11-19 更新 实时
首页/ AI 大模型/ 正文

IBM 发布 Nighthawk r3 量子处理器

AI 大模型 来源:MarkTechPost 2026-11-12 16:42 0 阅读 机器翻译

IBM 宣布了 IBM Quantum Nighthawk r3,这是其下一代超导量子处理器。它被设计为可在 520 个可编程量子比特上精确运行 10,000 次量子操作。2027 年,IBM 计划将 2 颗 r3 芯片连接成一个 1,040 量子比特的单一系统。对开发者而言,这一发布意味着速度。当前的 Nighthawk r2 运行工作负载的速度已经比 IBM 早期处理器快至多 19 倍。

Nighthawk 是 IBM 的方晶格处理器系列,专为量子优势工作负载打造。大多数量子比特与 4 个最近邻相连,而早期设计只有 2 或 3 个。这种连接性让电路能以更少的交换操作表达更多的纠缠结构。

当前量产芯片是 Nighthawk r2。它拥有 120 个可编程量子比特、218 个耦合器和 120 个重置元件,总计 458 个物理量子元件。

瓶颈在于量子比特重置。Heron 采用条件重置:先测量,再翻转任何处于激发态的量子比特。这种方法无法修复泄漏的量子比特,因此芯片在两次运行之间要空转数百微秒。

Nighthawk r2 用每个量子比特配备的耗散型重置器件取代了这一方法。可调耦合器按需将每个量子比特连接到一个冷环境。这使有效 T1 从约 200 微秒降至约 25 纳秒。两次运行之间的空闲时间最短降至 1 微秒。

结果是每秒可运行 100,000+ 个电路。Heron 大约能做到 4,000 个。IBM 还报告初始化误差降低约 25 倍,同时相邻量子比特不受干扰。

据 IBM 称,客户迄今已在 Nighthawk r2 上运行了 340 亿个电路。以下数据由 IBM 及其合作伙伴自行报告。

IBM 还报告称,在包含 7,500+ 个门的电路上实现了精确的可观测量估计,这是其 2026 年路线图中的一项里程碑。

Nighthawk r3 保留了 r2 的方形晶格和快速运算。它基于 2,034 个物理量子比特,扩展至 520 个可编程量子比特。它新增了 10 个名为 l-couplers 的芯片间互连。

这些连接是通往 1,040 个量子比特的路径。IBM 用 4 个 l-couplers 连接了 2 个 Nighthawk r1 芯片,每个 l-coupler 使用一根 80 cm 线缆。状态传输保真度最高达到 99.3%,平均为 99%。IBM 还连接了 2 个低温模块,在不到 5 天内降至 15 millikelvin 以下。

IBM Quantum Platform 现在可在同一个托管环境中运行 HPC 和量子作业。用户可以从笔记本电脑、超级计算机或平台本身接入经典计算资源。

IBM 和 RIKEN 重新运行了一个基于采样的量子对角化工作流,速度约提升 5 倍,计算量增加 100 倍。Cleveland Clinic 和 RIKEN 通过工作流自动化处理了 6.08 亿个样本,速度提升 33 倍。脉冲级控制也可用于纠错研究。

需要通过我们推广你的 GitHub 仓库、Hugging Face 页面、产品发布或网络研讨会等?请联系我们

文章《IBM 宣布 IBM Quantum Nighthawk r3:一款支持在 520 个可编程量子比特上执行 10k 量子操作的新处理器》首次出现在 MarkTechPost 上。

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(IBM Announces IBM Quantum Nighthawk r3: A New Processor to Support 10k Quantum Operations on 520 Programmable Qubits)

IBM has announced IBM Quantum Nighthawk r3, its next superconducting quantum processor. It is designed to run 10,000 quantum operations accurately across 520 programmable qubits. In 2027, IBM plans to link 2 r3 chips into a single 1,040-qubit system. For developers, the announcement is speed. The current Nighthawk r2 already runs workloads up to 19x faster than IBM’s earlier processors.

TL;DR

  • Size: Nighthawk r3: 520 programmable qubits, 2,034 physical qubits, 10 l-couplers. Nighthawk r2 (live today): 120 programmable qubits, 218 couplers.
  • Runs on: IBM Quantum Platform cloud (r2 as ibm_phoenix). Superconducting chips in dilution refrigerators, paired with laptop-to-supercomputer classical compute.
  • Performance: r2 executes 100,000+ circuits per second, about 25x Heron, with Heron-class gate fidelity.
  • Best: Up to 19x speedup over earlier IBM processors on partner workloads; E.ON reports 15x.
  • Worst: r3 is a design, not shipping hardware. Chip-to-chip links average 99% state transfer fidelity, short of the 99.9% target.
  • Bottom line: Best: real throughput gains on a cloud chip you can use now. Worst: the 1,040-qubit system is a 2027 plan, not a result.

