DasAtom: A Divide-and-Shuttle Atom Approach to Quantum Circuit Transformation

TL;DR

DasAtom tackles quantum circuit transformation for neutral-atom quantum computers by partitioning CZ-layer circuits into subcircuits and using atom shuttling to switch between optimized mappings. It leverages long-range interactions and in-computation qubit movement to execute all gates within a subcircuit directly, reducing idle time and gate infidelities. On a 30-qubit QFT, it achieves up to ~415x fidelity improvement over Enola and ~10x over Tetris, with results suggesting exponential scaling with qubit count. The approach demonstrates strong potential for scaling NA platforms and informs hardware-software co-design, while noting current limitations and possible improvements.

Abstract

Neutral atom (NA) quantum systems are emerging as a leading platform for quantum computation, offering superior or competitive qubit count and gate fidelity compared to superconducting circuits and ion traps. However, the unique features of NA devices, such as long-range interactions, long qubit coherence time, and the ability to physically move qubits, present distinct challenges for quantum circuit compilation. In this paper, we introduce DasAtom, a novel divide-and-shuttle atom approach designed to optimise quantum circuit transformation for NA devices by leveraging these capabilities. DasAtom partitions circuits into subcircuits, each associated with a qubit mapping that allows all gates within the subcircuit to be directly executed. The algorithm then shuttles atoms to transition seamlessly from one mapping to the next, enhancing both execution efficiency and overall fidelity. For a 30-qubit Quantum Fourier Transform (QFT), DasAtom achieves a 414x improvement in fidelity over the move-based algorithm Enola and a 10.6x improvement over the SWAP-based algorithm Tetris. Notably, this improvement is expected to increase exponentially with the number of qubits, positioning DasAtom as a highly promising solution for scaling quantum computation on NA platforms.

Figures (14)

• Figure 1: The architecture graph of a neutral atom quantum hardware, where $(x,y)$ ($0\leq x,y\leq 2$) denotes the location of an atom, and two atoms are connected if their distance is smaller than $R_\text{int}=r_\text{int}\times d$, where $r_\text{int}\in \set{1,\sqrt{2},2}$.
• Figure 2: Illustration of atom movement in NAQC: (a) program qubits are mapped to a $4\times 4$ SLM array, (b) a $2\times 2$ AOD array is activated, and two SLM atoms, carrying $q_1,q_2$, are loaded (i.e., transferred) to the AOD array, (c) AOD rows (columns) are moved to the right (upward), carrying $q_1$ and $q_2$ to positions (2,3) and (3,2). To apply a CZ gate between $q_0$ and $q_1$, we offload $q_1$ to the SLM array and then apply an individual Rydberg laser to interact $q_0$ and $q_1$. If $R_\text{restr}=R_\text{int}=d$, then $CZ(q_0,q_1)$ and $CZ(q_2,q_3)$ can be applied in parallel. However, this is not possible when $R_\text{restr}=\sqrt{2}d$ and $R_\text{int}=d$.
• Figure 3: The QFT-5 circuit, where $H$ is the Hadamard gate, $P(\theta)$ is a phase gate with phase $\exp(i\theta)$.
• Figure 4: The QFT-5 circuit decomposed in gate set $\set{CZ, R_x,R_y,R_z}$, where each $R$ denotes an $R_x$, $R_y$, or $R_z$ gate.
• Figure 5: The decomposed QFT-5 circuit with all single-qubit gates removed. The CZ circuit is partitioned in 12 CZ layers and divided into two subcircuits with the second consisting of the last four CZ gates.

Theorems & Definitions (4)

• Definition 1: Interaction Graph
• Example 1
• Remark 1
• Remark 2