Article

Suppressing quantum errors by scaling a

surface code logical qubit

https://doi.org/10.1038/s41586-022-05434-1 Google Quantum AI*

Received: 13 July 2022

Accepted: 10 October 2022 Practical quantum computing will require error rates well below those achievable
Published online: 22 February 2023 with physical qubits. Quantum error correction1,2 offers a path to algorithmically
relevant error rates by encoding logical qubits within many physical qubits,
Open access
for which increasing the number of physical qubits enhances protection against
Check for updates physical errors. However, introducing more qubits also increases the number
of error sources, so the density of errors must be sufficiently low for logical
performance to improve with increasing code size. Here we report the
measurement of logical qubit performance scaling across several code sizes,
and demonstrate that our system of superconducting qubits has sufficient
performance to overcome the additional errors from increasing qubit number.
We find that our distance-5 surface code logical qubit modestly outperforms an
ensemble of distance-3 logical qubits on average, in terms of both logical error
probability over 25 cycles and logical error per cycle ((2.914 ± 0.016)% compared
to (3.028 ± 0.023)%). To investigate damaging, low-probability error sources, we run
a distance-25 repetition code and observe a 1.7 × 10−6 logical error per cycle floor set
by a single high-energy event (1.6 × 10−7 excluding this event). We accurately model
our experiment, extracting error budgets that highlight the biggest challenges
for future systems. These results mark an experimental demonstration in which
quantum error correction begins to improve performance with increasing qubit
number, illuminating the path to reaching the logical error rates required for
computation.

Since Feynman’s proposal to compute using quantum mechanics3,

many potential applications have emerged, including factoring4, Surface codes with superconducting qubits
optimization5, machine learning6, quantum simulation7 and quan- Surface codes30–34 are a family of quantum error-correcting codes that
tum chemistry8. These applications often require billions of quantum encode a logical qubit into the joint entangled state of a d × d square
operations9–11 and state-of-the-art quantum processors typically have of physical qubits, referred to as data qubits. The logical qubit states
error rates around 10−3 per gate12–17, far too high to execute such large are defined by a pair of anti-commuting logical observables XL and ZL.
circuits. Fortunately, quantum error correction can exponentially For the example shown in Fig. 1a, a ZL observable is encoded in the joint
suppress the operational error rates in a quantum processor, at the Z-basis parity of a line of qubits that traverses the lattice from top to
expense of temporal and qubit overhead18,19. bottom, and likewise an XL observable is encoded in the joint X-basis
Several works have reported quantum error correction on codes parity traversing left to right. This non-local encoding of information
able to correct a single error, including the distance-3 Bacon–Shor20, protects the logical qubit from local physical errors, provided we can
colour21, five-qubit22, heavy-hexagon23 and surface24,25 codes, as well as detect and correct them.
continuous variable codes26–29. However, a crucial question remains of To detect errors, we periodically measure X and Z parities of adjacent
whether scaling up the error-correcting code size will reduce logical clusters of data qubits with the aid of d2 − 1 measure qubits interspersed
error rates in a real device. In theory, logical errors should be reduced if throughout the lattice. As shown in Fig. 1b, each measure qubit interacts
physical errors are sufficiently sparse in the quantum processor. In prac- with its neighbouring data qubits to map the joint data qubit parity
tice, demonstrating reduced logical error requires scaling up a device to onto the measure qubit state, which is then measured. Each parity
support a code that can correct at least two errors, without sacrificing measurement, or stabilizer, commutes with the logical observables of
state-of-the-art performance. In this work we report a 72-qubit super- the encoded qubit as well as every other stabilizer. Consequently, we
conducting device supporting a 49-qubit distance-5 (d = 5) surface can detect errors when parity measurements change unexpectedly,
code that narrowly outperforms its average subset 17-qubit distance-3 without disturbing the logical qubit state.
surface code, demonstrating a critical step towards scalable quantum A decoder uses the history of stabilizer measurement outcomes to
error correction. infer likely configurations of physical errors on the device. We can then

*A list of authors and their affiliations appears at the end of the paper.

676 | Nature | Vol 614 | 23 February 2023

a c Although device inhomogeneity makes this comparison difficult,
1Q CZ Meas. DD we can compare the distance-5 logical qubit to the average of four

