{"id":"6bedef9b-4b89-4f95-9f58-1b77dd661fa8","arxiv_id":"2501.18321","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Spin Hall nano-oscillator arrays with up to 105,000 elements achieve mutual synchronization, record output power of 9 nW, and record quality factor of 1.04e6.","lead":"Scientists synchronized 105,000 ultra-small microwave oscillators on a chip, over a thousand times more than the previous record, using spin currents and nano-scale constrictions. This opens a path toward brain-like computers, Ising machines, and wireless devices built from many coupled oscillators.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 105,000-oscillator synchronization claim rests on ensemble spectra and BLS maps that cannot resolve individual 24-nm-pitch constrictions; the data do not exclude a synchronized subnetwork, and the handling of partially synchronized arrays is not reported.","rationale":"The reader's weakest assumption focuses on frequency uniformity across constrictions. That concern is related but not identical to mine: the linewidth scaling with 1/N is already strong evidence that a collective phase-locked state exists, so a spread of free-running frequencies alone would not invalidate the claim. The more load-bearing issue is whether the collective state includes all N constrictions or only a large subnetwork. The paper provides no per-oscillator phase or frequency data; the BLS resolution cannot resolve individual oscillators; and the scaling plots are shown without error bars or free-exponent fits. The manuscript's own admission of partial synchronization in some arrays makes the handling of those data points important, but this is not described. I therefore agree with the reader's conditional assessment, but for a somewhat different reason. The central experiment is impressive and self-consistent; the missing piece is a control or direct measurement that ties the collective signal to the full nominal population.","tokens_in":10799,"tokens_out":9690,"duration_ms":112979,"concrete_test":"Fabricate two nominally identical large 10-nm arrays, then use focused-ion-beam milling to remove a known fraction p of the constrictions (e.g., one quarter of the rows) in one device while leaving the other intact. If the synchronized microwave power and linewidth scale with the remaining active number (1-p)N rather than with nominal N, full participation is confirmed; if they track nominal N, the signal is dominated by a subnetwork. Additionally, re-fit the Fig. 2g,h data with free exponents (P = A N^beta, Delta_f = B N^alpha) and report confidence intervals and the number of arrays per point, with a clear statement of how partially synchronized arrays were handled.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that all 105,000 constrictions are mutually synchronized. The evidence is a single narrow microwave peak, P scaling as N, linewidth scaling as 1/N, and BLS intensity maps. The weakest link is that none of these observables directly establishes that every fabricated constriction participates in the phase-locked state. The BLS spot is ~300 nm whereas the constriction pitch is 24 nm, so the maps are spatially averaged and cannot resolve individual oscillators or their relative phases. The electrical spectrum is a global measurement: a synchronized subset of size M = cN with a roughly constant fraction c would produce the same functional scalings as the full array, especially because the abscissa in Fig. 2g,h is the nominal N rather than the number of confirmed active oscillators. The paper itself states that 'only a few arrays showing multiple signals just above auto-oscillation onset due to partial synchronization' but does not say whether those arrays were excluded from the scaling plots or whether they eventually synchronized. In addition, the 'fits' to N^-1 and N appear to be forced to the expected exponents, with no reported error bars, no free-exponent fits, and no device-to-device scatter. This does not disprove the claim, but it leaves the quantitative 'all N participate' conclusion less secure than the headline suggests.