Article

High-threshold and low-overhead

fault-tolerant quantum memory

https://doi.org/10.1038/s41586-024-07107-7 Sergey Bravyi1, Andrew W. Cross1, Jay M. Gambetta1, Dmitri Maslov1 ✉, Patrick Rall2 &
Theodore J. Yoder1
Received: 25 August 2023

Accepted: 23 January 2024

The accumulation of physical errors1–3 prevents the execution of large-scale
Published online: 27 March 2024
algorithms in current quantum computers. Quantum error correction4 promises
Open access
a solution by encoding k logical qubits onto a larger number n of physical qubits,
Check for updates such that the physical errors are suppressed enough to allow running a desired
computation with tolerable fidelity. Quantum error correction becomes practically
realizable once the physical error rate is below a threshold value that depends on the
choice of quantum code, syndrome measurement circuit and decoding algorithm5.
We present an end-to-end quantum error correction protocol that implements
fault-tolerant memory on the basis of a family of low-density parity-check codes6.
Our approach achieves an error threshold of 0.7% for the standard circuit-based noise
model, on par with the surface code7–10 that for 20 years was the leading code in terms
of error threshold. The syndrome measurement cycle for a length-n code in our family
requires n ancillary qubits and a depth-8 circuit with CNOT gates, qubit initializations
and measurements. The required qubit connectivity is a degree-6 graph composed
of two edge-disjoint planar subgraphs. In particular, we show that 12 logical qubits
can be preserved for nearly 1 million syndrome cycles using 288 physical qubits in
total, assuming the physical error rate of 0.1%, whereas the surface code would
require nearly 3,000 physical qubits to achieve said performance. Our findings bring
demonstrations of a low-overhead fault-tolerant quantum memory within the reach
of near-term quantum processors.

Quantum computing attracted attention due to its ability to offer enables one to diagnose and correct errors by repeatedly measuring
asymptotically faster solutions to a set of computational problems syndromes of parity-check operators. However, error correction is
compared to the best known classical algorithms5. It is believed that only beneficial if the hardware error rate is below a certain threshold
a functioning scalable quantum computer may help solve computa- value that depends on a particular error correction protocol. The
tional problems in such areas as scientific discovery, materials research, first proposals for quantum error correction, such as concatenated
chemistry and drug design, to name a few11–14. codes21–23, focused on demonstrating the theoretical possibility of
The main obstacle to building a quantum computer is the fragility of error suppression. As understanding of quantum error correction and
quantum information, owing to various sources of noise affecting it. As the capabilities of quantum technologies matured, the focus shifted to
isolating a quantum computer from external effects and controlling it finding practical quantum error correction protocols. This resulted in
to induce a desired computation are in conflict with each other, noise the development of the surface code7–10 that offers a high error thresh-
appears to be inevitable. The sources of noise include imperfections old close to 1%, fast decoding algorithms and compatibility with the
in qubits, materials used, controlling apparatus, state preparation and existing quantum processors relying on two-dimensional (2D) square
measurement errors and a variety of external factors ranging from local lattice qubit connectivity. Small examples of the surface code with a
man-made, such as stray electromagnetic fields, to those inherent to the single logical qubit have already been demonstrated experimentally
Universe, such as cosmic rays. See ref. 15 for a summary. Whereas some by several groups24–28. However, scaling up the surface code to 100 or
sources of noise can be eliminated with better control16, materials17 and more logical qubits would be prohibitively expensive due to its poor
shielding18–20, several other sources appear to be difficult if at all pos- encoding efficiency. This spurred interest in more general quantum
sible to remove. The last kind can include spontaneous and stimulated codes known as low-density parity-check (LDPC) codes6. Recent pro-
emission in trapped ions1,2, and the interaction with the bath (Purcell gress in the study of LDPC codes suggests that they can achieve quan-
effect)3 in superconducting circuits—covering both leading quantum tum fault tolerance with a much higher encoding efficiency29. Here,
technologies. Thus, error correction becomes a key requirement for we focus on the study of LDPC codes, as our goal is to find quantum
building a functioning scalable quantum computer. error correction codes and protocols that are both efficient and pos-
The possibility of quantum fault tolerance is well-established 4. sible to demonstrate in practice, given the limitations of quantum
Encoding a logical qubit redundantly into many physical qubits computing technologies.

IBM Quantum, IBM T.J. Watson Research Center, Yorktown Heights, NY, USA. 2IBM Quantum, MIT-IBM Watson AI Laboratory, Cambridge, MA, USA. ✉e-mail: dmitri.maslov@ibm.com
1

778 | Nature | Vol 627 | 28 March 2024

 a b
X Data

Z X Z

Z X Z Checks

Surface _
Ancilla for X
code

L Z
Quasi-cyclic code
X R

_
Ancilla for Z = L data = R data = X check = Z check ‘A’ edge ‘B’ edge

Fig. 1 | Tanner graphs of surface and BB codes. a, Tanner graph of a surface edges indicate two planar subgraphs spanning the Tanner graph, see the
code, for comparison. b, Tanner graph of a BB code with parameters [[144, 12, 12]] Methods. c, Sketch of a Tanner graph extension for measuring Z and X
embedded into a torus. Any edge of the Tanner graph connects a data and a following ref. 50, attaching to a surface code. The ancilla corresponding to the
check vertex. Data qubits associated with the registers q(L) and q(R) are shown X measurement can be connected to a surface code, enabling load-store
by blue and orange circles. Each vertex has six incident edges including four operations for all logical qubits by means of quantum teleportation and some
short-range edges (pointing north, south, east and west) and two long-range logical unitaries. This extended Tanner graph also has an implementation in a
edges. We only show a few long-range edges to avoid clutter. Dashed and solid thickness-2 architecture through the A and B edges (Methods).

