oscillation frequency improves the speed and energy efficiency of oscillatory neural networks."},{"cited_title":"P´ osa, P","cited_arxiv_id":null,"evidence_quote":"Establishes the V-shaped confined active region that is the basis of the ultrafast device geometry."}],"review_version":1}