Surface code memory, distance 3, on C2LR (Yb-Ba) on QCCD — every design, ranked

The shuttle memory board's experiment, unchanged: three rounds of the distance-3 rotated surface code's syndrome extraction (eight ancillas, reset and reused every round), then the data read out. Here it runs on a C2LR swap-or-stay lattice of Yb-Ba crystals (arXiv:2403.00756): one crystal is one qubit, crystals move only by swap-or-stay steps (one shared junction waveform, one direction bit per leg), and a two-qubit gate runs on the two crystals of one leg, merged and split again. Judged by the C2LR rule set; ranked by the logical error rate per round under the noise model built from the machine's physics: the shuttle board's, plus the paper's step (400 us; 0.18 quanta to a crystal that moves, 0.165 to one that stays; reorder-or-not 93 us) and the platform's own merge and split. Every design ran the same experiment, compiled by the baseline C2LR router; its hardware program was turned into a noisy stim circuit by the noise model qccd-noise@1 (a text, on the Evaluation page) read over the board's own physics, which every design is entered with, and sampled (up to 200,000 shots or 200 logical errors, decoder pymatching, fixed seed). Ranked by the logical error rate per round, lower is better; click a dot to watch the ions and read the stim program that was sampled. Each dot is a design; pick what to rank by, click a dot for its numbers and the steps in the design tool where each claim can be checked.