A quantum error correcting code is of LDPC type if each check opera- degree-3 subgraphs. As we argue below, such qubit connectivity is well
tor of the code acts only on a few qubits and each qubit participates suited for architectures based on superconducting qubits.
in only a few checks. Several variants of the LDPC codes have been Our codes are a generalization of bicycle codes proposed by MacKay
proposed recently including hyperbolic surface codes30–32, hypergraph et al.41 and studied in more depth in refs. 35,36,42. We named our
product33, balanced product codes34, two-block codes based on finite codes bivariate bicycle (BB) because they are based on bivariate poly-
groups35–38 and quantum Tanner codes39,40. The latter were shown39,40 nomials, as detailed in the Methods. These are stabilizer codes of the
to be asymptotically ‘good’ in the sense of offering a constant encod- Calderbank–Shor–Steane (CSS) type43,44 that can be described by a
ing rate and linear distance: a parameter quantifying the number of collection of six-qubit check (stabilizer) operators composed of Pauli
correctable errors. By contrast, the surface code has an asymptoti- X and Z. At a high level, a BB code is similar to the two-dimensional toric
cally zero encoding rate and only square-root distance. Replacing code7. In particular, physical qubits of a BB code can be laid out on a
the surface code with a high-rate, high-distance LDPC code could two-dimensional grid with periodic boundary conditions such that all
have major practical implications. First, the fault-tolerance overhead check operators are obtained from a single pair of X and Z checks by
(the ratio between the number of physical and logical qubits) could applying horizontal and vertical shifts of the grid. However, in contrast
be reduced notably. Second, high-distance codes show a very sharp to the plaquette and vertex stabilizers describing the toric code, check
decrease in the logical error rate: as the physical error probability operators of BB codes are not geometrically local. Furthermore, each
crosses the threshold value, the amount of error suppression achieved check acts on six qubits rather than four qubits. We will describe the
by the code can increase by orders of magnitude even with a small code by a Tanner graph G such that each vertex of G represents either
reduction of the physical error rate. This feature makes high-distance a data qubit or a check operator. A check vertex i and a data vertex j are
LDPC codes attractive for near-term demonstrations that are likely connected by an edge if the ith check operator acts non-trivially on
to operate in the near-threshold regime. However, it was previously the jth data qubit (by applying Pauli X or Z). See Fig. 1a,b for example
believed that outperforming the surface code for realistic noise mod- Tanner graphs of surface and BB codes, respectively. The Tanner graph
els including memory, gate and state preparation and measurement of any BB code has vertex degree six and graph thickness29 equal to
errors may require very large LDPC codes with more than 10,000 two, which means it can be decomposed into two edge-disjoint planar
physical qubits31. subgraphs (Methods). Thickness-2 qubit connectivity is well suited
Here we present several concrete examples of high-rate LDPC codes for superconducting qubits coupled by microwave resonators. For
with a few hundred physical qubits equipped with a low-depth syn- example, two planar layers of couplers and their control lines can be
drome measurement circuit, an efficient decoding algorithm and a attached to the top and the bottom side of the chip hosting qubits,
fault-tolerant protocol for addressing individual logical qubits. These and the two sides mated.
codes show an error threshold close to 0.7%, show excellent perfor- A BB code with parameters [[n, k, d]] encodes k logical qubits into n
mance in the near-threshold regime and offer a 10 times reduction of data qubits offering a code distance d, meaning that any logical error
the encoding overhead compared with the surface code. Hardware spans at least d data qubits. We divide n data qubits into registers q(L)
requirements for realizing our error correction protocols are rela- and q(R) of size n/2 each. Any check acts on three qubits from q(L) and
tively mild, as each physical qubit is coupled by two-qubit gates with three qubits from q(R). The code relies on n ancillary check qubits
only six other qubits. Although the qubit connectivity graph is not to measure the error syndrome. We divide n check qubits into regis-
locally embeddable into a 2D grid, it can be decomposed into two planar ters q(X) and q(Z) of size n/2 that collect syndromes of X and Z types,

Nature | Vol 627 | 28 March 2024 | 779

Article
Table 1 | Performance of BB codes of the error syndrome. The time it takes to implement the syndrome
measurement circuit is proportional to its depth: the number of gate
[[n, k, d]] Net Circuit-level Pseudo- pL (10−3) pL (10−4) layers composed of non-overlapping CNOTs. As new errors continue to
encoding distance, threshold,
rate, r dcirc p0
occur while the syndrome measurement circuit is executed, its depth
should be minimized. The full cycle of syndrome measurement for a
[[72, 12, 6]] 1/12 ≤6 0.0048 7 × 10−5 7 × 10−8
BB code is illustrated on Fig. 2. The syndrome cycle requires only seven
[[90, 8, 10]] 1/23 ≤8 0.0053 5 × 10 −6
4 × 10−10 layers of CNOTs regardless of the code length. The check qubits are ini-
[[108, 8, 10]] 1/27 ≤8 0.0058 3 × 10 −6
1 × 10−10 tialized and measured at the beginning and at the end of the syndrome
[[144, 12, 12]] 1/24 ≤10 0.0065 2 × 10−7
8 × 10−13 cycle respectively (see the Methods for details). The circuit respects
[[288, 12, 18]] 1/48 ≤18 0.0069 2 × 10 −12
1 × 10−22 the cyclic shift symmetry of the underlying code.
The full error correction protocol performs Nc ≫ 1 syndrome meas-
Small examples of BB LDPC codes and their performance for the circuit-based noise model.
All codes have weight-6 checks, depth-7 syndrome measurement circuit, and the Tanner urement cycles and then calls a decoder: a classical algorithm that
graph composed of two planar subgraphs. A code with parameters [[n, k, d]] requires 2n takes as input the measured syndromes and outputs a guess of the final
physical qubits in total and achieves the net encoding rate r = k/2n (we round r down to error on the data qubits. Error correction succeeds if the guessed and
the nearest inverse integer). Circuit-level distance dcirc is the minimum number of faulty the actual error coincide modulo a product of check operators. In this
operations in the syndrome measurement circuit required to generate a logical error without
case, the two errors have the same action on any encoded (logical) state.
triggering any syndromes.
Thus, applying the inverse of the guessed error returns data qubits to
the initial logical sate. Otherwise, if the guessed and the actual error
differ by a non-trivial logical operator, error correction fails resulting
in a logical error. Our numerical experiments are based on the belief
respectively. In total, the encoding relies on 2n physical qubits. The net propagation with an ordered statistics decoder (BP-OSD) proposed by
encoding rate is therefore r = k/(2n). For example, the standard surface Panteleev and Kalachev36. The original work36 described BP-OSD in the
code architecture encodes k = 1 logical qubit into n = d2 data qubits for context of a toy noise model with memory errors only. Here we show
a distance-d code and uses n − 1 check qubits for syndrome measure- how to extend BP-OSD to the circuit-based noise model, see the Sup-
ments. The net encoding rate is r ≈ 1/(2d2), which quickly becomes plementary Information for details. Our approach closely follows
impractical as one is forced to choose a large code distance, due to, refs. 45–48.
for instance, the physical errors being close to the threshold value. A noisy version of the syndrome measurement circuit may include
By contrast, BB codes have encoding rate r ≫ 1/d2, see Table 1 for code several types of faulty operations such as memory errors on idle data
examples. To the best of our knowledge, all codes shown in Table 1 are or check qubits, faulty CNOT gates, qubit initializations and measure-
new. The distance-12 code [[144, 12, 12]] may be the most promising for ments. We consider the circuit-based noise model10 in which each
near-term demonstrations, as it combines large distance and high net operation fails independently with probability p. The probability of a
encoding rate r = 1/24. For comparison, the distance-11 surface code has logical error pL depends on the error rate p, details of the syndrome
a net encoding rate r = 1/241. Below, we show that the distance-12 BB measurement circuits, and the decoding algorithm. Let PL(Nc) be the
code outperforms the distance-11 surface code for the experimentally logical error probability after performing Nc syndrome cycles. Define
relevant range of error rates. the logical error rate as pL = 1 − (1 − PL(Nc))1/Nc ≈ PL(Nc)/Nc. Informally, pL
To prevent the accumulation of errors one must be able to measure can be viewed as the logical error probability per syndrome cycle.
the error syndrome frequently enough. This is accomplished by a syn- Following common practice, we choose Nc = d for a distance-d code.
drome measurement circuit that couples data qubits in the support of Figure 3 shows the logical error rate achieved by codes from Table 1.
each check operator with the respective ancillary qubit by a sequence The logical error rate was computed numerically for p ≥ 10−3 and
of CNOT gates. Check qubits are then measured revealing the value extrapolated to lower error rates using a fitting formula (Methods).

