arXiv:2610.03714
SpiderCSS: Scalable Fault-Tolerant CSS State Preparation
Boldizsár Poór, Andrey Boris Khesin, Sarah Meng Li, Benjamin Rodatz, John van de Wetering, Richie Yeung
Abstract
Fault-tolerant preparation of logical states is a critical primitive for the realization of large-scale fault-tolerant quantum computing. This paper introduces SpiderCSS, a scalable compilation pipeline that generates highly-optimized, fault-tolerant state preparation circuits for arbitrary Calderbank-Shor-Steane (CSS) codes. Starting from an idealized specification of a CSS state, we use the ZX-calculus and fault-equivalent rewrites to construct a diagram that prepares the given state while preserving the fault tolerance of the initial ideal ZX-diagram. From this diagram, SpiderCSS constructs a fault-tolerant CSS-state preparation circuit using a modular construction that is CNOT-minimal both within each component and also in the way these components are connected together. By only using fault-equivalent rewrites, the final circuit is provably fault-tolerant without needing expensive verification. Monte Carlo simulations across a variety of CSS codes up to distance 15 show that SpiderCSS scales efficiently and constructs circuits with substantially reduced depth and qubit footprint compared to existing scalable and heuristic methods, while simultaneously achieving an average decrease of 33.7% in logical error rate and an increase of 9.5% in acceptance rate.
AI Summary
SpiderCSS addresses automated fault-tolerant preparation of logical CSS states with low circuit overhead. It compiles a stabilizer specification into a postselected preparation circuit using ZX-calculus rewrites that preserve fault tolerance, avoiding exhaustive fault-propagation verification. The pipeline optimizes stabilizer generators, decomposes high-degree spiders into flagged CAT-state gadgets, and schedules their connections using dependency graphs. The construction introduces “well-ordered” CAT states whose interconnections can be extracted directly as CNOTs rather than Bell measurements. Within the specified modular construction, the minimum CNOT count is $\sum_{v\in V}C(\deg(v),t)-|E|$, where $C(n,t)$ is the minimum isolated CAT-preparation cost. This is not a global optimality guarantee over arbitrary preparation circuits. Additional reductions exploit hook errors equivalent, modulo state stabilizers, to at most one output error. Fault-tolerance guarantees use the CSS restriction of circuit-level noise. Benchmarks cover 7–144 data qubits and distances 3–15. Against Flag at Origin, reported mean improvements are 33.7% lower logical error rate, 9.5% higher acceptance, and, with depth-oriented reuse, 60.3% lower depth and 20.2% fewer simultaneous qubits. Comparisons use Stim Monte Carlo simulations with circuit-level depolarizing noise, matched scheduling, and lookup-table decoding; improvements are not uniform across codes. Limitations include increasing rejection rates at larger sizes, unavailable logical-error estimates when decoding tables exceed memory, and CAT gadgets that are not optimal in every parameter regime. Generator selection and scheduling use heuristics. Hardware-connectivity constraints are not incorporated, and extension beyond CSS states requires different rewrites.
OpenAI GPT-5.6 and GPT-6, accessed through ChatGPT since June 2026, were used for editing suggestions.
Central claim may not hold: The abstract's unconditional componentwise CNOT-minimality claim is contradicted by the implementation's acknowledged use of nonoptimal CAT-state constructions outside parameter regimes with proven optimal constructions. The claimed polynomial runtime for arbitrary inputs is not established because constructing well-ordered CAT states uses retries without a general existence theorem or a polynomial bound on their number. The construction proves fault tolerance under the CSS noise convention, whereas the abstract's unqualified guarantee also suggests the stronger ordinary-Pauli output-weight property, which the circuit-distance equivalence theorem does not establish.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 23 pages