Cumulative distribution
ZL 1
distance-3 logical qubit subgrids, each containing nine data qubits
and eight measure qubits. These distance-3 logical qubits cover the
four quadrants of the distance-5 code with minimal qubit overlap,
capturing the average performance of the full distance-5 grid.
0 In a single instance of the experiment, we initialize the logical qubit
10–3 10–2 state, run several cycles of error correction, and then measure the final
Pauli and measurement
error rates logical state. We show an example in Fig. 2a. To prepare a ZL eigenstate,
we first prepare each data qubit in |0⟩ or |1⟩ , an eigenstate of the
XL Data qubit (d2) Z stabilizers. The first cycle of stabilizer measurements then projects
Measure qubit (d2 – 1)
the data qubits into an entangled state that is also an eigenstate of the
Unused
Subset distance-3
X stabilizers. Each cycle contains CZ and Hadamard gates sequenced
to extract X and Z stabilizers simultaneously, and ends with the meas-
b
urement and reset of the measure qubits. In the final cycle, we also
measure the data qubits in the Z basis, yielding both parity information
and a measurement of the logical state. Preparing and measuring XL
eigenstates proceeds analogously. The instance succeeds if the
corrected logical measurement agrees with the known initial state;
Time
otherwise, a logical error has occurred.
Our stabilizer circuits contain a few modifications to the standard
Fig. 1 | Implementing surface code logical qubits. a, Schematic of a 72-qubit
gate sequence described above (see Supplementary Information),
Sycamore device with a distance-5 surface code embedded, consisting of 25 data
including phase corrections to correct for unintended qubit frequency
qubits (gold) and 24 measure qubits (blue). Each measure qubit is associated
with a stabilizer (blue coloured tile, dark: X, light: Z). Representative logical
shifts and dynamical decoupling gates during qubit idles43. We also
operators ZL (black) and XL (green) traverse the array, intersecting at the lower- remove certain Hadamard gates to implement the ZXXZ variant of the
left data qubit. The upper right quadrant (red outline) is one of four subset surface code44,45, which helps symmetrize the X- and Z-basis logical error
distance-3 codes (the four quadrants) that we compare to distance-5. rates. Finally, during initialization, the data qubits are prepared into
b, Illustration of a stabilizer measurement, focusing on one data qubit (labelled ψ) randomly selected bitstrings. This ensures that we do not preferentially
and one measure qubit (labelled 0), in perspective view with time progressing to measure even parities in the first few cycles of the code, which could
the right. Each qubit participates in four CZ gates (black) with its four nearest artificially lower logical error rates owing to bias in measurement error
neighbours, interspersed with Hadamard gates (H), and finally, the measure (see Supplementary Information).
qubit is measured and reset to |0⟩ (MR). Data qubits perform dynamical
decoupling (DD) while waiting for the measurement and reset. All stabilizers are
measured in this manner concurrently. Cycle duration is 921 ns, including 25-ns Error detectors
single-qubit gates, 34-ns two-qubit gates, 500-ns measurement and 160-ns reset
After initialization, parity measurements should produce the same
(see Supplementary Information for compilation details). The readout and reset
value in each cycle, up to known flips applied by the circuit. If we com-
take up most of the cycle time, so the concurrent data qubit idling is a dominant
source of error. c, Cumulative distributions of errors for single-qubit gates (1Q),
pare a parity measurement to the corresponding measurement in the
CZ gates, measurement (Meas.) and data qubit dynamical decoupling preceding cycle and their values are inconsistent, a detection event
(idle during measurement and reset), which we refer to as component errors. has occurred, indicating an error. We refer to these comparisons as
The circuits were benchmarked in simultaneous operation using random circuit detectors.
techniques, on the 49 qubits used in distance-5 and the 4 CZ layers from the The detection event probabilities for each detector indicate the
stabilizer circuit38,59 (see Supplementary Information). Vertical lines are means. distribution of physical errors in space and time while running the
surface code. In Fig. 2, we show the detection event probabilities in
the distance-5 code (Fig. 2b,c) and the distance-3 codes (Fig. 2d,e) run-
ning for 25 cycles, as measured over 50,000 experimental instances.
determine the overall effect of these inferred errors on the logical qubit, For the weight-4 stabilizers, the average detection probability is
thus preserving the logical state. Most surface code logical gates can 0.185 ± 0.018 (1σ) in the distance-5 code and 0.175 ± 0.017 averaged
be implemented by maintaining logical memory and executing differ- over the distance-3 codes. The weight-2 stabilizers interact with fewer
ent sequences of measurements on the code boundary35–37. Thus, we qubits and hence detect fewer errors. Correspondingly, they yield a
focus on preserving logical memory, the core technical challenge in lower average detection probability of 0.119 ± 0.012 in the distance-5
operating the surface code. code and 0.115 ± 0.008 averaged over the distance-3 codes. The relative
We implement the surface code on an expanded Sycamore device38 consistency between code distances suggests that growing the lattice
with 72 transmon qubits39 and 121 tunable couplers40,41. Each qubit is does not substantially increase the component error rates during error
coupled to four nearest neighbours except on the boundaries, with correction.
mean qubit coherence times T1 = 20 μs and T2,CPMG = 30 μs, in which The average detection probabilities exhibit a relative rise of 12% for
CPMG represents Carr–Purcell–Meiboom–Gill. As in ref. 42, we imple- distance-5 and 8% for distance-3 over 25 cycles, with a typical character-
ment single-qubit rotations, controlled-Z (CZ) gates, reset and measure- istic risetime of roughly 5 cycles (see Supplementary Information). We
ment, demonstrating similar or improved simultaneous performance attribute this rise to data qubits leaking into non-computational excited
as shown in Fig. 1c. states and anticipate that the inclusion of leakage-removal techniques
The distance-5 surface code logical qubit is encoded on a 49-qubit on data qubits would help to mitigate this rise42,46–48. We reason that
subset of the device, with 25 data qubits and 24 measure qubits. Each the greater increase in detection probability in the distance-5 code is
measure qubit corresponds to one stabilizer, classified by its basis due to increased stray interactions or leakage from simultaneously
(X or Z) and the number of data qubits involved (weight, 2 or 4). Ideally, operating more gates and measurements.
to assess how logical performance scales with code size, we would We test our understanding of the physical noise in our system by
compare distance-5 and distance-3 logical qubits under identical noise. comparing the experimental data to a simulation. We begin with a

Nature | Vol 614 | 23 February 2023 | 677

Article
a Initial data
Inferred
t=0 t=1 t=2 Final data t=3
parities
Spacelike
Hadamard pair
Spacetimelike pair

