{"id":"7da70f65-f16d-4f25-850e-3890adc46ae8","arxiv_id":"2507.08713","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Simulated distance-3 surface codes on silicon spin qubits convert slowly varying 1/f noise into memory-less logical noise, giving a quartic coherence-time scaling T*_2,L proportional to (T*_2)^4.","lead":"This paper simulates how well a small quantum error correcting code protects silicon spin qubits when the noise is slow and correlated in time, rather than random at every step. It finds the logical qubit's coherence time grows as the fourth power of the physical qubit's coherence time, much faster than the usual quadratic rule, as long as two-qubit gate noise is small.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The quartic law in Eq. (31) rests on an unvalidated quadratic logical-error scaling for temporally correlated noise; the 3-round decoder window and constant-noise-per-gate approximation need convergence checks.","rationale":"The paper's central numerical claim is that a distance-3 rotated surface code converts non-Markovian 1/f noise into Markovian logical noise, yielding T*_2,L ∝ (T*_2)^4 / t_QEC^3. The derivation of Eq. (31) is structurally sound given two ingredients: physical error per cycle scaling as p ≈ t_QEC^2/(2 T*_2^2) and logical error per cycle scaling as p_L ∝ p^2 for distance 3. The first ingredient follows from the observed Gaussian physical fidelity decay and is internally consistent. The second ingredient is standard for independent Pauli errors, but the whole point of the paper is that the physical noise is non-Markovian and temporally correlated. The simulation's own decoding procedure uses windows of only three syndrome extractions with one round of overlap, so it is not obvious that correlated errors spanning many rounds are being counted in the same way as independent per-round errors. This is the load-bearing gap: if the effective p_L is not quadratic in p under long-time correlations, the quartic law is not a general property of QEC with non-Markovian noise but an artifact of the decoder window and noise discretization. The constant-noise-per-gate approximation is a second, related gap because it replaces a continuous 1/f process by a zero-order hold over each gate; intra-gate fluctuations are discarded, and no test shows that refining the hold time leaves T*_2,L unchanged. These concerns do not invalidate the paper, but they make the headline scaling conditional on convergence checks that are currently absent. The reader's weakest_assumption identified the same issues, so agreement is full. The appropriate verdict remains CONDITIONAL, hence no adjustment to the reader's verdict is needed.","tokens_in":25498,"tokens_out":9152,"duration_ms":116870,"concrete_test":"Fix one no-J-noise operating point from Fig. 9 (e.g., T*_2 ≈ 71 μs) and recompute T*_2,L with decoder window lengths of 1, 3, 5, and 7 syndrome rounds and with a global minimum-weight matching decoder over the full run; then repeat with noise sampling periods t_s = 0.05, 0.1, and 0.2 μs, and with each gate split into 2 and 4 constant-noise sub-gates. If T*_2,L moves outside the reported 95% confidence interval or the fitted exponent departs from 4, the quartic law is not robust to these approximations.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq. (31) rests on p_L ∝ p^2 per QEC cycle, but the simulation never independently validates this quadratic law for temporally correlated 1/f noise. The decoder (Sec. III.B.5) matches syndromes only within windows of three rounds with one round of overlap. A slow noise excursion can persist over many rounds, so the same data qubit receives a coherent over-rotation in every round; these persistent errors are not independent per round, and a windowed matching decoder may either miss them or treat them as multiple independent errors, changing the effective logical error rate and the exponent in Eq. (31). The constant-noise-per-gate library (Sec. III.B.1) similarly truncates intra-gate fluctuations; no convergence test with smaller sampling time t_s or gate sub-sampling is reported. Without such checks, the quartic dependence in Fig. 9 could reflect the decoder memory length and the zero-order hold rather than intrinsic QEC suppression of non-Markovian noise. The paper itself notes the windowed decoder choice and the constant-noise assumption but does not quantify their impact, so the central scaling claim is conditionally supported at best.