New Advances Bring the Era of Quantum Computers Closer Than Ever
Kristina Armitage/ Quanta Magazine
克里斯蒂娜·阿米蒂奇/广达杂志
Some 30 years ago, the mathematician Peter Shor took a niche physics project — the dream of building a computer based on the counterintuitive rules of quantum mechanics — and shook the world.
大约 30 年前，数学家彼得·肖尔 (Peter Shor) 开展了一个小众物理项目——基于违反直觉的量子力学规则建造一台计算机的梦想——并震惊了世界。
Shor worked out a way for quantum computers to swiftly solve a couple of math problems that classical computers could complete only after many billions of years.
肖尔找到了一种让量子计算机能够快速解决经典计算机需要数十亿年才能完成的数学问题的方法。
Those two math problems happened to be the ones that secured the then-emerging digital world.
这两个数学问题恰好是保障当时新兴数字世界安全的问题。
The trustworthiness of nearly every website, inbox, and bank account rests on the assumption that these two problems are impossible to solve.
几乎每个网站、收件箱和银行账户的可信度都建立在这两个问题无法解决的假设之上。
Shor’s algorithm proved that assumption wrong.
肖尔的算法证明这个假设是错误的。
For 30 years, Shor’s algorithm has been a security threat in theory only.
30 年来，Shor 的算法仅在理论上构成了安全威胁。
Physicists initially estimated that they would need a colossal quantum machine with billions of qubits — the elements used in quantum calculations — to run it.
物理学家最初估计他们需要一台拥有数十亿量子位（量子计算中使用的元素）的巨大量子机器来运行它。
That estimate has come down drastically over the years, falling recently to a million qubits.
多年来，这一估计值已大幅下降，最近降至一百万个量子比特。
But it has still always sat comfortably beyond the modest capabilities of existing quantum computers, which typically have just hundreds of qubits.
但它仍然远远超出了现有量子计算机的适度能力，而现有量子计算机通常只有数百个量子位。
However, two different groups of researchers have just announced advances that notably reduce the gap between theoretical estimates and real machines.
然而，两个不同的研究小组刚刚宣布了一些进展，这些进展显着缩小了理论估计与实际机器之间的差距。
A star-studded team of quantum physicists at the California Institute of Technology went public with a design for a quantum computer that could break encryption with only tens of thousands of qubits and said that it had formed a company to build the machine.
加州理工学院的一个众星云集的量子物理学家团队公开了一种量子计算机的设计，该计算机只需数万个量子比特就可以破解加密，并表示已经成立了一家公司来建造这台机器。
And researchers at Google announced that they had developed an implementation of Shor’s algorithm that is ten times as efficient as the best previous method.
谷歌的研究人员宣布，他们已经开发出了 Shor 算法的实现，其效率是之前最佳方法的十倍。
Neither company has the hardware to break encryption today.
如今，两家公司都没有破解加密的硬件。
But the results underscore what some quantum physicists had already come to suspect: that powerful quantum computers may be years away, rather than decades.
但结果强调了一些量子物理学家已经开始怀疑的事情：强大的量子计算机可能还需要几年时间，而不是几十年。
“If you care about privacy or you have secrets, then you better start looking for alternatives,” said Nikolas Breuckmann , a mathematical physicist at the University of Bristol, who did not work on either of the papers.
“如果你关心隐私或有秘密，那么你最好开始寻找替代方案，”布里斯托大学数学物理学家尼古拉斯·布吕克曼 (Nikolas Breuckmann) 说，他没有参与这两篇论文的撰写。
While the new results may provide a jolt for the policymakers and corporations that guard our digital infrastructure, they also signal the rapid progress that physicists have made toward building machines that will let them more thoroughly explore the quantum world.
虽然新的结果可能会给保护我们数字基础设施的政策制定者和企业带来震动，但它们也标志着物理学家在构建机器方面取得的快速进展，这将使他们能够更彻底地探索量子世界。
“We’re going to actually do this,” said Dolev Bluvstein , a Caltech physicist and CEO of the new company, Oratomic.
“我们将真正做到这一点，”加州理工学院物理学家兼新公司 Oratomic 的首席执行官 Dolev Bluvstein 说。
Bluvstein and his collaborator Madelyn Cain came to Caltech last summer with a simple question: What is the smallest quantum computer one could imagine building that would have the juice to, say, hack a Bitcoin wallet?
