Heterogeneous quantum error-correcting codes Preprint: APS/123-QED
Omid Khosravani, Guillermo Escobar-Arrieta, Kenneth R. Brown, Mauricio Gutiérrez · 2026
Read Heterogeneous quantum error-correcting codes Preprint: APS/123-QED by Omid Khosravani, Guillermo Escobar-Arrieta, Kenneth R. Brown, Mauricio Gutiérrez (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-06817.txt.
Introduction
Khosravani and colleagues introduce mixed quantum error-correcting codes built from qubit types with error channels, decoded with maximum-likelihood tensor networks in the code-capacity regime. When both types share noise bias but differ in error frequency, placement matters. Putting noisier qubits in the bulk and cleaner qubits on the boundary yields higher thresholds than the opposite placement. Logical error improvements grow exponentially with code distance. When types share error rate but differ in bias, placement flips, with high-bias qubits on the boundary raising thresholds sharply.
Catalog record
| CREATOR | Omid Khosravani, Guillermo Escobar-Arrieta, Kenneth R. Brown, Mauricio Gutiérrez |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25405 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.06817
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Heterogeneous quantum error-correcting codes Preprint: APS/123-QED Omid Khosravani, Guillermo Escobar-Arrieta, Kenneth R. Brown, Mauricio Gutiérrez Abstract We introduce heterogeneous quantum error-correcting codes composed of qubit types with distinct error channels and study their performance in the code-capacity regime using maximum-likelihood tensor network decoding. In the regime where both qubit types share the same noise bias but differ in physical error rate, placing noisier qubits in the bulk—where each error triggers more syndrome bits—and cleaner qubits on the boundary yields thresholds exceeding 0.4 (compared to ∼ 0.2 \sim\!0.2 for the reverse placement) and improvements exceeding three orders of magnitude in logical error rate at high bias, with the advantage growing exponentially with code distance. In the regime where both types share the same error rate but differ in bias, the optimal strategy reverses: placing high-bias (more predictable) qubits on the boundary increases the threshold from 0.292 ( 5 ) 0.292(5) to 0.360 ( 9 ) 0.360(9) at η high = 100 \eta_{\text{high}}=100 , and from 0.29 ( 1 ) 0.29(1) to 0.398 ( 4 ) 0.398(4) at η high = 1000 \eta_{\text{high}}=1000 . We also observe a striking bias-inversion property: the logical error channel becomes strongly X X - and Y Y -biased despite the physical noise being Z Z -biased. We propose a stabilizer-ratio hypothesis that provides a unified information-theoretic explanation for both placement rules and predicts even larger advantages for code families such as color codes. I Introduction The threshold theorem for fault-tolerant quantum computation 2 establishes that arbitrarily long computations can be performed reliably provided that the physical error rate falls below a critical value, as recently demonstrated experimentally 1 . Historically, theoretical frameworks have predominantly considered homogeneous qubit systems with uniform error models, particularly the Pauli depolarizing channel 13 ; 24 . This simplified model, while mathematically tractable, fails to capture the complexity of real quantum systems where different qubits exhibit varying error magnitudes, error channels may be strongly biased toward specific error types, and physical implementations naturally lead to distinct error characteristics for different qubit types. These considerations are not merely theoretical: heterogeneous qubit systems are an emerging reality across multiple quantum computing platforms. In trapped-ion processors, different ionic species can serve as data and ancilla qubits with vastly different coherence times and gate fidelities 5 . Bosonic systems such as cat qubits exhibit strongly biased noise with bias ratios \eta that can exceed 100 19 ; 18 , while other qubits on the same chip may have near-depolarizing noise. Superconducting architectures increasingly integrate transmon qubits of varying quality, where fabrication variation alone produces a spread in T_{1} and T_{2} times across the device 17 . Rather than treating this heterogeneity as an obstacle to be mitigated, a natural question arises: can we design error-correcting codes and decoding strategies that actively exploit the differences between qubit types to achieve better logical error rates? Single-qubit Clifford deformations have proven to be a useful tool for improving the performance of codes