Experimental Quantum Randomness Processing Using Superconducting Qubits

Coherently manipulating multipartite quantum correlations leads to remarkable advantages in quantum information processing. A fundamental question is whether such quantum advantages persist only by exploiting multipartite correlations, such as entanglement. Recently, Dale, Jennings, and Rudolph negated the question by showing that a randomness processing, quantum Bernoulli factory, using quantum coherence, is strictly more powerful than the one with classical mechanics. In this Letter, focusing on the same scenario, we propose a theoretical protocol that is classically impossible but can be implemented solely using quantum coherence without entanglement. We demonstrate the protocol by exploiting the high-fidelity quantum state preparation and measurement with a superconducting qubit in the circuit quantum electrodynamics architecture and a nearly quantum-limited parametric amplifier. Our experiment shows the advantage of using quantum coherence of a single qubit for information processing even when multipartite correlation is not present.

Figure 1

Classical and quantum coin. For a given $p$ value, (a) classical and (b) quantum $p$-coin corresponds to two different ways of encoding $p$, see Eqs. (1) and (2), respectively. The key difference lies in whether there is coherence in the computational basis.

Figure 2

Experimental setup. (a) Optical image of a transmon qubit located in a trench, which dispersively couples to two 3D Al cavities. (b) Optical image of the single-junction transmon qubit. (c) Scanning electron microscope image of the Josephson junction. (d) Schematic of the device with the main parameters. In our experiment, the higher frequency cavity is not used and always remains in vacuum, which can be used as another $p$-quoin in future experiments [24]. Note that the highlighted boxes in (a) and (b) are not to scale and are intended for illustrative purposes only.

Figure 3

Theoretical and experimental results for the (a) $q$-coin and (b) $f(p)=4p(1-p)$-coin. Here, the number of experiment data for the $p$-quoins is on the order of ${10}^7$ and the number for the $f(p)$-coin is on the order of ${10}^6$. On average, we need about 20 $p$-quoins to construct an $f(p)$-coin. The standard deviations of $p$, $q$, and $f(p)$ are of the order of ${10}^{-4}$; thus, they are not plotted in the figure.