Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Automated All-RF Tuning for Spin Qubit Readout and Control

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read An all-RF machine-learning routine tunes spin qubits autonomously, finding 12 qubit-ready charge transitions in under 17 hours.

desk verdict A credible and useful all-RF spin-qubit tuning demonstration on one device, honestly reported, but the 'automated' claim is narrower than the title implies because the barrier search space is manually pre-set. read the letter →

arxiv 2506.10834 v1 pith:3HU5WE2Z submitted 2025-06-12 cond-mat.mes-hall

classification cond-mat.mes-hall
keywords automatedspinqubittuningradio-frequencychargesensingPauliblockadesinglet-tripletoscillationsmachinelearningforquantumdotsGe/SiGeheterostructuresinterdottransitionsexchangeinteraction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that spin qubit tuning, normally a slow manual search through gate voltages, can be fully automated using radio-frequency charge sensing and machine learning. The routine starts from a large charge stability diagram, locates interdot charge transitions, automatically searches barrier voltages for Pauli spin blockade, and verifies the qubit by detecting singlet-triplet oscillations. In one continuous run on a Ge/SiGe hole double quantum dot, it found qubits at 12 of 26 interdot transitions in under 17 hours, with a median successful tuning time near 15 minutes. It also automatically mapped how the exchange interaction, dephasing time, and quality factor vary across those charge configurations. The underlying message is that high-throughput, hands-off qubit characterisation is within reach for semiconductor quantum circuits.

What carries the argument

The central object is the three-stage verification loop: an Interdot CNN that locates interdot charge transitions in a wide charge stability diagram; a pair of Line and Angle CNNs that reconstruct the transition lines around each interdot transition and define the triangular metastable regions used for readout; and a PSB score built from Gaussian mixture models of single-shot radio-frequency readout histograms, combined with an oscillation finder on singlet-triplet traces. This loop is driven by Latin hypercube sampling of barrier voltages with a gradient-free simplex refinement, while all gates are virtualised to keep the interdot transition centred. The routine needs no knowledge of charge occupation parity because it pulses in both directions across the detuning axis, and it has explicit stopping criteria, including oscillation amplitude, $R^2$, PSB score, and outcome majority, that guard against noise and latching.

What would settle it

A concrete test: start a fresh device with no manually chosen barrier bounds, let the routine sweep a coarse window on its own, and count how many interdot transitions yield confirmed singlet-triplet oscillations; if the success rate collapses or the time balloons, the headline result is really measuring the quality of the manual pre-tuning rather than the autonomous routine.

Watch

Extended reading notes

Core claim

The central discovery is a closed-loop, all-RF autonomous tuning protocol that turns the detection of Pauli spin blockade and coherent singlet-triplet oscillations into a machine-driven search. The routine uses convolutional neural networks to find interdot transitions in charge stability diagrams, to segment the triangular metastable regions around each transition, and to parameterise transition lines by angles; a score function based on Gaussian mixture fits to single-shot readout histograms decides whether blockade is present. When oscillations are found with sufficient $R^2$, or when the PSB score and outcome majority agree, the qubit is declared present. On the experimental device the routine found 12 qubit-ready transitions among 26, and simultaneously produced gate-voltage dependence of exchange coupling, dephasing time, and quality factor, with $T_2^*$ varying by a factor of six and $Q$ by nearly an order of magnitude across transitions.

Load-bearing premise

The whole demonstration depends on a human first tuning the device into a double quantum dot and choosing the voltage window the machine is allowed to search; if the qubit-ready voltages fall outside that window, the routine will never find them.

