PaperlayerView exact revision on arXiv

Source-linked AI summary

Helios: A 98-qubit trapped-ion quantum computer

Anthony Ransford, M. S. Allman, Jake Arkinstall, J. P. Campora, Samuel F. Cooper, Robert D. Delaney, Joan M. Dreiling, Brian Estey, Caroline Figgatt, Alex Hall, Ali A. Husain, Akhil Isanaka, Colin J. Kennedy, Nikhil Kotibhaskar, Ivaylo S. Madjarov, Karl Mayer, Alistair R. Milne, Annie J. Park, Adam P. Reed, Riley Ancona, Molly P. Andersen, Pablo Andres-Martinez, Will Angenent, Liz Argueta, Benjamin Arkin, Leonardo Ascarrunz, William Baker, Corey Barnes, John Bartolotta, Jordan Berg, Ryan Besand, Bryce Bjork, Matt Blain, Paul Blanchard, Robin Blume-Kohout, Matt Bohn, Agustin Borgna, Daniel Y. Botamanenko, Robert Boutelle, Natalie Brown, Grant T. Buckingham, Nathaniel Q. Burdick, William Cody Burton, Varis Carey, Christopher J. Carron, Joe Chambers, John Children, Victor E. Colussi, Steven Crepinsek, Andrew Cureton, Joe Davies, Daniel Davis, Matthew DeCross, David Deen, Conor Delaney, Davide DelVento, B. J. DeSalvo, Jason Dominy, Ross Duncan, Vanya Eccles, Alec Edgington, Neal Erickson, Stephen Erickson, Christopher T. Ertsgaard, Bruce Evans, Tyler Evans, Maya I. Fabrikant, Andrew Fischer, Cameron Foltz, Michael Foss-Feig, David Francois, Brad Freyberg, Charles Gao, Robert Garay, Jane Garvin, David M. Gaudiosi, Christopher N. Gilbreth, Josh Giles, Erin Glynn, Jeff Graves, Azure Hansen, David Hayes, Lukas Heidemann, Bob Higashi, Tyler Hilbun, Jordan Hines, Ariana Hlavaty, Kyle Hoffman, Ian M. Hoffman, Craig Holliman, Isobel Hooper, Bob Horning, James Hostetter, Daniel Hothem, Jack Houlton, Jared Hout, Ross Hutson, Ryan T. Jacobs, Trent Jacobs, Melf Johannsen, Jacob Johansen, Loren Jones, Sydney Julian, Ryan Jung, Aidan Keay, Todd Klein, Mark Koch, Ryo Kondo, Chang Kong, Asa Kosto, Alan Lawrence, David Liefer, Michelle Lollie, Dominic Lucchetti, Nathan K. Lysne, Christian Lytle, Callum MacPherson, Andrew Malm, Spencer Mather, Brian Mathewson, Daniel Maxwell, Lauren McCaffrey, Hannah McDougall, Robin Mendoza, Michael Mills, Richard Morrison, Louis Narmour, Nhung Nguyen, Lora Nugent, Scott Olson, Daniel Ouellette, Jeremy Parks, Zach Peters, Jessie Petricka, Juan M. Pino, Frank Polito, Matthias Preidl, Gabriel Price, Timothy Proctor, McKinley Pugh, Noah Ratcliff, Daisy Raymondson, Peter Rhodes, Conrad Roman, Craig Roy, Ciaran Ryan-Anderson, Fernando Betanzo Sanchez, George Sangiolo, Tatiana Sawadski, Andrew Schaffer, Peter Schow, Jon Sedlacek, Henry Semenenko, Peter Shevchuk, Susan Shore, Peter Siegfried, Kartik Singhal, Seyon Sivarajah, Thomas Skripka, Lucas Sletten, Ben Spaun, R. Tucker Sprenkle, Paul Stoufer, Mariel Tader, Stephen F. Taylor, Travis H. Thompson, Raanan Tobey, Anh Tran, Tam Tran, Grahame Vittorini, Curtis Volin, Jim Walker, Sam White, Douglas Wilson, Quinn Wolf, Chester Wringe, Kevin Young, Jian Zheng, Kristen Zuraski, Charles H. Baldwin, Alex Chernoguzov, John P. Gaebler, Steven J. Sanders, Brian Neyenhuis, Russell Stutz, Justin G. Bohnet

arXiv:2511.05465v1quant-phphysics.atom-ph

TL;DR

Scaling quantum processors requires larger systems without sacrificing performance. Helios addresses this challenge with a transport-based trapped-ion architecture and real-time compilation, while benchmarks characterize its component and system-level operation.

