Resolving error floor limits with distributed quantum computing

Nu Quantum proposes a quantum networking-based solution to dealing with catastrophic quantum errors.

Researchers at Nu Quantum, a quantum networking vendor, published a pre-print research article to the arXiv in May 2026 that claims a new benefit for building distributed (networked) quantum computers: such computers may be more resilient to catastrophic failure errors than so-called monolithic quantum computers. [1] The proposed architecture was demonstrated in a simulation rather than a current working system.

Scale up or scale out

This claim cuts to the heart of the “scale up or scale out” debate within the quantum computing industry. Scale up refers to building larger computers by adding more qubits (the minimal unit structure of quantum information) onto one monolithic quantum processing unit (QPU). Essentially, scale up is analogous to packing more transistors onto one “classical computing” chip.

In contrast, scale out proposes building larger computers by networking large numbers of QPUs together. This is more analogous to a high-performance computer (HPC) consisting of many servers, each of which can be swapped in or out without negatively affecting the HPC (i.e., supercomputer) as a whole.

This debate is important because large scale fault tolerant quantum computers (FTQCs) will likely need hundreds of thousands or millions of physical qubits to tackle many of the problems of interest to scientists, governments, and industry. However, current quantum computers range from a few dozen qubits to the low thousands of qubits on a single QPU. Either the quantum computing industry needs to “scale up” these QPUs, or it needs to “scale out” the systems by networking many of these QPUs together.

Nu Quantum’s proposal is a notably new argument in the scale up or scale out debate. The company is saying its scale-out proposal leads to lower error rates for the computer compared to a monolithic (scale up) QPU. Until now, the debate has centered on the comparative technical difficulties of each approach, rather than on potential benefits. While each of the main qubit modalities (i.e., the various types of physical instantiation of a qubit) has its own challenges, I’ll describe those of superconducting qubits because to date they are the most widely deployed type of quantum computer.

Scale up challenges

Superconducting qubit QPUs traditionally face two key problems. The first is how to handle the huge volume of wired connections running from the QPU to the classical control system. The second is how to deal with qubit crosstalk – the interference caused by one qubit on a neighboring qubit – as qubit density increases on the QPU.

A superconducting qubit QPU sits inside an extreme cooling device called a dilution refrigerator (DR). The DR is the famous “golden chandelier” you often see in pictures of quantum computers. It cools in stages, with the end/bottom stage of the system (where the QPU sits) cooled down to only 10 – 15 degrees millikelvin (mK). Each qubit in the QPU is typically attached to 3 – 4 control wires [2] that traverse from the mK stage QPU out of the DR to the control system electronics sitting at room temperature (300 K).

This poses two problems. First, there isn’t enough room in the DR for millions of control wires. Second, the DR can only pump out an extremely small amount of unwanted heat at the mK stage to keep the temperature at that extreme level (mK temperatures are colder than deep space). Each wire, however, conducts unwanted heat down from outside the DR, and each necessary control pulse also dissipates some energy as unwanted heat. If more unwanted heat is introduced than the DR can remove at the mK stage, the QPU will lose quantum coherence and stop working.

As more qubits are packed more densely on a QPU there is an increasing chance that electromagnetic signals will mistakenly be coupled between nearby qubits. [3] This would result in control signals meant for one qubit also impacting a neighboring qubit. These mistakes degrade the overall fidelity of the chip to process a useful computation. Superconducting qubits typically each use a distinct microwave frequency carrier in the 4 – 8 GHz range. [4] The design challenge of making sure that nearby qubits each use different frequencies, and do so correctly, grows increasingly hard with increasing qubit density.

Scale out challenges

Focusing again on superconducting qubit quantum computers, there are two major challenges to scaling up systems via networking together separate QPU modules. First, it is extremely hard to convert microwave photons to optical photons and then back to microwave photons. [5] Second, synchronizing quantum entanglement between distributed QPUs requires sub-nanosecond timing. [6]

The easiest way to connect one superconducting qubit QPU to another is via a direct state transfer from a qubit in one QPU to a qubit in another QPU using microwave photons, the same frequency of photons (4 – 8 GHz) that these systems use for qubit control and readout. The challenge is that microwave photons only carry quantum coherent data well at mK temperatures; room temperature (300 K) swamps the quantum coherence with thermal noise. This means that the two QPUs must share a DR environment, limiting the range for true “networking” distances.

Entangling two qubits, one in each QPU, using optical photons (at 200 THz, or 200,000 GHz) enables room temperature networking. One key development challenge has been the inefficiency of converting (or “transducing”) microwave photons to optical photons. This challenge arises to the frequency mismatch between the two kinds of photons, the introduction of unwanted heat into the DR by lasers used in transduction, and a fabrication mismatch between the materials used for superconducting qubit QPUs and the materials used for the transducers.

A second challenge is that sub-nanosecond timing (picosecond timing) is needed to coordinate the entanglement process between the two separate QPUs. This process, known as “heralding”, involves coordinating both qubits (one in each QPU) to each send an optical photon to a central node, where, if the photons arrive at the exact same time, a beam splitter/photodetection setup leads to entanglement between the two separate qubits/QPUs.

