带来魔力：谷歌打破量子计算的瓶颈
内容总结：
谷歌量子人工智能团队在《科学》杂志发表最新研究成果，宣布在量子计算领域取得关键突破。研究团队成功在Willow处理器上验证了“魔法态培育”技术的可行性，将量子计算中的错误率降低了约40倍，为量子计算迈向实用化迈出重要一步。
所谓“魔法态”，是实现通用量子计算不可或缺的特殊量子态，但制备高质量、经过纠错的魔法态极其消耗资源。此前有估算认为，量子计算机高达95%的算力可能都需要用于制备魔法态，真正可用于计算的部分十分有限。而且魔法态一旦使用便无法重复利用，属于稀缺资源。过去20年来，研究人员一直在努力降低这一成本。
2024年，谷歌量子人工智能团队的研究人员提出“魔法态培育”方案，思路类似于“过滤器”：将带有噪声的量子态注入量子芯片的小区域，通过一系列高度协调的测量来检测并剔除错误，从而“培育”出极为纯净、低错误的量子态，并可同步转换为更大规模的有用量子编码。然而，从理论走向实践需要大量技术创新和工程攻关。研究团队在测试中发现，实验室中由电气设备漂移和温度变化引起的特定噪声干扰了测量基准的有效性，甚至一度出现看似违背物理规律的数据。为此，团队增加了系统控制以降低噪声影响，并发明了新的基准测试方法，利用培育过程本身来测量自身。
最终，团队在Willow处理器上实现了错误率相对未培育态降低40倍的成果，T态的逻辑保真度达到99.99%的纪录水平。这是首次在实际硬件上演示解决量子计算最严重资源瓶颈之一的可行方案。研究团队还首次展示了“颜色码”与“表面码”两种量子纠错方法之间的成功切换，证明净化后的量子态能够在硬件中平稳传输。
当然，要在更大规模上实现魔法态培育、真正走向实用化量子计算，仍需克服大量技术挑战。但这一成果已将关键理论构想转化为实验现实，使谷歌量子人工智能团队向着构建有用的、具备纠错能力的量子计算机的目标更近一步。
中文翻译：
带来魔法：谷歌突破量子计算瓶颈
谷歌量子人工智能已实证性地打通了量子计算的一个关键瓶颈：我们已证明，“魔法态培育”在我们的Willow处理器上切实可行，与此前不使用该技术相比，误差降低了约40倍。这项今日发表于《科学》的突破，是量子计算迈向实用化扩展的关键一步。
“魔法态”瓶颈
早在2004年——远在量子计算成为实际工程追求之前——研究人员就已预见到一个格外棘手的瓶颈。
要实现“通用”量子计算并运行有用的程序，你需要能够执行（并纠错）一组特定的量子操作。为此，你可以先创建所谓的“魔法态”，之所以如此命名，是因为它们能帮助你生成超越经典计算所需的操作。
但问题在于：创建高质量、经过纠错的魔法态极其耗费资源。一些估算表明，量子计算机全部容量的高达95%可能都需要专门用于制造它们——留给实际计算的空间所剩无几。另一个额外挑战是，魔法态一旦使用便无法重复利用，这使得魔法态成为一种稀缺资源。
因此，20年来，研究人员一直在逐步削减这种魔法的成本，使其越来越低。
我们如何实现这一突破
2024年，谷歌量子人工智能研究人员克雷格·吉德尼、诺亚·舒蒂和科迪·琼斯从文献中汲取了一个想法，加以改进，并开发出了我们现在所称的“魔法态培育”。他们的方法有望实现特别大幅度的降低。其理念是，培育就像一个过滤器。你将一个含噪态注入量子芯片上的一个小区域，然后运行一系列高度协调的测量来检查错误并剔除有缺陷的尝试。克雷格及其团队证明，这样一来，你就能培育出——也就是“培养”出——一个异常纯净、低误差的量子态，同时还能将其适配为更大、更有用的编码。
从纸面上看，这非常出色。如果我们能让这一想法真正在我们的量子硬件上运行，就能大幅缩小实际任务所需量子计算机的规模。
而在实践中——与开发许多新技术一样——它需要一整系列技术创新和工程攻关。
仅举一例，在首次测试培育协议时，谷歌量子人工智能研究科学家艾玛·罗森菲尔德看到的数据似乎违背了物理定律。经过数月的困惑、模拟、与同事讨论以及一致性检查，她发现这源于团队常用测量技术中的一个错误假设。具体来说，实验室中由电气设备漂移和温度变化引起的一种特定噪声，干扰了基准测试的有效性。解决方案需要我们在理论和工程方法上双双转向。团队为系统增加了额外控制以降低该噪声的影响，并发明了一种新的基准测试技术，通过将培育本身用作测量自身的工具，更根本地解决了这一问题。
凭借这一成果以及许多其他算法创新和工程解决方案，我们确立了量子计量学的新前沿。在谷歌量子人工智能的Willow处理器上，我们的协议相对于未经培育的态实现了惊人的40倍误差降低。我们还获得了逻辑保真度创纪录地达到99.99%的T态。
这项现已发表于《科学》的工作，首次展示了针对量子计算最严重资源瓶颈之一的实用解决方案。它在真实硬件上证明，我们能够高效地创建这些至关重要的魔法态。
我们的发现与未来
我们还迈出了第一步，证明克雷格及其他人的理论与我们的量子计算架构相兼容。我们学会了如何在“颜色码”（一种较新的量子纠错类型，非常适合培育）和“表面码”（另一种更成熟的方法）之间成功切换，将态从距离3的颜色码嫁接到距离5的表面码上。这一嫁接协议最初在2024年的理论论文中提出，它表明在实践中，纯化后的态可以在我们的硬件内部存活并顺利转移。
当然，我们仍需在更大规模上演示魔法态培育，才能使其对实际量子计算具有现实意义，而且仍有大量技术挑战。但这一成就将一个关键理论想法转变为经实验证明的现实。它让我们离谷歌量子人工智能的目标更近一步：构建一台有用的、具备纠错能力的量子计算机，以解决其他方式无法解决的问题。
英文来源：
Bringing the magic: Google breaks open a bottleneck in quantum computing
Google Quantum AI has demonstratively opened up a key bottleneck in quantum computing: we’ve shown that “magic state cultivation” works in practice on our Willow processor, reducing errors about 40x compared to without it. This breakthrough, published today in Science, is a crucial step towards scaling up quantum computing to practical use.
The "magic state" bottleneck
Since 2004, long before quantum computing was a practical engineering pursuit, researchers have foreseen a particularly tricky bottleneck.
To achieve “universal” quantum computing and execute useful programs, you need to be able to run (and error-correct) a specific set of quantum operations. You can do this by first creating what are called "magic states," so dubbed because they can help you create the operations you need for going beyond classical computing.
But there’s a catch: creating high-quality, error-corrected magic states is incredibly resource-heavy. Some estimates have shown that up to 95% of a quantum computer’s entire capacity might need to be dedicated to making them — leaving little room for actual computing. An extra challenge is that once you use a magic state, you can’t re-use it, making magic states a scarce resource.