  • Problem

    Quantum hardware must scale to much larger systems without sacrificing performance, including support for efficient arbitrary and dynamic quantum programs.

  • Method

    Helios combines a transport-based QCCD architecture with spatially separated memory and logic regions, all-to-all routing, and a runtime that compiles virtual-qubit operations to physical qubits in real time.

  • Results

    Helios establishes state-of-the-art component and system-level benchmark performance across single-qubit, two-qubit, SPAM, and circuit-based operation.

  • Takeaways & Limitations

    The benchmarks show that Helios supports arbitrary quantum programs and produces high-fidelity states whose classical sampling cost is vastly beyond practical reach.

  • Takeaways & Limitations

    The measured two-qubit error exceeds the predicted error budget, and classical sampling-cost estimates use approximate optimization heuristics.

Abstract

from arXiv · show

We report on Quantinuum Helios, a 98-qubit trapped-ion quantum processor based on the quantum charge-coupled device (QCCD) architecture. Helios features $^{137}$Ba$^{+}$ hyperfine qubits, all-to-all connectivity enabled by a rotatable ion storage ring connecting two quantum operation regions by a junction, speed improvements from parallelized operations, and a new software stack with real-time compilation of dynamic programs. Averaged over all operational zones in the system, we achieve average infidelities of $2.5(1)\times10^{-5}$ for single-qubit gates, $7.9(2)\times10^{-4}$ for two-qubit gates, and $4.8(6)\times10^{-4}$ for state preparation and measurement, none of which are fundamentally limited and likely able to be improved. These component infidelities are predictive of system-level performance in both random Clifford circuits and random circuit sampling, the latter demonstrating that Helios operates well beyond the reach of classical simulation and establishes a new frontier of fidelity and complexity for quantum computers.

I. INTRODUCTION

Helios scales trapped-ion QCCD computing with a 98-qubit architecture combining separated memory and logic regions, an X-junction, shared resources, and real-time control. These design choices target larger systems without sacrificing performance while supporting all-to-all connectivity.

  • I. INTRODUCTION: Helios is a 98-qubit trapped-ion QCCD processor designed to scale quantum systems without sacrificing performance.The paper frames scaling to larger systems as the central hardware challenge.
  • I. INTRODUCTION: An X-junction connects memory regions to the quantum logic regions without increasing electrical-control or device-fabrication complexity relative to Quantinuum H2.The junction routes qubits between ring storage, cache, and leg storage, while sorting can proceed alongside cooling.
  • I. INTRODUCTION: A new classical control implementation makes real-time transport and quantum-operation decisions, enabling arbitrary programs with all-to-all connectivity.The paper evaluates these capabilities through component-level and system-level benchmarks.
  • I. INTRODUCTION: The architecture separates qubit memory from quantum logic regions, transporting ions to isolated zones for low-crosstalk gate operations.The QPU uses individual ions as qubits and physically transports them to trapping zones for gates.
  • I. INTRODUCTION: Shared lasers across multiple operation zones improve the efficiency with which essential control resources scale.Helios uses eight operation zones, each supporting preparation, measurement, cooling, and quantum logic.

B. Ion Species - qubit and coolant

Helios uses ^137Ba+ hyperfine qubits with visible-wavelength control and ^171Yb+ sympathetic cooling. Its transport cycle combines memory routing, batched logic operations, cooling, and software-defined execution across the processor.

  • B. Ion Species - qubit and coolant: Helios uses ^137Ba+ hyperfine levels as qubit states, making it the first quantum computer to utilize ^137Ba+.The qubit states are defined in the electronic ground state.
  • B. Ion Species - qubit and coolant: Visible-wavelength optical transitions provide more mature, reliable, and cost-effective laser components for quantum operations.The paper links greater available laser power and improved phase performance to suppression of several gate-error sources.
  • B. Ion Species - qubit and coolant: ^171Yb+ ions provide sympathetic cooling for the ^137Ba+ qubits during QCCD mid-circuit recooling.^171Yb+ is chosen for its similar mass to ^137Ba+ and established control and measurement methods.
  • B. Ion Species - qubit and coolant: Each program layer removes qubits from ring storage, processes batches in the quantum logic region, and returns them to ring storage.The default configuration contains eight BYYB crystals in the logic region and 82 BY crystals in ring storage.
  • B. Ion Species - qubit and coolant: Cooling for two-qubit operations runs in parallel with moving the next batch from ring storage to cache.Two-qubit gates use four of the eight operation zones after crystals are merged and cooled.

