Shulou(Shulou.com)11/24 Report--

After achieving "quantum hegemony" three years ago, Google today announced the second milestone in reducing the calculation error rate by adding qubits for the first time.

In 2019, Google declared quantum hegemony for the first time, setting the first milestone.

Three years later, the company announced that it had reached its second key milestone (M2) on the road to building large quantum computers.

That is, for the first time in history, the calculation error rate is reduced by adding qubits!

According to the official blog, quantum error correction (QEC) encodes information through multiple physical qubits, or "logical qubits".

This method is considered to be the only way for large quantum computers to reduce the error rate.

The latest research results have been published in the journal Nature.

Paper link: https://www.nature.com/ articles / s41586,022,022-05434-1, if nothing else, just look at the number of authors.

More than 150 scientists participated in this study.

Physical qubits to logical qubits in 2020, Google released a roadmap for quantum computing with six key milestones.

Quantum hegemony comes first, and the latest achievements represent M2.

The final milestone, M6, is the realization of a quantum computer consisting of 1 million physical qubits, encoding 1000 logical qubits, by which time the value of commercial applications of quantum computers can be realized.

Why do you want to correct mistakes? To be clear, all computers can go wrong.

Error correction is inevitable if quantum computers are to be able to deal with problems that ordinary computers cannot solve, such as decomposing large integers into primes.

For an ordinary computer, the chip stores information in the form of bits (which can represent 0 or 1) and copies some information into redundant error correction bits.

When an error occurs, the chip can automatically find the problem and fix it.

However, in quantum computing, this cannot be done. Qubit is the basic unit of quantum information, and qubit is the quantum superposition of 0 and 1.

If the complete quantum state of a qubit is irrevocably lost, the information cannot be read, which means that its information cannot be simply copied to the redundant qubits.

Now, the Google Quantum team has found a new quantum error correction solution:

That is, qubits that encode information in a group of physical quanta, rather than in a single quantum, are called "logical qubits".

Quantum computers can use some physical qubits to check the status of logical qubits and correct errors. The more physical qubits, the more errors can be reduced.

In addition, the advantage of using multiple qubits for quantum error correction is that it can be continuously expanded (Sacling). Of course, things go to extremes, and adding more qubits will also lead to the chance that two of them will be affected by errors at the same time.

To solve this problem, Google researchers improved the qubits of the quantum chip Sycamore and studied two logical qubits of different sizes.

One is made up of 17 qubits that can be corrected from one error at a time, and the other is made up of 49 qubits that can be corrected from two simultaneous errors.

The experimental results show that its performance is better than that of 17 qubits.

Surface code logic qubit error correction how does the Google team achieve this result?

To take a simple example of classic communication: Bob wants to send a bit read as "1" to Alice over a noisy communication channel. He realized that if the bit was flipped to "0", the message would be lost, so he sent three bits "111" instead.

If a person flips incorrectly, Alice can make a majority vote on all received bits (a simple error correction code) and still understand the expected message.

If the information is repeated more than three times, that is, increasing the "size" of the code, the code will be able to correct more individual errors.

The surface code adopts this principle and envisions a practical quantum implementation. It must satisfy two additional constraints.

First, the surface code must be able to correct not only bit flipping (taking one qubit from 0 to 1), but also phase flipping. This error is unique to quantum states and converts qubits into superposition states, for example, from 0 to 0-1.

Secondly, checking the state of qubits will destroy their superposition states, so a method is needed to detect errors without directly measuring the states.

In order to break through these limitations, we arrange two types of qubits on the chessboard.

The "data" qubits on the vertices form logical qubits, while the "measurement" qubits in the center of each square are used for so-called stabilizer measurements.

These measurements tell us whether these qubits are exactly the same / different, indicating that an error has occurred, but in fact do not reveal the values of each data qubit.

Two types of stabilizers are measured in checkerboard mode to protect logical data from the effects of bit flipping and phase flipping.

If some stabilizer measurements record errors, the correlation in the stabilizer measurements is used to identify which errors occurred and where.

For example, in the above example, the message from Bob to Alice becomes more powerful as the code size increases, and a larger surface code can better protect the logical information it contains.

The surface code can withstand a certain number of bit and phase reversal errors, each of which is less than half of the distance, where the distance is the number of data qubits across the surface code in any dimension.

The problem is that every physical qubit is error-prone, so the more qubits are encoded, the greater the probability of error.

For this reason, the error of physical qubits must be lower than the so-called "fault tolerance threshold". For surface codes, this threshold is quite low.

The latest experiment proves this point.

The experiment runs on Google's most advanced third-generation Sycamore processor architecture, optimizes QEC and uses a fully improved surface code.

To this end, the researchers made seven major improvements to all parts of its quantum computer, including the quality of qubits, control software, and cryogenic devices used to cool the computer to near absolute zero.

The researchers used experiments to compare the logic error rate between distance-3 surface codes based on 17 physical qubits (ε 3) and distance-5 surface codes based on 49 physical qubits (ε 5).

The experimental results show that the larger surface code can achieve better logic qubit performance (2.914% logic error per cycle), which is better than the smaller surface code (3.028% logic error per cycle).

Google said that while this may seem like a small improvement, it has to stress that the result is the first in the field since Peter Shor's 1995 QEC proposal.

The advantage of larger coding over smaller coding is a key feature of QEC. All quantum computing architectures need to overcome this barrier in order to reduce the low error rate of quantum applications.

The above results show that we are entering a practical new era of QEC.

Over the past few years, Google's Quantum AI team has been thinking: how to define success in this new era and how to measure progress along the way?

Their ultimate goal is to demonstrate a low-error approach to using quantum computers in meaningful applications.

Therefore, the goal of experts is still to achieve a logic error rate of 1 / 10 ^ 6 or less in each QEC cycle.

Left: expected progress after improving the performance (quantified by 𝚲) and scale (quantified by code distance) of the superficial code. Right: the distance between the logic error rate of each cycle measured by the experiment and the one-dimensional repetition code and two-dimensional surface code.

In the picture on the left, experts outline the path to achieve this goal.

As they continue to improve the performance of physical qubits (and logical qubits), they hope to gradually increase 𝚲 from close to 1 to a larger number.

The figure shows that when 𝚲 = 4 and the code distance is 7 (that is, 577 physical qubits of good enough quality), a logic error rate of less than 1 / 10 ^ 6 will be generated.

Although it will take several years to achieve this result, it is a gratifying step forward that today's hardware can detect such a low error rate.

Although two-dimensional surface code allows experts to correct bit and phase flipping errors, they can also build one-dimensional repetitive code, which can only solve one type of error and is not strict.

In the figure on the right, you can see that the repetitive code with a distance of 25 can achieve an error rate of nearly 1 / 10 ^ 6 per cycle.

With such a low error rate, we can see a new type of error mechanism that can not be observed by the superficial code. By controlling these error mechanisms, the error rate of repeated code can be increased to nearly 1 / 10 ^ 7.

To reach this milestone, the whole team has been focused for three years.

After that, the team expects to introduce a fault-tolerant mechanism to suppress logic errors exponentially and unlock the first useful error correction quantum application.

And the team will continue to explore problems that quantum computers can solve, including condensed matter physics, chemistry, material science and machine learning.

Reference:

Https://www.nature.com/articles/d41586-023-00536-w

Https://www.nature.com/articles/s41586-022-05434-1

Https://ai.googleblog.com/2023/02/suppressing-quantum-errors-by-scaling.html

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.