In 2023, IBM unveiled Condor, a quantum chip that passed over the 1,000-qubit mark by reaching 1,121 superconducting qubits, and last year Big Blue laid out a roadmap that leads to Starling, which the venerable tech giant said will be the world’s first large-scale, fault-tolerant quantum system that will include 100 million quantum gates on 200 logical qubits.

IBM isn’t the only vendor or research group pushing to grow the number of qubits in a system. Atom computing in 2023 announced a quantum computer with more than 1,180 neutral atom qubits, and researchers at the California Institute of Technology last year unveiled a neutral atom array of 6,100 qubits.

The number of qubits a system can hold is a key market along the road to practical, commercial quantum computers, the kind that promise significant advancements in everything from pharmaceuticals to financial services. But to get there, those numbers will have to grow significantly.

“The only problem is that for a lot of the practical use cases that quantum needs to solve for, the number of qubits that we need to solve those problems is roughly around hundreds of thousands to a million qubits for a quantum computer,” Vijoy Pandey, senior vice president and general manager of Outshift, the incubation arm for emerging technologies at Cisco Systems, told journalists at a recent media briefing. “Right now, what's happening is we have quantum computers operating at roughly 100 to 1,000 qubits in size, and from the public roadmaps that all these companies have talked about, we believe this number going up to the low ten-thousands in the next three years.”

After that, the numbers jump to hundreds of thousands to millions of qubits, Pandey said. Such powerful systems are coming, but that raises other issues. One of them is what such large-scale quantum computing environments will look like. One option is to build increasingly massive systems that house all of these qubits. However, Cisco is looking at this through its historically network-focused lens.

“The other option – and this is borrowing from the paradigms that we have learned from cloud computing – is you can connect a whole bunch of these quantum computers together through a quantum network and make all of these entities operate as a singular, large, distributed quantum computer,” Pandey said. “You can connect, let's say, a hundred of the thousand-qubit computers through a quantum network, and get to a hundred thousand qubits in size.”

(Editor’s note: We would point out that the HPC crowd figured out distributed computing at scale long before the hyperscalers got around to it.)

Last year, Pandey and other Cisco executives announced that Outshift was building the network stack for quantum computing, which already includes the company’s quantum network entanglement chip, its network-aware Quantum Compiler to orchestrate the quantum algorithms across multiple quantum processors, and applications like Quantum Sync and Quantum Alert.

They are all part of what Pandey calls Cisco’s “North Star vision” of connecting “all of these quantum computing nodes through this quantum network to enable distributed quantum computing and to enable acceleration of that roadmap to get to hundreds of thousands of qubits and millions of qubits faster than what it would take through vertically scaling these compute nodes singularly.”

This is a classic scale out versus scale up architectural choice that came to ModSim supercomputing back in the 1990s and is now causing a certain amount of heartburn in AI systems right now. Although with AI, we are moving backwards, creating massive scale up architectures because scale out does not work well always, particularly if you want fast response times on AI inference with GenAI foundation models.

In any event, Cisco has unveiled its Universal Quantum Switch, a working prototype designed to connect quantum systems from different vendors that use different modalities and quantum sensors to create a single network. The goal is not only to scale quantum capabilities, but also to do so by linking multiple computers regardless of the modalities they’re based on IBM uses superconducting, for example, while QuEra runs systems with neutral atoms and IonQ uses trapped ions for its qubits.

They encode information in different ways, and the switch supports all four of the major encoding modalities – polarization (orientation of light waves), time-bin (timing of light pulses), frequency-bin (color or frequency of light), and path (physical or spatial path).

So far, the switch has been validated in experiments with polarization encoding, and time-bin and frequency-bin is already built into the design and will be up next in the validation process.

In the switch, a single photon carrying a qubit of quantum information enters it through a standard telecom optical fiber and arrives encoded to a quantum state converter (QSC) in one of the four encoding modalities. Inside the QSC, the qubit is converted into a format used for routing, and then enters the switch blocks, where the quantum information is routed to another QSC, where it is converted to the necessary encoding modality for the destination system. Qubits are extremely fragile and can become disentangled and destroyed by the slightest of noises, from light, sound, the activity of other qubits, and others means. In the Cisco switch, the converted qubit is not measured, so it remains entangled and intact, with its quantum state preserved.

“In the quantum world, entanglement states are actually encoded in single photons,” said Ramana Kompella, Cisco Fellow and vice president and head of research. “Any kind of polarization drift, any kind of like a loss, any kind noise just destroys this. This is actually foundational. The more of these devices that you put in the picture, the loss compounds. You have to build this switch, taking care of these very fragile quantum states from the ground up.”

Cisco’s switch is non-blocking, so multiple photons can run through the switch’s chip at the same time, with each being independently routed. It also does all this at room temperature, negating the need for cooling systems required by some quantum computers.

“It's not operating in a cryogenic chamber or what have you,” Kompella said. “It's actually operating in room temperatures and it's compatible with all the fiber that is already laid out. That actually makes it super-operable and super-easy to deploy and to build it out. It's also very tiny in terms of its power consumption. It uses less than a milliwatt of power.”

Included in the mix will be quantum sensors, which use quantum particles – like photons, ions, and atoms – to measure magnetic fields, gravity, time, rotation, and other physical properties with greater precision than classical systems can. Through the switch, quantum processing unites (QPUs) that are based on one modality will be able to communicates with sensors based built on another.

Proof-of-concept experimentation validated capabilities of the switch, the Cisco executives said. That includes not only the energy efficiency of the switch – consuming less than a milliwatt – but also its ability to preserve the quantum state of the information as it passes through the conversion processing, showing that a less-than 4 percent degradation of fidelity and entanglement.

In addition, the switching speed includes sub-nano-second electro-optic switching, with the reconfigured connections happening in as little as a nanosecond.

The ability to easily bridge the various modalities is getting more attention in various corners as quantum technology evolves. As we wrote about, Classiq through its platform, a Python-style Qmod (Quantum Modeling Language), and GitHub library of open code examples is giving software developers tools that allow them to write quantum software that can run on systems that are based on different modalities, like superconducting, photonics, and neutral atoms.

Earlier this month, the government’s Defense Advanced Research Projects Agency (DAPRA) launched the Heterogeneous Architectures for Quantum (HARQ) program, an initiative aimed at creating heterogeneous quantum computing architectures that bring together qubit types of different modalities into a single system. The organization sees this as another efficient way to scale quantum computing.

Cisco envisions similar results with its quantum networking stack. As in datacenters running classical computing environments, such scale-up systems and scale-out environments tied together through a network will give organizations options as they plan out their quantum futures, the executives said.

Here is what the Cisco quantum switch wafer looks like: 

“The universality and the power of actually transforming between modalities is actually pretty critical because we envision a datacenter of the future where you have quantum computers of potentially many different modalities coexisting,” Kompella said. “Part of it is because different modalities actually do better with certain types of applications. We don't know whether there'll be a universal winner, and chances are, just like we have CPUs, GPUs, DPUs, XPUs for different types of computation, we may end up very well with different types of modalities for quantum processors in a datacenter. Therefore, for us at Cisco, as we are actually looking at this quantum network, our vision has always been to enable this heterogeneous set of quantum computers coming together with one unified fabric.”