D. Real time compilation of sorting and gates

Helios adds a runtime that maps virtual-qubit operations to physical qubits during execution, supporting dynamic programs, real-time routing, and parallel gate execution. Benchmarks combine component-level and system-level tests to assess performance and limitations.

  • Runtime capabilities: Real-time virtual-to-physical qubit mapping supports dynamic programs with allocation, de-allocation, early termination, and classical control flow.The runtime performs translations while quantum state is live and executing.
  • Runtime capabilities: Gate streaming reduces shot time by omitting unnecessary basis-change gates and transport when measurement requests require no changes.The runtime uses real-time qubit identification to avoid extraneous operations.
  • Runtime capabilities: The runtime resolves virtual-qubit allocations, transforms gates into parallel physical operations, and transports qubit batches from ring storage into operation zones.These responsibilities are organized around allocation, gating, parallelization, and sorting.
  • Benchmarking: System evaluation combines component benchmarks with random Clifford circuits containing MCMR and mirrored random circuit sampling.The benchmark suite compares full-device behavior with predictions from component-level measurements.
  • Benchmarking: Component-level benchmark values are summarized as averages over all operation zones.The table provides the aggregate benchmark view used in the paper.

B. Component-level benchmarks

Component-level benchmarks quantify SPAM, single-qubit, and two-qubit performance across Helios’s operation zones, including leakage and correlated-error analyses. The reported averages are 2.5(1) × 10^-5 for 1Q error and 7.9(2) × 10^-4 for 2Q infidelity, with additional benchmarking methods revealing error composition and discrepancies with models.

  • State-preparation and measurement: Standard SPAM errors are 8(1) × 10^-4 for preparing |0⟩ and 1.6(5) × 10^-4 for preparing |1⟩, while leaked states can be misclassified as |1⟩.Ternary measurement finds average leakage probability 4.2(7) × 10^-3 and non-leakage SPAM errors of 7(1) × 10^-4 and 2.8(2) × 10^-3 for |0⟩ and |1⟩.
  • Single-qubit gates: 2.5(1) × 10^-5 is the zone-averaged single-qubit error, including a leakage rate of 1.12(6) × 10^-5.The measurement covers 16 qubits across 8 operation zones and agrees with a physical-error prediction of 2.6(6) × 10^-5.
  • Correlated errors: No evidence of correlated errors is found in simultaneous 1QRB at the 95% confidence level, and no significant correlated errors are found across 2QRB qubit pairs.The tests address correlated error channels associated with crosstalk.
  • Model comparison: The measured 2Q error of 7.9(2) × 10^-4 exceeds the predicted total error of 3.5(4) × 10^-4, with several possible unaccounted sources proposed.Suggested explanations include finite detection-beam extinction, non-thermal motion, crosstalk, and other effects.
  • Two-qubit gates: 8.1(2) × 10^-4 is the zone-averaged 2QCB infidelity, dominated by IZ and ZI errors, while its leakage estimate is about half the 2QRB estimate.2QCB reports a leakage rate of 1.14(4) × 10^-4.

4. Transport idle memory errors

Helios measures transport-induced memory errors with transport-1QRB circuits that vary transport rounds between Clifford gates. Leakage grows linearly with transport operations and accounts for nearly all of the linear memory error.

  • Measurement method: Transport-1QRB interleaves random transport operations between single-qubit Clifford gates to estimate memory errors under different transport schedules.The 98 qubits are partitioned into groups receiving Clifford operations after k = 1, 2, 4, or 8 depth-1 transport operations.
  • Measurement method: Transport-1QRB measures survival, leakage, and memory-error behavior across four groups with different numbers of depth-1 transport operations.The corresponding observables are shown as functions of Clifford sequence length and transport-operation count.
  • Error scaling: 5(1) × 10^-4 is the inferred linear memory error rate, while the quadratic memory error parameter is 7(2) × 10^-5.The fit uses a + bl + cl^2, with linear and quadratic terms associated with fast and slow noise, respectively.
  • Error scaling: 4.0(2) × 10^-4 per transport operation is the leakage rate, accounting for nearly all of the linear memory error.Leakage is extracted using ternary measurement during transport-1QRB.
  • Interpretation: The remaining coherent error may arise from phase-tracking imperfections or other unaccounted noise sources after accounting for magnetic-field drift.Typical drift between calibrations contributes approximately 3 × 10^-5 in a depth-1 circuit.