Editorial extensions

If this is right

  • A single unattended run can screen many charge transitions for qubit suitability, replacing a manual search that typically examines only a handful of transitions.
  • Qubit characterisation, including exchange coupling, dephasing time, and quality factor, becomes an automatic by-product of tuning, making statistical studies of qubit-to-qubit variability practical.
  • The routine's indifference to charge-occupation parity and its handling of latching should transfer to other spin-qubit materials and to larger dot arrays where hand tuning is not scalable.
  • The 15-minute median tuning time, if reproduced, would remove the tuning bottleneck that currently dominates the wall-clock cost of spin qubit experiments.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's fixed search bounds mean the honest scope of autonomy is tuning within a human-approved voltage region; a natural next test is to let the algorithm discover the search window itself from a coarse stability diagram, which the paper explicitly names as future work for a fully automated pipeline.
  • The observed breakdown of the conventional Pauli-spin-blockade checkerboard pattern suggests the routine can double as a disorder probe: the locations and gate-voltage ranges of successful transitions may map spurious dots or spin-orbit effects, a use the paper mentions only briefly.
  • Because the routine records exchange, dephasing, and quality factor at every success, its data could train a predictor that guesses which unvisited transitions are most likely to host a qubit, shortening later runs on similar devices.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper presents an automated tuning routine for spin qubits in gate-defined double quantum dots, using radio-frequency charge sensing throughout. The routine automatically locates interdot charge transitions (IDTs) in a charge stability diagram, segments each IDT to find Pauli spin blockade (PSB) readout points, evaluates PSB with a Gaussian-mixture score, and probes singlet-triplet oscillations. On a Ge/SiGe heterostructure device, a single continuous run visited 26 IDTs and identified 12 with PSB and singlet-triplet oscillations in under 17 hours, with a median successful tuning time of about 15 minutes. The routine also autonomously varied barrier voltages and magnetic fields to map the exchange interaction, dephasing time, and quality factor across the successful IDTs. The central claim is that this is the first automated all-RF tuning of spin qubits, including autonomous PSB detection and qubit characterization.

Significance. If the central claim holds, this is a valuable step toward high-throughput, automated spin-qubit tuning and characterization. The demonstration is strengthened by an external validation: the oscillation frequency varies with magnetic field in a manner consistent with singlet-triplet dynamics, and the machine-learning models are benchmarked on both simulated and hand-labelled experimental data. The Monte Carlo analysis of the search strategy and the detailed appendices on MLE, model selection, and out-of-distribution detection are also strong points. However, the autonomy claim is narrower than stated because the barrier-voltage search space is fixed by a manual initial tune-up, and the quantitative claims about T2* variability and quality factor lack reported uncertainties. These issues do not invalidate the measurements, but they do limit the generality of the headline result.

major comments (3)
  1. [Sec. IVA and Supplementary Sec. SV.B] The routine's barrier-voltage search space is user-defined: the bounds [-20,+90]/[-80,+90]/[-60,+80] mV on BL/BM/BR are chosen 'based on where double quantum dot confinement was maintained during the initial device tune-up.' Thus the headline 'automated all-RF tuning' and the statement that measurements are launched 'with no active user input' are not fully established: the algorithm tunes within a manually identified operating regime, and the quoted 12/26 yield and 17-hour timing are conditional on that prior knowledge. Supplementary Sec. SV.B explicitly concedes that false negatives arise if PSB configurations fall outside the search space and states that for a fully automated pipeline the bounds 'could be estimated algorithmically.' Please state this limitation in the main text, soften the abstract/introduction phrasing, and, if possible, add a demonstration with algorithmically estimated bounds or a second device to show that the search-space choice is not the dominant factor in the reported success rate.
  2. [Sec. IVB, Fig. 5(d-f), and Supplementary Sec. SVII.A] The reported T2* and Q values are presented without uncertainties, yet the paper claims a 'factor-of-six variation in T2* and nearly an order of magnitude variation in the Q factor.' The fits use A cos(2πft+φ) exp(-t/T2*), but no confidence intervals, goodness-of-fit statistics, or systematic-error budget are provided; the supplementary notes that charge switches occurred for interdots 13 and 16, and data with multiple oscillation frequencies were omitted. Without error bars or a sensitivity analysis, the quantitative variability claim is not supported. Please provide fit uncertainties, specify the number of traces per IDT, and state explicit criteria for omitting data.
  3. [Sec. IIIC3] The success criteria (R2 > 0.75, Fourier amplitude > 0.5, prominence > 0.2, PSB score > 0.1, outcome majority) are hand-picked thresholds, and the paper does not report a false-positive or false-negative characterization of the full stopping procedure. The B-field sweeps independently confirm the 12 successes, but the paper does not state how many configurations passed individual criteria yet failed B-field verification, nor how sensitive the 12/26 yield and median tuning time are to reasonable threshold variations. Given that the number and speed of identified qubits is the central quantitative claim, a threshold-sensitivity analysis or a control on known non-qubit configurations would materially strengthen the result.
minor comments (6)
  1. [Abstract and Sec. IVB] The abstract says '12 distinct charge transitions' while the text also refers to '12 different charge occupations'; please use consistent terminology for what was identified.
  2. [Appendix D] The reference acquisition assumes σ_b = σ_u for blocked and unblocked states because blocked-state standard deviations cannot be directly measured; this assumption is used in the MLE bounds of Eq. (E3c) but is not tested or discussed as a possible error source.
  3. [Appendix E2] The Latching outcome model forces the smaller blocked weight to equal the larger one, based on assumed symmetry of the interdot tunnelling rate with respect to pulsing direction; the paper's limitation section discusses asymmetric latching only in the context of false negatives, not as a modeling assumption.
  4. [Sec. IIIC3] The text contains a typo: 'succes' should be 'success'.
  5. [Fig. 5(c)] The caption states that the lower bound of ±5 mV is 'not visible for IDT 13'; please clarify what this means and whether this is a plotting artifact or a physical limit.
  6. [Data and code availability] The code is promised 'upon final publication'; for reproducibility, consider making at least the trained model weights and the routine's configuration files available at submission or in a permanent repository.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: detections are anchored by external magnetic-field verification and hand-labelled benchmarks; manual barrier bounds are an honest scope limitation, not a circular input.