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports spin Hall nano-oscillator arrays with 10- and 20-nm constrictions, up to N = 105,000 oscillators, and claims robust mutual synchronization of all oscillators in each array. The evidence presented includes single narrow microwave peaks, linewidth scaling as N^-1, power scaling as N, record values of 9 nW output power and Q = 1.04e6, micro-BLS intensity maps, and micromagnetic simulations. The paper also proposes a model in which coherent magnon exchange within the array and magnon losses at the array edges explain the unexpectedly strong array-size dependence of frequency-current tunability.","tokens_in":11091,"tokens_out":6573,"duration_ms":69557,"significance":"If fully supported, this is a significant experimental advance: it extends the previous record of 64 synchronized SHNOs by more than three orders of magnitude, demonstrates record microwave power and quality factor, and provides a large systematic dataset across 146 arrays. The fabrication reproducibility, resistance scaling, and the qualitative agreement between BLS, micromagnetic simulation, and the edge-loss model are notable strengths. However, the evidence for 'complete' synchronization is indirect: the BLS spot size is ~300 nm while the constriction pitch is 24 nm, and the electrical spectra are global measurements. The paper's central 'all N oscillators participate' claim therefore needs either stronger quantitative support or more careful wording.","major_comments":[{"comment":"The claim in the text that the single-peak spectra are 'consistent with complete mutual synchronization' and the conclusion that 'all our arrays exhibited robust mutual synchronization' are not reconciled with the statement that 'only a few arrays showing multiple signals just above auto-oscillation onset due to partial synchronization.' The manuscript does not state how many arrays showed such partial synchronization, whether those arrays were excluded from the scaling analyses in Fig. 2g,h, or whether they eventually locked at higher current. Since the scaling plots use the nominal N, inclusion or exclusion of partially synchronized devices directly affects the central claim; please report this information explicitly and either justify the inclusion or restrict the claim.","section":"Results: Auto-oscillations vs. array size (Fig. 2a-d)"},{"comment":"The scaling fits in Fig. 2g,h are constrained to the expected N^-1 and N exponents, with no error bars, no device-to-device scatter, and no free-exponent fits. Because a synchronized subset of size M = cN with a constant fraction c would produce the same functional dependence on nominal N, these plots as presented cannot exclude a synchronized subnetwork of constant fraction and do not establish that all 105,000 constrictions participate. Please provide per-device data, free-exponent fits with confidence intervals, and residual analysis, or soften the claim to 'large synchronized networks' with the participation fraction stated as an open question.","section":"Results: Auto-oscillations vs. array size (Fig. 2g,h)"},{"comment":"The BLS laser spot is ~300 nm in diameter while the constriction pitch is 24 nm, so the BLS maps are spatially averaged over many oscillators and cannot resolve individual constrictions or their relative phases. The intensity envelope is therefore consistent with, but does not directly prove, phase-locking or equal participation of every oscillator. The section's statement that BLS is used to 'directly visualize the auto-oscillations and the mutual synchronization' should be tempered, or supplemented with a phase-sensitive measurement, to avoid overstating the spatial evidence.","section":"Brillouin light scattering microscopy (Fig. 3)"},{"comment":"The model of coherent magnon exchange and edge losses is presented as explaining the N-dependent tunability, but no quantitative comparison is made between the model and the df/dI data in Fig. 2f. No calculated or fitted parameter values, uncertainties, or error bars are given for the BLS fits in Fig. 3h. Since this model is the basis for the 'unexpectedly strong array size dependence' claim, please specify the model equations, the parameter values used, and a direct comparison with both the measured tunability as a function of N and the measured edge fall-off.","section":"Micromagnetic simulations and model (Fig. 2f, Fig. 3h)"}],"minor_comments":[{"comment":"The typo 'SHNOS' appears in the sentence describing the center-to-center separation; it should read 'SHNOs.'","section":"Methods: Design and fabrication"},{"comment":"Please state whether the 'highest generated microwave output power' for each array is taken at a common operating condition (field, current, and field angle) and how the power is integrated from the measured spectrum; otherwise the comparison may mix different operating points.","section":"Fig. 2h"},{"comment":"The extraction of the maximum tunability (df/dI)_max from the derivative plot should specify the fit range and the uncertainty of the derivative; currently no error estimate is given.","section":"Fig. 2f inset"},{"comment":"The abstract and conclusion state 'mutually synchronized SHNO networks' and 'all our arrays exhibited robust mutual synchronization,' but the main text acknowledges partial synchronization in a few arrays. Please harmonize the wording so that the summary of results matches the detailed reporting of partial synchronization.","section":"Abstract and Conclusion"}],"recommendation":"major_revision","confidential_remarks":"I agree with the stress-test concern: the data are consistent with, but do not uniquely establish, full participation of all 105,000 constrictions in the phase-locked state. The main issue is not a fabrication or measurement error but an overclaim relative to the evidence. A major revision that either provides per-device scaling data with free-exponent fits and clear handling of partially synchronized arrays, or that softens the 'all N' claim, would bring the paper in line with its evidence base."