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports numerical emulation of a distance-3 rotated surface code (and its XZZX variant) on a silicon spin-qubit hardware model, with temporally correlated 1/f noise affecting both the Larmor frequency and the exchange coupling. A Ramsey-like logical memory experiment is simulated using the Qaptiva emulator, with a precomputed noisy-gate library and windowed minimum-weight perfect matching. The central result is Eq. (31): T*_2,L is proportional to (T*_2)^4 divided by (t_QEC)^3, derived from the code-distance relation p_L ∝ p^2, a Gaussian physical fidelity decay, and an exponential logical fidelity decay. The paper also analyzes fully spatially correlated noise and a sparse shuttling-based architecture.","tokens_in":25736,"tokens_out":4940,"duration_ms":60029,"significance":"If the quartic scaling is robust, it is a strong and practically relevant prediction: a distance-3 surface code would convert low-frequency Gaussian dephasing into exponentially decaying logical errors, yielding a fourth-power coherence-time enhancement that is qualitatively different from the quadratic scaling expected under Markovian noise. The work is valuable for using native silicon spin-qubit gates and a non-Pauli, temporally correlated noise model rather than the usual i.i.d. Pauli noise, and for explicitly addressing spatial correlation limits and shuttling-based sparse architectures. The central claim, however, rests on two numerical approximations whose convergence is not demonstrated: the constant-noise-per-gate library and the three-round decoder window. No code or data release is provided, and the emulator is proprietary, so the reported numerical results cannot currently be independently reproduced.","major_comments":[{"comment":"The precomputed noisy-gate library assumes that δω_L and δV_E remain constant during each gate. Since the π-pulse P gate lasts 3 μs while the noise is sampled every 0.1 μs, this zero-order hold truncates intra-gate fluctuations and could systematically bias the gate error rates that enter the logical fidelity simulations. No convergence test with smaller t_s or with sub-sampled Trotter integration is reported, so the exponent in Eq. (31) could be affected by this approximation. Please quantify the sensitivity of the quartic scaling to this assumption.","section":"Section III.B.1"},{"comment":"The decoder matches syndromes in windows of three consecutive rounds with two consecutive windows overlapping by one round. For temporally correlated 1/f noise, a slow noise excursion can produce coherent over-rotations on the same data qubit across many rounds; these errors are not independent per round, and a windowed matching decoder may either miss them or treat them as multiple independent errors, changing the effective logical error rate. The paper does not test longer decoder windows or a full matching decoder, so the quadratic relation p_L ∝ p^2 in Eq. (30) is not independently validated for this noise model. I request a check that the quartic scaling in Fig. 9 is stable under decoder-window length.","section":"Section III.B.5"},{"comment":"The derivation inserts p_L ∝ p^2 from the code distance, but the simulation can directly test this relation by plotting the per-cycle logical error probability p_L = 1 - f_L(t_QEC) against p = 1 - f(t_QEC) for the same noise realizations. Without such a plot, the quartic law is an inference from the slope of T*_2,L versus T*_2 on a log-log plot, which could in principle be shaped by the constant-noise-per-gate and decoder-window approximations. A direct p_L-versus-p plot would make the central claim machine-checkable and should be added.","section":"Eqs. (30)-(31) and Fig. 9"}],"minor_comments":[{"comment":"The claim that QEC 'Markovianizes' the noise is stronger than what the exponential fidelity decay demonstrates: an exponential decay of the ensemble-averaged fidelity is necessary but not sufficient to establish that the logical channel is Markovian. Consider either softening the terminology or adding a test for correlations between consecutive logical error events.","section":"Section IV.B"},{"comment":"The caption contains a typo: 'upped axis' should be 'upper axis'.","section":"Fig. 13 caption"},{"comment":"The caption states that error bars are 95% confidence intervals of the average logical fidelity, but the number of independent noise realizations used for each point and the procedure for propagating the confidence interval to T*_2,L are not given. Please report these details.","section":"Fig. 9 caption"},{"comment":"The shuttling noise model, which treats shuttling as idle dephasing scaled by a factor γ, is a phenomenological assumption; the text should state