full rationale

The central result is an experimental demonstration rather than a derived prediction, and its key detections are validated against an external physical handle. The stopping criteria (PSB score, R2 > 0.75) are not self-justifying: the paper shows the oscillation frequency changes with out-of-plane magnetic field (Fig. 4b and Supplementary SVII.B for each successful IDT), and this B-field dependence is not used to set the detection thresholds. The CNNs are trained on the authors' QArray simulator (refs. 36-37), but they are benchmarked on 10,200 hand-labelled experimental CSD patches and 308 hand-labelled IDT images (Supplementary SII.B and Table S1), so the simulator self-citation is not load-bearing. The sensor virtualisation method taken from ref. 22 is a supporting tool, not the claimed result, and it is peer-reviewed external work; it does not define the measured qubit properties. The blocked reference means are linearly extrapolated from unblocked traces (Appendix D), but the MLE must still find a second Gaussian component; if the data are unimodal the blocked weight goes to zero, and the score thresholds (0.1, 0.05), BIC outcome assignment, and the separate oscillation criterion provide discrimination. No equation reduces the detection to the reference construction. The manually pre-chosen barrier bounds are an acknowledged limitation ('we use bounds of [−20,+90]/[−80,+90]/[−60,+80] mV on BL/BM/BR, based on where double quantum dot confinement was maintained during the initial device tune-up', Sec. IVA; Supplementary SV.B concedes false negatives if PSB lies outside the search space). This narrows the meaning of 'fully automated' but is not circular: the bounds are an input condition, not an output of the routine, and the reported timings and yields are honestly conditional on that input. No self-definitional, fitted-input-as-prediction, uniqueness-imported-from-authors, ansatz-smuggled, or renaming patterns are present.

Assumptions & free parameters 8 free parameters · 7 assumptions · 0 invented entities

The routine introduces no new physical entities. Its output depends on a set of hand-chosen detection thresholds, calibration constants, and domain assumptions about the device and simulator. The most load-bearing free parameters are the stopping-criteria thresholds and the barrier voltage search bounds inherited from manual pre-tuning.

