Fig. 6: Large-scale high-fidelity coherent transfer between static and dynamic traps.

a, Layout of the transfer experiment showing 195 dynamic AOD traps (bright blue) overlapped with 1,061-nm static SLM traps (pale blue). Atoms are repeatedly picked up and moved away by 2.4 μm, then held for 100 μs. During this time, the SLM traps are turned off to ensure that atoms left behind in SLM traps are removed (this way, atom survival correctly corresponds to a successful pick-up and drop-off). SLM traps are subsequently turned back on and atoms held in AOD tweezers are moved back and dropped off into the SLM traps. For IRB data shown in d, gates are interleaved between each round-trip transfer. A pick-up and split-move operation (or equivalently a merge-move and drop-off operation) is considered a ‘one-way transfer’. b, Best hand-optimized trajectory for trap transfer (Methods), using a quadratic depth profile and a constant jerk movement. Here we implement the pick-up and the tweezer separation move in sequence, without overlap. c, To speed up atom transfer between static and dynamic traps while preserving high survival, we optimize, through machine learning, a trajectory in which dynamic AOD traps are simultaneously ramped and moved. The dashed lines and black dots represent the values that are optimized by the algorithm. d, Top, atom survival as a function of the number of repeated one-way transfers for various one-way ‘pick-up and split' total durations. A 400-μs trajectory is optimized through machine learning. Middle, return probability after IRB for the machine-learning-optimized trajectory. Bottom, extracted instantaneous fidelity of a coherent one-way transfer as a function of the number of previous one-way transfers.