Timelike pair

Meas. error

b c d e f g h
0.25 0.05 Exp.
Pauli+
Detection prob., pd

0.20 0.04 Pauli

Correlation, pij
0.15 0.03
0.10 0.02
0.05 Weight-4 Weight-4 Weight-4 0.01
Weight-2 Weight-2 Weight-2
0 0
0 5 10 15 20 25 0.10 0.25 0 5 10 15 20 25 0.10 0.25 0 5 10 15 20 25

Unexp.
Time
Space
Spacetime
0.10 0.25
QEC cycle, t pd QEC cycle, t pd QEC cycle, t pd

Fig. 2 | Error detection in the surface code. a, Illustration of a surface code correction. c, Detection probability heatmap, averaging over t = 1 to 24.
experiment, in perspective view with time progressing to the right. We begin d,e, Similar to b,c for four separate distance-3 experiments covering the four
with an initial data qubit state that has known parities in one stabilizer basis quadrants of the distance-5 code. f,g, Similar to b,c using a simulation with Pauli
(here, Z). We show example errors that manifest in detection pairs: a Z error errors plus leakage, crosstalk and stray interactions (Pauli+). h, Bar chart
(red) on a data qubit (spacelike pair), a measurement error (purple) on a measure summarizing the detection correlation matrix pij, comparing the distance-5
qubit (timelike pair), an X error (blue) during the CZ gates (spacetimelike pair) experiment from b to the simulation in f (Pauli+) and a simpler simulation with
and a measurement error (green) on a data qubit (detected in the final inferred only Pauli errors. We aggregate four groups of correlations: timelike pairs;
Z parities). b, Detection probability for each stabilizer over a 25-cycle distance-5 spacelike pairs; spacetimelike pairs expected for Pauli noise; and spacetimelike
experiment (50,000 repetitions). Darker lines: average over all stabilizers with pairs unexpected for Pauli noise (Unexp.), including correlations over two
the same weight. There are fewer detections at timestep t = 0 because there is timesteps. Each bar shows a mean and standard deviation of correlations from a
no preceding syndrome extraction, and at t = 25 because the final parities are 25-cycle, 50,000-repetition dataset.
calculated from data qubit measurements directly. QEC, quantum error

depolarizing noise simulation based on the component error informa- and averaged according to the different classes of pairs. In addition to
tion in Fig. 1c, and then extend to a Pauli simulation with qubit-specific T1 the expected pairs, we also quantify how often detection pairs occur
and T2,CPMG, transitions to leaked states, and stray interactions between that are unexpected in a local depolarizing circuit model. Overall, the
qubits during CZ gates (see Supplementary Information). We refer Pauli simulation systematically underpredicts these probabilities
to this simulation as Pauli+. Figure 2f shows that this second simula- compared to experimental data, whereas the Pauli+ simulation is closer
tor accurately predicts the average detection probabilities, finding and predicts the presence of unexpected pairs, which we surmise are
0.180 ± 0.013 for the weight-4 stabilizers and 0.116 ± 0.011 for the related to leakage and stray interactions. These errors can be espe-
weight-2 stabilizers, with average detection probabilities increasing cially harmful to the surface code because they can generate multiple
7% over 25 cycles (distance-5). detection events distantly separated in space or time, which a decoder
might wrongly interpret as multiple independent component errors.
We expect that mitigating leakage and stray interactions will become
Understanding errors through correlations increasingly important as error rates decrease.
We next examine pairwise correlations between detection events, which
give us fine-grained information about which types of error are occur-
ring during error correction. Figure 2a illustrates a few examples of Decoding and logical error probabilities
pairwise detections that are generated by X or Z errors in the surface We next examine the logical performance of our surface code qubits.
code. Measurement and reset errors are detected by the same stabilizer To infer the error-corrected logical measurement, the decoder requires
in two consecutive cycles, which we classify as a timelike pair. Data a probability model for physical error events. This information may
qubits may experience an X (Z) error while idling during measurement be expressed as an error hypergraph: detectors are vertices, physical
that is detected by its neighbouring Z (X) stabilizers in the same cycle, error mechanisms are hyperedges connecting the detectors they trig-
forming a spacelike pair. Errors during CZ gates may cause a variety ger, and each hyperedge is assigned its corresponding error mecha-
of pairwise detections to occur, including spacetimelike pairs that are nism probability. We use a generalization of pij to determine these
separated in both space and time. More complex clusters of detection probabilities42,50.
events arise when a Y error occurs, which generates detection events Given the error hypergraph, we implement two different decoders:
for both X and Z errors. belief-matching, an efficient combination of belief propagation and
To estimate the probability for each detection event pair from our minimum-weight perfect matching51; and tensor network decoding,
data, we compute an appropriately normalized correlation pij between a slow but accurate approximate maximum-likelihood decoder. The
detection events occurring on any two detectors i and j (refs. 42,49; belief-matching decoder first runs belief propagation on the error
see Supplementary Information). In Fig. 2h, we show the estimated prob- hypergraph to update hyperedge error probabilities based on nearby
abilities for experimental and simulated distance-5 data, aggregated detection events51,52. The updated error hypergraph is then decomposed