free parameters (8)
  • R2 success threshold = 0.75 (primary), 0.55-0.75 with checks
    Stopping criterion in Sec. IIIC3: oscillation peak with R2 > 0.75 is accepted; 0.55-0.75 requires additional PSB checks. Hand-chosen threshold affects which configurations are deemed qubits.
  • Oscillation detection amplitude and prominence = normalized amplitude > 0.5, prominence > 0.2
    Used to identify a Fourier component before fitting R2. Hand-chosen.
  • PSB score threshold = 0.1 (success), 0.05 (optimization qualifier)
    Sec. IIIC3 and Appendix F; hand-chosen.
  • Number of sample pairs per IDT = N = 4
    Sec. IIIB: chosen to reduce measurement time compared to >50 points in prior work.
  • Latin hypercube sample count = 40
    Sec. IVA and SI SIV; justified by Monte Carlo as a compromise between success probability and measurement overhead.
  • Barrier voltage search bounds = BL [-20,+90], BM [-80,+90], BR [-60,+80] mV
    Sec. IVA; based on initial manual device tune-up. Strongly affects which IDTs can be tuned.
  • Pulse timing parameters = trelax=10 us, ramp=16 ns, treadout=8 us, tidle=148-300 ns
    Sec. IIIC1; typical for Ge/SiGe qubits; not optimized per IDT.
  • MLE constraint bounds = Eq. E3: 0.1, 0.5, 0.2
    Appendix E; constrain fitted mixture parameters to stay near reference values.
assumptions (7)
  • domain assumption The constant-capacitance model implemented in QArray accurately simulates experimental charge stability diagrams, including noise and latching.
    Used to generate 5e5 training images for the Interdot CNN (Sec. IIIA). Benchmarking on experimental data supports this, but it remains a model assumption.
  • domain assumption Charge transition lines in the CSD have angles less than 45 degrees from the axes.
    Appendix B; used to constrain nearest-neighbour search. Assumes dot-plunger capacitance exceeds cross-capacitance, which may fail for spurious dots or strong cross-talk.
  • domain assumption PSB metastable regions are triangular when singlet-triplet splitting exceeds other energy scales.
    Sec. II, citing [33]; the routine samples readout points inside these triangles. Deviations could bias placement of readout points.
  • ad hoc to paper Blocked-state reference standard deviations equal unblocked-state ones (sigma_b = sigma_u).
    Appendix D; needed because blocked references cannot be measured directly. If false, MLE bounds are mis-calibrated.
  • ad hoc to paper Interdot tunnelling rate is symmetric with respect to pulsing direction for the Latching outcome.
    Appendix E; used to reduce parameters in the BIC Latching model. In spin-orbit-coupled systems this may not hold.
  • domain assumption The device remains stable over the full 17-hour autonomous run.
    The routine assumes no significant drift or charge switching; a charge switch was noted for IDTs 13 and 16 in SI SII.A (Further Data).
  • domain assumption The initial manual tune-up provides a suitable barrier voltage search space.
    Sec. IVA; the barrier bounds are set manually. If they do not contain PSB regions, the routine will fail regardless of its internal logic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Automated All-RF Tuning for Spin Qubit Readout and Control." pith.science (2026). https://pith.science/paper/3HU5WE2Z

@misc{pith2026250610834,
  author       = {Pith},
  title        = {Pith review of: Automated All-RF Tuning for Spin Qubit Readout and Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3HU5WE2Z}},
  note         = {Machine review of arXiv:2506.10834}
}
read the original abstract

Efficient tuning of spin qubits remains a major bottleneck in scaling semiconductor quantum dot-based quantum processors. A key challenge is the rapid identification of gate voltage regimes suitable for qubit initialisation, control, and readout. Here, we leverage radio-frequency charge sensing to automate spin qubit tuning, achieving a median tuning time of approximately 15 minutes. In a single continuous run, our routine identifies spin qubits at 12 distinct charge transitions in under 17 hours. Beyond tuning, our routine autonomously acquires data revealing the gate-voltage dependence of the exchange interaction, dephasing time, and quality factor -- quantities that vary substantially between charge configurations. These results represent a step change in high-throughput spin qubit tuning and provide a foundation for a systematic and automated exploration of semiconductor quantum circuits.

Figures

Figures reproduced from arXiv: 2506.10834 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. a left. Using this spatial information, IDTs are in￾dexed and assigned unique voltage ranges and detuning amplitudes for their respective image segmentation scans (next section) and pulsed measurements (Sec. III C 1). Further details on these sub-routines are provided in the Supplementary Materials. B. Image Segmentation To define potential meta-stable regions, all dot￾reservoir transition lines around an IDT must b… view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: a shows the interplay of our three stopping criteria over an example tuning loop that ends in suc￾cess. Prominent oscillations are found at iteration 9 (R2 = 0.929), accompanied by a high PSB score (0.159) and unanimous ‘Bottom PSB’ outcomes across sample pairs. The mo…
Figure 5
Figure 5. Figure 5: c maps the barrier voltage configurations at which our stopping criteria were met. The variety in dis￾tances from the starting barrier configuration (red cross) highlights the appropriacy of our exploratory tuning ap￾proach (see Supplementary Materials). Furthermore, t…
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: (a) illustrates the virtualisation procedure about IDT 20 from the autonomous run of our routine presented in the main text. After the scan window is centred on the IDT, each barrier then makes three volt￾age steps in increments of 1/8 of the window size. Based on hist…
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10 [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Action-Factored Multi-Agent Reinforcement Learning for Scalable Quantum Device Tuning

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Online action-space factorization via Kalman-refined cross-capacitance lets shared multi-agent policies zero-shot tune larger quantum-dot arrays with near-constant steps.

