Ninety-six qubits, thirty quiet days

A quantum computing laboratory in the city's new coastal district announced this week that the 96-qubit superconducting processor it developed has run steadily for 30 days in a row, with no need to cool it down again and no calibration by hand. The team says the advance lies not in the number of qubits but in the steadiness: keeping a quantum computer switched on and free of errors for long stretches, as an ordinary server is, is a test it must pass before it can leave the laboratory.

The processor itself is the size of a fingernail. It sits at the bottom of a dilution refrigerator more than two metres tall; with the cover off there are six gold-plated copper plates, one above another, and a dense run of coaxial cable. Each stage is colder than the one above, and the lowest is held at minus 273.14 degrees Celsius, 0.01 of a degree above absolute zero.

Three measures

• Stability: an automatic calibration routine ran every six hours throughout the 30 days, and every qubit stayed usable.
• Precision: the average fidelity of two-qubit gates is 99.6 per cent, 0.3 percentage points better than the previous generation.
• Integration: multiplexed control lines have cut the cabling from more than 400 lines to 150.

The engineer who leads the team told a briefing that each cool-down used to be followed by two or three days of tuning by hand, so that the machine was really usable for less than half the year. "Making 96 qubits steady is harder than joining several hundred together, and more useful. Users do not care about records. They care whether a job submitted today brings back a result they can trust tomorrow."

The laboratory is frank about the limits. The processor cannot yet solve a practical problem that is beyond a classical computer: its error rate is still several orders of magnitude above what general computing requires, and error correction needs far more qubits than it has.

Next: time for local users

From next month machine time will be offered through a cloud platform to the city's universities and companies. The first 12 projects are already scheduled, among them a trial calculation with the port on the scheduling of containers and a university's project in molecular simulation. A paper has been submitted for peer review, and part of the experimental data will be released with it so that other laboratories can check the results.

The lead engineer says the team plans to pass 200 qubits by the end of next year, and to test a small error-correction scheme on this machine.