5. Mid-circuit measurement and reset crosstalk

Helios characterizes measurement-and-reset crosstalk with target and spectator qubits, then tests its system-level impact in random Clifford circuits containing MCMRs. Component-level errors and effective system-level errors are broadly consistent, although the MCMR comparison remains heuristic.

  • Crosstalk characterization: MCMR crosstalk arises when unmeasured or unreset qubits absorb stray measurement or reset light, producing bit-flip, leakage, or dephasing errors.Target qubits are repeatedly measured and reset while spectator qubits are prepared in |0⟩ or |1⟩ and measured ternarily.
  • Crosstalk characterization: 2.1(1) × 10^-4 local and 4.8(1) × 10^-5 global crosstalk infidelities are measured per MCMR.Local crosstalk concerns the three laser-adjacent spectators, whereas global crosstalk averages over all 97 spectators.
  • System-level benchmark: Random Clifford-with-MCMR layers apply random 1Q gates, randomly paired RZZ(π/2) gates, and MCMR operations on a random subset of qubits.A random stabilizer is tracked through each circuit to classify each shot as success or failure and compute polarization.
  • System-level benchmark: Layer fidelity increases slightly from nm = 8 to nm = 16, with overlapping error bars, because 16 measurements use a protected measure scheme in the operation zones.The protected scheme mitigates MCMR crosstalk in those zones.
  • System-level benchmark: 2.0(3) × 10^-3 is the effective 2Q gate error from the no-MCMR data, compared with a predicted 2.2(1) × 10^-3 from 2Q and memory errors.The effective error includes 1Q gates, 2Q gates, and memory errors.
  • System-level benchmark: For nm = 8 and nm = 16, the measured effective MCMR errors are (2.6 ± 1.3) × 10^-3 and 1.0(7) × 10^-3, respectively.The corresponding component-level predictions are 2.2(1) × 10^-3 and 1.7(1) × 10^-3; tighter error bars are needed to assess consistency.

2. RCS mirror benchmarking

Random circuit sampling uses 98-qubit mirrored circuits to evaluate Helios fidelity and classical sampling difficulty. The fitted effective error agrees with random-Clifford and component benchmarks, while estimated classical costs remain vastly beyond existing supercomputers.

  • Benchmark design: RCS benchmarks whether a quantum processor generates computationally complex states while probing agreement with component-level performance.Its classical sampling difficulty makes it a well-vetted system-level benchmark.
  • Circuit construction: Helios RCS circuits interleave l layers of random-regular-graph RZZ(π/2) gates with l + 1 layers of Haar-random single-qubit gates.Each 2Q layer contains N/2 gates and each 1Q layer contains N gates.
  • Results: The measured fidelity follows an exponential decay with circuit depth under the gate-counting model.Figure 13 reports fidelity for N = 98 mirrored RCS circuits as a function of depth.
  • Experimental procedure: Mirrored circuits start from random computational-basis states and use 1000–2500 shots across 100 random circuit connectivities at each depth.Random initialization prevents unequal SPAM errors between the two basis states from biasing the fidelity estimate.
  • Results: 2.00(6) × 10^-3 is the fitted effective average 2Q error, consistent with random-Clifford circuits and component benchmarks.The fit also estimates pspam = 5.3(51) × 10^-4 using the gate-counting model.
  • Classical cost: Tensor-network estimates indicate that sampling comparable forward circuits is vastly beyond existing supercomputers, although the contraction-cost optimization is approximate.Additional optimization could mildly improve the estimates without changing the overall conclusion.

IV. OUTLOOK

Helios demonstrates state-of-the-art capabilities at approximately 100 qubits while retaining several clear improvement paths. Its architecture and junction support larger QCCD processors with all-to-all connectivity and fault-tolerance-relevant design options.

