Skip To Content

A Programmable Quantum Chip, via Silicon Photonics

artist conception of waveguides on programmable chip

[Image: Xiaogang Qiang/University of Bristol]

A team led by researchers at the University of Bristol, U.K., has demonstrated a silicon-photonics-based chip that reportedly implements a fully programmable, two-qubit quantum processor (Nat. Photon., doi: 10.1038/s41566-018-0236-y). The researchers were able to use the chip, which packs in more than 200 separate photonic components, to program and run 98 different two-qubit operations, as well as an optimization algorithm and a quantum simulation.

The system—which relies on a different model of quantum information processing (QIP) than the conventional “quantum circuit” model—isn’t ready to scale to a full-fledged universal quantum computer. But the Bristol-led group suggests that its proof-of-concept device concretely demonstrates the potential of silicon as a platform for “full-scale universal quantum technologies using light.”

Problematic entanglements

From one perspective, silicon photonics seems the ideal QIP vehicle. It’s compatible with long-standing CMOS manufacturing processes honed in the classical computing business, for example, and features an increasing array of compact, reconfigurable optical components on which to draw for quantum operations. One challenge, though, lies in combining all of the elements necessary for QIP—generating photons, encoding quantum information on them, manipulating them and reading out their quantum state—on a single, programmable device.

Even more fundamental than these engineering challenges is the issue of using photons for operations that include quantum entanglement, a fundamental requirement for QIP. That’s because, in the conventional circuit model of quantum computing, each arbitrary two-qubit operation requires the equivalent of three consecutive entangling logic gates for control of the quantum system. That’s something that’s been too complex to implement practically using free-space optics or combinations of free-space and integrated photonic systems.

Changing the model

The research team, which included not only Bristol scientists but researchers in China and Australia, attacked the problem by focusing on a different QIP model. Rather than one that implements quantum processing as a multiplication of quantum logic gates in series, as under the conventional circuit model, the group used a “linear combination of quantum operators” scheme.

In this approach, an arbitrary two-qubit unitary operation is reframed as a linear combination, or weighted sum, of four easier-to-implement unitaries. Previous work at Bristol and elsewhere had suggested that the linear-combination approach could simplify control of quantum operations in a programmable QIP framework (Nat. Commun., doi: 10.1038/ncomms1392).

Even under that ostensibly simpler model, however, putting arbitrarily programmable QIP into practice calls for a formidably complex array of optical components. To achieve the required complexity, the research team turned to silicon photonics, and set their sights on a single photonic chip that could handle the quantum functions end to end.

The monolithically integrated, silicon-based device that the researchers fabricated to do the job includes four spontaneous four-wave-mixing photon-pair sources, four pump rejection filters, 58 thermo-optical phase shifters, 82 multimode interferometer beamsplitters, 18 waveguide crossers and 40 optical-grating couplers. All of these elements are crowded onto a device with an effective footprint of less than 14 mm2.

Nearly 100,000 reprogrammed settings

This small platform packs in sufficient complexity, according to the researchers, to allow them to perform end-to-end reprogrammable two-qubit operations—generating two photons; turning the photons into qubits by encoding quantum information on them; performing arbitrary unitary operations on those qubits, including entanglement; and reading out the resulting quantum state via quantum tomography. (The light source for the experiments was an externally connected, tunable laser, tied to the chip via a fiber array.)

The Bristol-led team found that it could program the device to implement 98 different unitary quantum operations, with an average quantum process fidelity of 93.2 ± 4.5 percent. The researchers also programmed the chip to realize a previously developed quantum optimization algorithm, and to simulate a specific kind of “quantum walk,” the quantum analogue to a classical mathematical random walk. Together, the experiments involved some 98,480 different reprogrammed settings.

Taking on larger tasks

The team is careful to stress that the chip—or, more precisely, the linear-combination protocol it’s based on—can’t, in its present form, scale to universal quantum computing. That’s because the protocol’s probability of success is inversely proportional to the number of terms, so that “it would achieve exponentially small success probability for a universal quantum computer.” Still, the researchers believe that, through certain optimization and scaling efforts, the platform could expand to handle families of large-scale QIP tasks “with considerable success probability.”

The researchers also see the system as an important step toward proving the mettle of silicon photonics in the drive for a universal quantum computer. “What we’ve demonstrated is a programmable machine that can do lots of different tasks,” the paper’s lead author, Xiaogang Qiang—previously a University of Bristol Ph.D. student, and now a researcher at the National University of Defence Technology, China—said in a press release accompanying the work.

“It’s a very primitive processor, because it only works on two qubits, which means there is still a long way before we can do useful computations with this technology,” Qiang continued. “But what is exciting is that it the different properties of silicon photonics that can be used for making a quantum computer have been combined together in one device. This is just too complicated to physically implement with light using previous approaches.”

Publish Date: 28 August 2018

Add a Comment