quantum face/off

Results of the first head-to-head throwdown in the nascent field of quantum computing are in, and delighting technology enthusiasts and soccer fans alike, it was a tie. In one corner, the stalwart ytterbium atom ion traps of Chris Monroe at the University of Maryland/ionQ (of which I previously blogged about) squared off against the superconducting circuits of niobium and aluminum of IBM's publicly available Quantum Experience (of which I've been meaning to blog about). The comparative preprint performed four quantum algorithms on the respective five-qubit processors and gauged their performance. The gist of the matchup comes down to the superior coherence time and connectivity of the ion trap processor compared to the speed of the superconducting version [it should probably be mentioned now that yours truly works on the superconducting one].

The weigh in: a) the IBM Quantum Experience processor consisting of five superconducting qubits and b) Chris Monroe's five qubit ion trap. Qubit connectivity is depicted in the insets. [N.M. Linke et al, arXiv:1702.01852]

arXiv:1702.01852arXiv:1702.01852

arXiv:1702.01852arXiv:1702.01851702.01852

The implementation of the algorithms, while at an assembly-level language, varies quite a lot between the two systems. IBM offers a finite number of operations, the X, Y, Z, S, T, H, and CNOT (based on ZX implemented by cross resonance) gates [IBM has expanded the gate selection and connectivity since the writing of this manuscript], while the ion trap uses arbitrary single-qubit rotations and the XX entangling gate. The decomposition of each of these into physical operations varies even further and requires another post to address, in addition to a description of the operations under test: the Margolus and Toffoli gates, and Bernstein-Vazirani and Hidden Shift algorithms. While the ion trap system bests the superconducting circuits in average probability of success, the superconducting operations are approximately a thousand times faster, meaning each experiment could be repeated to improve the likelihood of success. This comes at no surprise to me, although I am quite fascinated by the patterns of errors in the quantum state tomography.

Ion traps have been the forerunners of quantum computing since the very first experiments in the field, while superconducting circuits, who entered the race quite feebly, have vastly improved in the 20 years since their realization. Now roughly on equal footing, and with most large companies aiming for commercialization of quantum computers favoring the superconducting style, it seems ion trap groups are sweating a bit. And in order to corner the systems engineering required to build a useful quantum computer, an ion trap group is already proposing building one the size of a football [they're British, so they probably mean soccer] pitch [British for 'field']. While this contest resulted in a tie, it seems pretty clear who's on defense.