arXiv:2609.29267
MagiCFirm: A Runtime for Magic-State Cultivation with Algorithm-Hardware Co-Design
Jubo Xu, Abbas B. Ziad, Prakash Murali, Hongxiang Fan
Abstract
Magic-state cultivation offers a promising alternative for lowering the cost of non-Clifford operations in fault-tolerant quantum computing (FTQC). However, realizing cultivation in practice exposes two challenges: (i) the lack of an open-source classical runtime layer between logical software and physical control, and (ii) the latency constraints on protocol-specific decisions that determine magic-state readiness. To address these challenges, we adopt an algorithm--hardware co-design approach. At the algorithm level, we develop a two-stage early-escape scheme that identifies an informative subset of detectors offline, constructs a compact decoding problem, and performs partial decoding in parallel with complete decoding at runtime, allowing high-confidence attempts to advance before full decoding completes. At the hardware level, we present MagiCFirm, a configurable runtime that combines offline-compiled microprograms with dedicated datapaths for detector construction, event processing, and protocol control, enabling end-to-end execution of magic-state cultivation. Across evaluated configurations, MagiCFirm reduces wall-clock magic-state preparation time by up to 39.3% at matched logical error rate. For a representative magic-state-bound workload, this translates to an estimated 11% reduction in overall application runtime.
AI Summary
Magic-state cultivation requires detector-driven retries and decoding-based acceptance, making classical control latency part of state-preparation time. This paper develops MagiCFirm, a microprogrammed FPGA runtime for these operations, and a partial-decoding scheme intended to release high-confidence states before complete decoding finishes. Offline, logical-ambiguity profiling selects detectors associated with competing logical explanations, structural closure adds missing error context, and graph contraction constructs a smaller decoding problem. At runtime, partial and complete complementary-gap decoders run concurrently. A calibrated partial-gap threshold permits early acceptance; otherwise, acceptance awaits complete decoding. Complete decoding always finishes to provide the Pauli-frame correction. Evaluation combines noisy-circuit simulation, RTL tests, FPGA place-and-route results, and cycle-accurate timing models using Micro Blossom decoder latencies. Reconstructed detectors match Stim on 1,000 noisy shots per design. For evaluated distributed configurations, modeled post-selection feedback remains below $0.78\,\mu\mathrm{s}$. For $d_{\mathrm{cultiv}}=3$, two-stage acceptance reduces mean preparation time, including retries, by up to $39.3\%$ at matched logical error rate. A cultivation-plus-distillation supply model gives approximately $12\%$ lower supply time at a representative operating point, implying roughly $11\%$ lower application runtime when supply accounts for $90\%$ of execution. The partial gap’s conservative bias is empirical, not guaranteed, and thresholds require calibration. Benefits diminish at stringent output error rates and for larger cultivation codes. End-to-end gains are modeled rather than demonstrated on quantum hardware. The runtime currently supports precompiled paths; dynamic circuits, online error-model updates, and adaptation to device drift remain future work.
Error found: The DEM-contraction algorithm retains only one shortest hidden-space path per endpoint pair without distinguishing logical parity, so its pre-truncation construction need not preserve the two logical-class minimum weights required for the complementary gap.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.