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe headline is real: this group has pushed mutually synchronized spin Hall nano-oscillator arrays from 64 to 105,000 elements, a three-orders-of-magnitude jump, with record linewidth (25.3 kHz, Q = 1.04e6) and output power (9 nW). If the claim holds, it is the biggest step yet toward practical oscillator networks for Ising machines and neuromorphic hardware. I think it will hold up in substance, but read the fine print.\n\nWhat is genuinely new: the sheer scale, the clean N-scaling of power and linewidth, and an interesting array-size-dependent frequency-current tunability that they explain via magnon exchange and edge losses. The BLS maps and micromagnetic simulations support the qualitative picture. The resistance scaling and threshold-current density independence across 146 arrays show careful fabrication. This is a serious experimental effort, not a fishing expedition.\n\nThe soft spots are proportionate but real. First, \"complete mutual synchronization\" is inferred from a single narrow microwave peak and BLS intensity maps. The BLS spot is ~300 nm, while the constriction pitch is 24 nm, so the maps cannot resolve individual oscillators or their relative phases. A synchronized subnetwork of constant fraction c would produce the same functional scalings of power and linewidth vs. nominal N. The paper does not report free-exponent fits, error bars, or device-to-device scatter in the scaling plots; the lines are drawn to the expected N and N^-1. Second, the paper admits that \"only a few arrays showing multiple signals just above auto-oscillation onset due to partial synchronization\" but never says whether those arrays are included in the scaling plots or eventually synchronize. That should be clarified. Third, the tunability model is qualitative and post-hoc, with no fitted parameters or independent prediction; it is consistent with the data but not strongly tested.\n\nThat said, I would not call any of this disqualifying. The central claim—large-scale phase locking—is supported by the dramatic linewidth narrowing and power increase, which are hard to explain without synchronization. The uniformity of the arrays is evidenced by resistance scaling and consistent threshold current. The lack of per-constriction frequency measurements is a limitation, not a contradiction.\n\nVerdict: this deserves serious peer review. The authors need to add a few paragraphs on array inclusion/exclusion, error bars, and ideally a free-exponent fit on the scaling, but the core result is significant and likely reproducible. I would cite it, and I'd bring it to our reading group to talk about what \"synchronized\" can mean at this scale.\n\nRecommendation: send to a strong referee, not desk reject.","headline":"Record-scale SHNO synchronization is real and important, but the 'all N participate' claim needs stronger evidence than ensemble spectra.","tokens_in":11594,"tokens_out":2267,"would_cite":true,"duration_ms":20936,"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":"Mutually synchronized networks of 105,000 10-nm spin Hall nano-oscillators are demonstrated, with microwave power and quality factor scaling linearly with oscillator number.","keywords":["spin Hall nano-oscillator","mutual synchronization","nano-constriction array","magnon exchange","spin wave","microwave power scaling","quality factor","Brillouin light scattering microscopy"],"falsifier":"Measure the auto-oscillation frequency of many individually addressed constrictions from the same fabrication run; if the spread of their free-running frequencies is much larger than the 25-kHz linewidth seen in the arrays, the single-peak reading would demand an extremely strong coupling, while a spread comparable to the linewidth would mean the narrow peak could arise without true synchronization. A second check: fabricate arrays with a deliberate gradient in constriction width so the natural frequencies differ, and observe whether a single synchronized peak still