FPGA Control Cuts Atom Rearrangement Latency

Neutral-atom processors increasingly depend on fast classical control. Rearranging stochastically loaded optical tweezers into defect-free arrays is already a feedback problem, and mid-circuit measurement or quantum error correction will tighten the timing budget further. The paper addresses that electronics layer directly: it replaces a conventional PC-mediated control loop with an FPGA-based PXIe system that handles photon counting, atom-presence decisions, path planning, and AOD waveform triggering inside the hardware stack.

Core Contribution

The main contribution is not a new rearrangement algorithm by itself, but a hardware integration that removes the PC from the feedback path. The authors build a PXIe-QC100 electronic control system with counter and AWG cards, an ARM processor, FPGA programmable logic, optical channel mapping, fiber arrays, and SPD arrays. In the demonstrated setup, fluorescence from a one-dimensional tweezer array is mapped to detector channels, binarized or read as counts, converted into rearrangement instructions, and sent to the AWG card that drives the AOD.

The practical point is latency. Atom rearrangement can tolerate slow control when it is only a preparation step, but real-time correction during computation cannot. The authors show that their system can assemble defect-free arrays from 24 loaded tweezers and can repeat rearrangement rounds within a single MOT loading cycle, using surplus atoms as a reservoir for defects in the target zone.

Technical Approach

Figure 1 gives the clearest view of the control loop: atom fluorescence is detected through optical channel mapping and detector arrays; a counter card and ARM processor perform detection and path planning; and the AWG card drives both static tweezers and transport waveforms for rearrangement.

The latency optimizations are mostly about what does not move through the system. First, the system avoids PC communication over PCIe or USB during feedback. Second, it can transmit binarized comparison results rather than raw photon counts, although the reported experiments use real-value mode because it is easier to debug during characterization. Third, it sends compact rearrangement instructions rather than full time-domain waveforms.

The measured control path is decomposed into several sub-links. For raw count readout, the read latency scales as $t_{read}=279.01N_{trap}+809$ ns; for comparison-value readout, it is about 1393 ns for $N_{trap}\leq7$ and then scales as $15.14N_{trap}+1305.1$ ns. The path assignment step for rearranging atoms from a 24-site array into a 10-site target takes $890\pm60$ ns over 70,000 samples. Transmission from the ARM processing system to the counter-card programmable logic scales as $0.795f+4$ μs, where $f=15N_{move}+12N_{close}$, and counter-card-to-AWG transmission scales as $0.549S+0.96$ μs for the longest segment number $S$.

Results and Analysis

The headline result is a measured total feedback latency of $282\pm19$ μs over 8,000 repeated measurements. The paper’s latency table gives a representative sub-link budget of 7.51 μs for readout at $N_{trap}=24$, 0.89 μs for path calculation, about 261.58 μs for ARM-to-counter-card transmission at $f\approx324$, and 4.80 μs for counter-card-to-AWG transmission at $S\approx7$, summing to about 274.78 μs. That breakdown matters: most of the remaining delay is not photon counting or path planning, but a routing choice in the electronics architecture.

On the atom-array side, the system reaches about a 96% filling fraction after single-round rearrangement. Iterative rearrangement improves defect-free assembly within one loading cycle: after five rounds, success probabilities rise from baselines of 66.67%, 65.68%, and 55.78% to 99.01%, 95.38%, and 88.78% for target sizes $N_{target}=8$, 10, and 12. For the 10-atom target emphasized in the abstract, the reported success rate is 95.4% from 24 tweezers.

The evidence supports the narrower claim: the authors built a working low-latency control system and validated it on physical atom rearrangement. The broader claim about infrastructure for mid-circuit measurement and real-time error correction is plausible but not yet demonstrated in a full quantum-circuit setting. The authors also identify a concrete path to lower latency: moving the ARM function into a higher-performance AWG-card FPGA could remove the dominant transmission step, with projected latencies below 35 μs for 64 atoms and about 400 μs for 1000 atoms. Those are projections, not measured scaling results.

Limits and Significance

The experiment is confined to one-dimensional rearrangement from 24 tweezers, with demonstrated targets up to 12 atoms and the headline success rate reported for 10 atoms. The authors state that the remaining single-round limit is set by atomic physics, including temperature and transport-trajectory smoothness, rather than by the control electronics. That is a useful distinction: the electronics are no longer the only bottleneck, but larger arrays will still require demonstrated synchronization, transport quality, and internal-state readout under computation-like conditions.