arXiv:2610.30747
Demand-Driven Routing Space for Lattice-Surgery Compilation
Hanyu Wang, Wan-Hsuan Lin, Zihan Chen, Eddy Z. Zhang, Jason Cong
Abstract
Lattice-surgery compilation places logical qubits on a grid of surface-code tiles and routes their interactions. Sparse static layouts limit routing contention but reserve space that often remains idle. Reducing this reservation requires balancing area savings against additional gate serialization and qubit movement. In this paper, we present a co-designed dynamic architecture and compiler that allocates routing space according to circuit demand. The architecture preserves a fixed virtual qubit floorplan while allowing its physical spacing to change between execution stages. The compiler coordinates placement, scheduling, and routing to implement lattice surgeries using fewer physical qubits. Across our application benchmarks, our method reduces space-time volume by 55.6% in geometric mean and compiles 75.5 times faster than the evaluated DasCot configuration. A circuit-synthesis case study shows how our compiler exposes optimization tradeoffs.
AI Summary
Lattice-surgery layouts reserve routing space that may remain idle, but shrinking it can increase contention and qubit movement. This paper asks how to allocate that space according to circuit demand while minimizing fixed-chip space–time volume. The proposed architecture fixes qubit assignments, orientations, and relative row/column order on a virtual floorplan, while allowing physical spacing to change between execution stages. The compiler combines demand-based shape selection, spectral placement and simulated annealing, criticality-weighted selection of tile-disjoint routes, and stage compaction under lane budgets. Its objective is $HW(\sum_s T_s+\sum_s\delta_s)$, charging the maximum chip dimensions throughout execution and including interstage movement. Evaluation uses 60 circuits spanning arithmetic, QFT/QPE, QAOA, oracles, kernels, and random stress tests. Across application benchmarks, reported geometric-mean volume is 55.6% lower than the evaluated DasCot configuration; compilation is 75.5 times faster on the timing subset. A ten-circuit architecture comparison gives volume ratios of 0.61 for static on-demand layouts and 0.49 for dynamic layouts relative to DasCot, with the dynamic layouts taking 13% more execution steps. These results come from compiled tile-level routes and modeled walks, rather than physical experiments. The model assigns no execution cost to single-qubit Cliffords, assumes on-demand magic states in the main comparisons, and uses an optimistic one-step-per-tile parallel-movement estimate. Routing and placement are heuristic and can miss feasible routes; main comparisons use one seed. Baseline endpoint rules and volume units differ, limiting physical-resource comparisons. Extensions examine finite magic supply, hierarchical reuse, local resynthesis, and Pauli-based computation.
Error found: The supplied paper does not expose numerical results sufficient to independently verify either headline aggregate: the relevant body text contains unresolved macros, the figure descriptions omit values, and the ten-circuit table establishes a different, explicitly narrower aggregate.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 21 pages, 25 figures, 5 tables