under biased noise 28 ; 3 ; 25 ; 9 ; 11 . The XY surface code 27 replaces the Z stabilizers of the standard surface code with Y stabilizers so that Z errors anticommute with both the X and Y stabilizer types, providing more syndrome information about the dominant error. In the code-capacity regime with a tensor network (TN) decoder, the XY surface code achieves thresholds that increase monotonically with the noise bias, from 18.7\% for the depolarizing channel to 43.7\% for pure dephasing 27 . The XZZX surface code 3 achieves high thresholds even with a simple matching decoder. As shown in Ref. 25 , it is possible to tailor surface codes to heterogeneous single-qubit noise by carefully aligning the Pauli noise and the stabilizer checks at each site. Crucially, these ideas extend well beyond the surface code. The domain wall color code 26 achieves a 50% code-capacity threshold at infinite bias and matches the XZZX surface code thresholds at finite bias, while retaining the color code’s transversal Clifford gate advantage. The XYZ color code 21 exhibits fracton-like properties under infinite dephasing bias, eliminating string-like logical operators entirely. Clifford deformations have been applied to three-dimensional topological codes including the 3D color code 14 , to compass codes 6 , and to quantum LDPC codes 20 . However, all of these studies consider homogeneous noise across the code—no prior work has examined the strategic placement of distinct qubit types within any of these bias-tailored code families. The concept of heterogeneous quantum error-correcting codes, in which qubits of fundamentally different types are combined within a single code block, was introduced in Ref. 16 and presented in Ref. 15 . We note that the term “heterogeneous” has been used in recent QEC literature to describe architectures that combine different code families within a single fault-tolerant system 22 ; 23 ; our usage is distinct, referring to physically different qubit types—with different error rates or noise biases—strategically placed within a single code block. In this work, we study the problem of qubit placement within Clifford-deformed surface codes for heterogeneous systems with two types of qubits. We investigate two regimes: (a) qubits with the same bias but different physical error rates, and (b) qubits with the same physical error rate but different bias values. Using tensor network decoding, we find that strategic placement of qubits can yield improvements exceeding three orders of magnitude in logical error rate and substantial threshold gains. We observe a striking bias-inversion property in the logical error channel and propose a stabilizer-ratio hypothesis that explains both placement rules and predicts even larger advantages for other code families. II Setup II.1 Biased noise model The biased noise error channel is modeled as \begin{split}\mathcal{E}_{\text{biased}}(\rho)&=(1-p)\rho\\ &\quad+p\!\left(\frac{\eta}{1+\eta}Z\rho Z+\frac{X\rho X+Y\rho Y}{2(1+\eta)}\right)\!,\end{split} (1) where p is the total error probability and \eta=p_{Z}/(p_{X}+p_{Y}) is the bias ratio, following the convention of Ref. 28 . This gives p_{Z}=p\eta/(1+\eta) and p_{X}=p_{Y}=p/[2(1+\eta)] . When \eta=1/2 , this reduces to the standard depolarizing channel; as \eta\to\infty , the channel approaches pure dephasing. II.2 Clifford deformations and the XY surface code Clifford deformations consist of conjugating the stabilizers of a QEC code by single-qubit Clifford operators, transforming a code into a topologically equivalent one with vastly different performance against biased noise. Among the Clifford-deformed versions of the rotated surface code catalogued in Ref. 11 , the XY code does not have the highest threshold at all bias values—at the moderate biases ( \eta=10 – 100 ) considered here, other deformations such as C1 and C2 can outperform it in the homogeneous setting. We nevertheless focus on the XY surface code because it applies the same single-qubit Clifford to every data qubit, making it the natural starting point for heterogeneous codes: the optimal Clifford deformation for a site with one error channel need not be optimal for a site with a different channel, and the uniform structure of the XY code avoids conflating the effects of heterogeneous placement with those of non-uniform deformations. Single-qubit deformations do not change the overall code distance d but can change the individual Pauli distances d_{x} , d_{y} , and d_{z} , allowing the distance along the dominant error type to be increased. II.3 Tensor network decoding Maximum-likelihood decoding can