Extended Data Fig. 2: Single-qubit performance and inter-qubit crosstalk data.

a, Single-qubit performance for the P1, P2, P3 qubit with M1 readout. The ‘blind’ RB measurement³⁵ yields an average single qubit gate error of (1.1 ± 0.1) × 10⁻³ and leakage error of (3 ± 1) × 10⁻⁴. Using the methods of Blumoff et al.⁴⁶, we infer a SPAM fidelity on the M1 side from these data of 96.0% ± 0.1%. See Methods for more detailed discussion of this figure. b, Crosstalk characterization. IRB error on qubit Q₀ (using exchange energies J_X1 and J_X2) while interleaving between each Clifford either periods of idle (blue) or composite Clifford gates (orange) on qubit Q₁ (using exchange energies J_X5 and J_X4). The Q₁ composite sequences used here are X, XX, XXX, Y, YY, YYY, S^†, S^†S^† and S^†S^†S^†. These gates, respectively, take 90 ns, 180 ns, 270 ns, 120 ns, 240 ns, 360 ns, 30 ns, 60  ns and 90 ns. The x-axis is the duration of the interleaved idle period or total duration of the composite pulse sequence on qubit Q₁. The black dashed line indicates modelled RB error due to hyperfine dephasing at each idle duration in a magnetic field of 20 μT as generated by a numeric simulation. As discussed in Methods, these data serve as a validation of our approach for limiting crosstalk error, which includes strict avoidance of simultaneous exchange pulses and low susceptibility to spurious signals in the device ‘idle’ configuration. All error bars in this figure correspond to 1σ standard deviation intervals.