  • Current performance and improvements: Helios already exhibits state-of-the-art capabilities at the scale of approximately 100 qubits, with expected improvements in gate, memory, transport, and compilation performance.The authors identify halving 2Q gate error, dynamic decoupling, faster transport, and better compilation as improvement paths.
  • Scaling outlook: Parallelized cooling operations reduce time spent in 2Q gate zones and suggest increasing the ratio of cooling zones to gate zones to raise clock speed.Earlier generations used the same space for cooling and gating, with cooling operations up to two orders of magnitude slower.
  • Scientific reach: Helios is far beyond classical simulation abilities in the reported RCS demonstration, while its power and limitations are not yet fully understood.The authors also connect the platform to quantum simulations of superconductivity and certified-randomness protocols.
  • Scaling outlook: The integrated four-way junction supports larger QCCD processors that maintain all-to-all connectivity for many qubits.This architecture opens design space for high-efficiency encodings, transversal logic, low-overhead magic-state factories, and single-shot error correction.

A. Quantum logic

Helios combines trapped-ion quantum operations, transport, cooling, and real-time control to execute arbitrary programs with all-to-all connectivity. Program profiling identifies transport, especially ring rotations and global shifts, as the dominant contributors to layer time.

  • Quantum operations: 2Q gates use Mølmer–Sørensen interactions in four operation zones, with wrapper pulses converting the interaction into ZZ gates.The interaction uses Raman beams and the axial stretch mode at 1.86 MHz.
  • Ground-state cooling: Approximately 3 ms is required for sideband-cooling sequences to achieve ground-state cooling.The cooling configuration uses counter-propagating Raman beams with projections on all three principal axes.
  • Qubit stability: Helios uses an externally imposed 3.95 G bias field to make the qubit states approximate clock states.Their second-order magnetic-field coefficient is 488.8 Hz/G^2 at zero field.
  • Qubit stability: A real-time spatial phase-tracking routine mitigates qubit-frequency variations from magnetic-field drift, gradients, and AC Zeeman shifts.The routine applies corrections based on measurements of the average magnetic field in the quantum operation zones.
  • Program profiling: Ring rotations dominate transport time in fully dense random programs, with global shifts the second-largest contributor.Compiler optimizations and faster transport are identified as routes to reducing transport time and improving depth-1 time.

A3. BENCHMARKING DETAILS

The correlated-error analysis compares joint and individual subsystem decay factors in simultaneous randomized benchmarking. It finds no statistically significant correlated errors, while simulations quantify detection thresholds for such errors.

  • Test procedure: The analysis estimates individual λ{i} and pairwise λ{i,j}, then applies statistical tests with Bonferroni correction across subsystem pairs.The tests use a normal approximation and bootstrap-estimated standard deviation.
  • Test statistic: a = log(λ{i,j}) − log(λ{i}) − log(λ{j}) tests whether pairwise decay exceeds the product expected without correlated errors.Under the null hypothesis, a = 0; correlated errors imply a > 0.
  • Statistical calibration: The family-wise error rate is at most 5% when no correlated errors are present.This reflects the stated hypothesis-testing procedure.
  • Results: No a values are larger than zero at 95% confidence, indicating no statistically significant evidence for correlated errors.The analysis may not detect sufficiently small correlated errors.
  • Detection sensitivity: Two-subsystem correlated errors would need to contribute approximately 10% of total 2QRB error for at least 50% detection probability.For 1QRB, the corresponding threshold is approximately 50% of total error.

B. Detailed component benchmarking data and experimental details

The supplementary benchmarks define average infidelity and detail randomized-benchmarking protocols for single- and two-qubit operations, including standard and leakage-sensitive SPAM measurements.

  • Metric definition: Average infidelity is defined for an error process E composed with the intended operation U, averaged over pure computational-Hilbert-space states.The state average uses the Haar measure.
  • SPAM benchmarking: 2.7(8) × 10^-3 and 5.7(1) × 10^-3 are ternary-measurement leakage probabilities for prepared |0⟩ and |1⟩ states, respectively.Conditioned on non-leakage, the SPAM errors are 7(1) × 10^-4 and 2.8(2) × 10^-3.
  • Randomized benchmarking: 1QRB applies random single-qubit Clifford sequences followed by an inverse Clifford that randomly includes an X gate.The protocol is run simultaneously on 16 qubits in eight operation zones.
  • Randomized benchmarking: 2QRB applies random two-qubit Clifford sequences to eight qubit pairs, with two-qubit gates executed in parallel in four operation zones.The supplementary table reports leakage rates and average infidelities for these pairs.

4. Two-qubit cycle benchmarking

Two-qubit cycle benchmarking estimates Pauli fidelities by fitting repeated Pauli-twirled gate experiments, then converts them into Pauli error probabilities and average infidelity.