So for 20 years, researchers have been chipping away at this cost of magic, bringing it lower and lower.
How we made this breakthrough
In 2024, Google Quantum AI researchers Craig Gidney, Noah Shutty, and Cody Jones harnessed an idea from the literature, refined it, and developed what we now call “magic state cultivation.” Their approach promised an especially large reduction. The idea was that cultivation would act like a filter. You would inject a noisy state into a small region on a quantum chip, and then run a series of highly coordinated measurements to check for errors and remove any flawed attempts. Craig and team showed how that would let you grow, or “cultivate,” an exceptionally pure, low-error quantum state that you could simultaneously adapt into a larger, useful encoding.
On paper, it looked great. If we could make this idea actually work on our quantum hardware, it could substantially reduce the size of the quantum computer needed for practical tasks.
In practice — and on par with developing many new technologies — it required a whole series of technical innovations and engineering.
As just one example, when first testing the cultivation protocol, Google Quantum AI research scientist Emma Rosenfeld was seeing data that seemed to defy the laws of physics. After months of confusion, simulations, discussions with colleagues, and consistency checks, she found that it arose from an incorrect assumption in the team’s go-to measurement technique. Specifically, a particular kind of noise in the lab caused by electrical equipment drift and temperature variation was interfering with the benchmark’s validity. The solution required a pivot in both our theoretical and engineering approaches. The team added extra control to the system to reduce the impact of that noise, and invented a new benchmarking technique to solve the issue more fundamentally, by using cultivation as a tool to measure itself.
With this and many more algorithmic innovations and engineering solutions, we established a new state-of-the-art in quantum metrology. On Google Quantum AI's Willow processor, our protocol achieved an extraordinary 40-fold reduction in errors relative to un-cultivated states. And we got T-states with a record logical fidelity of 99.99%.
Now published in Science, this work is the first demonstration of a practical solution to one of quantum computing's most severe resource bottlenecks. It demonstrates, on real hardware, that we can efficiently create these crucial magic states.
Our findings and the future
We also took the first steps in demonstrating that Craig and others’ theories are compatible with our quantum computing architecture. We learned how to successfully switch between “color codes” (one newer type of quantum error correction, ideal for cultivation) and “surface codes” (another, more established method), grafting states from a distance-3 color code to a distance-5 surface code. Originally developed in the 2024 theory paper, this grafting protocol showed that in practice, purified states can survive and transfer smoothly inside our hardware.
Of course, we still need to demonstrate magic state cultivation at a larger scale to make it realistic for practical quantum computing, and plenty of technical challenges remain. But this achievement takes a key theoretical idea and turns it into an experimentally demonstrated reality. It brings us one step closer to our goal at Google Quantum AI: building a useful, error-corrected quantum computer to solve otherwise unsolvable problems.