More efficient exchange-only quantum gates via reinforcement learning

There has recently been rapid progress in the research of spin qubits [1], including the realization of exchange-only qubits [2,3]. Here, we use reinforcement learning to optimize the efficiency of exchange-based pulse sequences that encode the universal two-qubit gates CNOT and CZ with nearest-neighbor interaction for quantum dot arrangements in a chain and in a 2 by 3 grid. We improve on gate sequences currently known in the literature. Specifically, with our reinforcement learning framework, we manage to find a gate sequence encoding CNOT with a shorter total time than the Fong-Wandzura sequence [4] which is currently state of the art. Moreover, the flexibility of our approach makes it applicable for gate-sequence optimization for a variety of desired quantum gates and a variety of different connection topologies.

[1] Burkard, Ladd, Pan, Nichol, Petta, Rev. Mod. Phys. 95, 025003 (2023)

[2] DiVincenzo, Bacon, Kempe, Burkard, Whaley, Nature 408, 339 (2000)

[3] Weinstein et al., Nature 615, 817 (2023)

[4] Fong, Wandzura, Quantum Info. Comput. 11, 1003 (2011)