Bluvstein 和他的合作者 Madelyn Cain 去年夏天带着一个简单的问题来到加州理工学院：人们可以想象建造的最小的量子计算机是什么，它有能力破解比特币钱包？
Finding the answer would require them and their new colleagues to project where two major trends in quantum computing might collide.
找到答案需要他们和他们的新同事预测量子计算的两个主要趋势可能发生冲突的地方。
Madelyn Cain (left) and Dolev Bluvstein asked themselves: What is the smallest quantum computer one could imagine creating that could hack something like a Bitcoin wallet?
Madelyn Cain（左）和 Dolev Bluvstein 问自己：人们可以想象创造出能够破解比特币钱包之类的东西的最小量子计算机是什么？
The first trend was the rise of a flexible new type of qubit: the neutral atom .
第一个趋势是灵活的新型量子位的兴起：中性原子。
Over the last decade, physicists have refined their ability to suspend dozens, hundreds and, more recently, thousands of neutral atoms in laser beams and arrange them as they like.
在过去的十年中，物理学家已经改进了他们在激光束中悬浮数十个、数百个以及最近数千个中性原子并按自己喜欢的方式排列它们的能力。
Other qubits, such as the superconducting circuits championed by Google and IBM, operate much more quickly but sit locked in place, like traditional transistors.
其他量子位，例如谷歌和 IBM 倡导的超导电路，运行速度要快得多，但像传统晶体管一样被锁定在适当的位置。
Bluvstein and Cain had been working in the lab of the Harvard physicist Mikhail Lukin , where they arranged for 280 neutral atoms to run sophisticated quantum algorithms in 2023.
Bluvstein 和 Cain 一直在哈佛大学物理学家 Mikhail Lukin 的实验室工作，他们在那里安排了 280 个中性原子，在 2023 年运行复杂的量子算法。
Soon afterward, a group led by Manuel Endres at Caltech set a record when it demonstrated the ability to manipulate 6,100 neutral atoms at once , although it did not perform any calculations with them.
不久之后，加州理工学院的曼努埃尔·恩德雷斯 (Manuel Endres) 领导的一个小组创造了一项记录，他们展示了同时操纵 6,100 个中性原子的能力，尽管它没有对它们进行任何计算。
The second quantum trend was an increase in the potency of error correcting codes.
第二个量子趋势是纠错码效力的提高。
Qubits of any sort are extremely error-prone, and computing with them will require constant vigilance.
任何类型的量子位都非常容易出错，使用它们进行计算需要时刻保持警惕。
The gold-standard protocol for error correction is called the surface code .
纠错的黄金标准协议称为表面代码。
You lay qubits out in a rectangular grid, with each one linked to its neighbor, and use the whole block to store one virtual qubit of information.
您将量子位放在一个矩形网格中，每个量子位都与其相邻的量子位相连，并使用整个块来存储一个虚拟量子位信息。
Then, when some of the real qubits go haywire, the virtual qubit stays protected long enough for you to find and fix the rogue qubits.
然后，当一些真实的量子位失控时，虚拟量子位会受到足够长的保护，以便您找到并修复恶意量子位。
The surface code is completely reliable and thoroughly understood, but it would take thousands of real qubits to create one reliable virtual qubit.
表面代码是完全可靠且透彻理解的，但需要数千个真实量子位才能创建一个可靠的虚拟量子位。
And the virtual qubits are the ones you need to perform an accurate calculation.
虚拟量子位是执行精确计算所需的量子位。
Mikhail Lukin’s lab arranged for 280 neutral atoms to run sophisticated quantum algorithms in 2023.
Mikhail Lukin 的实验室在 2023 年安排了 280 个中性原子来运行复杂的量子算法。
From left: Simon Evered, Sophie Li, Alexandra Geim, Mikhail Lukin, Dolev Bluvstein and Markus Greiner.
左起：Simon Evered、Sophie Li、Alexandra Geim、Mikhail Lukin、Dolev Bluvstein 和 Markus Greiner。
But over the last few years, physicists have found a way to dramatically reduce the number of real qubits they need to create the virtual ones, using quantum “low-density parity-check” (qLDPC) codes .