Round 1 2 3 4 5 6 7 8

|+〉
|+〉
|+〉
X |+〉 X
|+〉
|+〉

L
e
y cl
xt c

R
Ne

|0〉
|0〉
|0〉
Z |0〉
|0〉 Z
|0〉

Fig. 2 | Syndrome measurement circuit. Full cycle of syndrome measurements under horizontal and vertical shifts of the Tanner graph. Each data qubit is
relying on seven layers of CNOTs. We provide a local view of the circuit that only coupled by CNOTs with three X-check and three Z-check qubits: see the Methods
includes one data qubit from each register q(L) and q(R). The circuit is symmetric for more details.

780 | Nature | Vol 627 | 28 March 2024

 a Noise performance of small BB codes the surface code would require more than 3,000 physical qubits to
10–1 suppress the error rate from 10−3 to 10−7 (Fig. 3). In this example, the
distance-12 BB code offers 10 times saving in the number of physical
qubits compared with the surface code.
10–3
A proposal for quantum error correction is only useful if the logical
qubits are accessible. Fortunately, BB LDPC codes possess the required
10–5 features to act as a logical memory. As shown in Fig. 1c, extensions
Logical error rate, pL

of the Tanner graph leveraging techniques by Cohen et al.50 enable

fault-tolerant measurement operations involving an ancillary surface
10–7 code. These measurements enable fault-tolerant load-store operations.
See the Supplementary Information for details.
10–9 [[72,12,6]] Our work highlights key hardware challenges to enable the new codes
with superconducting qubits: (1) the development of low-loss second
[[90,8,10]]
layer in the thickness-2 architecture; (2) the development of qubits that
[[108,8,10]]
10–11 can be coupled to seven connections (six buses and one control line);
[[144,12,12]]
and (3) the development of long-range couplers.
[[288,12,18]] These are all difficult to solve but not impossible. For the first chal-
10–13
10–4 10–3 10–2 lenge, we can imagine a small change to the packaging51 that was devel-
Physical error rate, p oped for the IBM Quantum Eagle processor52. The simplest would be
to place the extra buses on the opposite side of the qubit chip. This
b would require the development of high Q through substrate vias that
10–1 would be part of the coupling buses and as such would require inten-
Surface [[972,12,9]] sive microwave simulation to make sure these through substrate vias
Surface [[1452,12,11]] could support microwave propagation while not introducing large
10–3
Surface [[2028,12,13]] unwanted crosstalk.
Surface [[2700,12,15]] The second challenge is an extension of the number of couplers from
10–5 LDPC [[144,12,12]] the heavy hex lattice arrangement53, which is four (three couplers and
Logical error rate, pL

one control) to seven. The implication of this is that the cross-resonance

gate, which has been the core gate used in large quantum systems
10–7
for the past few years, would not be the path forward. Qubits in
cross-resonance gates are not tuneable and as such for a large device
10–9 with many connections the probability of a energy collisions (not just
the qubit levels but also higher levels of the transmon) trends to one
very quickly54. However, with the tuneable coupler55,56 in IBM Quan-
10–1 1 tum Egret and now being developed for the IBM Quantum Heron, this
problem no longer exists as qubit frequencies can be designed to be
farther apart. This new gate is also similar to the gates used by Google
10–1 3
10–4 10–3 10–2 Quantum AI57, which have shown that a square lattice arrangement is
Physical error rate, p possible. Extending the coupling map to seven connections will require
Fig. 3 | Noise properties of BB codes. a, Logical versus physical error rate for
notable microwave modelling; however, typical transmons have about
small examples of BB LDPC codes. A numerical estimate of pL (diamonds) was 60 fF of capacitance and each gate is around 5 fF to get the appropri-
obtained by simulating d syndrome cycles for a distance-d code. Most of the ate coupling strengths to the buses, so it is fundamentally possible to
data points have error bars roughly equal to pL/10 due to sampling errors. develop this coupling map without altering the long coherence times
b, Comparison between the BB LDPC code [[144, 12, 12]] and surface codes with and stability of transmon qubits.
12 logical qubits and distance d ∈ {9, 11, 13, 15}. The distance-d surface code with The final challenge is the most difficult. For the buses that are short
12 logical qubits has the length n = 12d2 because each logical qubit is encoded enough so that the fundamental mode can be used, the standard circuit
into a separate d × d patch of the surface code lattice. quantum electrodynamics model holds. However, to demonstrate
the 144-qubit code some of the buses will be long enough that we will
require frequency engineering. One way to achieve this is with filtering
resonators, and a proof of principle experiment was demonstrated
The pseudo-threshold p0 is defined as a solution of the break-even in ref. 58.
equation pL(p) = kp. Here kp is an estimate of the probability that at least In summary, we offer a new perspective on how a fault-tolerant
one of k unencoded qubits suffers from an error. BB codes offer a quantum memory could be realized using near-term quantum pro-
pseudo-threshold close to 0.7%, see Table 1, which is nearly the same cessors with a small qubit overhead. Although these LDPC codes are
as the error threshold of the surface code49 and exceeds the threshold not geometrically local, qubit connectivity required for syndrome
of all high-rate LDPC codes known to the authors. measurements is described by a thickness-2 graph that can be imple-
For example, suppose that the physical error rate is p = 10−3, which mented using two planar degree-3 layers of qubit couplers. This is a
is a realistic goal for near-term demonstrations. Encoding 12 logical valid architectural solution for platforms based on superconducting
qubits using the distance-12 code from Table 1 would offer the logical qubits. Numerical simulations performed for the circuit-based noise
error rate 2 ×10−7, which is enough to preserve 12 logical qubits for model indicate that the proposed LDPC codes compare favourably with
nearly 1 million syndrome cycles. The total number of physical qubits the surface code in the practically relevant range of error rates p ≥ 0.1%
required for this encoding is 288. The distance-18 code from Table 1 offering the same level of error suppression with 10 times reduction in
would require 576 physical qubits whereas suppressing the error rate the qubit overhead. Meanwhile, it remains unclear whether our code
from 10−3 to 2 ×10−12 enabling nearly hundred billion syndrome cycles. examples can be scaled up while retaining the high encoding rate in
For comparison, encoding 12 logical qubits into separate patches of the limit of large code length.