Reference graph

Works this paper leans on

54 extracted references · 35 canonical work pages · cited by 1 Pith paper

  1. [1]

    Calculating scores In our case, the probability distribution of interest is theGaussianmixturemodel(GMM)givenbyEq.1ofthe main text. The likelihood function of a GMM,Lj, given observed data,X j ={x j 1, xj 2, ..., xj m}, is the product of individual probabilities, Lj(wj, ˜µj, ˜σj|X j) = Y m p(xj m|wj, ˜µj, ˜σj).(E1) If we letθ j ={w j, ˜µj, ˜σj}, then MLE ...

  2. [2]

    Assigning outcomes As stated in Sec. IIIC3 of the main text, we use the Bayesian Information Criterion (BIC) to assign readout signal distributions of each sample pair to an outcome among ‘No PSB’, ‘Top PSB’, ‘Bottom PSB’ and ‘Latch- ing’. The BIC rewards outcomes with high likelihoods and mitigates over-fitting by penalising outcomes with more distributi...

  3. [3]

    Detecting anomalies Anomalous measurements may occur if a sample’s readout voltage point lands within the thermally broad- ened range of the interdot charge transition, or if latching occurs from a nearby transition that is not the interdot charge transition, either during reference acquisition or during pulsing. To catch these instances, we conduct the f...

  4. [24]

    S. R. Katiraee-Far, Y. Matsumoto, B. Undseth, M. De Smet, V. Gualtieri, C. V. Meinersen, I. F. de Fuentes, K. Capannelli, M. Rimbach-Russ, G. Scap- pucci,et al., Unified evolutionary optimization for high-fidelity spin qubit operations, arXiv preprint arXiv:2503.12256 (2025)

  5. [25]

    Berritta, T

    F. Berritta, T. Rasmussen, J. A. Krzywda, J. Van Der Heijden, F. Fedele, S. Fallahi, G. C. Gardner, M. J. Manfra, E. Van Nieuwenburg, J. Danon,et al., Real-time two-axis control of a spin qubit, Nature Communications 15, 1676 (2024). 9

  6. [26]

    Saez-Mollejo, D

    J. Saez-Mollejo, D. Jirovec, Y. Schell, J. Kukucka, S. Calcaterra, D. Chrastina, G. Isella, M. Rimbach-Russ, S. Bosco, and G. Katsaros, Exchange anisotropies in microwave-driven singlet-triplet qubits, Nature Commu- nications16, 3862 (2025)

  7. [27]

    Lundberg, D

    T. Lundberg, D. J. Ibberson, J. Li, L. Hutin, J. C. Abadillo-Uriel, M. Filippone, B. Bertrand, A. Nun- nenkamp, C.-M. Lee, N. Stelmashenko,et al., Non- symmetric pauli spin blockade in a silicon double quan- tum dot, npj Quantum Information10, 28 (2024)

  8. [28]

    A. Sen, G. Frank, B. Kolok, J. Danon, and A. Pályi, Clas- sification and magic magnetic field directions for spin- orbit-coupled double quantum dots, Physical Review B 108, 245406 (2023)

Show all 54 references
  1. [29]

    P. M. Mutter and G. Burkard, g-tensor resonance in dou- ble quantum dots with site-dependent g-tensors, Materi- als for Quantum Technology1, 015003 (2020)

  2. [30]

    Hendrickx, L

    N. Hendrickx, L. Massai, M. Mergenthaler, F. J. Schupp, S. Paredes, S. Bedell, G. Salis, and A. Fuhrer, Sweet-spot operation of a germanium hole spin qubit with highly anisotropic noise sensitivity, Nature Materials23, 920 (2024)

  3. [31]

    I. K. Jin, J. Hillier, S. D. Liles, Z. Wang, A. Shamim, I. Vorreiter, R. Li, C. Godfrin, S. Kubicek, K. De Greve, et al., Probing g-tensor reproducibility and spin-orbit ef- fects in planar silicon hole quantum dots, arXiv preprint arXiv:2411.06016 (2024)

  4. [32]

    Schuff, D

    J. Schuff, D. T. Lennon, S. Geyer, D. L. Craig, F. Fedele, F. Vigneau, L. C. Camenzind, A. V. Kuhlmann, G. A. D. Briggs, D. M. Zumbühl,et al., Identifying pauli spin blockade using deep learning, Quantum7, 1077 (2023)