678 | Nature | Vol 614 | 23 February 2023

a 0.45 By contrast, a maximum-likelihood decoder considers all possible
d = 3 average sets of errors consistent with the detection events, splits them into
d = 3 individual quadrants two groups on the basis of whether they flip the logical measurement,
0.40
d=5 and chooses the group with the greater total likelihood. The two likeli-
Fit to experimental average
Pauli+ simulation
hoods are each expressed as a tensor network contraction51,54,55 that
0.35
exhaustively sums the probabilities of all sets of errors within each
group. We can contract the network approximately, and verify that the
0.30 approximation converges. This yields a decoder that is nearly optimal
Logical error probability, pL

given the hypergraph error priors, but is considerably slower. Further

0.25 improvements could come from a more accurate prior, or by incorpo-
rating more fine-grained measurement information47,56.
Figure 3 shows a comparison of the logical error performance
0.20
of the distance-3 and distance-5 codes using the approximate
maximum-likelihood decoder. As the ZXXZ variant of the surface code
0.15 symmetrizes the X and Z bases, differences between the two bases’
logical error per cycle are small and attributable to spatial variations
0.10 in physical error rates. Thus, for visual clarity, we report logical error
probabilities averaged between the X and Z basis; the full dataset may
0.05 be found in the Supplementary Information. Note that we do not
post-select on leakage or high-energy events to capture the effects
of realistic non-idealities on logical performance. Over all 25 cycles of
b 0
error correction, the distance-5 code realizes lower logical error prob-
1.0
Logical fidelity, F = 1 – 2pL

abilities pL than the average of the subset distance-3 codes.

0.8
We fit the logical fidelity F = 1 − 2pL to an exponential decay. We start
0.6 the fit at t = 3 to avoid two phenomena that advantage the larger code:
the lower detection probability during the first cycle relative to subse-
0.4 d = 3 average quent cycles (Fig. 2b,d), and the higher effective threshold caused by
Fit, H3 = (3.028 ± 0.023)% the confinement of errors to thin time slices in few-cycle experiments31.
d=5
We obtain a logical error per cycle ε5 = (2.914 ± 0.016)% (1σ statistical
Fit, H3 = (2.914 ± 0.016)%
0.2 and fit uncertainty) for the distance-5 code, compared to an average
0 5 10 15 20 25 of ε3 = (3.028 ± 0.023)% for the subset distance-3 codes, a relative error
Quantum error correction cycle, t reduction of about 4%. When decoding with the faster belief-matching
c 0.98 decoder, we fit a logical error per cycle of (3.056 ± 0.015)% for the
d = 3 average distance-5 code, compared to an average of (3.118 ± 0.025)% for the
d = 3 individual distance-3 codes, a relative error reduction of about 2%. We note that
the distance-5 logical error per cycle is slightly higher than those of
0.96
two of the distance-3 codes individually, and that leakage accumula-
tion may cause distance-5 performance to degrade faster than that of
distance-3 as logical error probability approaches 50%.
s
ent
1 – H5

0.94 In principle, the logical performance of a distance-5 code should improve

em

faster than that of a distance-3 code as physical error rates decrease33.

rov

Over time, we improved our physical error rates, for example by opti-
imp

0.92 H5 mizing single- and two-qubit gates, measurement and data qubit idling
tem

=
H3 (see Supplementary Information). In Fig. 3c, we show the corresponding
Sys

performance progression of distance-5 and distance-3 codes. The larger

0.90 code improved about twice as fast until finally overtaking the smaller
0.90 0.92 0.94 0.96 0.98 code, validating the benefit of increased-distance protection in practice.
1 – H3 To understand the contributions of individual components to our
logical error performance, we follow ref. 42 and simulate the distance-5
Fig. 3 | Logical error reduction. a, Logical error probability pL versus cycle and distance-3 codes while varying the physical error rates of the vari-
comparing distance-5 (blue) to distance-3 (pink: four separate quadrants, ous circuit components. As the logical-error-suppression factor
red: average), all averaged over ZL and XL. Each individual data point represents
100,000 repetitions. Solid line: fit to experimental average, t = 3 to 25 (see main Λd /(d +2) = εd /εd +2 (1)
text). Dotted line: comparison to Pauli+ simulation. b, Logical fidelity F = 1 − 2pL
versus cycle, semilog plot. The datapoints and fits are the experimental averages
is approximately inversely proportional to the physical error rate, we
and fits from a. c, Summary of experimental progression comparing logical error
can budget how much each physical error mechanism contributes to
per cycle εd (specifically plotting 1 − εd) between distance-3 and distance-5, for
which system improvements lead to faster improvement for distance-5 (see main
1/Λ3/5 (as shown in Fig. 4a) to assess scaling. This error budget shows
text). Each open circle is a comparison to a specific distance-3 code, and filled that CZ error and data qubit decoherence during measurement and
circles average over several distance-3 codes measured in the same session. reset are dominant contributors.
Markers are coloured chronologically from light to dark. Typical 1σ statistical
and fit uncertainty is 0.02%, smaller than the points.
Algorithmically relevant error rates
into a pair of disjoint error graphs, one each for X and Z errors31. These Even as known error sources are suppressed in future devices, new
graphs are decoded efficiently using minimum-weight perfect match- dominant error mechanisms may arise as lower logical error rates are
ing53 to select a single probable set of errors. realized. To test the behaviour of codes with substantially lower error

Nature | Vol 614 | 23 February 2023 | 679

Article
a 1.0 b 10–1
Towards large-scale quantum error correction
/ 3/5

1Q 10–2 To understand how our surface code results project forwards to future
Surface code error budget, 1//