Searching for solutions

Of course, the industry is working hard on finding solutions to the challenges described above. One scale up solution is simply to use qubit modalities other than superconducting qubits. For example, Diraq, an Australian maker of spin qubit quantum computers, has stated they believe that can leverage standard semiconductor manufacturing processes to achieve up to 10,000,000 qubits per chip, though this is likely to take until sometime in the next decade. [7]

IBM has partnered with the DR maker Bluefors in the development of IBM’s System Two architecture. [8] System Two leverages Bluefors Kide DRs to collocate a cluster of hexagonal DRs physically next to each other, connected at the mK cooling stage by cryogenic coaxial lines transmitting microwave photons for direct state transfer rather than coordinated entanglement. Each Kide can currently hold 1,000 physical qubits, and Bluefors says that with cable miniaturization this could rise to 10,000 qubits. [9] By implication, a single 6 DR cluster could connect 60,000 physical qubits using microwave photons.

Regarding scale out, Nu Quantum has announced a “Quantum Network Interface Card” targeting 98-99% fidelity. [10] Additionally, the company leverages CERN’s “White Rabbit” open-source sub-nanosecond precision timing system for quantum networking. [11]

Scale out as a benefit

There are two key benefits to Nu Quantum’s proposed architecture. First, individual QPUs in a networked system can be swapped out without the entire system needing to go offline. Scheduled node swap out is achieved by “teleporting” the quantum states from the QPU to be removed over to another QPU being swapped in as a replacement. The company states this process leads to minimal impact on the logical error rate.

Second, the architecture’s robustness to catastrophic error in any specific QPU removes the “error floor” risk inherent in monolithic quantum computers. For example, in Google’s famous Willow announcement in December 2024, the company noted that it couldn’t push theoretical error rates below 10^-10 due to unidentified correlated errors (errors impacting a large contiguous section of the chip). [12] In Nu Quantum’s proposal, if the QPU is small enough relative to the overall system, the distributed quantum computer can treat the QPU as “a spatially correlated error” (meaning an error involving a collocated subset of qubits). A new QPU can be swapped in (presumably automatically) and the lost data can essentially be “inferred” onto the new QPU using the information contained in the rest of the systems error correcting code.

In short, the proposed architecture suggests a distributed quantum computer that behaves at least somewhat “elastically” in the way that data center server infrastructure is elastic. The distributed system should experience more uptime during scheduled maintenance and be more resistant to certain types of errors.

Citations

[1] Sutcliffe, Evan and Westoby, Coral M., Tolerating Device Failure in Distributed Quantum Computing” (May 2026). https://arxiv.org/abs/2605.11088

[2] Croot, Xanthe, et al. “Enabling Technologies for Scalable Superconducting Quantum Computing” (December 2025). https://arxiv.org/pdf/2512.15001

[3] Wesdorp, Jaap J., et al. “Mitigating crosstalk errors for simultaneous single-qubit gates on a superconducting quantum processor” (March 2026). https://arxiv.org/pdf/2603.11018v1

[4] Ivezic, Marin, “Quantum Computing Modalities: Superconducting Qubits” (October 2023). https://postquantum.com/quantum-modalities/superconducting-qubits/#coupling-and-gates

[5] Han, Xu, et al. “Microwave-optical quantum frequency conversion” (August 2021). https://opg.optica.org/optica/fulltext.cfm?uri=optica-8-8-1050

[6] Silva, Augustin and Orgaz Fuertes, Alvaro. “Manarat: A scalable QICK-based control system for superconducting quantum processors supporting synchronized control of 10 flux-tunable qubits’ (February 2026). https://pubs.aip.org/aip/rsi/article/97/2/024702/3378494/Manarat-A-scalable-QICK-based-control-system-for

[7] Diraq, “One Dollar Per Qubit: The Number That Changes Everything” (June 2026). https://www.diraq.com/newsdesk/one-dollar-per-qubit-the-number-that-changes-everything

[8] IBM, “IBM scientists cool down the world’s largest quantum-ready cryogenic concept system” (September 2022). https://www.ibm.com/quantum/blog/goldeneye-cryogenic-concept-system

[9] Bluefors, “We Made 1000 Qubits for Quantum Computing Possible” (February 2023). https://bluefors.com/stories/we-made-1-000-qubits-for-quantum-computing-possible/#:~:text=KIDE%20is%20ready%20to%20house,through%20its%20flat%2C%20hexagonal%20shape.

[10] IEEE Spectrum, “New Interface Uses Light to Scale Up Quantum Computers” (October 2024). https://spectrum.ieee.org/quantum-network-interface

[11] CERN, “Nu Quantum is the First Industrial Partner to use White Rabbit for Quantum Technology Applications” (November 2024). https://report2024-kt.web.cern.ch/nu-quantum/

[12] Google Research, “Making quantum error correction work” (December 2024). https://research.google/blog/making-quantum-error-correction-work/