  5. [33]

    Petta, A

    J. Petta, A. Johnson, A. Yacoby, C. Marcus, M. Hanson, andA.Gossard,Pulsed-gatemeasurementsofthesinglet- triplet relaxation time in a two-electron double quantum dot, Physical Review B—Condensed Matter and Materi- als Physics72, 161301 (2005)

  6. [34]

    Lapointe-Major, O

    M. Lapointe-Major, O. Germain, J. Camirand Lemyre, D. Lachance-Quirion, S. Rochette, F. Camirand Lemyre, and M. Pioro-Ladrière, Algorithm for automated tuning ofaquantumdotintothesingle-electronregime,Physical Review B102, 085301 (2020)

  7. [35]

    Mills, M

    A. Mills, M. Feldman, C. Monical, P. Lewis, K. Larson, A. Mounce, and J. R. Petta, Computer-automated tun- ing procedures for semiconductor quantum dot arrays, Applied Physics Letters115(2019)

  8. [36]

    van Straaten, J

    B. van Straaten, J. Hickie, L. Schorling, J. Schuff, F. Fedele, and N. Ares, QArray: A GPU-accelerated con- stant capacitance model simulator for large quantum dot arrays, SciPost Phys. Codebases , 35 (2024)

  9. [37]

    van Straaten, J

    B. van Straaten, J. Hickie, L. Schorling, J. Schuff, F. Fedele, and N. Ares, Codebase release 1.3 for QAr- ray, SciPost Phys. Codebases , 35 (2024)

  10. [38]

    H. Liu, B. Wang, N. Wang, Z. Sun, H. Yin, H. Li, G. Cao, andG.Guo, Anautomatedapproachforconsecutive tun- ing of quantum dot arrays, Applied Physics Letters121 (2022)

  11. [39]

    Van Riggelen, N

    F. Van Riggelen, N. Hendrickx, W. Lawrie, M. Russ, A. Sammak, G. Scappucci, and M. Veldhorst, A two- dimensional array of single-hole quantum dots, Applied Physics Letters118(2021)

  12. [40]

    D. L. Craig, H. Moon, F. Fedele, D. T. Lennon, B. van Straaten, F. Vigneau, L. C. Camenzind, D. M. Zumbühl, G. A. D. Briggs, M. A. Osborne,et al., Bridging the real- ity gap in quantum devices with physics-aware machine learning, Physical Review X14, 011001 (2024)

  13. [41]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R.Weiss, V.Dubourg, J.Vanderplas, A.Passos, D.Cour- napeau, M. Brucher, M. Perrot, and E. Duchesnay, Scikit-learn: Machine learning in Python, Journal of Ma- chine Learning ...

  14. [42]

    Y. Muto, T. Nakaso, M. Shinozaki, T. Aizawa, T. Ki- tada, T.Nakajima, M.R.Delbecq, J.Yoneda, K.Takeda, A. Noiri,et al., Visual explanations of machine learning model estimating charge states in quantum dots, APL Machine Learning2(2024)

  15. [43]

    Jirovec, A

    D. Jirovec, A. Hofmann, A. Ballabio, P. M. Mutter, G. Tavani, M. Botifoll, A. Crippa, J. Kukucka, O. Sagi, F. Martins,et al., A singlet-triplet hole spin qubit in pla- nar ge, Nature Materials20, 1106 (2021)

  16. [44]

    Botzem, M

    T. Botzem, M. D. Shulman, S. Foletti, S. P. Harvey, O. E. Dial, P.Bethke, P. Cerfontaine, R.P.McNeil, D.Mahalu, V. Umansky,et al., Tuning methods for semiconductor spin qubits, Physical Review Applied10, 054026 (2018)

  17. [45]

    J. Z. Blumoff, A. S. Pan, T. E. Keating, R. W. Andrews, D. W. Barnes, T. L. Brecht, E. T. Croke, L. E. Euliss, J. A. Fast, C. A. Jackson,et al., Fast and high-fidelity state preparation and measurement in triple-quantum- dot spin qubits, PRX Quantum3, 010352 (2022)

  18. [46]

    Zhang, E

    X. Zhang, E. Morozova, M. Rimbach-Russ, D. Jirovec, T.-K. Hsiao, P. C. Fariña, C.-A. Wang, S. D. Oost- erhout, A. Sammak, G. Scappucci, M. Veldhorst, and L. M. K. Vandersypen, Universal control of four sin- glet–triplet qubits, Nature Nanotechnology , 1 (2024)

  19. [47]

    E. G. Kelly, L. Massai, B. Hetényi, M. Pita-Vidal, A. Orekhov, C. Carlsson, I. Seidler, K. Tsoukalas, L. Sommer, M. Aldeghi, S. W. Bedell, S. Paredes, F. J. Schupp, M. Mergenthaler, A. Fuhrer, G. Salis, and P. Harvey-Collard, Identifying and mitigating errors in hole spin qubi...

