Source-linked AI summary
Low-distance Surface Codes under Realistic Quantum Noise
Yu Tomita, Krysta M. Svore
TL;DR
The paper addresses limited threshold analyses for modified low-distance layouts and realistic non-Clifford noise. It evaluates distance-three layouts through depolarizing and damping simulations, finding pessimistic Pauli-twirled error estimates and practical regimes favoring Surface-17.
Problem
Thresholds for modified surface-code layouts and realistic non-Clifford noise models had received limited analysis despite the need to assess near-term implementations.
Method
The study simulates distance-three layouts under depolarizing, amplitude-and-phase-damping, and Pauli-twirled channels using Monte Carlo methods and a small lookup-table decoder.
Results
Pauli twirling produces a pessimistic estimate of logical error rates, while Surface-17 improves logical error rates at 1 µs T1 under suitable gate settings.
Takeaways & Limitations
Surface-17 is the preferred layout for early implementations because it uses fewer qubits and gates while achieving improved logical error rates in supported regimes.
Abstract
from arXiv · showhide
We study the performance of distance-three surface code layouts under realistic multi-parameter noise models. We first calculate their thresholds under depolarizing noise. We then compare a Pauli-twirl approximation of amplitude and phase damping to amplitude and phase damping. We find the approximate channel results in a pessimistic estimate of the logical error rate, indicating the realistic threshold may be higher than previously estimated. From Monte-Carlo simulations, we identify experimental parameters for which these layouts admit reliable computation. Due to its low resource cost and superior performance, we conclude that the 17-qubit layout should be targeted in early experimental implementations of the surface code. We find that architectures with gate times in the 5-40 ns range and T1 times of at least 1-2 us range will exhibit improved logical error rates with a 17-qubit surface code encoding.
I. INTRODUCTION
Surface codes offer a nearest-neighbor route to fault-tolerant quantum computation, but modified low-distance layouts and realistic non-Clifford noise require further analysis. This work evaluates distance-three layouts, realistic noise models, and simulation methods for near-term implementations.
- Surface codes use a 2-D planar layout with nearest-neighbor interactions and can tolerate error rates up to approximately 1%.
- Thresholds for modified surface-code layouts had not been analyzed, while realistic non-Clifford noise models remained costly to simulate.
- The study determines distance-three thresholds under depolarizing, amplitude-and-phase-damping, and Pauli-twirled noise models.
- Distance-three surface codes encode one logical qubit using stabilizer measurements and classical decoding to detect and correct physical errors.
- The paper examines Surface-25, Surface-17, and Surface-13 layouts and compares their stabilizers, logical operators, and measurement circuits.
B. 13- and 17-qubit Layouts
Surface-17 and Surface-13 retain distance three while reducing qubit resources, with Surface-13 trading fewer qubits for greater circuit depth. A lookup-table decoder uses short syndrome histories to identify likely low-weight corrections for these small codes.
- Surface-17: Surface-17 uses 9 data qubits and 8 syndrome qubits, totaling 17 while retaining distance three.It is formed by rotating Surface-25 and removing four corner data qubits.
- Surface-13: Surface-13 uses 4 syndrome qubits by reusing each for both X- and Z-stabilizer measurements, reducing qubits but adding 4 time steps per round.
- Layout comparison: Surface-17 and Surface-13 reduce resources by 32–48% while remaining distance-three surface codes.
- Lookup-table decoder: The decoder prioritizes short error chains, ignores repeated temporal flips as likely measurement errors, and applies corrections from neighboring spatial or temporal syndrome flips.
- Lookup-table decoder: For these small surface codes, restricting matching to neighboring syndrome pairs performs as well as considering more distant pairs.
- Lookup-table decoder: The lookup table maps syndrome-volume measurement flips to probable data-qubit errors using constant time and 2n space, where n is the number of data qubits.
B. Improved Stabilizer Measurement Circuits
Using the same CNOT ordering for X- and Z-stabilizers can propagate syndrome-qubit errors into logical errors in the reduced layouts. Reordering Z-stabilizer CNOTs prevents this while preserving circuit depth and size.
- Failure of the original ordering: A syndrome-qubit Z error can propagate to two data-qubit Z errors and then form a logical Z chain because distance-three codes require only three data qubits.
- Reordered circuits: The proposed circuit measures X- and Z-type stabilizers in different CNOT orders to prevent creation of a logical error.
- Reordered circuits: The reordered Z-stabilizer sequence preserves commutation relations, circuit depth, and circuit size while mapping the example to a single Z error instead of a logical error.