但在过去的几年里，物理学家找到了一种方法，可以使用量子“低密度奇偶校验”（qLDPC）码来大幅减少创建虚拟量子位所需的真实量子位数量。
The codes are tricky because they require linking up the real qubits to other real qubits that are far away in the array, as opposed to just their neighbors.
这些代码很棘手，因为它们需要将真实量子位连接到数组中远离的其他真实量子位，而不仅仅是它们的邻居。
But in return, they let you pack many more virtual qubits into an array of a given size.
但作为回报，它们允许您将更多的虚拟量子位打包到给定大小的数组中。
Neutral atoms are a natural fit for these codes, because physicists can freely move one atom across an array to meet a distant atom.
中性原子非常适合这些代码，因为物理学家可以自由地在阵列中移动一个原子以与远处的原子相遇。
Bluvstein and Cain’s question about the simplest code-cracking quantum computer became a challenge: How far could the Caltech physicists tailor qLPDC codes to the neutral atom technology?
Bluvstein 和 Cain 关于最简单的代码破解量子计算机的问题变成了一个挑战：加州理工学院的物理学家能够在多大程度上将 qLPDC 代码定制到中性原子技术？
They started working with Qian Xu , an expert in those codes; Robert Huang , an expert in quantum theory and machine learning; and Endres, for experimental feedback.
他们开始与这些代码方面的专家钱旭合作； Robert Huang，量子理论和机器学习专家；和 Endres，获取实验反馈。
John Preskill , a senior theoretical physicist at the university with a long history in the field of quantum error correction, advised the group.
该小组的顾问是该大学的高级理论物理学家约翰·普雷斯基尔（John Preskill），他在量子纠错领域拥有悠久的历史。
These funky new qLDPC codes come in many forms, and choosing which one to use typically involves a tradeoff.
这些时髦的新 qLDPC 码有多种形式，选择使用哪一种通常需要权衡。
Some are efficient, in that each virtual qubit doesn’t require many real qubits; others are effective, in that they can withstand many simultaneous errors.
有些是高效的，因为每个虚拟量子位不需要很多真实的量子位；其他的则有效，因为它们可以承受许多同时发生的错误。
But small tweaks can lead to big changes in performance.
但小的调整可能会导致性能的巨大变化。
Breuckmann, who has done pioneering work on qLDPC codes, likens it to cooking: Sometimes a pinch of just the right ingredient can go a long way.
Breuckmann 在 qLDPC 码方面做出了开创性的工作，他将其比作烹饪：有时一小撮正确的原料就能发挥很大的作用。
The team knew that the key to a smaller, more powerful quantum computer would be finding a code that balanced both virtues.
该团队知道，制造更小、更强大的量子计算机的关键是找到一种能够平衡这两种优点的代码。
Xu identified a particularly promising recipe, and Huang set off to perfect it.
徐发现了一个特别有前途的配方，黄开始完善它。
Huang and his students recruited a large language model (LLM) designed by mathematicians to help.
黄和他的学生聘请了数学家设计的大型语言模型（LLM）来提供帮助。
They gave it a mathematical description of qLDPC codes and set it loose.
他们给出了 qLDPC 码的数学描述并对其进行了释放。
Eventually it came back with a code efficient enough to make one virtual qubit from just four atoms and effective enough to withstand 20 to 24 catastrophic errors.
最终，它带来了一种足够高效的代码，只需四个原子就可以生成一个虚拟量子位，并且足以承受 20 到 24 个灾难性错误。
(By contrast, an earlier high-performing qLDPC code needed 12 real qubits for each virtual qubit and could handle up to 12 catastrophic errors.) The LLM also found an efficient decoder, an algorithm for figuring out what kinds of errors have occurred and devising a plan to correct them.
（相比之下，早期的高性能 qLDPC 代码每个虚拟量子位需要 12 个真实量子位，并且最多可以处理 12 个灾难性错误。）法学硕士还找到了一种高效的解码器，一种用于找出发生了哪些类型的错误并制定纠正计划的算法。
In 2018, researchers in Paris demonstrated their ability to maneuver neutral atoms by arranging a swarm of atoms in the shape of the Eiffel Tower.
