01 · ONE SHOT
The experimental cycle, on the camera
Press Run. The square is what an EMCCD/sCMOS camera would see; the panel keeps score.
Interactive
- Load
- Image
- Rearrange
- Verify
- Entangle
- Read out
02 · YOUR APPARATUS
Settings
Green tags are published numbers; amber tags are illustrative defaults to replace with your own system's.
Array
Loading
Imaging
Rearrangement
Entangling gate
03 · A THOUSAND SHOTS
Where the defects come from
The same simulation without the animation, repeated. Every number below comes from these settings.
Interactive
Shots spoiled by…
A shot counts under every cause that left at least one hole in the target, so the bars can add to more than 100%.
Photon counts per site
Defect-free probability vs target size
Press “Sweep target size” to fill in the Monte Carlo points (200 shots per size).
04 · WHY REARRANGE, AND WHY RELOAD
The exponential wall
Without rearrangement
Each tweezer loads independently with probability $p$, so a chosen $K\times K$ block is full with probability $p^{K^2}$. At $p = 0.5$ an $8\times 8$ block is full once in $2^{64} \approx 1.8\times10^{19}$ tries. Blockade loading caps $p$ near one half (Schlosser 2001); enhanced loading raises it, but even $p = 0.89$ gives $0.89^{64} \approx 6\times 10^{-4}$.
Rearrangement turns this around: take an image, then use a steerable tweezer to carry spare atoms into the holes (Barredo et al. 2016; Endres et al. 2016). Now the hard limit is having enough atoms, $P\big(\mathrm{Bin}(M^2, p) \ge K^2\big)$, which is close to 1 once the array holds comfortably more than $K^2/p$ traps.
With rearrangement, a slower wall
Each move risks the atom being carried, and the whole sequence takes time while every atom can be knocked out by background gas. If a shot needs $n$ moves of duration $t$, then roughly $$P_{\text{defect-free}} \approx (1-\varepsilon)^{\,n}\; e^{-K^2 n t/\tau}\;\times(\text{imaging terms}),$$ and $n$ grows with $K^2$. The decay is still exponential in the number of atoms, just with a far smaller exponent. Try it: raise $K$ with 1% loss per move and watch the Monte Carlo fall away from the “enough atoms” curve.
Beyond one shot: keep the array full
The newest machines stop treating each shot as a fresh start. Norcia et al. filled 1225-site ¹⁷¹Yb arrays incrementally from a repeatedly refilled reservoir, reaching 99% per-site occupancy (PRX Quantum 5, 030316, 2024). Chiu et al. kept an array of over 3,000 atoms running for more than two hours by reloading 300,000 atoms per second into tweezers (Nature 646, 1075, 2025). Manetsch et al. trapped 6100 Cs atoms in one array (Nature, 2025). This lab models the single-shot cycle those systems are built from.
05 · WHAT THE MODEL DOES
Assumptions
- Loading: each tweezer holds one atom with probability $p$, independently; never two.
- Imaging: a site's count is Poisson with mean $N_{bg}$ (empty) or $N_{bg} + N_{at}$ (occupied). The threshold is the integer that minimises the sum of the false-positive and false-negative rates. An occupied site survives the image with the set probability.
- Assignment: empty target sites, taken from the centre outward, each claim the nearest unused atom detected outside the block. Atoms already detected inside the block stay. Paths that cross other traps are allowed.
- Moves: a move from a false-positive or already-lost site delivers nothing. A real atom is lost with probability ε. If it lands on a site that held an undetected atom, light-assisted collisions eject both.
- Vacuum: after the moves, every atom survives with $e^{-T/\tau}$, where $T$ = moves × time per move.
- Defect-free is judged from the true occupancy. The verification image is drawn for show; its own errors are not counted.
- Bell pairs: neighbouring columns in the block are paired. A pair with both atoms is a good Bell pair with probability F, giving equal Z-basis bits; otherwise its bits are random. The expected agreement is $F + (1-F)/2$. Agreement in one basis is necessary but does not prove the fidelity.