Randomized Benchmarking of Single-Qubit Gates in a 2D Array of Neutral-Atom Qubits

We characterize single-qubit Clifford gate operations with randomized benchmarking in a 2D array of neutral-atom qubits and demonstrate global and site selected gates with high fidelity. An average fidelity of $F^2=0.9983(14)$ is measured for global microwave-driven gates applied to a 49-qubit array. Single-site gates are implemented with a focused laser beam to Stark shift the microwaves into resonance at a selected site. At Stark selected single sites we observe $F^2=0.9923(7)$ and an average spin-flip crosstalk error at other sites of 0.002(9).

Figure 1

Qubit array measurements. (a) False-color fluorescence image of single-atom qubits filtered with an independent component analysis. Each pixel views a region $0.62\times 0.62\mu \mathrm{m}$. The image is a composite averaged over 500 exposures. (b) Histogram of the number of occupied sites for 500 array loading measurements. The inset shows the single-atom photoelectron counts at a single site for 2000 measurements. The average overlap of the Gaussian fits at all 49 sites was 0.0004.

Figure 2

(a) Microwave Rabi oscillations at a single site during global addressing. Each point is the average of 50 measurements and the fitted Rabi frequency is $\mathrm{\Omega }=2\pi \times 4.74\mathrm{kHz}$. (b) Oscillations on site 31 using Stark addressing with the average probability of measuring a neighboring site in the wrong state due to crosstalk during a $3\pi$ oscillation indicated numerically. The Rabi frequency for this data was $\mathrm{\Omega }=2\pi \times 8.5\mathrm{kHz}$, and the sites are numbered from 0 in the upper left corner to 48 in the lower right corner.

Figure 3

Probability of measuring the correct output state at site 27 of the array for seven RB sequences. Each sequence was truncated at ten different lengths $\ell =\{1,12,23,34,45,56,67,78,89,100\}$. Each data point is an average of 50 measurements. The inset shows a histogram of gate fidelities for 47 of the 49 array sites. Two sites were dropped due to poor loading statistics.

Figure 4

Crosstalk error for $R_x(\pi /2)$ (solid blue line), $R_x(\pi )$ (dashed yellow line), $R_z(\pi /2)$ (dotted blue line), and $R_z(\pi )$ (dash-dotted yellow line) rotations. The vertical dashed lines mark values of $\sqrt{16n^2-1}$. Panel (a) is the error from Eq. (2) averaged over input states, and panel (b) is the error for the initial state $|1⟩$.

Figure 5

RB data at Stark beam addressed site 31 for ten RB sequences. Each data point is an average of 50 measurements.