Microsecond-scale high-survival and number-resolved detection of ytterbium atom arrays

Abstract

Scalable atom-based quantum platforms for simulation, computing, and metrology require fast high-fidelity, low-loss imaging of individual atoms. Standard fluorescence detection methods rely on continuous cooling, limiting the detection range to one atom and imposing long imaging times that constrain the experimental cycle and mid-circuit conditional operations. Here, we demonstrate fast and low-loss single-atom imaging in optical tweezers without active cooling, enabled by the favorable properties of ytterbium. Collecting fluorescence over microsecond timescales, we reach single-atom discrimination fidelities above 99.9% and single-shot survival probabilities above 99.5%. Through interleaved recooling pulses, as short as a few hundred microseconds for atoms in magic traps, we perform tens of consecutive detections with constant atom-retention probability per image - an essential step toward fast atom re-use in tweezer-based processors and clocks. Our scheme does not induce parity projection in multiply-occupied traps, enabling number-resolved single-shot detection of several atoms per site. This allows us to study the near-deterministic preparation of single atoms in tweezers driven by blue-detuned light-assisted collisions. Moreover, the near-diffraction-limited spatial resolution of our low-loss imaging enables number-resolved microscopy in dense arrays, opening the way to direct site-occupancy readout in optical lattices for density fluctuation and correlation measurements in quantum simulators.

Figures (4)

• Figure 1: Fast imaging of ytterbium arrays. (a) An array of ytterbium atoms is illuminated with high-intensity ($s\simeq40$) counter-propagating alternated pulses. The scattered photons are collected through an objective with $\text{NA}=0.6$. Right inset: relevant Yb transitions with dashed gray lines indicating transitions to dark states. Left inset: histogram of filling fraction for a 30-site tweezer array over 1000 realizations. (b) Mid-circuit readout scheme, single-shot image (top) and average (bottom) over 1000 realizations of the tweezer array with 10.8 illumination time. The gray squares indicate the typical $3\times3$ px ROIs for photon integration. (c) ROI photon counts histogram from $5000$ experimental images with 10.8 illumination time. (d) Mean photon counts per atom versus illumination time. Points are the array-average value and error bars (not visible) are the standard deviation. Dashed line: linear fit to the first 6 data points.
• Figure 2: High-fidelity low-loss single-atom detection. (a) Top: Sequence used to measure imaging losses and detection fidelity. Bottom: correlation of photons collected in two consecutive images with illumination time $t_\mathrm{ill} = 6.3\,\text{\textmu s}$ and trap depth $U_0 \simeq2.27\,\text{mK}$. (b) Infidelity (top) and loss probability (bottom) versus illumination time and mean-photon counts. Purple and orange circles show the array median and best-tweezer infidelity. Gray circles and blue diamonds indicate the array-averaged loss probability from the equal-images and the fidelity-bias method, respectively. Shaded region: fit to a loss model with $2\sigma$ confidence interval. (c) Loss probability versus trap depth for $t_\mathrm{ill} = 5.4\,\text{\textmu s}$. The gray dotted and dashed lines show expected losses from heating and off-resonant scattering, respectively; the solid blue line combines both effects. (d) Top: Repeated imaging sequence with interleaved reset pulse. Center: single-shot images after 1, 10, 20 and 30 repetitions. Empty sites in gray. Bottom: survival probability versus number of images ($t_\mathrm{ill} = 6.3\,\text{\textmu s}$) without (gray) and with (green) reset pulse. The green solid (dashed) line is a fit of the measured survival after repeated detection-reset (reset alone) pulses, assuming uniform survival probability across repetitions. Gray line is a guide to the eye. Inset: survival probability after 10 images versus reset pulse duration for ${}^{174}\text{Yb}$. Data points in (b-c), (d) and inset are averaged over $5000$, $3000$ and $1000$ shots respectively. Error bars are the standard deviation across a 10-tweezers array.
• Figure 3: Number-resolved detection and LACs-driven near-deterministic loading dynamics. (a) Multi-atom photon-count histograms recorded after different LACs-pulse durations (0, 5 and 14 ms), using $t_\mathrm{ill} = 20.7\,\text{\textmu s}$ and integrating over circular ROIs with $7$-px radius. Solid lines are multi-peak fits. (b) Top: single-shot image after a $7\,\text{ms}$ LACs pulse. ROI colors indicate the detected occupations. Bottom: filling probability versus blue-detuned LACs-pulse duration. Solid lines are fits with the rate-equation model SM. Error bars are not visible. (c) Loss rates versus LACs-pulse detuning from the ${}^1\text{S}_0\rightarrow{}^3\text{P}_1(F' = 3/2)$ free-space resonance at fixed intensity $I_{L} = 170\,I_\text{sat}$. Solid lines are visual guides. Vertical dashed line marks the optimal detuning employed in the other panels. (d) Filling fraction versus magnetic field and detuning for a $80\,\text{ms}$ LACs-pulse intensity $I_L\simeq215\,I_\text{sat}$. Dashed (solid) lines show the splitting of $m_F' = \pm3/2 (\pm1/2)$ states. The yellow cross marks the optimal parameters used in all other panels ($\Delta_L/2\pi = 9.4\,\text{MHz}$ and $7\,\text{G}$ magnetic field). (e) Good-to-bad collisions ratio $R$ as a function of LACs-pulse intensity. (f) Optimal filling fraction versus LACs-pulse intensity obtained from the rate equation fits; dashed line marks the maximum filling $\simeq 0.87$. Each point is obtained for the LACs-pulse duration that maximizes single-atom occupancy. (g) LACs dynamics with repeated multi-atom imaging. Data are obtained by post-selecting for $N=3$ initial occupation and correcting for losses induced by the first detection SM. All data except in (d) are obtained from $\sim$1000 shots; data in (d) from 100 shots.
• Figure 4: Fast microscopy of tightly-spaced arrays. (a) Size of the atomic signal from Monte Carlo simulations for different imaging conditions: $t_\mathrm{ill} = 5\,\text{\textmu s}$, $U_0=1.1\,\text{mK}$ (blue); $t_\mathrm{ill} = 20\,\text{\textmu s}$, $U_0=2.3\,\text{mK}$ (yellow); $t_\mathrm{ill} = 20\,\text{\textmu s}$, $U_0=1.1\,\text{mK}$ (green). Colored lines are Gaussian fits; black dotted line is the analytical PSF. Insets: experimental (top) and simulated (bottom) average signal from a single atom. (b) Comparison between photon counts in $1\times3$ ROIs and logL separation between filled and empty configurations SM for 1.5-spacing arrays with $t_{\mathrm{ill}}=5.4\,\text{\textmu s}$. (c) Infidelity versus spacing for $t_{\mathrm{ill}}=5.4\,\text{\textmu s}$. Blue circles and yellow diamonds indicate experimental and simulated data respectively, obtained from $>10^4$ data points. Error bars not visible. (d) logL separation of the 2-1 atoms hypotheses for occupied sites in a multiply-filled array with 1.5 spacing and $t_{\mathrm{ill}}=20.7\,\text{\textmu s}$. (e) Single-shot MLE reconstruction with 0, 1 or 2 atoms per site in a 30-site array for $t_{\mathrm{ill}}=20.7\,\text{\textmu s}$.