Autonomously Designed Pulses for Precise, Site-Selective Control of Atomic Qubits

TL;DR

Site-selective control in neutral-atom arrays is hampered by motion-induced amplitude fluctuations of tightly focused control beams. The authors develop an AI-driven framework that autonomously designs composite pulses CP(n) to implement precise single-qubit rotations despite these fluctuations. CP(3) and CP(4) outperform conventional sequences (BB1, SK1) in fidelity, and remain robust to optical aberrations and misalignment, thanks to a bias-cancellation mechanism and tailored spectral filtering that suppress motion harmonics. The method generalizes to arbitrary SU(2) rotations, integrates with existing hardware, and can be extended to other atom-like platforms such as trapped ions and solid-state color centers.

Abstract

Quantum computers based on cold-atom arrays offer long-lived qubits with programmable connectivity, yet their progress toward fault-tolerant operation is limited by the relatively low fidelity of site-selective local control. We introduce an artificial-intelligence (AI) framework that overcomes this limitation. Trained on atom-laser dynamics, a deep neural network autonomously designs composite pulses that improve local control fidelities tenfold while remaining compatible with existing control hardware. We further demonstrate the robustness of these pulses against optical aberrations and beam misalignment. This approach establishes AI-trained pulse compilation for high-fidelity qubit control and can be readily extended to other atom-like platforms, such as trapped ions and solid-state color centers.

Figures (6)

• Figure 1: Site-Selective Control of Atomic Qubits in Optical Tweezers(a) Addressing target atoms with tightly focused beams enables efficient site-selective qubit control. (b) Atom motions in optical tweezers induce amplitude fluctuations in the local control field. (c) With time-varying amplitude (green), conventional composite pulses become ineffective at mitigating amplitude errors (orange).
• Figure 2: Trained Composite Pulses(a) A target qubit rotation (left) on the Bloch sphere is encoded into a series of pulses (right). The trained deep neural network (middle) identifies the required pulse parameters. (b) and (c) Pulse parameters of trained CP(3) and CP(4) implementing the target rotation ($A_\text{tg}$, $\theta_\text{tg}$), respectively.
• Figure 3: Control Fidelities of Trained Composite Pulses(a) and (b) Fidelity of CPs implementing target rotations with $\theta_\text{tg}=\pi/2$ and $A_\text{tg}=\pi$, respectively. (c) Control fidelities over the training range.
• Figure 4: Control Fidelities under Optical Focus Array(a) Camera image of a $40\times40$ focus array generated by a phase SLM. (b) Distributions of peak intensity (top) and $1/e^2$ radius (bottom); solid lines are Gaussian fits with a standard deviation of 1.3% for the intensity and 6.8% (5.1%) for the radius along $x$ ($y$) axis. $R_\text{eff}=\sqrt{R_xR_y}$ where $R_x$ ($R_y$) is the average radius along $x$($y$) axis. (c)-(f)$\pi$-pulse fidelities plotted against deviations in peak intensity and radius for the control beam (c, d) and tweezer (e, f).
• Figure 5: Control Fidelities under Misalignments(a) and (b)$\pi$-pulse fidelities plotted against misalignment in the radial and axial directions, respectively. Insets show the control beam (red) offset from the tweezer (blue).