A tweezer array with 6100 highly coherent atomic qubits

Abstract

Optical tweezer arrays have transformed atomic and molecular physics, now forming the backbone for a range of leading experiments in quantum computing, simulation, and metrology. Typical experiments trap tens to hundreds of atomic qubits, and recently systems with around one thousand atoms were realized without defining qubits or demonstrating coherent control. However, scaling to thousands of atomic qubits with long coherence times, low-loss, and high-fidelity imaging is an outstanding challenge and critical for progress in quantum science, particularly towards quantum error correction. Here, we experimentally realize an array of optical tweezers trapping over 6,100 neutral atoms in around 12,000 sites, simultaneously surpassing state-of-the-art performance for several metrics that underpin the success of the platform. Specifically, while scaling to such a large number of atoms, we demonstrate a coherence time of 12.6(1) seconds, a record for hyperfine qubits in an optical tweezer array. We show room-temperature trapping lifetimes of 23 minutes, enabling record-high imaging survival of 99.98952(1)% with an imaging fidelity of over 99.99%. We present a plan for zone-based quantum computing and demonstrate necessary coherence-preserving qubit transport and pick-up/drop-off operations on large spatial scales, characterized through interleaved randomized benchmarking. Our results, along with recent developments, indicate that universal quantum computing and quantum error correction with thousands to tens of thousands of physical qubits could be a near-term prospect.

Figures (16)