more explicitly the regime in which motional narrowing or shuttling-induced extra noise would invalidate this approximation.","section":"Section III.D"},{"comment":"The fitting form for T*_2(t_m, S_0) is written with a square root inside the logarithm in a way that is easy to misread; adding an explicit bracket or parentheses would improve clarity.","section":"Eq. (27)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the scope of a quantum information journal and the quartic-scaling prediction is interesting and falsifiable. My main concern is reproducibility and validation: the proprietary emulator, the absence of data/code release, and the missing convergence checks for the constant-noise-per-gate approximation and the windowed decoder. These are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the quick read. The paper does something genuinely useful: it takes a realistic non-Markovian (1/f) noise model for silicon spin qubits—covering both Larmor and exchange noise—and plugs it into a distance-3 surface code using native gates. The headline result is that logical coherence time scales as the fourth power of T2*, not the quadratic law you'd get from Markovian noise. That quartic law is internally consistent: it follows from the distance-3 p_L ∝ p^2 suppression plus the observed Gaussian-to-exponential fidelity decay. The numerical figures show the trend with confidence intervals, and the analysis of J-noise saturation is a useful practical pointer. Credit where due: this goes beyond earlier spin-qubit QEC papers that used Pauli or quasistatic noise.\n\nThe soft spots are real but not disqualifying. The main one is exactly the stress-test note: the derivation assumes p_L ∝ p^2 per QEC cycle holds for temporally correlated noise, but the decoder only matches syndromes within a three-round window (with one round overlap). For 1/f noise with long correlations, a slow excursion can persist over many rounds; the windowed decoder might either miss it or treat it as multiple independent errors, which would change the exponent. The paper states the window choice but never validates that the quadratic law survives longer correlation times. Second, the constant-noise-per-gate assumption (δω_L and δV_E fixed during each gate) truncates intra-gate fluctuations, and there's no convergence test as sampling time t_s is reduced. Both are flagged in the paper, but the impact is unquantified. That keeps the central claim conditional.\n\nThe absence of released code and data is a real hindrance, especially since Qaptiva is proprietary. The sparse-architecture scaling formula (32) is an empirical fit, not a derived prediction; the paper doesn't mislabel it, but it shouldn't be quoted as a law. The intentional exclusion of measurement and initialization errors makes absolute logical times optimistic, though the scaling law might survive.\n\nWho's it for: people working on silicon spin qubit QEC architectures and anyone trying to set hardware noise budgets. The paper deserves a serious referee—the question is substantive and the simulation appears carefully done—but I'd want a revision that adds convergence checks on the decoder window and the constant-noise approximation, plus a clear statement of what artifacts would be released.","headline":"A serious simulation study with a plausible but unproven quartic scaling claim; worth refereeing after convergence checks and a reproducibility pass.","tokens_in":26340,"tokens_out":2180,"would_cite":true,"duration_ms":25091,"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 distance-3 surface code on silicon spin qubits turns non-Markovian noise into memoryless logical errors, giving a logical coherence time that scales as the fourth power of the physical coherence time.","keywords":["silicon spin qubits","surface code","quantum error correction","non-Markovian noise","1/f noise","logical coherence time","XZZX code","sparse architecture"],"falsifier":"Keep $t_{\\mathrm{QEC}}$ fixed and run the same emulation with only Larmor-frequency $1/f$ noise, sweeping $T^*_2$ over at least a decade; the log-log slope of $T^*_{2,L}$ versus $T^*_2$ must be 4 and the logical fidelity decay must remain exponential. If the slope is 2 instead, or if widening the decoding window changes the slope, the Markovianization assumption is the part that fails.","tokens_in":25217,"feed_emoji":"⚛️","tokens_out":11059,"duration_ms":113957,"temperature":0.7,"pith_summary":"Silicon spin qubits are limited by temporally correlated (non-Markovian) noise in the Larmor frequency and the exchange coupling between neighbors. This paper