2018 年，巴黎的研究人员通过将一群原子排列成埃菲尔铁塔的形状，展示了操纵中性原子的能力。
With a superior code and decoder in hand, Cain, Xu, and Huang developed ways to pull off the intricate manipulations of the real qubits required to perform calculations, all while keeping them protected.
凭借先进的代码和解码器，Cain、Xu 和 Huang 开发出了对执行计算所需的真实量子位进行复杂操作的方法，同时保护它们。
The team assembled a string of protocols and made its best estimates for how fast it would execute them.
该团队组装了一系列协议，并对执行这些协议的速度做出了最佳估计。
Finally, the researchers simulated their design to see how well it would run Shor’s algorithm.
最后，研究人员模拟了他们的设计，看看它运行 Shor 算法的效果如何。
“We put together a lot of things,” Preskill said.
“我们整合了很多东西，”普雷斯基尔说。
“When you do it right, the answer turns out to be surprisingly encouraging.”
“当你做得正确时，答案会令人惊讶地鼓舞人心。”
The team members simulated different atomic arrays to get a sense of how fast each size could crack the two main encryption schemes, called Rivest–Shamir–Adleman (RSA) and elliptic curve cryptography (ECC).
团队成员模拟了不同的原子阵列，以了解每种尺寸破解两种主要加密方案的速度，即 Rivest-Shamir-Adleman (RSA) 和椭圆曲线加密 (ECC)。
They concluded that they would be able break the common form of RSA in about a century using 10,000 atoms.
他们得出的结论是，他们将能够在大约一个世纪内使用 10,000 个原子破解 RSA 的常见形式。
Using 100,000 atoms, however, it would take only three months.
然而，使用 100,000 个原子，只需要三个月的时间。
The team found that the easier-to-crack ECC encryption, which is also widely used, would be overcome by an array of 10,000 atoms in about three years, or to 26,000 atoms in a few days.
研究小组发现，同样广泛使用的更容易破解的 ECC 加密将在大约三年内被 10,000 个原子的阵列所克服，或者在几天内被 26,000 个原子的阵列所克服。
At the same time the Caltech team was designing its dream machine, researchers at Google led by Craig Gidney were continuing work they had done for years, coming up with ever more efficient ways of executing Shor’s algorithm.
在加州理工学院团队设计其梦想机器的同时，克雷格·吉德尼 (Craig Gidney) 领导的谷歌研究人员正在继续他们多年来所做的工作，想出更有效的方法来执行肖尔算法。
In 2019, Gidney and a collaborator detailed a quantum program that could break RSA encryption in eight hours with 20 million qubits.
2019 年，Gidney 和一位合作者详细介绍了一个量子程序，该程序可以在 8 小时内使用 2000 万个量子位破解 RSA 加密。
Last year, he came up with a way of doing it with fewer than a million qubits .
去年，他想出了一种用不到一百万个量子比特来做到这一点的方法。
In Mikhail Lukin’s lab at Havard, physicists manipulate qubits made of atoms with finely controlled laser beams.
在米哈伊尔·卢金 (Mikhail Lukin) 哈佛大学的实验室中，物理学家用精细控制的激光束操纵由原子组成的量子位。
In a white paper posted the same day as the Caltech paper, Gidney and his collaborators announced that they had developed a new quantum procedure specifically for breaking ECC that was at least 10 times as efficient than previous procedures.
在与加州理工学院论文同一天发布的一份白皮书中，吉德尼和他的合作者宣布，他们开发了一种专门用于破解 ECC 的新量子程序，其效率至少是以前程序的 10 倍。
They estimated that most cryptocurrencies would yield in minutes to a machine with fewer than 500,000 qubits.
他们估计，大多数加密货币将在几分钟内生成少于 500,000 个量子位的机器。
“That tenfold reduction in the actual space-time cost of elliptic curve code breaking is hugely significant,” said Jeff Thompson , a physicist at Princeton University and CEO of the neutral atoms startup Logiqal.
普林斯顿大学物理学家、中性原子初创公司 Logiqal 的首席执行官杰夫·汤普森 (Jeff Thompson) 表示：“椭圆曲线密码破译的实际时空成本降低了十倍，意义重大。”
Google’s efficient implementation of Shor’s algorithm and Caltech’s new protocol suggest that smaller quantum computers will be able to pull off bigger feats than many researchers had realized.