A. Symmetric and Asymmetric Depolarizing Channels
The paper distinguishes depolarizing noise from amplitude and phase damping and considers Pauli-twirl approximations for realistic decoherence. Amplitude damping models energy dissipation, while phase damping corresponds to pure dephasing or phase flips.
- Depolarizing channels: The depolarizing channel applies identity or discrete X, Y, and Z errors, with the symmetric case satisfying pX = pY = pZ.
- Depolarizing channels: The asymmetric depolarizing channel allows independent probabilities for X, Y, and Z errors.
- Amplitude damping: Amplitude damping models energy dissipation, including spontaneous photon emission, with pAD denoting the probability of emitting a single photon.
- Amplitude damping: The amplitude-damping simulation samples whether the qubit measures 0 or 1, then applies either Ry(θ) or damping to |0⟩ without an extra ancilla.
- Phase damping: Phase damping, also called pure dephasing, is equivalent to a phase-flip channel and can be expressed using Z-error probabilities.
- Combined decoherence: Amplitude and phase damping together model single-qubit decoherence using gate time t and relaxation and dephasing times T1 and T2.
C. Approximate Amplitude and Phase Damping Channel
Pauli twirling approximates the amplitude and phase damping channel by removing off-diagonal terms and expressing the result as an asymmetric depolarizing channel. Its failure probabilities depend on gate execution time and qubit relaxation and dephasing times, with independent errors used to approximate two-qubit failures.
- Pauli twirling removes off-diagonal terms, yielding an asymmetric depolarizing noise channel.
- The channel’s failure probabilities depend on gate execution time t, relaxation time T1, and dephasing time T2.
- Two-qubit error probabilities are approximated assuming independent errors.
- The approximation assigns joint two-qubit Pauli probabilities using products of the corresponding single-qubit probabilities.
V. EXPERIMENTAL SETUP
The simulations use LIQUi|⟩’s Universal simulation environment to model arbitrary noisy quantum circuits. Large-memory HPC nodes support roughly 30 qubits each, while identity gates represent idle periods and different timestep location types.
- LIQUi|⟩ provides Universal simulation for full simulation of arbitrary quantum circuits.The software also includes Stabilizer simulation for Clifford circuits, but the study uses Universal simulation for consistency.
- Universal simulation is limited by machine memory, with each HPC node simulating roughly 30 qubits.
- Figure 7 inserts identity gates according to whether the other timestep location is preparation, a single-qubit gate, a two-qubit gate, or measurement.
- The simulations required thousands of hours of compute time.
A. Monte-Carlo Simulation
Monte-Carlo simulations apply noise after each circuit location to estimate logical error rates for depolarizing, approximate damping, and realistic damping models. Idle qubits receive identity gates whose durations follow the active timestep location.
- Monte-Carlo simulation computes logical error rates by applying a noise model after each qubit location at every timestep.
- Depolarizing and approximate damping noise replace non-measurement locations with probabilistic X, Y, or Z errors.
- Amplitude and phase damping, together with its Pauli-twirl approximation, is applied after every location using the current timestep duration.
- Each syndrome-volume window contains three surface-code rounds, with one layer overlapping the previous window.
B. Logical Error Rate Calculation
Logical error rates are estimated by repeatedly decoding three-layer syndrome volumes, applying tracked corrections, and checking the resulting logical state. Three rounds are optimal for distance-three layouts, and the study reports rates per window or per round depending on the noise analysis.
- The simulation initializes a quiescent encoded state before executing noisy syndrome-measurement rounds.
- Each iteration records syndrome flips, decodes a three-layer volume, applies noise-free corrections, checks for a logical error, and repeats.
- Three rounds are the optimal window size for distance-three layouts, while m ranges from 10 to 200 detected logical errors.
- The logical error rate is calculated per window because experiments measure the logical qubit after completing a window.
- Because only distance d = 3 is simulated, the reported threshold is a pseudothreshold defined by the crossing of p and Pr.
- For depolarizing noise the study calculates P1 and P3, whereas damping and Pauli-twirl analyses calculate P3.
C. Architectural Settings
The study evaluates distance-three layouts across superconducting and ion-trap settings, then identifies Surface-17 as the preferred layout under depolarizing noise. Its advantage reflects lower resource requirements and slightly higher pseudothresholds than the alternatives.
- Architectural settings: Six architecture settings span superconducting and ion-trap platforms with differing relaxation, dephasing, gate, preparation, and measurement times.The time per round depends on the layout and architectural parameters.