Nature | Vol 627 | 28 March 2024 | 781

Article
32. Higgott, O. & Breuckmann, N. P. Constructions and performance of hyperbolic and semi-
Online content hyperbolic floquet codes. Preprint at https://doi.org/10.48550/arXiv.2308.03750 (2023).
33. Tillich, J.-P. & Zémor, G. Quantum LDPC codes with positive rate and minimum distance
Any methods, additional references, Nature Portfolio reporting summa- proportional to the square root of the blocklength. IEEE Trans. Inf. Theory 60, 1193–1202
ries, source data, extended data, supplementary information, acknowl- (2013).
34. Breuckmann, N. P. & Eberhardt, J. N. Balanced product quantum codes. IEEE Trans. Inf.
edgements, peer review information; details of author contributions Theory 67, 6653–6674 (2021).
and competing interests; and statements of data and code availability 35. Kovalev, A. A. & Pryadko, L. P. Quantum kronecker sum-product low-density parity-check
are available at https://doi.org/10.1038/s41586-024-07107-7. codes with finite rate. Physi. Rev. A 88, 012311 (2013).
36. Panteleev, P. & Kalachev, G. Degenerate quantum LDPC codes with good finite length
performance. Quantum 5, 585 (2021).
37. Lin, H.-K. & Pryadko, L. P. Quantum two-block group algebra codes. Phys. Rev. A 109,
1. Wu, Y., Wang, S.-T. & Duan, L.-M. Noise analysis for high-fidelity quantum entangling gates 022407 (2024).
in an anharmonic linear Paul trap. Phys. Rev. A 97, 062325 (2018). 38. Wang, R., Lin, H.-K. & Pryadko, L. P. Abelian and non-Abelian quantum two-block codes.
2. Boguslawski, M. J. et al. Raman scattering errors in stimulated-Raman-induced logic Preprint at https://doi.org/10.48550/arXiv.2306.16400 (2023).
gates in 133Ba+. Phys. Rev. Lett. 131, 063001 (2023). 39. Panteleev, P. & Kalachev, G. Asymptotically good quantum and locally testable classical
3. Houck, A. A. et al. Controlling the spontaneous emission of a superconducting transmon LDPC codes. In Proc. 54th Annual ACM SIGACT Symposium on Theory of Computing
qubit. Phys. Rev. Lett. 101, 080502 (2008). 375–388 (ACM, 2022).
4. Shor, P. W. Scheme for reducing decoherence in quantum computer memory. Phys. Rev. A 40. Leverrier, A. & Zémor, G. Quantum Tanner codes. In Proc. 2022 IEEE 63rd Annual
52, R2493 (1995). Symposium on Foundations of Computer Science (FOCS) 872–883 (IEEE, 2022).
5. Nielsen, M. A. & Chuang, I. Quantum Computation and Quantum Information (Cambridge 41. MacKay, D. J. C., Mitchison, G. & McFadden, P. L. Sparse-graph codes for quantum error
Univ Press, 2002). correction. IEEE Trans. Inf. Theory 50, 2315–2330 (2004).
6. Gottesman, D. Fault-tolerant quantum computation with constant overhead. Quant. Inf. 42. Panteleev, P. & Kalachev, G. Quantum LDPC codes with almost linear minimum distance.
Comput. 14, 1338–1372 (2014). IEEE Trans. Inf. Theory 68, 213–229 (2021).
7. Kitaev, A. Y. Fault-tolerant quantum computation by anyons. Ann. Phys. 303, 2–30 (2003); 43. Steane, A. Multiple-particle interference and quantum error correction. Proc. R. Soc.
preprint at https://doi.org/10.48550/arXiv.quant-ph/9707021 (1997). London A Math. Phys. Eng. Sci. 452, 2551–2577 (1996).
8. Bravyi, S. B. & Kitaev, A. Y. Quantum codes on a lattice with boundary. Preprint at https:// 44. Calderbank, A. R. & Shor, P. W. Good quantum error-correcting codes exist. Phys. Rev. A
doi.org/10.48550/arXiv.quant-ph/9811052 (1998). 54, 1098 (1996).
9. Dennis, E., Kitaev, A., Landahl, A. & Preskill, J. Topological quantum memory. J. Math. Phys. 45. Delfosse, N. & Paetznick, A. Spacetime codes of Clifford circuits. Preprint at https://doi.
43, 4452–4505 (2002). org/10.48550/arXiv.2304.05943 (2023).
10. Fowler, A. G., Stephens, A. M. & Groszkowski, P. High-threshold universal quantum 46. McEwen, M., Bacon, D. & Gidney, C. Relaxing hardware requirements for surface code
computation on the surface code. Phys. Rev. A 80, 052312 (2009). circuits using time-dynamics. Preprint at https://doi.org/10.48550/arXiv.2302.02192
11. Lloyd, S. Universal quantum simulators. Science 273, 1073–1078 (1996). (2023).
12. Wang, H., Kais, S., Aspuru-Guzik, A. & Hoffmann, M. R. Quantum algorithm for obtaining 47. Higgott, O., Bohdanowicz, T. C., Kubica, A., Flammia, S. T. & Campbell, E. T. Improved
the energy spectrum of molecular systems. Phys. Chem. Chem. Phys. 10, 5388–5393 decoding of circuit noise and fragile boundaries of tailored surface codes. Phys. Rev. X 13,
(2008). 031007 (2023).
13. Reiher, M., Wiebe, N., Svore, K. M., Wecker, D. & Troyer, M. Elucidating reaction 48. Geher, G. P., Crawford, O. & Campbell, E. T. Tangling schedules eases hardware
mechanisms on quantum computers. Proc. Natl Acad. Sci. USA 114, 7555–7560 (2017). connectivity requirements for quantum error correction. In 2023 IEEE International
14. Alexeev, Y. et al. Quantum computer systems for scientific discovery. PRX Quantum 2, Conference on Quantum Computing and Engineering (QCE) 342–343 (IEEE, 2023).
017001 (2021). 49. Groszkowski, P., Fowler, A. & Stephens, A. M. High-threshold surface code quantum
15. Gambetta, J. M., Chow, J. M. & Steffen, M. Building logical qubits in a superconducting computing: threshold calculation. In Proc. APS March Meeting Abstracts W17–002 (APS,
quantum computing system. npj Quant. Info. 3, 2 (2017). 2009).
16. Mundada, P. S. et al. Experimental benchmarking of an automated deterministic error 50. Cohen, L. Z., Kim, I. H., Bartlett, S. D. & Brown, B. J. Low-overhead fault-tolerant quantum
suppression workflow for quantum algorithms. Preprint at https://doi.org/10.48550/ computing using long-range connectivity. Sci. Adv. https://doi.org/10.1126/sciadv.abn1717
arXiv.2209.06864 (2023). (2022).
17. De Leon, N. P. et al. Materials challenges and opportunities for quantum computing 51. Bravyi, S., Dial, O., Gambetta, J. M., Gil, D. & Nazario, Z. The future of quantum computing
hardware. Science 372, eabb2823 (2021). with superconducting qubits. J. Appl. Phys. 132, 160902 (2022).
18. Córcoles, A. D. et al. Protecting superconducting qubits from radiation. Appl. Phys. Lett. 52. Chow, J., Dial, O. & Gambetta, J. IBM Quantum breaks the 100-qubit processor barrier.
99, 181906 (2011). IBM https://research.ibm.com/blog/127-qubit-quantum-processor-eagle (2021).
19. Vepsäläinen, A. P. et al. Impact of ionizing radiation on superconducting qubit coherence. 53. Nation, P., Paik, H., Cross, A. & Nazario, Z. The IBM Quantum heavy hex lattice. IBM https://
Nature 584, 551–556 (2020). research.ibm.com/blog/heavy-hex-lattice (2021).
20. Thorbeck, T., Eddins, A., Lauer, I., McClure, D. T. & Carroll, M. Two-level-system dynamics 54. Hertzberg, J. B. et al. Laser-annealing Josephson junctions for yielding scaled-up
in a superconducting qubit due to background ionizing radiation. PRX Quant. 4, 020356 superconducting quantum processors. npj Quantum Inf. 7, 129 (2021).
(2023). 55. McKay, D. C. et al. Universal gate for fixed-frequency qubits via a tunable bus. Phys. Rev.
21. Aharonov, D. & Ben-Or, M. Fault-tolerant quantum computation with constant error. In Appl. 6, 064007 (2016).
Proc. 29th Annual ACM Symposium on Theory of Computing 176–188 (ACM, 1997). 56. Stehlik, J. et al. Tunable coupling architecture for fixed-frequency transmon
22. Kitaev, A. Y. Quantum computations: algorithms and error correction. Russ. Math. Surv. superconducting qubits. Phys. Rev. Lett. 127, 080505 (2021).
52, 1191 (1997). 57. Arute, F. et al. Quantum supremacy using a programmable superconducting processor.
23. Aliferis, P., Gottesman, D. & Preskill, J. Quantum accuracy threshold for concatenated Nature 574, 505–510 (2019).
distance-3 codes. Quant. Inf. Comput. 6, 97–165 (2006); preprint at https://doi. 58. McKay, D. C., Naik, R., Reinhold, P., Bishop, L. S. & Schuster, D. I. High-contrast qubit
org/10.48550/arXiv.quant-ph/0504218 (2005). interactions using multimode cavity QED. Phys. Rev. Lett. 114, 080501 (2015).
24. Takita, M. et al. Demonstration of weight-four parity measurements in the surface code
architecture. Phys. Rev. Lett. 117, 210505 (2016). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
25. Marques, J. F. et al. Logical-qubit operations in an error-detecting surface code. Nat. Phys. published maps and institutional affiliations.
18, 80–86 (2022).
26. Krinner, S. et al. Realizing repeated quantum error correction in a distance-three surface Open Access This article is licensed under a Creative Commons Attribution
code. Nature 605, 669–674 (2022). 4.0 International License, which permits use, sharing, adaptation, distribution
27. Zhao, Y. et al. Realization of an error-correcting surface code with superconducting and reproduction in any medium or format, as long as you give appropriate
qubits. Phys. Rev. Lett. 129, 030501 (2022). credit to the original author(s) and the source, provide a link to the Creative Commons licence,
28. Google Quantum AI. Suppressing quantum errors by scaling a surface code logical qubit. and indicate if changes were made. The images or other third party material in this article are
Nature 614, 676–681 (2023). included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
29. Tremblay, M. A., Delfosse, N. & Beverland, M. E. Constant-overhead quantum error to the material. If material is not included in the article’s Creative Commons licence and your
correction with thin planar connectivity. Phys. Rev. Lett. 129, 050504 (2022). intended use is not permitted by statutory regulation or exceeds the permitted use, you will
30. Breuckmann, N. P. & Terhal, B. M. Constructions and noise threshold of hyperbolic need to obtain permission directly from the copyright holder. To view a copy of this licence,
surface codes. IEEE Trans. Inf. Theory 62, 3731–3744 (2016). visit http://creativecommons.org/licenses/by/4.0/.
31. Higgott, O. & Breuckmann, N. P. Subsystem codes with high thresholds by gauge fixing
and reduced qubit overhead. Phys. Rev. X 11, 031039 (2021). © The Author(s) 2024

