A QUANTUM CAT THAT STAYS PUT FOR MINUTES

A quantum computer will only be useful if it can correct its own errors. That costs a great deal of hardware — unless the errors are lopsided. If one kind of error almost never happens, the correcting code only has to fight the other kind, and it can be much simpler.

So-called cat qubits are built to be lopsided. They store information in a superconducting circuit that holds a microwave field, as two opposite oscillations of that field. Turning a 0 into a 1 — a bit flip — means jumping from one oscillation to the other. The more photons in the field, the further apart the two states are: bit flips become exponentially rarer with photon number, while the other error, the phase flip — which scrambles the quantum relationship between the two states instead of swapping them — only grows in proportion to the photon number.

Two valleys, one weak spot

There are two ways to keep a cat in place. “Dissipative” cats are constantly pushed back towards their two states by a lossy partner; they already hold for more than 100 seconds. “Kerr” cats instead sit in an energy landscape with two valleys separated by a barrier, created by feeding a microwave tone (a “drive”) into a nonlinear circuit. Their advantage is fast control. Their weakness: bit flips after only milliseconds — one recent improvement reached 3.6 ms, still five orders of magnitude short of dissipative cats.

The reason is that nothing removes stray energy. Noise kicks the cat into excited states inside its valley, and from there, slipping over to the other valley becomes much easier. Worse, at strong driving, the qubit usually enters resonances with its readout circuit that sharply damage the protection.

The readout circuit as a drain

Sergey Hazanov, Mikhail Svintsov and colleagues at the Weizmann Institute of Science in Israel turned this problem around. The same nonlinear interaction that digs the two valleys also links the excitations inside each valley to the readout resonator, a circuit that constantly leaks photons into the measurement line.

When the energy of an excitation matches the difference in frequency between the drive and the readout resonator — an “autoparametric” resonance — the excitation is handed over as a readout photon, which escapes. The 0 or 1 itself is left untouched. A second drain appears through the filter that sits on the output line.

The team measured it directly: on resonance, the first excited state empties as fast as the readout circuit allows, and higher states drain in a cascade. Their chip — aluminium on silicon, cooled to 10 millikelvin — needed no extra cooling drive and no extra circuit element.

Seven orders of magnitude

Away from the resonances, bit flips still came within milliseconds. Near them, the protection soared:

• 519 milliseconds on the readout resonance (22.4 photons on average);
• 15.4 seconds between the two resonances (27.3 photons);
• no measurable decay of the stored state within a 100-second measurement window on the filter resonance (33.8 photons), meaning a bit-flip time of more than 100 seconds.

By counting rare flips in long records of repeated readouts — 62 flips in 15 hours — the team estimated the bit-flip time at that last point: 448 ± 57 seconds, more than five orders of magnitude beyond previous Kerr cats.

Bit flips that fall exponentially with photon number are a defining prediction for Kerr cats, but they had never been observed. Here it was: each extra photon multiplied the bit-flip time by 1.82 ± 0.06, from 52 microseconds at 7.4 photons to 448 seconds at 33.8, about seven orders of magnitude, with no sign of levelling off.

Fast gates, gentle readout

Protection is worthless if the qubit can no longer be used. A basic rotation of the cooled qubit took 8 nanoseconds, and leaving that rotation drive on throughout did not shorten the bit-flip time (0.65 s with it, 0.64 s without). Reading the qubit out in 4 microseconds caused a flip only 5 times in 10 million readouts.

The other error

The price is the phase flip. It follows the expected limit set by the loss of single photons, about T₁/(2n̄), where T₁ is the circuit’s photon lifetime (about 8 microseconds) and n̄ the photon number. At 24.2 photons, the cat held 3.55 seconds against bit flips but only 113 nanoseconds against phase flips: a bias of about 30 million to one.

The authors point to two ways to close the gap. Photon lifetimes of 55.7 microseconds have already been demonstrated elsewhere, about seven times longer than here. And colder circuitry would leave fewer stray excitations, possibly giving the same protection with fewer photons — and therefore fewer phase flips.