- Depolarizing-noise comparison: Surface-13 has slightly lower pseudothresholds because its circuit depth is higher, while Surface-25’s larger data and syndrome resources also reduce its pseudothreshold.These comparisons are made under symmetric depolarizing noise.
- Depolarizing-noise comparison: Logical X and Z pseudothresholds for Surface-17 are comparable, with similar behavior observed for Surface-13 and Surface-25.The comparison uses logical |1L⟩ for bit flips and |+L⟩ for phase flips.
- Preferred layout: Surface-17 requires roughly half the depth of Surface-13 and significantly fewer qubits and gates than Surface-25.It also exhibits slightly higher pseudothresholds than the other layouts.
- Preferred layout: Surface-17 is selected for the remaining experiments because of its resource savings and slightly higher pseudothresholds.The simulations therefore use the 17-qubit layout for realistic-noise studies.
B. Amplitude and Phase Damping vs. Pauli Twirling
The paper compares Pauli-twirled and physical amplitude-and-phase damping for Surface-17. Pauli twirling matches phase-flip behavior but overestimates bit-flip errors, making decoherence thresholds appear worse than they may be.
- Channel comparison: For T1 = 10 µs, the Pauli-twirl approximation gives per-round logical error rates PZ,1 = 4.27 × 10^-3 and PX,1 = 4.41 × 10^-3.These values closely reproduce earlier Surface-25 results, with small differences expected from the layout change.
- Phase-flip errors: The approximate and physical channels produce closely matching logical Z error rates per window under the SCH setting.The comparison concerns an encoded |+L⟩ state in Surface-17.
- Bit-flip errors: Pauli twirling produces much higher logical X error rates than amplitude and phase damping, especially as T1 increases.The comparison concerns an encoded |1L⟩ state under SCH.
- Practical comparison: The physical-channel improvement regime is larger than the Pauli-twirl regime, indicating that Pauli twirling pessimistically estimates Surface-17’s logical error rate.The regime is defined by the unencoded error surface exceeding the encoded logical-error surfaces.
C. Amplitude and Phase Damping
Under physical amplitude and phase damping, Surface-17’s benefit depends strongly on gate speed, T1, and memory duration. Faster gates and longer relaxation times expand the regime where encoding lowers logical error rates.
- Architecture dependence: As T1 increases, Surface-17 improves logical error rates over a larger range of memory durations across all six architecture settings.Amplitude damping probability increases monotonically with memory duration.
- Superconductor settings: At T1 = 1 µs, SCS encoding provides no improvement, whereas SCF encoding improves the logical error rate for at least 8 µs of memory.At T1 = 10 µs, SCF improves performance for memories of 2 µs or longer.
- Superconductor settings: At T1 = 1 µs and 10 µs, SCH encoding lowers logical error rates for memory durations longer than 150 ns and 20 ns, respectively.SCH uses substantially faster preparation, measurement, and gate times than SCF.
- Superconductor settings: At T1 = 1 µs, SCD requires memory durations roughly three times longer than SCH for encoding to reduce logical error rates.The difference is associated with SCD’s four-times-longer CNOT gate.
- Ion-trap settings: Ion-trap encoding improves logical error rates above roughly 300–400 µs for ITS and around 15 µs for ITF.ITF assumes ten-times-faster CNOT gates than ITS.
- Overall performance: Improving gate times from SCF to SCH can yield an order-of-magnitude reduction in logical error rate.The comparison assumes a three-round memory duration.
VII. CONCLUSION
The paper finds that Surface-17 is promising under realistic noise, while Pauli twirling overestimates logical bit-flip rates. Simulations identify architecture regimes where near-term experiments may demonstrate surface-code error correction.
- Conclusions: Surface-13 has a slightly lower depolarizing pseudothreshold than Surface-17 and Surface-25, while Pauli twirling pessimistically estimates logical bit-flip rates.The latter result suggests realistic-noise thresholds may exceed earlier Pauli-twirling estimates.
- Conclusions: Gate durations between SCF and SCH with current T1 times improve Surface-17 logical error rates, whereas SCS gates are too slow for significant improvement.For example, SCF with T1 around 10 µs may produce experimentally detectable improvement, and SCH can work at shorter T1 times.
- Future work: Clifford-gate decoherence approximations have been more accurate than Pauli twirling at the gate-operation level, but circuit-level studies remain a future direction.The cited prior studies did not analyze complete code circuits.