New Codes Could Make Quantum Computing 10 Times More Efficient | Quanta Magazine

Through these checks and more subtle tweaks of the iffy qubits you can also hide a reliable qubit throughout the square blocks data-carrying qubits, not exactly here or there but sort of everywhere. As long as the iffy qubits keep the Minesweeper operations humming along smoothly, the hidden qubit stays safe and can be manipulated to perform operations. In this way, the surface code elegantly fuses many shoddy qubits into a single qubit that rarely errs.

The slightly annoying thing for me is that the surface code is the simplest thing you can think of, said Nikolas Breuckmann, a physicist turned mathematician at the University of Bristol who has spent years trying to improve on the scheme. And it performs remarkably well.

The code became the gold standard for error correction; it was highly tolerant of misbehaving qubits, and the grid was easy to visualize. As a result, the surface code influenced the design of quantum processors and quantum road maps.

Its been the thing to do, said Barbara Terhal, a quantum information theorist at the QuTech research institute in the Netherlands. This is the chip you have to make.

The downside of the surface code, which has not yet been fully demonstrated in practice, is an insatiable appetite for qubits. Bigger blocks of shoddy qubits are needed to more strongly protect the reliable qubit. And to make multiple protected qubits, you need to stitch together multiple blocks. For researchers dreaming of running quantum algorithms on many protected qubits, these are onerous burdens.

In 2013, Gottesman saw a potential way out of this mess.

Researchers including Terhal and Bravyi had found evidence suggesting that, for a flat code that only connected neighbors to neighbors, the surface code did as well as you could hope. But what if you allowed each check to link far-flung qubits together? Quantum information theorists had already begun to explore codes featuring such nonlocal connections, which are casually called LDPC codes. (Confusingly, the surface code is technically an LDPC code too, but in practice the term often refers to the more exotic clan members with nonlocal checks.)

Gottesman then showed that certain LDPC codes could be far less ravenous: They could cram multiple protected qubits into a single block, which would help avoid the surface codes ballooning qubit requirements for larger algorithms.

But Gottesmans work was highly idealized and considered essentially infinite swarms of qubits. The practical challenge was seeing whether researchers could scale down LDPC codes to work in real quantum devices, while preserving their oomph.

Demonstrating Virtual Protection

Over the last two years, Breuckmann and other researchers have started scrutinizing the performance of LDPC codes that can run on smaller and smaller systems. The hope was that some might fit into todays devices, which can furnish perhaps 100 raw qubits.

Last week, a team of researchers at IBM led by Bravyi unveiled a simulation of the smallest and most concrete LDPC blueprint yet, based on an LDPC code from a little-known paper published in 2012. It started with the surface codes check of four neighboring qubits and added two carefully chosen nonlocal qubits.

They simulated the various errors that could ariseif the code were run on a real circuit, a process that is like sticking a digital fighter jet in a digital wind tunnel and seeing how it flies. And they found that their code could protect its reliable qubits far more efficiently than the surface code. In one test run, the code took 288 raw qubits that failed 0.1% of the time and used them to create 12 protected qubits with a failure rate 10,000 times lower. For the same task, the team estimated, the surface code would have required more than 4,000 input qubits.

We were very surprised by that, said Andrew Cross, a researcher on the IBM team.

The simulation teases the possibility of getting tomorrows error correction today, because while no one has access to 4,000 qubits, devices with hundreds of qubits are right around the corner.

You could see quite a substantial amount of fault tolerance with devices which have a number of qubits that we have today, Gottesman said.

A day after the IBMs preprint appeared, a multi-institution collaboration of researchers headed by Mikhail Lukin of Harvard University andLiang Jiangof the University of Chicago posted similar results. (The researchers declined to discuss their work, which has been submitted to a peer-reviewed journal.) They had dusted off two other LDPC codes, modified them for simulation, and found that they too required roughly one-tenth the number of input qubits to make dozens to hundreds of good qubits, when compared to the surface code.

But building an F-35 is harder than simulating an F-35, and building an LDPC code-ready device will will also be extremely challenging. Two main things could stop these things from actually taking over, Gottesman said.

First, creating nonlocal connections between qubits is tough, especially for companies like IBM that make qubits out of immobile superconducting circuits. Connecting those circuits with their neighbors is natural, but creating links between distant qubits is not.
Suivez Actusduweb sur Google News

Ce site utilise des cookies pour améliorer votre expérience. Nous supposerons que cela vous convient, mais vous pouvez vous désinscrire si vous le souhaitez. J'accepteLire la suite