  20. [48]

    Jirovec, P

    D. Jirovec, P. M. Mutter, A. Hofmann, A. Crippa, M. Rychetsky, D. L. Craig, J. Kukucka, F. Martins, A. Ballabio, N. Ares,et al., Dynamics of hole singlet- triplet qubits with large g-factor differences, Physical re- view letters128, 126803 (2022)

  21. [49]

    Schwarz, Estimating the dimension of a model, The annals of statistics , 461 (1978)

    G. Schwarz, Estimating the dimension of a model, The annals of statistics , 461 (1978)

  22. [50]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. van der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, and others, SciPy 1.0: Fundamental Algorithms for Scientific Computing in Pyth...

  23. [51]

    J. A. Nelder and R. Mead, A simplex method for function minimization, The computer journal7, 308 (1965)

  24. [52]

    A. C. Johnson, J. Petta, C. Marcus, M. Hanson, and A.Gossard,Singlet-tripletspinblockadeandchargesens- ing in a few-electron double quantum dot, Physical Re- view B—Condensed Matter and Materials Physics72, 165308 (2005)

  25. [53]

    Stano and D

    P. Stano and D. Loss, Review of performance metrics of spin qubits in gated semiconducting nanostructures, Nature Reviews Physics4, 672 (2022)

  26. [54]

    S. D. Ha, E. Acuna, K. Raach, Z. T. Bloom, T. L. Brecht, J. M. Chappell, M. D. Choi, J. E. Christensen, I. T. Counts, D. Daprano,et al., Two-dimensional si spin qubitarrayswithmultilevelinterconnects,arXivpreprint 10 arXiv:2502.08861 (2025)

  27. [55]

    Liles, D

    S. Liles, D. Halverson, Z. Wang, A. Shamim, R. Eggli, I. K. Jin, J. Hillier, K. Kumar, I. Vorreiter, M. Rendell, et al., A singlet-triplet hole-spin qubit in mos silicon, Na- ture Communications15, 7690 (2024)

  28. [56]

    N.Ares, V.N.Golovach, G.Katsaros, M.Stoffel, F.Four- nel, L. I. Glazman, O. G. Schmidt, and S. De Franceschi, Nature of tunable hole g factors in quantum dots, Phys- ical review letters110, 046602 (2013)

  29. [57]

    Liles, F

    S. Liles, F. Martins, D. Miserev, A. Kiselev, I. Thorvald- son, M. Rendell, I. Jin, F. Hudson, M. Veldhorst, K. Itoh, et al., Electrical control of the g tensor of the first hole in a silicon mos quantum dot, Physical Review B104, 235303 (2021)

  30. [58]

    Mauro, E

    L. Mauro, E. A. Rodríguez-Mena, M. Bassi, V. Schmitt, and Y.-M. Niquet, Geometry of the dephasing sweet spotsofspin-orbitqubits,PhysicalReviewB109,155406 (2024)

  31. [59]

    Bassi, E.-A

    M. Bassi, E.-A. Rodrıguez-Mena, B. Brun, S. Zihlmann, T. Nguyen, V. Champain, J. C. Abadillo-Uriel, B. Bertrand, H. Niebojewski, R. Maurand,et al., Op- timal operation of hole spin qubits, arXiv preprint arXiv:2412.13069 (2024)

  32. [60]

    Geyer, B

    S. Geyer, B. Hetényi, S. Bosco, L. C. Camenzind, R. S. Eggli, A. Fuhrer, D. Loss, R. J. Warburton, D. M. Zum- bühl, and A. V. Kuhlmann, Anisotropic exchange inter- action of two hole-spin qubits, Nature Physics20, 1152 (2024)

  33. [61]

