During last decades, the development of large-scale fault-tolerant quantum computers has been a central aim of research in quantum information processing. The availability of such a device could fundamentally impact not only a large number of research fields but also the whole of society. A quantum device could perform quantum-enhanced algorithms to solve problems, which are practically inaccessible using a classical machine. A key challenge lies in qualities of quantum logic gates and any errors occurring during their execution.
In this work, the investigation and improvements on the quantum-gate fidelity of an ion-trap-based quantum computer are presented. This is a small-scale quantum processor based on laser-cooled ions in a macroscopic linear Paul trap. Quantum bits (qubits) are realized by the ions’ pseudo-spins using ionic hyperfine states in a static magnetic field gradient following the MAgnetic Gradient Induced Coupling (MAGIC) scheme. Quantum states of individual ions are manipulated using radio frequency (RF) radiation and addressing in frequency space.
This work reports the investigation and improvement of the Bell-state fidelity, which is an essential ingredient of quantum gates, in several aspects. (i) We have achieved sideband cooling on a single ion near the motional ground state using RF radiation with a minimum motional excitation of 0.30(12). The cooling technique is also extended to a two-ion system. This realization shows sympathetic cooling using RF sideband cooling of an ion crystal. (ii) We have outlined and explored possible sources of qubit dephasing, which are limitations of qubit control, fluctuations of magnetic fields, fluctuations of electric fields, and consequences of ions being in motionally excited states. (iii) Using dynamical decoupling sequences to protect from qubit dephasing, we have explored the limitations of this technique in numerical simulations and experiments, in which we have achieved a Bell-state fidelity of 0.95(3).
The implementation of all these improvements allowed the proof-of-principle demonstration of a quantum-enhanced learning agent to speed up the deliberation process within the reinforcement learning paradigm -- a learning scheme in the field of machine learning. The deliberation time of the quantum learning agent has been quadratically improved with respect to its classical counterpart. We have demonstrated that the algorithm takes O(epsilon^-0.57(5)) steps instead of O(epsilon^-1), where epsilon represents a probability to sample an action in the probability distribution of the learning process. This demonstration highlights the potential of a quantum computer in the fields of machine learning and artificial intelligence, which have grown into fundamental components for modern autonomous machines. Furthermore, we have discussed some prerequisite elements for a static quantum register using four ions. In addition, preliminary demonstrations of quantum state transfer have been presented.