Reinforcement learning approach for finding exchange-only gate sequences for CNOT with optimized gate time

Exchange-only quantum computation is a version of spin-based quantum computation that entirely avoids the difficulty of controlling individual spins by a magnetic field and instead functions by sequences of exchange pulses. The challenge for exchange-only quantum computation is to find short sequences that generate the required logical quantum gates. A reduction of the total gate time of such synthesized quantum gates can help to minimize the effects of decoherence and control errors during the gate operation and thus increase the total gate fidelity. We apply reinforcement learning to the optimization of exchange-gate sequences realizing the CNOT and CZ two-qubit gates which lend themselves to the construction of universal gate sets for quantum computation. We obtain a significant improvement regarding the total gate time compared to previously published results.