0.8 devices, we simulate the logical error performance of surface codes

Logical error per cycle, Hd

Leakage

10–3 ranging from distance-3 to 25, while also scaling the physical error
Measure
0.6 rates shown in Fig. 1c. For efficiency, the simulation considers only Pauli
DD 10–4 errors. Figure 4c,d illustrates the contours of this parameter space,
which has three distinct regions. When the physical error rate is high
0.4 CZ
stray 10–5 (for example, the initial runs of our surface code in Fig. 3c), logical error
int. probability increases with increasing system size (εd+2 > εd). On the other
0.2 10–6 Surface code hand, low physical error rates show the desired exponential suppression
CZ Repetition code of logical error (εd+2 < εd). This threshold behaviour can be subtle58, and
Removed high-energy event
0 10–7 there exists a crossover regime in which, owing to finite-size effects,
3 5 15 25 increasing system size initially suppresses the logical error per cycle
Code distance, d before later increasing it. We believe our experiment lies in this regime.
c d
Although our device is close to threshold, reaching algorithmically
regime crossover

1.3 0.06 s = 1.3

Crossover Above

Hd < Hd+2 relevant logical error rates with manageable resources will require an
Error model scale factor, s

Logical error per cycle, Hd

Hd =
H3 = H5 0.0 0.05 s = 1.2
1.2 50 error-suppression factor Λd/(d+2) ≫ 1. On the basis of the error budget
H7 = H9 Hd = 0.0
Hd = 0.0 6
3 0.04 s = 1.1 and simulations in Fig. 4, we estimate that component performance
1.1 30
must improve by at least 20% to move below threshold, and substan-
Hd = 0.020 0.03
1.0 H23 = H25 tially improve beyond that to achieve practical scaling. However, these
H d = 0.014 0.02 s = 1.0 projections rely on simplified models and must be validated experi-
threshold

.010
Below

0.9 Hd = 0
0.01 s = 0.9 mentally, testing larger code sizes with longer durations to eventu-
Hd > Hd+2
ally realize the desired logical performance. This work demonstrates
0.8 0
3 5 15 25 3 5 15 25 the first step in that process, suppressing logical errors by scaling a
Surface code distance, d Surface code distance, d quantum error-correcting code—the foundation of a fault-tolerant
quantum computer.
Fig. 4 | Towards algorithmically relevant error rates. a, Estimated error
budget for the surface code, based on component errors (see Fig. 1c) and Pauli+
simulations. Λ 3/5 = ε 3/ε 5. CZ, contributions from CZ error (excluding leakage
Online content
and stray interactions). CZ stray int., CZ error from unwanted interactions.
DD, dynamical decoupling (data qubit idle error during measurement and reset). Any methods, additional references, Nature Portfolio reporting summa-
Measure, measurement and reset error. Leakage, leakage during CZs and due ries, source data, extended data, supplementary information, acknowl-
to heating. 1Q, single-qubit gate error. b, Logical error for repetition codes. edgements, peer review information; details of author contributions
Inset: schematic of the distance-25 repetition code, using the same data and and competing interests; and statements of data and code availability
measure qubits as the distance-5 surface code. Smaller codes are subsampled are available at https://doi.org/10.1038/s41586-022-05434-1.
from the same distance-25 data42. A high-energy event resulted in an apparent
error floor around 10 −6. After removing the instances nearby (light blue),
1. Shor, P. W. Scheme for reducing decoherence in quantum computer memory. Phys. Rev.
error decreases more rapidly with code distance. The dataset has 50 cycles,
A 52, R2493 (1995).
5 × 105 repetitions. We also plot the surface code error per cycle from Fig. 3b in 2. Gottesman, D. Stabilizer Codes and Quantum Error Correction. PhD thesis, California
black. c, Contour plot of simulated surface code logical error per cycle ε d as a Institute of Technology (1997).
function of code distance d and a scale factor s on the error model in Fig. 1c 3. Feynman, R. P. Simulating physics with computers. Int. J. Theor. Phys. 21, 467–488 (1982).
4. Shor, P. W. Polynomial-time algorithms for prime factorization and discrete logarithms on
(Pauli simulation, s = 1.0 corresponds to the current device error model).
a quantum computer. SIAM Rev. 41, 303–332 (1999).
d, Horizontal slices from c, each for a value of error-model scale factor s. s = 1.3 5. Farhi, E. et al. A quantum adiabatic evolution algorithm applied to random instances of an
is above threshold (larger codes are worse), and s = 1.2 to 1.0 represent the NP-complete problem. Science 292, 472–475 (2001).
crossover regime, for which progressively larger codes get better until a 6. Biamonte, J. et al. Quantum machine learning. Nature 549, 195–202 (2017).
7. Lloyd, S. Universal quantum simulators. Science 273, 1073–1078 (1996).
turnaround. s = 0.9 is below threshold (larger codes are better).
8. Aspuru-Guzik, A., Dutoi, A. D., Love, P. J. & Head-Gordon, M. Simulated quantum
computation of molecular energies. Science 309, 1704–1707 (2005).
9. Reiher, M., Wiebe, N., Svore, K. M., Wecker, D. & Troyer, M. Elucidating reaction
rates, we use the bit-flip repetition code, a one-dimensional version mechanisms on quantum computers. Proc. Natl Acad. Sci. USA 114, 7555–7560 (2017).
10. Gidney, C. & Ekera, M. How to factor 2048 bit RSA integers in 8 hours using 20 million
of the surface code. The bit-flip repetition code does not correct for noisy qubits. Quantum 5, 433 (2021).
phase-flip errors and is thus unsuitable for quantum algorithms. How- 11. Kivlichan, I. D. et al. Improved fault-tolerant quantum simulation of condensed-phase
ever, correcting only bit-flip errors allows it to achieve much lower correlated electrons via trotterization. Quantum 4, 296 (2020).
12. Ballance, C., Harty, T., Linke, N., Sepiol, M. & Lucas, D. High-fidelity quantum logic gates
logical error probabilities. using trapped-ion hyperfine qubits. Phys. Rev. Lett. 117, 060504 (2016).
Without post-selection, we achieve a logical error per cycle of 13. Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 569, 532–536
(1.7 ± 0.3) × 10−6 using a distance-25 repetition code decoded with (2019).
14. Rol, M. et al. Fast, high-fidelity conditional-phase gate exploiting leakage interference in
minimum-weight perfect matching (Fig. 4b). We attribute many of these weakly anharmonic superconducting qubits. Phys. Rev. Lett. 123, 120502 (2019).
logical errors in the higher-distance codes to a high-energy impact, which 15. Jurcevic, P. et al. Demonstration of quantum volume 64 on a superconducting quantum
can temporarily impart widespread correlated errors to the system57. computing system. Quantum Sci. Technol. 6, 025020 (2021).
16. Foxen, B. et al. Demonstrating a continuous set of two-qubit gates for near-term quantum
These events may be identified by spikes in detection event counts42, and algorithms. Phys. Rev. Lett. 125, 120504 (2020).
such error mechanisms must be mitigated for scalable quantum error cor- 17. Wu, Y. et al. Strong quantum computational advantage using a superconducting quantum
rection to succeed. In this case, there was one such event; after removing it processor. Phys. Rev. Lett. 127, 180501 (2021).
18. Knill, E., Laflamme, R. & Zurek, W. H. Resilient quantum computation. Science 279,
(0.15% of trials), we observe a logical error per cycle of (1.6 ± 0.8) × 10−7 (see 342–345 (1998).
Supplementary Information). The repetition code results demonstrate 19. Aharonov, D. & Ben-Or, M. Fault-tolerant quantum computation with constant error rate.
that low logical error rates are possible in a superconducting system, but SIAM J. Comput. 38, 1207–1282 (2008).
20. Egan, L. et al. Fault-tolerant control of an error-corrected qubit. Nature 598, 281–286 (2021).
finding and mitigating highly correlated errors such as cosmic ray impacts 21. Ryan-Anderson, C. et al. Realization of real-time fault-tolerant quantum error correction.
will be an important area of research moving forwards. Phys. Rev. X 11, 041058 (2021).