survives.","tokens_in":1599,"feed_emoji":"📶","tokens_out":2349,"duration_ms":66978,"temperature":0.7,"pith_summary":"This paper reports that spin Hall nano-oscillators—tiny constrictions that convert a direct current into a microwave signal through spin-orbit torque—can be mutually synchronized in dense arrays of up to 105,000 oscillators, three orders of magnitude beyond the previous limit of 64. Across every array tested, the emitted microwave power and the signal quality factor grow linearly with the number of oscillators, reaching 9 nW and $1.04 \\times 10^{6}$, respectively. The authors also find that the frequency-current tunability becomes stronger and more nonlinear as arrays grow, and they trace this to a balance between coherent magnon exchange inside the array and magnon loss at the array edges. If these claims hold, the result would make large-scale oscillator networks practical for wireless communication, neuromorphic computing, and Ising machines, and it suggests the method may scale to even larger arrays.","feed_headline":"105,000 nano-oscillators lock into one microwave signal","feed_subtitle":"Microwave power and signal purity scale linearly with oscillator count, surpassing the previous limit of 64.","key_machinery":"The central object is the nano-constriction SHNO array: electron-beam-patterned constrictions in W-Ta/CoFeB/MgO on Si with an Al$_2$O$_3$ thermal and electrical buffer, spaced closely enough (24 or 40 nm) for spin-wave coupling. The load-bearing mechanism is mutual synchronization through exchange of coherent propagating magnons between constrictions, whose strength grows with array size because interior oscillators receive coherent spin waves from many neighbors while edge oscillators lose magnons into unpatterned mesa areas. The quantitative signatures are the scaling laws $\\Delta f \\propto 1/N$, $P \\propto N$, and $Q \\propto N$, plus the model prediction that collective radiative-loss reduction makes the frequency-current response increasingly quadratic for larger arrays.","core_discovery":"The central claim is that mutual synchronization in spin Hall nano-oscillator arrays is not limited to small groups but persists in dense arrays of up to 105,000 nano-constrictions (10 nm wide with 24 nm center-to-center spacing, and 20 nm wide with 40 nm spacing). The evidence is a single ultranarrow microwave peak in every array, with linewidth decreasing as $1/N$ and output power increasing as $N$; the best observed linewidth is 25.3 kHz at 26.2 GHz, giving $Q = 1.04 \\times 10^{6}$, and the highest output power is 9 nW. The paper explains the unexpected array-size dependence of frequency-current tunability through a model in which coherent propagating magnons are constructively exchanged among interior oscillators while magnons are lost to unpatterned magnetic material at the short edges, a picture supported by micro-Brillouin light scattering maps and micromagnetic simulations.","pith_inferences":["A direct test of the synchronization claim would be to pattern isolated single constrictions with the same process and measure their free-running frequency spread; if the spread is much smaller than the observed linewidth, part of the sharp peak could be statistical, whereas a large spread would strengthen the synchronization interpretation.","The model's prediction that the quadratic current-frequency dependence saturates beyond one magnon decay length could be checked by fabricating arrays with different aspect ratios at fixed $N$ and comparing tunability slopes.","The BLS intensity is proportional to the square of the dynamic magnetization amplitude, so the edge fall-off serves as a diagnostic of coupling strength; measuring it as a function of spacing would map the effective coupling length.","The same array geometry could be extended to non-square network topologies, such as random or small-world connectivity, by positioning constrictions accordingly, connecting directly to network science."],"forward_implications":["Any SHNO-based application needing output power or spectral purity, such as wireless transmission or ultrafast spectrum analysis, can now use arrays of $10^5$ oscillators instead of 64, directly improving signal strength and coherence.","Neuromorphic and reservoir computing schemes requiring many phase-locked nonlinear oscillators become feasible at array sizes where the network itself, not the oscillator count, sets the computational capability.","Sparse Ising machines, which need very large numbers of interacting bistable oscillators for combinatorial problems, can in principle be built from these arrays.","The demonstrated absence of an upper size limit at 105,000 suggests that further reducing spacing or increasing the magnetization and thickness of the ferromagnet could