    Froning, M

    F. Froning, M. Rančić, B. Hetényi, S. Bosco, M. Rehmann, A. Li, E. P. Bakkers, F. A. Zwanenburg, D. Loss, D. Zumbühl,et al., Strong spin-orbit interac- tion and g-factor renormalization of hole spins in ge/si nanowire quantum dots, Physical Review Research3, 013081 (2021)

  34. [62]

    Schorling, P

    L. Schorling, P. Vaidhyanathan, J. Schuff, M. J. Car- ballido, D. Zumbühl, G. Milburn, F. Marquardt, J. Foer- ster, M. A. Osborne, and N. Ares, Meta-learning char- acteristics and dynamics of quantum systems, arXiv preprint arXiv:2503.10492 (2025)

  35. [63]

    Gebhart, R

    V. Gebhart, R. Santagati, A. A. Gentile, E. M. Gauger, D. Craig, N. Ares, L. Banchi, F. Marquardt, L. Pezze, and C. Bonato, Learning quantum systems, Nature Re- views Physics5, 141 (2023)

  36. [64]

    Tosato, A

    A. Tosato, A. Elsayed, F. Poggiali, L. Stehouwer, D. Costa, K. Hudson, D. D. Esposti, and G. Scappucci, Qarpet: A crossbar chip for benchmarking semiconduc- tor spin qubits, arXiv preprint arXiv:2504.05460 (2025)

  37. [65]

    J. C. Dunn, Well-separated clusters and optimal fuzzy partitions, Journal of cybernetics4, 95 (1974)

  38. [66]

    Ziegler, F

    J. Ziegler, F. Luthi, M. Ramsey, F. Borjans, G. Zheng, and J. P. Zwolak, Automated extraction of capacitive coupling for quantum dot systems, Physical Review Ap- plied19, 054077 (2023)

  39. [67]

    X. Ding, X. Zhang, J. Han, and G. Ding, Scaling up your kernels to 31x31: Revisiting large kernel design in cnns, inProceedings of the IEEE/CVF conference on computer vision and pattern recognition(2022) pp. 11963–11975

  40. [68]

    Czischek, V

    S. Czischek, V. Yon, M.-A. Genest, M.-A. Roux, S. Ro- chette, J. C. Lemyre, M. Moras, M. Pioro-Ladrière, D. Drouin, Y. Beilliard,et al., Miniaturizing neural networks for charge state autotuning in quantum dots, Machine Learning: Science and Technology3, 015001 (2021)

  41. [69]

    Ziegler, T

    J. Ziegler, T. McJunkin, E. Joseph, S. S. Kalantre, B. Harpt, D. Savage, M. G. Lagally, M. Eriksson, J. M. Taylor, and J. P. Zwolak, Toward robust autotuning of noisy quantum dot devices, Physical Review Applied17, 024069 (2022). Appendix A: Clustering Prior tok-means clusteri...

  42. [70]

    best guess

    is used to score the goodness of clustering, and is given by, DI = min 2≤i<j≤k δ(C i, Cj) min 2≤m≤k ∆m ,(A1) whereδ(C i, Cj)istheinter-clusterdistancebetweenclus- tersC i andC j, and∆ m is the intra-cluster distance. A high DI signifies a clustering intokclusters that have bot...

  43. [71]

    This has been accounted for in a previous work by weighting towards patches with transition lines in the training data [S68]

    Interdot CNN Locating IDTs is an imbalanced classification problem, because the number of CSD patches that contain an IDT constitute a minority. This has been accounted for in a previous work by weighting towards patches with transition lines in the training data [S68]. Anothe...

  44. [72]

    The Angle CNN’s training data, on the other hand, is generated in two ways

    Angle and Line CNNs The Line CNN is trained on1×105 simulated images of charge-sensed data [S36]. The Angle CNN’s training data, on the other hand, is generated in two ways. One subset includes predictions made by the trained Line CNN on 1×10 5 simulated images. A second subse...

  45. [73]

    In all of the above cases, the stopping criteria are not met

    Decoherence occurs faster than the oscillation frequency or the period of the oscillations is larger or comparable to the maximumtidle. In all of the above cases, the stopping criteria are not met. The first case could result from initialisation and readout errors caused by La...

  46. [74]

    checkerboard

    Oscillationsattheelectrostaticconfiguration for which the stopping criteria were met are shown in middle plots. We note that for interdots 13 and 16, a charge switch occurred in the device between the calibration of the voltage readout point and magnetic field sweep. 0 200 400...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.