680 | Nature | Vol 614 | 23 February 2023

22. Abobeih, M. et al. Fault-tolerant operation of a logical qubit in a diamond quantum 57. McEwen, M. et al. Resolving catastrophic error bursts from cosmic rays in large arrays of
processor. Nature 606, 884–889 (2022). superconducting qubits. Nat. Phys. 18, 107–111 (2022).
23. Sundaresan, N. et al. Matching and maximum likelihood decoding of a multi-round 58. Stephens, A. M. Fault-tolerant thresholds for quantum error correction with the surface
subsystem quantum error correction experiment. Preprint at https://arXiv.org/abs/ code. Phys. Rev. A 89, 022321 (2014).
2203.07205 (2022). 59. Emerson, J., Alicki, R. & Życzkowski, K. Scalable noise estimation with random unitary
24. Krinner, S. et al. Realizing repeated quantum error correction in a distance-three surface operators. J. Opt. B 7, S347 (2005).
code. Nature 605, 669–674 (2022).
25. Zhao, Y. et al. Realization of an error-correcting surface code with superconducting Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
qubits. Phys. Rev. Lett. 129, 030501 (2022). published maps and institutional affiliations.
26. Ofek, N. et al. Extending the lifetime of a quantum bit with error correction in
superconducting circuits. Nature 536, 441–445 (2016). Open Access This article is licensed under a Creative Commons Attribution
27. Flühmann, C. et al. Encoding a qubit in a trapped-ion mechanical oscillator. Nature 566, 4.0 International License, which permits use, sharing, adaptation, distribution
513–517 (2019). and reproduction in any medium or format, as long as you give appropriate
28. Campagne-Ibarcq, P. et al. Quantum error correction of a qubit encoded in grid states of credit to the original author(s) and the source, provide a link to the Creative Commons license,
an oscillator. Nature 584, 368–372 (2020). and indicate if changes were made. The images or other third party material in this article are
29. Grimm, A. et al. Stabilization and operation of a Kerr-cat qubit. Nature 584, 205–209 included in the article’s Creative Commons license, unless indicated otherwise in a credit line
(2020). to the material. If material is not included in the article’s Creative Commons license and your
30. Kitaev, A. Y. Fault-tolerant quantum computation by anyons. Ann. Phys. 303, 2–30 intended use is not permitted by statutory regulation or exceeds the permitted use, you will
(2003). need to obtain permission directly from the copyright holder. To view a copy of this license,
31. Dennis, E., Kitaev, A., Landahl, A. & Preskill, J. Topological quantum memory. J. Math. visit http://creativecommons.org/licenses/by/4.0/.
Phys. 43, 4452–4505 (2002).
32. Raussendorf, R. & Harrington, J. Fault-tolerant quantum computation with high threshold © The Author(s) 2023
in two dimensions. Phys. Rev. Lett. 98, 190504 (2007).
33. Fowler, A. G., Mariantoni, M., Martinis, J. M. & Cleland, A. N. Surface codes: towards
practical large-scale quantum computation. Phys. Rev. A 86, 032324 (2012). Google Quantum AI
34. Satzinger, K. et al. Realizing topologically ordered states on a quantum processor.
Science 374, 1237–1241 (2021).
35. Horsman, C., Fowler, A. G., Devitt, S. & Meter, R. V. Surface code quantum computing by Rajeev Acharya1, Igor Aleiner1,2, Richard Allen1, Trond I. Andersen1, Markus Ansmann1,
lattice surgery. New J. Phys. 14, 123011 (2012). Frank Arute1, Kunal Arya1, Abraham Asfaw1, Juan Atalaya1, Ryan Babbush1, Dave Bacon1,
36. Fowler, A. G. & Gidney, C. Low overhead quantum computation using lattice surgery. Joseph C. Bardin1,3, Joao Basso1, Andreas Bengtsson1, Sergio Boixo1, Gina Bortoli1,
Preprint at https://arXiv.org/abs/1808.06709 (2018). Alexandre Bourassa1, Jenna Bovaird1, Leon Brill1, Michael Broughton1, Bob B. Buckley1,
37. Litinski, D. A game of surface codes: large-scale quantum computing with lattice surgery. David A. Buell1, Tim Burger1, Brian Burkett1, Nicholas Bushnell1, Yu Chen1, Zijun Chen1,
Quantum 3, 128 (2019). Ben Chiaro1, Josh Cogan1, Roberto Collins1, Paul Conner1, William Courtney1,
38. Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Alexander L. Crook1, Ben Curtin1, Dripto M. Debroy1, Alexander Del Toro Barba1,
Nature 574, 505–510 (2019). Sean Demura1, Andrew Dunsworth1, Daniel Eppens1, Catherine Erickson1, Lara Faoro1,
39. Koch, J. et al. Charge-insensitive qubit design derived from the Cooper pair box. Phys. Edward Farhi1, Reza Fatemi1, Leslie Flores Burgos1, Ebrahim Forati1, Austin G. Fowler1,
Rev. A 76, 042319 (2007). Brooks Foxen1, William Giang1, Craig Gidney1, Dar Gilboa1, Marissa Giustina1,
40. Neill, C. A Path towards Quantum Supremacy with Superconducting Qubits. PhD thesis, Alejandro Grajales Dau1, Jonathan A. Gross1, Steve Habegger1, Michael C. Hamilton1,4,
Univ. California Santa Barbara (2017). Matthew P. Harrigan1, Sean D. Harrington1, Oscar Higgott1, Jeremy Hilton1,
41. Yan, F. et al. Tunable coupling scheme for implementing high-fidelity two-qubit gates. Markus Hoffmann1, Sabrina Hong1, Trent Huang1, Ashley Huff1, William J. Huggins1,