782 | Nature | Vol 627 | 28 March 2024

Methods also describes a code [[126, 12, 10]] that has parameters similar to ours.
This code has a form QC(A, B) with A = 1 + x43 + x37, B = 1 + x59 + x31, ℓ = 63
Code construction and m = 1. We note that the recent work by Wang, Lin and Pryadko37,38
We begin with a formal definition of BB codes. Let Iℓ and Sℓ be the ide­ described examples of group-based codes closely related to the codes
ntity matrix and the cyclic shift matrix of size ℓ × ℓ respectively. The considered here. Some of the group-based codes with weight-8 checks
ith row of Sℓ has a single non-zero entry equal to one at the column found in ref. 37 outperform our BB codes with weight-6 checks in terms
i + 1 (mod ℓ). For example, of the parameters n, k, d. It remains to be seen whether group-based
codes can achieve a similar or better level of error suppression for the
0 1 0 circuit-based noise model.
0 1 
S2 =  and S3 = 0 0 1  .
 1 0   In the following, we partition the set of data qubits as [n] = LR, where
 1 0 0 L ≔ q(L) and R ≔ q(R) are the left and right blocks of n/2 = ℓm data qubits.
Then, data qubits L and R and checks X and Z may each be labelled by
Consider matrices integers Zℓm = {0, 1, …, ℓm − 1}, which are indices into the matrices A, B.
Alternatively, qubits and checks can be labelled by monomials from
x = Sℓ ⊗ Im and y = Iℓ ⊗ Sm . M = {1, y, …, y m−1 , x , xy, …, xy m−1 , …, x ℓ −1y m−1 } in this order, so that
i ∈ Zℓm labels the same qubit or check as x ai yi − mai for ai = floor(i/m).
Note that xy = yx, xTx = yTy = Iℓm, and xℓ = ym = Iℓm. A BB code is defined Using the monomial labelling, L data qubit α ∈ M is part of X checks
by a pair of matrices ATi α and Z checks Biα for i = 1, 2, 3. Similarly, R data qubit β ∈ M is part
of X checks BTi β and Z checks Aiβ. A unified notation assigns each qubit
A = A1 + A2 + A3 and B = B1 + B2 + B3 (1) or check a label q(T, α) where T ∈ {L, R, X, Z} denotes its type and α ∈ M
its monomial label. (The monomial notations should not be confused
where each matrix Ai and Bj is a power of x or y. Here and below the addi- with the matrix notations used earlier in this section. For example,
tion and multiplication of binary matrices is performed modulo two, multiplication of monomials such as Biα is different from multiplying
unless stated otherwise. Thus, we also assume the Ai are distinct and the a vector α by a matrix Bi).
Bj are distinct to avoid cancellation of terms. For example, one could One drawback of high-rate LDPC codes is that their Tanner graphs
choose A = x3 + y + y2 and B = y3 + x + x2. Note that A and B have exactly may not be locally embeddable into the 2D grid60,61. This poses a chal-
three non-zero entries in each row and each column. Furthermore, lenge for hardware implementation with superconducting qubits cou-
AB = BA because xy = yx. The above data defines a BB quantum code pled by microwave resonators. A useful very-large-scale integration
denoted QC(A, B) with length n = 2ℓm and check matrices (VLSI) design concept is graph thickness, see refs. 29,62 for details. A
graph G = (V, E) is said to have thickness θ if one can partition its set of
H X = [A B] and H Z = [BT AT ]. (2) edges E into disjoint union of θ sets E1⊔E2⊔…⊔Eθ = E such that each sub-
graph (V, Ei) is planar. Informally, a graph with thickness θ can be viewed
Here the vertical bar indicates stacking matrices horizontally and T as a vertical stack of θ planar graphs. Qubit connectivity described
stands for the matrix transposition. Both matrices HX and HZ have size by a planar graph (thickness θ = 1) is the simplest one from hardware
(n/2) × n. Each row v ∈ Fn2 of HX defines an X-type check operator perspective because the couplers do not cross.
n v
X (v) = ∏ j =1 X j j. Each row v ∈ Fn2 of HZ defines a Z-type check operator Here we show that the Tanner graph of any BB code has thickness-2.
n v
Z (v) = ∏ j =1 Z j j . Any X and Z checks commute as they overlap on even This result may be surprising as it is known that a general degree-6
number of qubits (note that H X (H Z )T = AB + BA = 0 (mod 2)). By con- graph can have thickness θ = 3 (ref. 62). Graphs with thickness θ = 2
struction, the code QC(A, B) has weight-6 check operators and each might still be implementable with superconducting qubits because
qubit participates in six checks (three X-type plus three Z-type checks). two planar layers of couplers and their control lines can be attached
Accordingly, the code QC(A, B) has a degree-6 Tanner graph. One can to the top and the bottom side of the chip hosting qubits.
view the matrices A and B as bivariate polynomials over the variables Lemma 2. The Tanner graph G of the code QC(A, B) has thickness θ ≤ 2.
x and y. Specializing BB codes to the case m = 1 and B = AT gives the A decomposition of G into two planar layers can be computed in time
original bicycle codes41 based on univariate polynomials. Likewise, BB O(n). Each planar layer of G is a degree-3 graph.
codes are a specialization of the generalized bicycle codes35, two-block Proof. Let G = (V, E) be the Tanner graph. Partition G into subgraphs
group-based codes37,42 and polynomial-based codes59. Given a binary GA = (V, EA) and GB = (V, EB) that describe CSS codes with check matrices
matrix M, let ker(M ) be its nullspace spanned by all binary vectors v
such that M v = 0 (mod 2). Let rs(M) be the row space of M spanned by Tanner graph GA : H XA = [A2 + A3 B3] and H ZA = [BT3 AT2 + AT3 ] (3)
rows of M.
Lemma 1. The code QC(A, B) has parameters [[n, k, d]], where Tanner graph GB : H BX = [A1 B1 + B2] and H BZ = [BT1 + BT2 AT1 ]. (4)
n = 2ℓm, k = 2 × dim(ker(A) ∩ ker(B)) and d As A = A1 + A2 + A3 and B = B1 + B2 + B3, every edge of G appears either in
= min{|v| : v ∈ ker(H X )\ rs(H Z )}. GA or GB, in which the two subgraphs are named by whether they contain
more Ai edges or more Bi edges. Then GA and GB are regular degree-3
The code offers equal distance for X-type and Z-type errors. graphs (because Ai and Bj are permutation matrices).
The proof, relying on elementary linear algebra, is deferred to Consider the graph GA. Each X-check vertex is connected to a pair of
the Supplementary Information. Extended Data Table 1 describes data vertices i1, i2 ∈ L by means of the matrices A2, A3 and a data vertex
the polynomials A and B that give rise to examples of high-rate, i3 ∈ R by means of the matrix B3. Each Z-check vertex is connected to a
high-distance BB codes found by a numerical search. This includes pair of data vertices i1, i2 ∈ R by means of the matrices AT2 , AT3 and a data
all codes from Table 1 and two examples of higher distance codes. vertex i3 ∈ L by means of the matrix BT3 .
To the best of our knowledge, all these examples are new. The code We claim that each connected component of GA can be represented
[[360, 12, ≤24]] improves on a code [[882, 24, ≤24]] with weight-6 checks by a ‘wheel graph’ illustrated in Extended Data Fig. 1. A wheel graph
found by Panteleev and Kalachev in ref. 36 (assuming that our distance consists of two disjoint cycles of the same length p interconnected
upper bound is tight). Indeed, taking two independent copies of the by p radial edges. The outer cycle alternates between X-check and L