  • Protocol: 2QCB prepares Pauli eigenstates, repeatedly applies a Pauli-twirled two-qubit gate, and measures in the corresponding Pauli basis.Pauli twirling permits treating the error channel as a stochastic Pauli channel.
  • Channel model: For a stochastic Pauli channel, Pauli operators are eigenoperators satisfying E(P_i) = f_i P_i.The eigenvalues f_i are called Pauli fidelities.
  • Error reconstruction: The Pauli error probabilities are obtained from Pauli fidelities using commutation or anticommutation relations between Pauli operators.The relation is expressed through the indicator ⟨i,j⟩, which is 0 for commuting and 1 for anticommuting operators.
  • Estimation: E_l(P_i) = Tr(P_i C_l(P_i)) is fit to an exponential model to estimate Pauli fidelities.The fitted fidelities are then used to compute Pauli error probabilities via Eq. (A6).
  • Infidelity: The average infidelity excluding leakage is the sum-based quantity over all non-identity Pauli error probabilities.The analysis includes symmetry assumptions and unlearnable degrees of freedom when estimating these quantities.

5. Transport-1QRB

The section develops benchmarking methods for transport and MCMR crosstalk, including state-resolved error estimation and simultaneous-target experiments. These measurements quantify crosstalk contributions for operation zones and the storage ring.

  • Transport-1QRB: Transport-1QRB used sequence lengths l ∈ {8, 64, 128}, with 10 circuits and 100 shots per sequence length.Here, sequence length denotes the number of depth-1 transport operations.
  • MCMR crosstalk: MCMR crosstalk is estimated by fitting spectator-qubit survival probabilities to a linear decay model and relating fit parameters to effective quantum-jump error rates.For ^137Ba+ qubits, additional spectator states and ternary measurement separate bit-flip and leakage errors.
  • MCMR crosstalk: The average infidelity combines conditional state-transition probabilities and phase-flip probability across the computational and leakage states.The expression includes p(0|1), p(1|0), p(L|0), p(L|1), and pZ.
  • MCMR crosstalk: Estimating pZ uses an approximation based on transition probabilities and an elastic-scattering term under a weak-crosstalk assumption.The X/Y-eigenstate circuits needed to measure pZ incur additional memory error, while the elastic contribution remains subject to future study on Helios.
  • MCMR crosstalk: 5.2(2) × 10^-5 and 1.21(4) × 10^-5 are the average MCMR crosstalk errors per qubit in operation zones and the storage ring, respectively.The experiment applied MCMRs simultaneously to eight target qubits, with spectators in operation zones and the storage ring.

C. Random Clifford circuits with mid-circuit measurements

This section evaluates random Clifford circuits containing mid-circuit measurements and resets using stabilizer tracking and polarization fits. It compares effective circuit fidelities with predictions assembled from component-level gate, memory, SPAM, and crosstalk measurements.

  • Benchmark method: Random Clifford circuits are classically verified by tracking a random stabilizer and testing whether measured parity matches the evolved stabilizer sign.A trial succeeds when the parity of final and mid-circuit measurement outcomes agrees with the tracked stabilizer.
  • Benchmark method: For nm ∈ {0, 8, 16} and l ∈ {2, 4, 6, 8}, polarization is fit to ypol(l, nm) = A F(nm)^l to estimate process fidelity per circuit layer.Here nm is the number of MCMRs per circuit layer, with 10 circuits and 100 shots for each parameter pair.
  • Results: The process-fidelity results are reported for 0, 8, and 16 mid-circuit measurements and resets per circuit layer.These values are plotted in Fig. 12 and summarized in the corresponding process-fidelity table.
  • Results: The effective fidelities estimated from random-Clifford data are compared with values predicted from component-level benchmarking data.The comparison is summarized in Table A8.
  • Prediction procedure: Effective two-qubit fidelity is predicted by combining the two-qubit gate process infidelity with twice the depth-1 memory error per qubit.The component values used are ϵavg,2Q = 7.9(2) × 10^-4 and ϵmem = 6.0(3) × 10^-4.
  • Prediction procedure: Effective MCMR fidelity predictions combine standard SPAM error with measured crosstalk errors for the relevant simultaneous-measurement configurations.For nm = 16, the protected-measure scheme omits operation-zone crosstalk and uses storage-ring crosstalk in the prediction.
Loading 2511.05465v1…