Phys. Rev. Appl. 10, 054062 (2018). Lev B. Ioffe1, Sergei V. Isakov1, Justin Iveland1, Evan Jeffrey1, Zhang Jiang1, Cody Jones1,
42. Chen, Z. et al. Exponential suppression of bit or phase errors with cyclic error correction. Pavol Juhas1, Dvir Kafri1, Kostyantyn Kechedzhi1, Julian Kelly1, Tanuj Khattar1,
Nature 595, 383–387 (2021). Mostafa Khezri1, Mária Kieferová1,5, Seon Kim1, Alexei Kitaev1,6, Paul V. Klimov1,
43. Kelly, J. et al. Scalable in situ qubit calibration during repetitive error detection. Phys. Rev. Andrey R. Klots1, Alexander N. Korotkov1,7, Fedor Kostritsa1, John Mark Kreikebaum1,
A 94, 032321 (2016). David Landhuis1, Pavel Laptev1, Kim-Ming Lau1, Lily Laws1, Joonho Lee1, Kenny Lee1,
44. Wen, X.-G. Quantum orders in an exact soluble model. Phys. Rev. Lett. 90, 016803 Brian J. Lester1, Alexander Lill1, Wayne Liu1, Aditya Locharla1, Erik Lucero1, Fionn D. Malone1,
(2003). Jeffrey Marshall8,9, Orion Martin1, Jarrod R. McClean1, Trevor McCourt1, Matt McEwen1,10,
45. Bonilla Ataides, J. P., Tuckett, D. K., Bartlett, S. D., Flammia, S. T. & Brown, B. J. The XZZX Anthony Megrant1, Bernardo Meurer Costa1, Xiao Mi1, Kevin C. Miao1, Masoud Mohseni1,
surface code. Nat. Commun. 12, 2172 (2021). Shirin Montazeri1, Alexis Morvan1, Emily Mount1, Wojciech Mruczkiewicz1, Ofer Naaman1,
46. Aliferis, P. & Terhal, B. M. Fault-tolerant quantum computation for local leakage faults. Matthew Neeley1, Charles Neill1, Ani Nersisyan1, Hartmut Neven1 ✉, Michael Newman1,
Quantum Inf. Comput. 7, 139–156 (2007). Jiun How Ng1, Anthony Nguyen1, Murray Nguyen1, Murphy Yuezhen Niu1, Thomas E. O’Brien1,
47. Suchara, M., Cross, A. W. & Gambetta, J. M. Leakage suppression in the toric code. Proc. Alex Opremcak1, John Platt1, Andre Petukhov1, Rebecca Potter1, Leonid P. Pryadko1,11,
2015 IEEE International Symposium on Information Theory (ISIT) 1119–1123 (2015). Chris Quintana1, Pedram Roushan1, Nicholas C. Rubin1, Negar Saei1, Daniel Sank1,
48. McEwen, M. et al. Removing leakage-induced correlated errors in superconducting Kannan Sankaragomathi1, Kevin J. Satzinger1, Henry F. Schurkus1, Christopher Schuster1,
quantum error correction. Nat. Commun. 12, 1761 (2021). Michael J. Shearn1, Aaron Shorter1, Vladimir Shvarts1, Jindra Skruzny1, Vadim Smelyanskiy1,
49. Spitz, S. T., Tarasinski, B., Beenakker, C. W. & O’Brien, T. E. Adaptive weight estimator for W. Clarke Smith1, George Sterling1, Doug Strain1, Marco Szalay1, Alfredo Torres1,
quantum error correction in a time-dependent environment. Adv. Quantum Technol. 1, Guifre Vidal1, Benjamin Villalonga1, Catherine Vollgraff Heidweiller1, Theodore White1,
1800012 (2018). Cheng Xing1, Z. Jamie Yao1, Ping Yeh1, Juhwan Yoo1, Grayson Young1, Adam Zalcman1,
50. Chen, E. H. et al. Calibrated decoders for experimental quantum error correction. Phys. Yaxing Zhang1 & Ningfeng Zhu1
Rev. Lett. 128, 110504 (2022).
51. Higgott, O., Bohdanowicz, T. C., Kubica, A., Flammia, S. T. & Campbell, E. T. Fragile 1
Google Research, Mountain View, CA, USA. 2Department of Physics, Columbia University,
boundaries of tailored surface codes and improved decoding of circuit-level noise. New York, NY, USA. 3Department of Electrical and Computer Engineering, University of
Preprint at https://arXiv.org/abs/2203.04948 (2022). Massachusetts, Amherst, MA, USA. 4Department of Electrical and Computer Engineering,
52. Criger, B. & Ashraf, I. Multi-path summation for decoding 2D topological codes. Quantum Auburn University, Auburn, AL, USA. 5Centre for Quantum Computation and Communication
2, 102 (2018). Technology, Centre for Quantum Software and Information, Faculty of Engineering and
53. Fowler, A. G., Whiteside, A. C. & Hollenberg, L. C. Towards practical classical processing Information Technology, University of Technology Sydney, Sydney, New South Wales, Australia.
for the surface code. Phys. Rev. Lett. 108, 180501 (2012). 6
Department of Physics, Institute for Quantum Information and Matter, and Walter Burke
54. Bravyi, S., Suchara, M. & Vargo, A. Efficient algorithms for maximum likelihood decoding Institute for Theoretical Physics, California Institute of Technology, Pasadena, CA, USA.
in the surface code. Phys. Rev. A 90, 032326 (2014). 7
Department of Electrical and Computer Engineering, University of California, Riverside, CA,
55. Chubb, C. T. & Flammia, S. T. Statistical mechanical models for quantum codes with USA. 8USRA Research Institute for Advanced Computer Science, Mountain View, CA, USA.
correlated noise. Ann. Inst. Henri Poincaré D 8, 269–321 (2021). 9
QuAIL, NASA Ames Research Center, Mountain View, CA, USA. 10Department of Physics,
56. Pattison, C. A., Beverland, M. E., da Silva, M. P. & Delfosse, N. Improved quantum error University of California, Santa Barbara, CA, USA. 11Department of Physics and Astronomy,
correction using soft information. Preprint at https://arXiv.org/abs/2107.13589 (2021). University of California, Riverside, CA, USA. ✉e-mail: [email protected]