360-qubit code gives parameters [[720, 24, ≤24]]. Appendix C in ref. 36 data vertices.
Article
Edges of the outer cycle alternate between those generated by A3 elements of the cosets of this subgroup. This implies the theorem’s
(as one moves from a check to a data vertex) and AT2 (as one moves from second statement.
a data to a check vertex). The length of the outer cycle is equal to the For the next part, we establish some terminology. A spanning sub-
order of the matrix A3 AT2 , that is, the smallest integer Ord such that graph of a graph G is a subgraph containing all the vertices of G. Also,
(A3 AT2 )Ord = Iℓm . For example, consider the code [[144, 12, 12]] from the undirected Cayley graph of a finite Abelian group G (with identity
Extended Data Table 1. Then A = x3 + y + y2, A2 = y and A3 = y2. Thus element 0) generated by set S ⊂ G is the graph with vertex set G and
A3 AT2 = y 2 y−1 = y that has order m = 6. The inner cycle of a wheel graph undirected edges (g, g + s) for all g ∈ G and all s ∈ S, s ≠ 0. We say the
alternates between Z-check and R-data vertices. Cayley graph of Za × Zb when we mean the Cayley graph of Za × Zb gen-
Edges of the inner cycle alternate between those generated by AT3 erated by {(1, 0), (0, 1)}. The order ord(g) of an element g in a multiplica-
(as one moves from a check to a data vertex) and A2 (as one moves from tive group is the smallest positive integer such that g ord(g) = 1.
a data to a check vertex). The length of the inner cycle is equal to the Definition 1. Code QC(A, B) is said to have a toric layout if its Tanner
order of the matrix AT3 A2 that is the transpose of A3 AT2 considered ear- graph has a spanning subgraph isomorphic to the Cayley graph of
lier. Thus both inner and outer cycles have the same length m. The two Z2μ × Z2λ for some integers μ and λ.
cycles are interconnected by m radial edges as shown in Extended Data Note that only codes with connected Tanner graphs can have a toric
Fig. 1a. Radial edges are generated by the matrix B3, as one moves layout according to this definition. An example toric layout is depicted
towards the centre of the wheel. The wheel graph contains four cycles in Fig. 1b.
generated by tuples of edges (B3, A2 , BT3 , AT2 ) and (BT3 , A3 , B3, AT3 ). Com- Lemma 4. A code QC(A, B) has a toric layout if there exist i, j, g, h ∈ {1, 2, 3}
mutativity between Ai and Bj ensures that traversing any of these four such that
cycles implements the identity matrix, that is, the graph is well defined. 1. Ai ATj , Bg BTh  = M and
Clearly, the wheel graph is planar. As GA is a disjoint union of wheel 2. ord(Ai ATj )ord(Bg BTh ) = ℓm.
graphs, GA is planar. The same argument shows that GB is planar Proof. We let μ = ord(Ai ATj ) and λ = ord(Bg BTh ). We associate qubits and
(Extended Data Fig. 1b). The visualization of the [[144, 12, 12]] code in checks in the Tanner graph of QC(A, B) with elements of G = Z2μ × Z2λ.
Fig. 1b shows the edges of GA and GB as dashed ‘A’ edges and solid ‘B’ For L qubit with label α ∈ M, because of (1), there is (a, b) ∈ Zμ × Zλ such
edges, respectively. that α = (Ai ATj )a (Bg BTh )b . Because of (2) and the pigeonhole principle,
We leave optimization of the code layout satisfying specific hardware this choice of (a, b) is unique. We associate L qubit α with (2a, 2b) ∈ G .
constraints for future work. For now, it is sufficient to note that any Similarly, an R qubit with label α ATj Bg is associated with
planar graph admits a planar embedding without edge crossings for (2a + 1, 2b + 1) ∈ G , X-check α ATj with (2a + 1, 2b) and Z-check αBg with
any prescribed vertex locations, see for example theorem 1 in ref. 63. (2a, 2b + 1). Edges in the Tanner graph Ai, ATj , Bg and BTh can now be drawn
Moreover, this embedding can be efficiently computed63. Accordingly, as in Extended Data Fig. 2b and correspond to edges in the Cayley graph
both planar layers in the thickness-2 decomposition of the Tanner graph of G. For instance, to get from (2a + 1, 2b + 1), an R qubit, to (2a + 2, 2b + 1),
can be simultaneously embedded into a plane for any fixed vertex loca- a Z check, we apply Ai, taking R qubit labelled α ATj Bg to the Z check
tions such that edges do not cross within each layer. labelled (α ATj Bg )Ai = α(Ai ATj )Bg .
Another example of thickness-2 graphs in the literature is the bilayer All codes in Extended Data Table 1 have a toric layout with μ = m and
architecture of ref. 64. This connectivity is described by two planar λ = ℓ. Most of these codes satisfy Lemma 4 with i = g = 2 and j = h = 3.
graphs with additional transversal edges between them. It can be veri- The exception is the [[90, 8, 10]] code, for which we can take i = 2, g = 1
fied that bilayer graphs are thickness-2 by placing transversally con- and j = h = 3.
nected nodes next to each other in one of the two planes and placing However, we also note two interesting cases. First, there are codes
the transversal edges in that same plane. with connected Tanner graphs that do not satisfy the conditions for a
The definition of code QC(A, B) does not guarantee that its Tanner toric layout given in Lemma 4. One example of such a code is QC(A, B)
graph is connected. Some choices of A and B lead to a code that is actu- with ℓ, m = 28, 14, A = x26 + y6 + y8 and B = y7 + x9 + x20 that has parameters
ally several separable code blocks. This manifests as a Tanner graph [[784, 24, ≤24]]. Second, for a code satisfying the conditions of Lemma
with several connected components. For instance, although all codes 4, it need not be the case that the set {ord(Ai ATj ), ord(Bg BTh )} and the set
in Extended Data Table 1 are connected, taking any of them with even {ℓ, m} are equal. For example, the [[432, 4, ≤22]] code with ℓ, m = 18, 12
ℓ and replacing every instance of x with x2 creates a code with two con- and A = x + y11 + y3, B = y2 + x15 + x only satisfies Lemma 4 with μ, λ = 36, 6
nected components. (take i = g = 1 and j = h = 2 for instance).
Lemma 3. The Tanner graph of the code QC(A, B) is connected if and only
if S = {Ai ATj : i , j ∈ {1, 2, 3}} ∪ {Bi BTj : i , j ∈ {1, 2, 3}} generates the group Summary of other capabilities
M. The number of connected components in the Tanner graph is For the remainder of this section, we summarize important details of
ℓm/∣⟨S⟩∣, and all components are graph isomorphic to one another. additional capabilities of BB LDPC codes. For more details on these
Proof. Extended Data Fig. 2 is helpful for following the arguments in topics, see the Supplementary Information.
this proof. We start by proving the reverse implication of the first state-
ment. Note that there is a length 2 path in the Tanner graph from L qubit Syndrome circuit. Our syndrome measurement circuit relies on 2n
α ∈ M to L qubit Ai ATj α and another length 2 path to L qubit Bi BTj α . physical qubits, comprising of n data qubits and n ancillary check qu-
These travel through X and Z checks, respectively. Thus, because the bits used to record the measured syndromes. It repeatedly measures
Ai ATj and Bi BTj generate M, there is some path from α to any other the syndrome of each check operator. The single syndrome cycle is
