Bosonic Code Architectures

• Bosonic code architectures are quantum error-correcting designs that embed logical information into infinite-dimensional Hilbert spaces, exploiting symmetry and phase-space geometry to counter photon loss and dephasing.
• They leverage diverse code families—such as rotation-symmetric, GKP, and group-theoretic codes—to enable scalable implementations in superconducting circuits, photonics, and continuous-variable platforms.
• These architectures integrate advanced techniques like dissipative stabilization, analog syndrome extraction, and hybrid error correction protocols, paving the way for fault-tolerant quantum computation.

Bosonic code architectures are families of quantum error-correcting codes that embed logical information into subspaces of the infinite-dimensional Hilbert space of bosonic modes (quantized harmonic oscillators). These architectures leverage the large Hilbert space per mode to realize robust error correction with minimal hardware overhead, supporting scalable quantum computation, communication, and sensing in superconducting circuits, photonics, and other continuous-variable platforms. Key design principles exploit symmetry (rotation, translation, group-theoretic), phase-space geometry, hardware constraints, dissipative stabilization, and analog decoding, yielding a broad taxonomy of code families addressing photon loss, dephasing, and control errors in both single-mode and multimode settings.

1. Foundational Code Families and Symmetry Principles

Bosonic code architectures capitalize on discrete or continuous symmetries imposed on mode quadratures or number operators, which underpin the error-correcting structure and dictate logical gate and syndrome extraction methodologies.

Rotation-symmetric codes define logical subspaces as simultaneous eigenspaces of discrete rotation operators in phase space, $\hat R_N = \exp(i \frac{2\pi}{N} \hat n)$, with $N$-fold symmetry. Generic codewords in the Fock basis are

$$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$

where $f_{mN}$ are code-dependent amplitudes. Explicit realizations include cat codes (superpositions of coherent states on a regular phase-space grid) and binomial codes (finite Fock support, binomial-weighted coefficients) (Grimsmo et al., 2019, Michael et al., 2016, Marinoff et al., 2023, Totey et al., 2023).

Translational-symmetric (GKP) codes encode information in grid states that are eigenstates of two commuting displacement operators, typically $S_q = \exp(2i\sqrt\pi\, \hat q)$ and $S_p = \exp(-2i\sqrt\pi\, \hat p)$, yielding code lattices in phase space. GKP codes correct small displacement errors and exhibit leading thresholds when concatenated with surface codes (Brady et al., 2023, Lemonde et al., 2024).

Group-theoretic and multimode constructions generalize these principles by designing codes via projectors onto subspaces invariant under group actions, permitting logical encoding with native implementation of Clifford or Pauli groups via linear optics or passive Gaussian operations. Notable constructions include multi-mode rotationally symmetric codes (Ahmed et al., 28 Aug 2025), two-mode Fourier cat codes (Leverrier, 22 May 2025), and 2T-qutrit codes (Denys et al., 2022).

2. Error Models, Code Distance, and Syndrome Extraction

Canonical bosonic error channels include photon loss (amplitude damping), photon gain, dephasing (phase damping), and, for grid codes, small coherent displacements. The Knill–Laflamme conditions for correctability translate, for rotation codes, into the requirement that photon-number shifts $< N$ ($d_n = N$) and phase rotations $|\phi| < \pi/N$ ($d_\phi = \pi / N$) are detectable and correctable (Marinoff et al., 2023).

Syndrome extraction is intimately tied to the code symmetry. For binomial and cat codes, error subspaces are labeled by photon number modulo an integer, permitting syndrome discrimination via number-selective parity measurements with dispersive ancilla coupling and SNAP gates (Michael et al., 2016, Ma et al., 2021). For GKP codes, modular quadrature measurements using echoed conditional displacements and high-resolution homodyne detection extract analog syndrome information (Brady et al., 2023, Lemonde et al., 2024).

In multimode codes, syndrome measurement may involve parity or number-difference operators across modes or projection into group-defined subspaces (e.g., via integer-matrix constraints for “tiger codes” (Xu et al., 2024)). Syndrome extraction can be performed in tandem with dissipative stabilization in current superconducting architectures.

Code Type	Dominant Syndrome	Distance $N$0
Single-mode cat	Photon parity	$N$1
Binomial	Number mod $N$2	$N$3
GKP (1-mode)	Modular $N$4	$N$5 (grid spacing)
2-mode Fourier-cat	Joint parity, parity diff.	$N$6, $N$7 (see (Leverrier, 22 May 2025))

3. Code Construction and State Engineering

Binomial codewords with finite Fock support are engineered through recursive application of multiphoton interactions (multiphoton Jaynes–Cummings Hamiltonians) between a bosonic oscillator and a two-level system. The ability to reduce required multiphoton order (via cascading lower-order processes) enhances experimental feasibility (Laha et al., 11 Jul 2025). Cat codes are stabilized via tailored two-photon or four-photon driven dissipative processes, creating comb-like Fock state structures or phase-space legs (Ma et al., 2021, Guo et al., 2024).

Quantum cubature codes (QCCs) provide a general geometric design framework, where codewords are weighted superpositions of phase-space points chosen as a cubature (Euclidean or spherical design), enforcing moment-cancellation conditions to guarantee correctability up to a designed order of photon loss or general excitation-changing errors (Yang et al., 28 Nov 2025). This unifies cat codes, binomials, and new families (e.g., multi-shell arrangements), with separation metrics such as geometric resolution $N$8 optimizing overlap and error rates under photon loss.

Group-theoretic architectures project codewords onto subspaces supporting representations of finite groups, using, e.g., passive linear optics networks and beam splitters, to simultaneously realize high-dimensional code spaces and logical gate sets with minimal control overhead (Ahmed et al., 28 Aug 2025, Leverrier, 22 May 2025, Denys et al., 2022).