push synchronization beyond one million oscillators.","The size-dependent current tunability and nonlinearity provide a controllable, task-adaptive parameter for physical reservoir computing."],"supporting_citations":[{"why":"Sets the previous record of 64 mutually synchronized SHNOs and provides the scaling expectation that microwave power and quality factor grow with $N$.","marker":"[17]"},{"why":"Supplies the 10-nm SHNO fabrication and the Si/Al$_2$O$_3$ thermal-management stack enabling dense low-power arrays.","marker":"[11]"},{"why":"Defines the W-Ta spin-orbit alloy and stack optimized for spin-orbit efficiency used in these arrays.","marker":"[49]"},{"why":"Demonstrates spin-orbit torque-driven propagating spin waves from SHNOs, the basis for spin-wave-mediated coupling.","marker":"[5]"},{"why":"Shows long-range mutual synchronization of SHNOs, establishing the coupling mechanism this paper scales up.","marker":"[16]"},{"why":"Provides an in-depth demonstration of spin-wave-mediated mutual synchronization and phase tuning in SHNOs.","marker":"[6]"},{"why":"Demonstrates robust synchronization in long SHNO chains, the precursor for scaling to dense 2D arrays.","marker":"[18]"},{"why":"Supports the phase-noise and linewidth behavior underlying the quality-factor scaling.","marker":"[48]"}],"fun_headline_variants":["105,000 nano-oscillators sync into one signal","Record 105k nano-oscillators sync into one signal","From 64 to 105k: nano-oscillators sync","105k nano-oscillators lock together in sync","105,000 oscillators sync: new record"],"cache_read_input_tokens":13824,"weakest_assumption_plain":"The load-bearing assumption is that the 105,000 constrictions are nearly identical in their magnetic behavior, so that a single narrow microwave peak can be read as collective locking rather than as a narrow distribution of independent frequencies; the paper's evidence for this uniformity is indirect, based on resistance scaling and SEM images rather than on measuring each oscillator's frequency separately.","fun_headline_variants_meta":{"raw":{"variants":["105,000 nano-oscillators sync into one signal","Record 105k nano-oscillators sync into one signal","From 64 to 105k: nano-oscillators sync","105k nano-oscillators lock together in sync","105,000 oscillators sync: new record"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001396,"raw_usage":{"total_tokens":5631,"prompt_tokens":912,"completion_tokens":4719,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":528,"completion_tokens_details":{"reasoning_tokens":4648}},"tokens_in":528,"tokens_out":4719,"duration_ms":31865,"temperature":1.0,"reasoning_tokens":4648,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T23:53:41.284986+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the auto-oscillation frequency of many individually addressed constrictions from the same fabrication run; if the spread of their free-running frequencies is much larger than the 25-kHz linewidth seen in the arrays, the single-peak reading would demand an extremely strong coupling, while a spread comparable to the linewidth would mean the narrow peak could arise without true synchronization. A second check: fabricate arrays with a deliberate gradient in constriction width so the natural frequencies differ, and observe whether a single synchronized peak still survives.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Sets the previous record of 64 mutually synchronized SHNOs and provides the scaling expectation that microwave power and quality factor grow with $N$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the 10-nm SHNO fabrication and the Si/Al$_2$O$_3$ thermal-management stack enabling dense low-power arrays."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the W-Ta spin-orbit alloy and stack optimized for spin-orbit efficiency used in these arrays."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Demonstrates spin-orbit torque-driven propagating spin waves from SHNOs, the basis for spin-wave-mediated coupling."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows long-range mutual synchronization of SHNOs, establishing the coupling mechanism this paper scales up."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides an in-depth demonstration of spin-wave-mediated mutual synchronization and phase tuning in SHNOs."},{"cited_title":"Nano Lett.23(14), 6720– 6726 (2023)","cited_arxiv_id":null,"evidence_quote":"Demonstrates robust synchronization in long SHNO chains, the precursor for scaling to dense 2D arrays."},{"cited_title":"Applied Physics Letters 122(22) (2023)","cited_arxiv_id":null,"evidence_quote":"Supports the phase-noise and linewidth behavior underlying the quality-factor scaling."}],"review_version":1}