L qubit β. A similar argument shows existence of a path connecting any illustrated in Fig. 2. The entire syndrome measurement circuit was
pair of R qubits. As each X check and each Z check are connected to at composed to simultaneously minimize the number of gates used, op-
least one L qubit and at least one R qubit, this implies that the entire timize depth (including parallelizing register measurement with state
Tanner graph is connected. The forward implication of the first state- initialization and with gate application, whenever possible), limit the
ment follows after noticing that, for all T ∈ {L, R, X, Z}, the path from a propagation of errors and comply with the qubit-to-qubit connectivity
type T node to any other T node is necessarily described as a product layout offered by the Tanner graph. We refer the interested reader to
of elements from S. Connectivity of the Tanner graph implies the exist- Supplementary Information for the complete circuit description and
ence of all such paths, and so S must generate M. proof of its correctness.
If S does not generate M, it necessarily generates a subgroup ⟨S⟩ and We used a computer search to find a total of 936 low-depth syndrome
nodes in connected components of the Tanner graph are labelled by measurement circuit alternatives. For the [[144, 12, 12]] code, the circuit
shown in Fig. 2 achieves circuit distance of less than or equal to ten and measurement of X and Z for all qubits by acting on the original measure-
we conjecture it equals ten. This syndrome measurement circuit was ment by conjugation, and have fault-tolerant circuit implementations
used to compile the data for all codes reported in Fig. 3, leaving the within the existing connectivity of the Tanner graph.
possibility that tailoring each of the 936 alternatives to specific codes
would yield better results.
Data availability
Decoder. We adapt the BP-OSD 36,65,66
to the circuit noise model. This The simulation software to generate data reported in this paper is avail-
involves both an offline and online stage. In the offline stage, we take able at https://github.com/sbravyi/BivariateBicycleCodes/.
as input the syndrome measurement circuit and the error rate p. For
every distinct single fault, we simulate the circuit efficiently using the 59. Haah, J. Algebraic methods for quantum codes on lattices. Rev. Colomb. de Mat. 50,
299–349 (2016).
stabilizer formalism, tracking the probability of the fault, the syndrome 60. Baspin, N. & Krishna, A. Quantifying nonlocality: how outperforming local quantum
observed, and a final ideal syndrome. We also record in each case the codes is expensive. Phys. Rev. Lett. 129, 050505 (2022).
logical syndrome, which indicates the logical operators anticommuting 61. Bravyi, S., Poulin, D. & Terhal, B. Tradeoffs for reliable quantum information storage in 2D
systems. Phys. Rev. Lett. 104, 050503 (2010).
with the final error. In the online stage, we take a syndrome instance 62. Mutzel, P., Odenthal, T. & Scharbrodt, M. The thickness of graphs: a survey. Graphs Comb.
and determine a likely set of faults that occurred. Using the results of 14, 59–73 (1998).
63. Pach, J. & Wenger, R. Embedding planar graphs at fixed vertex locations. Graphs Comb.
the offline stage, we can formulate this as an optimization problem,
17, 717–728 (2001).
which is solved heuristically by BP-OSD. 64. Pattison, C. A., Krishna, A. & Preskill, J. Hierarchical memories: simulating quantum LDPC
We also leverage BP-OSD to perform two additional useful tasks that codes with local gates. Preprint at https://doi.org/10.48550/arXiv.2303.04798 (2023).
65. Roffe, J., White, D. R., Burton, S. & Campbell, E. Decoding across the quantum low-density
can be framed as appropriate optimization problems. First, given a
parity-check code landscape. Phys. Rev. Res. 2, 043423 (2020).
code, we can find an upper bound on the code distance. Second, given a 66. Roffe, J. LDPC: Python tools for low density parity check codes. PyPI https://pypi.org/
code and a syndrome measurement circuit, we can determine an upper project/ldpc/ (2022).
67. Breuckmann, N. P. & Burton, S. Fold-transversal Clifford gates for quantum codes.
bound on the circuit distance.
Preprint at https://doi.org/10.48550/arXiv.2202.06647 (2022).
68. Landahl, A. J., Anderson, J. T. & Rice, P. R. Fault-tolerant quantum computing with color
Logical memory capabilities. As depicted in Fig. 1c a BB LDPC code codes. Preprint at https://doi.org/10.48550/arXiv.1108.5738 (2011).
can be used as a data storage unit for, for example, a small surface code
quantum computer. To this end we demonstrate two capabilities: joint Acknowledgements We thank B. Brown, O. Dial, A. Ivrii, T. J. O’Connor, Y. Nam, M. Steffen,
logical XX measurements between a surface code qubit and any qubit K. Tien and J. Blue for stimulating discussions at various stages of this project. Financial
support: we acknowledge the IBM Research Cognitive Computing Cluster service for
within the BB code, and logical Z measurements on any qubit in the BB
providing resources that have contributed to the research results reported within this paper.
code. These measurements facilitate quantum teleportation circuits
implementing load-store operations, transporting qubits into and Author contributions S.B., A.W.C., J.M.G. and T.J.Y. designed quantum codes and decoding
algorithms used in this study. S.B. and D.M. designed the syndrome measurement circuit.
out of the BB code.
A.W.C. checked the circuit-level distance. S.B., A.W.C. and T.J.Y. developed the code
These measurements are facilitated by the combination of two tech- visualization and layout. P.R. designed the logical memory capabilities. All authors contributed
niques. A construction based on ref. 50 enables fault-tolerant logical to writing and editing the manuscript.
measurement of one logical X and one logical Z operator. The main
Competing interests US Patent Application 18/527304 (filed on 3 December 2023 and naming
idea, as illustrated in Fig. 1c, is to extend the Tanner graph of the BB S.B., A.W.C., J.M.G., D.M., P.R., Kevin Tien and T.J.Y. as co-inventors) contains technical aspects
code to a larger code that features the desired logical operators as from this paper.
stabilizers. We show that this extended Tanner graph is compatible
Additional information
with a thickness-2 architecture while simultaneously connecting the Supplementary information The online version contains supplementary material available at
logical X extension to an ancillary surface code. https://doi.org/10.1038/s41586-024-07107-7.
Correspondence and requests for materials should be addressed to Dmitri Maslov.
To extend the reach of these logical measurements beyond a sin-
Peer review information Nature thanks Michael Vasmer and the other, anonymous, reviewer(s)
gle X and Z operator, we leverage techniques from ref. 67 to derive for their contribution to the peer review of this work.
several fault-tolerant unitary operations. These operations achieve Reprints and permissions information is available at http://www.nature.com/reprints.
Article
a Extraction of ‘A’ wheels in GA
X L