• Figure 1: Large-scale tweezer array.a, Representative single-shot image of single cesium atoms across a 11,998-site tweezer array. Inset: magnified view of a subsection of the stochastically loaded array. b, Averaged image of single atoms across a 11,998-site tweezer array. Inset: magnified view of a subsection of the averaged array. Atoms are spaced by $7.2 \ \upmu$m and held in 1061 nm and 1055 nm optical tweezers. The contrast is enhanced for visual clarity. c, Schematic of the optical tweezer array generation. Tweezer arrays, generated by two spatial light modulators (SLM), at 1061 nm and 1055 nm are combined with orthogonal polarization, and focused through an objective with a numerical aperture (NA) of 0.65 and a field of view (FOV) of 1.5 mm in diameter. The direction of gravity is along $\hat{y}$. We collect scattered photons from single atoms through the same objective and image them on a qCMOS camera. d, Histogram of filling fraction. We load 6,139 single atoms on average per experimental iteration (51.2% of the array on average), with a relative standard deviation of 1.13% over 16,000 iterations. e, Summary of the key metrics demonstrated in this work.
• Figure 1: Experiment apparatus and sequence.a, Simplified view of the vacuum chamber. The 2D MOT cell (Infleqtion CASC) containing an electrically heated cesium dispenser, shown inside its integrated photonics assembly. It is attached to a stainless steel vacuum chamber on which an ion pump is mounted. We further use two titanium sublimation pumps (one mounted from the top, as shown, and one mounted from the bottom, not visible), sputtering almost the entire surface area of the chamber, except the rectangular part of the science glass cell and the ion pump. We use the following conventions for the laser beams: thick red for MOT beams, thin red for PGC beams, dark red (along $\hat{x}$) for state preparation beam, and purple for tweezer beam. b, Summary of the relevant states and transitions used in this work. c, Summary of a typical experimental sequence, as described in the Methods.
• Figure 2: High-fidelity atom detection in a large-scale tweezer array. Imaging histogram showing the number of photons collected per site and per image. Note that the horizontal axes are weighted photon counts (see text); for non-weighted photon counts, see Ext. Data Fig. \ref{['fig:imaging-ed']}b. a, Imaging histogram of three randomly selected sites in the array (where $x$ and $y$ respectively denote the horizontal and vertical site indices in the array), and b, averaged over all sites in the array. Per-site histograms are fitted with a Poissonian model that integrates losses during imaging (Methods). The wide separation of peaks for empty and filled tweezers enables the high imaging fidelity presented in this work. The binarization threshold used to determine tweezer occupation is indicated by the vertical dashed line, and the average point-spread functions for the two classifications (atom absent and atom detected) are shown next to their corresponding peaks. Note that we detect no more than one atom in each tweezer. Inset: the same histogram presented with a log-scale vertical axis. The weighted average relative error bar per bin is 0.08% (0.05% for the log-scale inset due to the smaller number of bins).
• Figure 2: Tweezer uniformity details.a, The tweezers created by two fiber amplifiers are labeled on the averaged atomic image shown in Fig. \ref{['Fig1']}b. We create 11,513 (488) tweezers with laser light at 1061 nm (1055 nm), for a total of 12,001 tweezers. The 1055-nm tweezers fill the gap created by the spatial filtering of the 0th order in the 1061-nm tweezer pathway, as described further in the tweezer generation section. b, The WGS weights given to tweezers during the tweezer homogenization procedure, as a function of angular distance from the 0th-order reflection off the SLM, with the physical distance this corresponds to given our optical setup shown on the upper axis. In teal is plotted the weights obtained after the tweezer depths are uniformized based on loading probability. In yellow is shown the weight compensation that would be expected based on diffraction efficiency calculations assuming blazed gratings are utilized for displacement. The weight increases with larger angle in order to compensate for the diminishing diffraction efficiency as a function of tweezer distance to the center. This additionally informs our decision to create a circularly shaped array. c, The per-site loading probability array map and its histogram. We feedback on the WGS trap depths based on the loading rate per site to uniformize the trap depth. We see an average loading probability per site of 51.2% with a relative standard deviation of 3.4%. The lowest loading probability is 25.1% for one tweezer, which is the only tweezer not shown in the histogram, but included in the quoted average. This tweezer, however, does not exhibit a significant difference in imaging survival probability, coherence time, or single-qubit gate fidelity (see Ext. Data Fig. 5a and Ext. Data Fig. 8). Therefore, this tweezer site could also be utilized in the experiment. Note that three tweezers in the array are excluded for the data shown in this work, since they are affected by leakage from the 0th order of the SLM on the 1061-nm tweezer pathway that is not completely extinguished via the spatial filtering, resulting in 11,998 usable sites out of 12,001 generated sites. We note that quantum computation can be done despite defects in an array, and that thus one can choose to avoid a small number of traps strikis_quantum_2023. d, The per-site tweezer depth map and its histogram. This is obtained by measuring the differential light shift on $F=4 \leftrightarrow F'= 4$$D2$ transition levine_quantum_2021. We see an average trap depth of $k_B \times0.18(2)$ mK with a standard deviation of 11.4% across the sites.
• Figure 3: Long vacuum-limited lifetime and high imaging survival probabilitya, Vacuum-limited lifetime. Array-averaged survival fraction as a function of hold time is plotted. Three experiments are shown in the figure: pulsed cooling, continuous cooling, and no cooling. The green markers show data with a 10-ms 2D PGC block applied every 2 s during the wait time (pulsed PGC). The red markers show data with 2D PGC block continuously applied during the wait time (continuous PGC), while the blue markers show the data without cooling during the wait time (no PGC). The error bars are smaller than the markers. We find a $1/e$ lifetime of around 2.2 min without cooling. When the pulsed PGC block is applied, by fitting the data with $p(t) \propto \exp(-t/\tau)$, we find a vacuum lifetime of $\tau=22.9(1)$ min. When the 2D PGC is applied continuously, we obtain $\tau=17.7(2)$ min. b, Array-averaged survival fraction after many successive images. Between each image, we hold the atoms for 10 ms, without applying any cooling beams. We fit the data with $p(N) \propto p_{1}^N$, where $p(N)$ is the survival fraction after imaging $N$ times. From the fit, we find a steady-state imaging survival probability of $p_{1}=0.9998952(1)$. The light purple fill shows the estimated 68% confidence interval.