High-Fidelity Single-Qubit Gates on Neutral Atoms in a Two-Dimensional Magic-Intensity Optical Dipole Trap Array

As a conventional approach, optical dipole trap (ODT) arrays with linear polarization have been widely used to assemble neutral-atom qubits for building a quantum computer. However, due to the inherent scalar differential light shifts (DLS) of qubit states induced by trapping fields, the microwave-driven gates acting on single qubits suffer from errors on the order of ${10}^{-3}$. Here, we construct a DLS compensated ODT array based upon a recently developed magic-intensity trapping technique. In such a magic-intensity optical dipole trap (MI-ODT) array, the detrimental effects of DLS are efficiently mitigated so that the performance of global microwave-driven Clifford gates is significantly improved. Experimentally, we achieve an average error of $(4.7\pm 1.1)\times {10}^{-5}$ per global gate, which is characterized by randomized benchmarking in a $4\times 4$ MI-ODT array. Moreover, we experimentally study the correlation between the coherence time and gate errors in a single MI-ODT with an optimum error per gate of $(3.0\pm 0.7)\times {10}^{-5}$. Our demonstration shows that MI-ODT array is a versatile platform for building scalable quantum computers with neutral atoms.

Figure 1

Experimental setup for producing ODTs array. A 830 nm laser beam is deflected in two orthogonal directions to produce an ODT array by a dual-axis acousto-optic deflector (AOD) which is driven by a two-channel radio frequency (rf) with several tones. A liquid crystal retarder (LCR) is used to change the linearly polarized traps to circularly polarized magic-intensity traps. An objective lens not only provides a 3D tight confinement but also collects fluorescence from the trapped ${}^{87}\mathrm{Rb}$ atoms. The fluorescence (denoted by dashed lines) is then counted by a single photon counting module (SPCM). A microwave horn external to the vacuum glass cell is used to deliver the microwave radiation.

Figure 2

Results of the single-qubit Clifford gate benchmarking experiment. The data are probabilities of measuring the $|0⟩$ after applying five different random RB sequences truncated at different lengths $\ell =1$, 200, 400, 600, 800, 1000, 1300. The closed points are the average values of five different RB sequences at each value of $\ell$. Error bars are statistical, and represent the standard deviation of the mean. The solid line is a fit to Eq. (1), yielding, respectively, errors per gate of ${ϵ}_g\approx (3.0\pm 0.7)\times {10}^{-5}$ and state preparation errors and readouts errors $d_{if}\approx 0.03\pm 0.01$.

Figure 3

Dependence of values of errors per gate on the coherence times $T_{2s}$. All values of errors per gate are measured by using the RB method. All of the accompanying error bars of coherence times and the values of errors per gate are fitting errors. The solid curve is fit to Eq. (2). The inset shows the dependence of coherence time $T_{2s}$ on the ratio of varied trap depths ($U$) to magic working trap depth ($U_m$). The accompanying solid curve is to guide the eye.

Figure 4

Array experiments. (a) Measurement sequence. Specifically, each site of the array is labeled from 1 to 16. The dashed arrows indicate the path of scanning. (b) The recorded values of error per gate as a function of site number. All values of error per gate are measured by using RB method. All the accompanying error bars error per gate are fitting errors. The lowest and highest gate errors are $(2.7\pm 1.8)\times {10}^{-5}$ and $(7.0\pm 1.4)\times {10}^{-5}$ respectively.