4. Error Correction Protocols and Decoding Strategies

Measurement-based correction employs analog syndrome readout at the bosonic level (e.g., continuous homodyne for GKP, parity analogs for cats), enabling quasi-single-shot decoding and soft-decision integration into outer DV decoders (Berent et al., 2023, Hillmann, 17 Dec 2025). Decoders exploiting analog information (likelihood ratios, soft parity) yield thresholds and latencies superior to repeated hard-syndrome sampling.

Teleportation-based error correction—especially for rotation codes—employs code-agnostic entangling gates (CROT) and phase-basis measurements, shifting error tracking entirely into Pauli frame software updates (Grimsmo et al., 2019). This approach enables efficient concatenation with subsystem (Bacon–Shor, surface, or LDPC) codes.

Dissipative (autonomous) error correction in paradigms such as squeezed cat codes, pair-cat, and higher-mode codes leverages engineered jump operators to stabilize the code manifold and continually pump leakage back to the logical subspace, with passive recovery of correctable errors (Ma et al., 2021, Hillmann, 17 Dec 2025, Guo et al., 2024).

Decoding Approach	Features	Key Thresholds
Analog quasi-single-shot	Uses continuous syndrome, low latency	$N$9 reduced overhead, up to $$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$0 for 3D surface-QLDPC (Berent et al., 2023)
Teleportation-based	Pauli frame tracking, software correction	Near-optimal under noise
Autonomous (dissipative)	Engineered jumps, zero-latency correction	$$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$1 for squeezed-cats (Hillmann, 17 Dec 2025)

5. Multimode and Group-Theoretic Code Architectures

Multimode constructions fundamentally expand the correctable error domain and hardware efficiency. Pair-cat, extended cat, and group-theoretic codes (e.g., via homological products, finite group representations) can achieve:

• Simultaneous first-order correction of several error types (independent or correlated dephasing/loss), often eliminating the trade-off between protection against loss and dephasing present in single-mode codes (Ahmed et al., 28 Aug 2025).
• Linear-optics implementation of all logical gates required for Clifford or Pauli universality (e.g., via passive beam splitter networks and phase shifters) (Ahmed et al., 28 Aug 2025, Leverrier, 22 May 2025).
• Exact correction of correlated phase noise by circuit-level recovery protocols employing controlled-X gates between data and auxiliary codes (Ahmed et al., 28 Aug 2025, Xu et al., 2024).

Surface-like multimode codes (“tiger codes”) constructed by the homological product of pair-cat and repetition codes realize fully two-dimensional bosonic topological codes, permitting geometric locality and simultaneous syndrome extraction and stabilization (Xu et al., 2024).

6. Physical Realization, Hardware Scaling, and Performance

Superconducting cavity QED is the prevalent platform for implementing bosonic codes, offering high-$$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$2 resonators ($$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$3–$$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$4 ms), dispersive ancilla coupling for syndrome extraction, and parametric drives or nonlinear couplers for dissipative stabilization and gate operations (Ma et al., 2021, Lemonde et al., 2024, Hillmann, 17 Dec 2025). Single-mode codes require only one cavity per logical qubit; GKP and surface-GKP concatenations offer logical error rates $$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$5 at hardware overhead an order of magnitude less than surface-code transmon arrays (Lemonde et al., 2024).

Cat code engineering via Josephson nonlinearity supports fast lattice gates (sub-ns) and direct Floquet-engineered Hamiltonian preparation, removing the need for long SNAP sequences (Guo et al., 2024).

Hybrid CV–DV architectures employing cat-coded continuous-variable modes and discrete-variable (photon) qubits yield fault-tolerant logicals with high loss thresholds ($$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$6) and resource efficiency in photonic and superconducting systems (Lee et al., 2023).

Performance trade-offs are dictated by mean photon number $$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$7, code symmetry order $$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$8, cubature degree $$|0_N\rangle = \sum_{k=0}^\infty f_{2kN} |2kN\rangle,\qquad |1_N\rangle = \sum_{k=0}^\infty f_{(2k+1)N} |(2k+1)N\rangle,$$9 (for QCCs), and the interplay between loss and dephasing rates. Optimal $f_{mN}$0 is typically a few photons per mode, suppressing loss while constraining dephasing-induced logical error (Totey et al., 2023). For rotation codes, the number–phase distance product $f_{mN}$1 imposes a fundamental trade-off, but multimode group-theoretic codes can bypass this (Ahmed et al., 28 Aug 2025).

7. Future Directions and Open Challenges

Frontiers in bosonic code architectures include:

• Systematic search and benchmarking of novel phase-space geometries (e.g., higher-dimensional cubature codes, multi-shell QCCs) to optimize separation and error threshold under arbitrary noise (Yang et al., 28 Nov 2025).
• Expanding multimode code constructions to higher logical dimensions (qudits, rotors) and topological codes with local hardware interactions (Xu et al., 2024, Ahmed et al., 28 Aug 2025).
• Integration of fast analog decoding, quasi-single-shot/reinforcement-learned feedback, and fault complexes for the design of dynamic, low-latency error correction in large-scale circuits (Hillmann, 17 Dec 2025).
• Extension to hybrid photonic/superconducting platforms, realizing the necessary nonlinearities and integrating bosonic modes with high-quality photon-number-resolving detectors and ancilla reset mechanisms (Lee et al., 2023).

The evolving landscape of bosonic code architectures is marked by the convergence of symmetry-based code design, hardware-adapted engineering, analog information utilization, and resource-efficient concatenation strategies. These advances indicate a clear and scalable pathway toward fault-tolerant quantum computation and communication in continuous-variable architectures.