Arctic: A Field Programmable Quantum Array Scheduling Technique

TL;DR

This work introduces Arctic, a field-programmable compiler pass for zoned neutral-atom quantum arrays that optimizes reconfigurable coupling through movement-based scheduling. It formulates qubit mapping and movement as a max-cut and layered cross-minimization problem and adds a stacking mechanism to balance array dimensions with algorithmic parallelism, delivering seconds-level compilation and substantial fidelity and pulse-count improvements over prior movement-based methods. Evaluated against state-of-the-art benchmarks on both neutral-atom and superconducting models, Arctic achieves up to 5x pulse reductions and up to 7x fidelity gains, highlighting the practical impact of zoning-aware, movement-focused compilation for scalable quantum computation. The results suggest significant potential for extending to multi-dimensional configurations and additional AOD resources to further enhance parallelism and reduce exposure to laser fields in real devices.

Abstract

Advancements in neutral atom quantum computers have positioned them as a valuable framework for quantum computing, largely due to their prolonged coherence times and capacity for high-fidelity gate operations. Recently, neutral atom computers have enabled coherent atom shuttling to facilitate long-range connectivity as a high-fidelity alternative to traditional gate-based methods. However, these inherent advantages are accompanied by novel constraints, making it challenging to create optimal movement schedules. In this study I present, to the best of my knowledge, the first compiler pass designed to optimize reconfigurable coupling in zoned neutral atom architectures, while adhering to the reconfigurability constraints of these systems. I approach qubit mapping and movement scheduling as a max-cut and layered cross-minimization problem while enhancing support for spatially complex algorithms through a novel "stacking" feature that balances the qubit array's spatial dimensions with algorithmic parallelism. I compare the method across various algorithms sourced from Supermarq and Qasmbench where the compiler pass represents the first exclusively movement-based technique to achieve compilation times consistently within seconds. Results also demonstrate that the approach reduces pulse counts by up to 5x and increases fidelity by up to 7x compared to existing methods on currently available technology.

Figures (7)

• Figure 1: Top) A zoning scheme for an array of neutral atom qubits. Qubits are stored in SLM traps at the start of computation and shuttled from storage to execution for entanglement. One qubit gates can be executed in the storage zone, while mid-circuit measurements can be preformed in the measurement zone. Botttom) The three phases of Arctic: Phase 1) An example of a circuit connectivity graph being partitioned to maximize qubit trajectories from one layer to another (along one dimension). Phase 2) A swapping of horizontal positions to minimize crossing edges that connect the top and bottom layers that were determined in the previous phase. Phase 3) Atoms being shuffled into the execution zone by an AOD to be coupled by the global Rydberg Laser.
• Figure 2: Cross minimization through calculation of the barycenter of neighbors residing in the other layer of qubits. In an ideal scenario, qubits are aligned perfectly with neighbors in the other layer
• Figure 3: a) A minimized two-dimensional layered graph being transitioned into a four-layer graph where the horizontal ordering between qubits is preserved. Edge weights represent operational frequency, where total edge weighted degree represents the operational frequency of the qubit examined. b) An example of how vertical gates must be serialized due to the constraint that neutral atoms move as whole rows or columns.
• Figure 4: Compilation Time in Seconds
• Figure 5: Number of Two Qubit Gates