What is IBM Quantum Nighthawk?

Nighthawk is IBM’s square-lattice processor family built for quantum advantage workloads. Most qubits connect to 4 nearest neighbors, versus 2 or 3 on earlier designs. That connectivity lets circuits express more entangling structure with fewer swap operations.

The current production chip is Nighthawk r2. It has 120 programmable qubits, 218 couplers and 120 reset elements. That is 458 physical quantum elements in total.

How does Nighthawk r2 get 25x more throughput?

The bottleneck was qubit reset. Heron uses conditional reset: measure, then flip any qubit found in the excited state. That method cannot fix leaked qubits, so the chip idles for hundreds of microseconds between shots.

Nighthawk r2 replaces this with a dissipative reset gadget per qubit. A tunable coupler links each qubit to a cold environment on demand. This drops effective T1 from about 200 microseconds to roughly 25 nanoseconds. Idle time between runs falls to as little as 1 microsecond.

The result is 100,000+ circuits per second. Heron manages roughly 4,000. IBM also reports about 25x lower initialization error, while neighbors stay undisturbed.

What speedups are partners reporting?

According to IBM, clients have run 34 billion circuits on Nighthawk r2 so far. The figures below are self-reported by IBM and its partners.

  • E.ON: 15x speedup in quantum machine learning for energy grid optimization, versus Heron.
  • Oak Ridge National Laboratory: 12x speedup on a magnetic materials simulation matching neutron scattering data.
  • Algorithmiq: circuits ran 8x faster than the original Heron experiment. IBM estimates tensor-network methods need 240x more memory to match.
  • Q-CTRL: a 60-qubit fermionic simulation with 4,497 two-qubit gates. Runtime fell from 4 minutes 25 seconds to 23 seconds. Deviation from the Heron r3 baseline stayed under 1%.

IBM also reports accurate observable estimation on circuits with 7,500+ gates, a 2026 roadmap milestone.

What changes in Nighthawk r3?

Nighthawk r3 keeps r2’s square lattice and fast operations. It scales to 520 programmable qubits from 2,034 physical qubits. It adds 10 chip-to-chip interconnects called l-couplers.

Those links are the path to 1,040 qubits. IBM connected 2 Nighthawk r1 chips with 4 l-couplers, each using an 80 cm cable. State transfer fidelity reached up to 99.3%, averaging 99%. IBM also linked 2 cryogenic modules, reaching below 15 millikelvin in under 5 days.

How does IBM Quantum Platform mix quantum and classical compute?

IBM Quantum Platform now runs HPC and quantum jobs in one managed environment. Users can attach classical compute from a laptop, a supercomputer or the platform itself.

IBM and RIKEN reran a sample-based quantum diagonalization workflow roughly 5x faster with 100x more computations. Cleveland Clinic and RIKEN processed 608 million samples 33x faster with workflow automation. Pulse-level controls are also available for error correction research.

How does Nighthawk compare with other quantum processors?

SpecIBM Nighthawk r2IBM Nighthawk r3 (planned)IBM Heron r3Google WillowQuantinuum Helios
Qubit typeSuperconductingSuperconductingSuperconductingSuperconductingTrapped ion
Programmable qubits12052015610598
Physical elements4582,034 physical qubitsNot disclosedNot disclosedNot disclosed
ConnectivitySquare lattice, 4 neighborsSquare lattice + 10 l-couplers2 to 3 neighborsAvg 3.47All-to-all
Circuit throughput100,000+ circuits/sNot disclosed~4,000 circuits/sNot disclosedNot disclosed
Accurate circuit size7,500+ gates10,000 operations (target)Not disclosedNot disclosedNot disclosed
2-qubit gate fidelityHeron-class (no figure)Not disclosedNot disclosedNot disclosed99.921%
AvailabilityIBM Quantum Platform2027 (2-chip system)IBM Quantum PlatformNot disclosedNot disclosed

Key Takeaways

  • Nighthawk r3 targets 10,000 operations on 520 programmable qubits.
  • 2 linked r3 chips should form a 1,040-qubit system in 2027.
  • Nighthawk r2 runs 100,000+ circuits per second, about 25x Heron.
  • Partners report up to 19x speedups across 34 billion circuits.
  • Chip-to-chip links average 99% fidelity, still below the 99.9% goal.


Check out the IBM press release, the Nighthawk r2 technical blog and the l-coupler paper. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us

The post IBM Announces IBM Quantum Nighthawk r3: A New Processor to Support 10k Quantum Operations on 520 Programmable Qubits appeared first on MarkTechPost.

正文保留英文原文,标题与摘要由机器翻译。

本文由采集器自动同步

原文发布于 MarkTechPost(IBM Announces IBM Quantum Nighthawk r3: A New Processor to Support 10k Quantum Operations on 520 Programmable Qubits),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