B3 B2 B3 R Z
X R X R L Z R X

A3 A3 A3 A3 R Z
B3 B2 B3 X L
L Z L Z

A2 A2 A2 A2
B3 B2 B3
X R X R

A3 A3 A3 A3
B3 B2 B3
L Z L Z

b Extraction of ‘B’ wheels in GB

L Z
B1 B2 B1
X R X R X R

Z R X L
A1 A1 A1 A1
X R
B1 B2 B1
L Z L Z L Z

A2 A2 A2 A2
B1 B2 B1
X R X R

A1 A1 A1 A1
B1 B2 B1
L Z L Z

Extended Data Fig. 1 | Decomposition of the Tanner graph into planar

graphs. Two different grids over a torus defined using different subsets of
A 1, A 2, A 3, B1, B2, B3. Edge labels indicate adjacency matrices that generate the
respective edges. By extracting either horizontal or vertical strips from these
grids, we obtain planar ‘wheel graphs’ whose union contains all edges in the
Tanner graph. To avoid clutter, each grid shows only a subset of edges present
in the Tanner graph. a, The A wheels (dashed lines) cover A 2, A 3, B3. b, The B
wheels (solid lines) cover B1, B2, A 1.
 a b
Bg Bh
X R X

A X
B Ai Ai Ai
Bg Bh
L R
L Z L

B Z A
Aj Aj Aj
Bg Bh
X R X

Extended Data Fig. 2 | Navigating the Tanner graph. a, A ‘compass’ diagram

that shows the direction in which matrices A and B are applied to travel between
different nodes. b, The unit cell of the construction of a toric layout in the proof
of Lemma 4.
Article
Extended Data Table 1 | Code parameters of Bivariate Bicycle
codes

Small examples of Bivariate Bicycle LDPC codes and their parameters. All codes have
weight-6 checks, thickness-2 Tanner graph, and a depth-7 syndrome measurement circuit.
Code distance was computed by the mixed integer programming approach of ref. 68. The
notation ≤d indicates that only an upper bound on the code distance is known at the time of
this writing. We round r down to the nearest inverse integer. The codes have check matrices
HX = (A∣B) and HZ = (BT∣AT) with A and B defined in the last two columns. The matrices x, y obey
xℓ = ym = 1 and xy = yx.