谷歌对肖尔算法和加州理工学院新协议的有效实施表明，较小的量子计算机将能够实现比许多研究人员意识到的更大的成就。
They also mark a turning point at which researchers are beginning to conceal crucial details that competitors or bad actors might find useful.
它们还标志着一个转折点，研究人员开始隐瞒竞争对手或不良行为者可能认为有用的关键细节。
For the first time, Google described their work using a “ zero knowledge proof ,” a technique for revealing that a program works without revealing exactly how it works.
谷歌首次使用“零知识证明”来描述他们的工作，这是一种揭示程序工作原理但不揭示其具体工作原理的技术。
Robert Huang used a large language model to create a qLDPC code efficient enough to make one virtual qubit from only four atoms.
Robert Huang 使用大型语言模型创建了一种 qLDPC 代码，其效率足以仅由四个原子生成一个虚拟量子位。
Given the rapid quantum progress, physicists say that switching out RSA and ECC for new cryptographic schemes that quantum computers can’t break is essential.
鉴于量子技术的快速进步，物理学家表示，将 RSA 和 ECC 替换为量子计算机无法破解的新加密方案至关重要。
In 2024, the National Institute of Standards and Technology published new codes that can keep secrets safe from both classical and quantum computers.
2024 年，美国国家标准与技术研究所发布了新代码，可以保护经典计算机和量子计算机的秘密。
And the U.S. government has laid out a plan to completely switch to these new codes by 2035.
美国政府已制定计划，到 2035 年完全改用这些新代码。
But some researchers believe that key players may need to act more quickly.
但一些研究人员认为，关键参与者可能需要更快地采取行动。
Google, for instance, recently announced that it aims to stop relying on RSA and ECC by 2029 .
例如，谷歌最近宣布，其目标是到 2029 年停止依赖 RSA 和 ECC。
“If you were thinking about when you were going to do a post-quantum crypto transition, you should not be waiting any longer,” Thompson said.
汤普森说：“如果你正在考虑何时进行后量子加密货币转型，那么你不应该再等待了。”
“This is the time to do it.”
“现在是时候这样做了。”
Opinions range on how plausible it will be for Oratomic to build a quantum computer as formidable as the one the physicists have described on paper.
对于 Oratomic 建造一台像物理学家在纸上描述的那样强大的量子计算机的可能性有多大，人们的意见不一。
To one leader of neutral atom computing, the Caltech team’s projections are not particularly surprising.
对于中性原子计算领域的一位领导者来说，加州理工学院团队的预测并不特别令人惊讶。
“They are broadly in line with what we and others have estimated,” said Lukin of Harvard, a founder of the neutral atom startup QuEra Computing.
“它们与我们和其他人的估计基本一致，”中性原子初创公司 QuEra Computers 的创始人、哈佛大学的卢金 (Lukin) 说。
“But in these resource estimates details matter and it is important to work them out carefully.”
“但在这些资源估算中，细节很重要，仔细计算也很重要。”
“We just have to build these machines and see if they work,” said John Preskill.
“我们只需建造这些机器，看看它们是否有效，”约翰·普雷斯基尔说。
And a few key details remain vague — notably error correction steps crucial to the Caltech team’s rosiest projections — making it hard for external researchers to fully evaluate their claims.
一些关键细节仍然模糊——尤其是对加州理工学院团队最乐观的预测至关重要的纠错步骤——这使得外部研究人员很难充分评估他们的说法。
Other researchers question some of the team’s mechanical expectations.
其他研究人员质疑该团队的一些机械期望。
For example, the Caltech group made “aggressive assumptions about the speed of operations they can do,” Thompson said.
例如，加州理工学院的团队“对他们可以执行的操作速度做出了激进的假设，”汤普森说。
The group claims in its paper that the machine will eventually be able pull off the entire error correction process — check for errors, interpret what it finds, fix the errors, replace any atoms that have gone astray, and prepare to do it all over again — once every millisecond.
该小组在其论文中声称，该机器最终将能够完成整个纠错过程——检查错误、解释发现的内容、修复错误、替换任何误入歧途的原子，并准备重新开始——每毫秒一次。
The machine would also have to keep up that cadence of error correction for days or even weeks while a computation runs, a feat no group has accomplished.