Nature | Vol 614 | 23 February 2023 | 681

Article
Author contributions The Google Quantum AI team conceived and designed the
Data availability experiment. The theory and experimental teams at Google Quantum AI developed the data
analysis, modelling and metrological tools that enabled the experiment, built the system,
The data that support the findings of this study are available at https:// performed the calibrations and collected the data. The modelling was carried out jointly
doi.org/10.5281/zenodo.6804040. with collaborators outside Google Quantum AI. All authors wrote and revised the
manuscript and the Supplementary Information.

Acknowledgements We are grateful to S. Brin, S. Pichai, R. Porat, J. Dean, E. Collins and Competing interests The authors declare no competing interests.
J. Yagnik for their executive sponsorship of the Google Quantum AI team, and for their
continued engagement and support. A portion of this work was performed in the University Additional information
of California, Santa Barbara Nanofabrication Facility, an open access laboratory. J.M. Supplementary information The online version contains supplementary material available at
acknowledges support from the National Aeronautics and Space Administration (NASA) https://doi.org/10.1038/s41586-022-05434-1.
Ames Research Center (NASA-Google SAA 403512), NASA Advanced Supercomputing Correspondence and requests for materials should be addressed to Hartmut Neven.
Division for access to NASA high-performance computing systems, and NASA Academic Peer review information Nature thanks Barbara Terhal, Boris Varbanov and the other,
Mission Services (NNA16BD14C). D.B. is a CIFAR Associate Fellow in the Quantum Information anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer
Science Program. reports are available.
Reprints and permissions information is available at http://www.nature.com/reprints.