asks whether a small distance-3 surface code can still protect such qubits, and how much the logical coherence time improves over the physical one. The numerical answer is that the surface code \"Markovianizes\" the noise: the logical qubit's fidelity decays exponentially in time even though the physical qubit's fidelity decays as a Gaussian, and in the regime where two-qubit exchange noise is not the bottleneck the logical coherence time grows as the fourth power of the physical coherence time. The practical consequence is that error correction on silicon spin qubits is far more powerful than a naive quadratic estimate would suggest, provided the exchange-noise channel is controlled and syndrome extraction is fast.","feed_headline":"Surface code lifts spin-qubit coherence time to the fourth power","feed_subtitle":"Error correction turns correlated 1/f noise into memoryless logical errors and yields a quartic coherence-time gain.","key_machinery":"The load-bearing object is the distance-3 rotated surface code run as a 17-qubit quantum memory, with the syndrome extraction circuit compiled into silicon spin-qubit native gates and the two-qubit gate $P=\\mathrm{CZ}(S\\otimes S)$ as the entangling operation. The main results use the $\\pi$-pulse version of $P$, which inserts spin-refocusing $X$ pulses and thereby cancels Larmor-frequency deviations better than the symmetry-corrected version. Noise is injected through discrete-time traces of $\\delta\\omega_L$ and $\\delta V_E$ with a $1/f$ power spectrum; each gate is replaced by a precomputed noisy-gate library entry, and decoding is performed on overlapping windows of three syndrome rounds with a minimum-weight perfect-matching decoder. The conceptual mechanism behind the headline result is the observed Markovianization: physical fidelity decays as a Gaussian while logical fidelity decays exponentially, so the per-cycle logical error rate is linear in $t_{\\mathrm{QEC}}/T^*_{2,L}$ while the physical per-cycle error is quadratic in $t_{\\mathrm{QEC}}/T^*_2$; combining that pair of scalings with the distance-3 quadratic error suppression gives $T^*_{2,L}\\propto (T^*_2)^4/(t_{\\mathrm{QEC}})^3$.","core_discovery":"The paper claims that quantum error correction converts temporally correlated non-Markovian noise into Markovian (memoryless) logical noise for silicon spin qubits. In numerical emulations of a distance-3 rotated surface code subject to $1/f$ Larmor-frequency noise and exchange-energy noise, the logical qubit's Ramsey-like fidelity decays exponentially even though the physical qubit's fidelity decays as a Gaussian. In the regime where two-qubit exchange noise is not the bottleneck, the logical coherence time obeys $T^*_{2,L}\\propto (T^*_2)^4/(t_{\\mathrm{QEC}})^3$: the quadratic error-rate suppression of the distance-3 code combines with the Gaussian-to-exponential decay conversion to raise the naive quadratic scaling to a quartic one. Exchange-energy noise with small $T^*_J$ saturates this gain, fully spatially correlated noise leaves it nearly intact, and a sparse shuttling architecture preserves it up to shuttling times of about $0.1\\,\\mu\\mathrm{s}$.","pith_inferences":["Generalizing the paper's $d=3$ relation $p_L\\propto p^2$ under the same Markovianization assumption predicts $T^*_{2,L}\\propto (T^*_2)^{d+1}/t_{\\mathrm{QEC}}^d$ for a distance-$d$ surface code, a concrete target for future $d=5$ emulations.","Because the logical noise is memoryless, standard Pauli-noise simulation tools may be adequate for predicting surface-code performance on silicon spin qubits even though the physical noise is non-Markovian; the paper suggests this but does not prove it.","An experimental test of the quartic law is within reach of a 17-qubit device: with exchange noise suppressed, sweeping the single-qubit $T^*_2$ by an order of magnitude should move $T^*_{2,L}$ by four orders of magnitude if the claim holds."],"forward_implications":["Under the $\\pi$-pulse $P$-gate syndrome circuit, the logical qubit fidelity decays exponentially even when physical qubit fidelity decays as a Gaussian, so the logical error process is effectively memoryless.","In the regime where exchange-energy noise is negligible and $T^*_2\\gg t_{\\mathrm{QEC}}$, the logical coherence time follows $T^*_{2,L}\\propto (T^*_2)^4/(t_{\\mathrm{QEC}})^3$, a quartic gain over the physical coherence time.","When two-qubit exchange noise is present with a small $T^*_J$, it caps the logical coherence time; improving $T^*_2$ alone no longer helps once the two-qubit error contribution dominates.","Fully spatially correlated $1/f$ noise degrades the logical coherence time only slightly and preserves the near-quartic scaling, because the syndrome circuit exposes different qubits to different gate sequences.","In a sparse shuttling architecture, performance remains close to nominal up to shuttling times around $0.1\\,\\mu\\mathrm{s}$, and since shuttling cost does not grow with code distance, robustness is expected to persist or improve for larger codes."],"supporting_citations":[{"why":"Supplies the surface-code distance scaling $p_L \\propto p^2$ used to derive the quartic coherence-time law.","marker":"[18]"},{"why":"Provides the improved syndrome extraction circuit for the distance-3 code and the overlapping-window minimum-weight perfect-matching decoder used in the emulations.","marker":"[41]"},{"why":"Gives the Gaussian decay form and $T^*_2$ scaling for $1/f^\\alpha$ noise that defines the physical coherence-time metric.","marker":"[34]"},{"why":"Establishes the $1/f$ noise characterization of Larmor-frequency and exchange fluctuations in silicon spin qubits.","marker":"[33]"},{"why":"Supplies the silicon spin-qubit hardware model, including the single- and two-qubit Hamiltonians used for native gates.","marker":"[17]"},{"why":"Provides the experimental $J$ versus $V_E$ data used to fit the exponential conversion of exchange-potential noise into coupling noise.","marker":"[36]"},{"why":"Defines the XZZX variant of the rotated surface code compared in the simulations.","marker":"[22]"},{"why":"Supports the $T^*_2$ dependence on noise intensity and machine time used in the Ramsey fits.","marker":"[57]"}],"fun_headline_variants":["Spin qubits get quartic coherence gain from surface code","Surface code turns 1/f noise into Markovian, boosts spin qubits","Non-Markovian noise tamed: qubit coherence scales to fourth power","Silicon spin qubits: error correction yields T2^4 scaling"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quartic law assumes that the distance-3 error-rate relation $p_L\\propto p^2$ holds per QEC cycle even when the $1/f$ noise is correlated over times comparable to or longer than the three-round decoding window, and that $\\delta\\omega_L$ and $\\delta V_E$ can be treated as frozen during each gate.","fun_headline_variants_meta":{"raw":{"variants":["Spin qubits get quartic coherence gain from surface code","Surface code turns 1/f noise into Markovian, boosts spin qubits","Non-Markovian noise tamed: qubit coherence scales to fourth power","Silicon spin qubits: error correction yields T2^4 scaling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000175,"raw_usage":{"total_tokens":1263,"prompt_tokens":899,"completion_tokens":364,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":515,"completion_tokens_details":{"reasoning_tokens":285}},"tokens_in":515,"tokens_out":364,"duration_ms":4489,"temperature":1.0,"reasoning_tokens":285,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:12:29.163530+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Keep $t_{\\mathrm{QEC}}$ fixed and run the same emulation with only Larmor-frequency $1/f$ noise, sweeping $T^*_2$ over at least a decade; the log-log slope of $T^*_{2,L}$ versus $T^*_2$ must be 4 and the logical fidelity decay must remain exponential. If the slope is 2 instead, or if widening the decoding window changes the slope, the Markovianization assumption is the part that fails.","supporting_citations":[{"cited_title":"Preskill, California institute of technology16, 1 (1998)","cited_arxiv_id":null,"evidence_quote":"Provides the improved syndrome extraction circuit for the distance-3 code and the overlapping-window minimum-weight perfect-matching decoder used in the emulations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Gaussian decay form and $T^*_2$ scaling for $1/f^\\alpha$ noise that defines the physical coherence-time metric."},{"cited_title":"Pataki, ´A","cited_arxiv_id":null,"evidence_quote":"Establishes the $1/f$ noise characterization of Larmor-frequency and exchange fluctuations in silicon spin qubits."},{"cited_title":"Aspuru-Guzik, A","cited_arxiv_id":null,"evidence_quote":"Supplies the silicon spin-qubit hardware model, including the single- and two-qubit Hamiltonians used for native gates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the experimental $J$ versus $V_E$ data used to fit the exponential conversion of exchange-potential noise into coupling noise."},{"cited_title":"Rojas-Arias, A","cited_arxiv_id":null,"evidence_quote":"Supports the $T^*_2$ dependence on noise intensity and machine time used in the Ramsey fits."}],"review_version":1}