机器还必须在计算运行时保持几天甚至几周的纠错节奏，这是任何团队都没有完成的壮举。
“I’d like to see a demonstration on a smaller scale, say, 100 or 1,000 qubits,” said Mark Saffman , a physicist at the University of Wisconsin-Madison and chief scientist for quantum information at Infleqtion, another neutral atom startup.
威斯康星大学麦迪逊分校的物理学家、另一家中性原子初创公司 Infleqtion 的量子信息首席科学家马克·萨夫曼 (Mark Saffman) 表示：“我希望看到更小规模的演示，比如 100 或 1,000 个量子比特。”
“Show me that you can do a million rounds or something.”
“让我看看你可以进行一百万轮之类的。”
The Caltech team knows its plan is ambitious and that integrating all the parts it has in mind will require a tremendous engineering and technological effort.
加州理工学院团队知道其计划雄心勃勃，整合其想到的所有部分将需要巨大的工程和技术努力。
At the same time, the physicists don’t see any insurmountable obstacles.
与此同时，物理学家并没有看到任何无法克服的障碍。
“We just have to build these machines and see if they work,” Preskill said.
“我们只需建造这些机器，看看它们是否有效，”普雷斯基尔说。
If any group succeeds at building a quantum computer that can realize Shor’s algorithm, it will mark the end an era — specifically, the “ Noisy Intermediate Scale Quantum ” era, as Preskill dubbed the pre-error-correction period in a 2018 paper.
如果任何一个团队成功构建出能够实现 Shor 算法的量子计算机，这将标志着一个时代的结束——具体来说，就是“嘈杂的中尺度量子”时代，普雷斯基尔在 2018 年的一篇论文中将其称为“前纠错期”。
Each researcher has a vision for what to pursue first with a machine in the new “fault-tolerant” era.
每个研究人员对于新的“容错”时代的机器首先要追求什么都有一个愿景。
Huang said he would start by running Shor’s algorithm, just to prove that the device works.
黄说，他将从运行 Shor 的算法开始，只是为了证明该设备有效。
After that, he said he would try to use it to speed up machine learning — an application to be detailed in coming work.
之后，他表示他将尝试使用它来加速机器学习——该应用程序将在接下来的工作中详细介绍。
Most of the architects building quantum computers, whether at Oratomic or other startups, are physicists at heart.
大多数构建量子计算机的架构师，无论是在 Oratomic 还是其他初创公司，本质上都是物理学家。
They’re interested in physics, not cryptography.
他们对物理学感兴趣，而不是密码学。
Specifically, they’re interested in all the things a computer fluent in the language of quantum mechanics could teach them about the quantum realm, such as what sort of materials might become superconductors even at warm temperatures.
具体来说，他们对精通量子力学语言的计算机可以教给他们有关量子领域的所有知识感兴趣，例如哪种材料即使在温暖的温度下也可能成为超导体。
Preskill, for his part, would like to simulate the quantum nature of space-time.
普雷斯基尔则希望模拟时空的量子本质。
The Caltech group knows it has years of work ahead before any of its dreams have a chance of coming true.
加州理工学院的团队知道，在其梦想有机会实现之前，还需要多年的努力。
But the researchers can’t wait to get started.
但研究人员迫不及待地想开始。
“Pick a cooler life quest than building the world’s first quantum computer with your friends!” said a jubilant Bluvstein, reached by phone shortly before their paper went live, before rushing off to celebrate.
“选择一个比与朋友一起建造世界上第一台量子计算机更酷的人生追求！”在他们的报纸上线前不久，布鲁夫斯坦通过电话联系到了他，他兴高采烈地说道，然后就冲出去庆祝了。
Get highlights of the most important news delivered to your email inbox
获取发送到您的电子邮件收件箱的最重要新闻的摘要
Quanta Magazine moderates comments to facilitate an informed, substantive, civil conversation.
《Quanta》杂志对评论进行审核，以促进知情、实质性、文明的对话。
Abusive, profane, self-promotional, misleading, incoherent or off-topic comments will be rejected.
辱骂、亵渎、自我推销、误导、语无伦次或偏离主题的评论将被拒绝。
Moderators are staffed during regular business hours (New York time) and can only accept comments written in English.
版主在正常工作时间（纽约时间）有工作人员，并且只能接受英文评论。