Quantum Modularity: Where Should a Quantum Computer Come Apart?

For years, discussions about scaling quantum computers have centred on qubit counts. Roadmaps have often been measured in terms of processor size, with each new generation promising larger devices containing ever more qubits. Yet as the industry moves closer to fault-tolerant quantum computing, a different metric is emerging. Rather than asking how large a single processor can become, researchers and companies are increasingly asking how a quantum computer should be divided into smaller, manageable pieces.

Last week, IBM announced their latest development in this field: modular cryogenics. The announcement, when viewed alongside developments in modularity across superconducting, trapped ion, neutral atom, photonic, and silicon spin platforms, suggests that modularity is becoming a defining theme of next-generation quantum-computing architectures. What is less clear is what form that modularity should take. While there is growing agreement that future systems will be built from modules, there is little consensus on where the boundaries between those modules should be drawn.

IBM’s latest step: modularising the cryostat

IBM's modularisation approach focuses not on a new processor, but on something arguably more important for long-term scaling: the cryogenic infrastructure that houses their superconducting qubits.

The company unveiled a modular cryogenic system based on cryogenic cells. Each cell contains its own vacuum chamber, cooling hardware, and thermal shielding. Adjacent cells are connected via a protected cryogenic tunnel designed to support short quantum interconnects between processors located in different cells while maintaining the necessary cryogenic temperatures.

At first glance, this might appear to be a relatively incremental engineering development. In reality, it represents an attempt to address several of the most significant obstacles facing superconducting quantum computing. IBM is not merely modularising the processor itself, but rather the physical infrastructure surrounding the processor. This distinction matters because many of the challenges associated with large-scale quantum computing arise not only from the qubits, but from the systems required to control, cool and connect them. As processors grow larger and more sophisticated, they require increasing numbers of control lines, more complex wiring arrangements, larger refrigeration systems, and tighter control of unwanted thermal and electromagnetic interactions. Simply building ever larger cryostats eventually becomes both technically and economically unattractive.

IBM's proposed solution is to shift from monolithic scaling towards modular scaling. Rather than constructing a single enormous refrigerator containing every component of the quantum computer, the modular cryogenic cell approach enables consistent cool down times and temperature stability, providing scalability without disrupting computational performance.

Convergence and divergence

IBM’s announcement highlights a broader trend that is becoming increasingly apparent across the quantum industry. There is now a growing consensus that modularity will be required to reach large-scale, fault-tolerant quantum computing. Yet there is remarkably little agreement on what form that modularity should take.

IBM have chosen to draw the boundary between modules at the cryogenic level. Other superconducting companies are drawing the boundary elsewhere. Rigetti, for example, has focused on chiplet-style approaches, constructing larger processors from smaller quantum chips integrated within a common cryogenic environment. Similarly, Google are developing a modular chip stack which combines qubits and superconducting control electronics. Still others are experimenting with architectural modularity rather than physical modularity. IQM's resonator-based designs, for example, aim to create highly connected processing “constellations” as an alternative to traditional square lattice designs.

What emerges is a picture of an industry that is converging on the necessity of modularity while diverging on where the module boundary should sit. Should modularity occur at the level of the architecture, the chip, the processor, the cryostat? At present, there is no clear winner. Different companies are optimising for different constraints, and the eventual answer may involve several layers of modularity operating simultaneously.

Indeed, it may prove misleading to think of modularity as a single architectural choice. Future fault-tolerant systems could combine chiplets within processors, processors within cryogenic cells, and cryogenic cells within larger quantum data-centre installations. In that sense, the debate is gradually shifting away from whether quantum computers should be modular and towards identifying the most effective hierarchy of modules.

Beyond superconducting qubits

The diversity of approaches becomes even more striking when one looks beyond superconducting quantum computing. Nearly every major modality is embracing some form of modularity, and the way in which modularity is implemented is often shaped by the underlying physics of the platform itself.

Trapped ions: modularity through networking – Among trapped-ion developers, modularity has long been viewed as a natural route to scalability. Rather than attempting to construct a single trap containing vast numbers of ions, many research groups and commercial developers envision networks of smaller ion-trap processors connected through photonic links. Each module performs local computation, while entanglement between modules enables distributed quantum operations. In this way, the modular boundary is placed at the communication interface. The processor is designed from the outset with networking in mind. As a result, modularity is not simply an engineering response to scaling challenges; it is embedded within the architectural philosophy of the platform.

Neutral atoms: modularity through reconfiguration – Neutral atom systems offer a rather different perspective. Here, modularity often arises not from physically separate hardware modules but from the ability to dynamically rearrange qubits within a single apparatus. Using optical tweezers, atoms can be moved between storage zones, processing zones, and measurement regions. The architecture therefore becomes highly reconfigurable, with resources allocated dynamically according to computational requirements. The boundaries between modules are not necessarily defined by physical structures but by the role that groups of qubits play at a given moment. In some ways, neutral atom systems blur the distinction between modular and monolithic architectures.

Photonics: modularity by design – Photonic quantum computing may represent the clearest example of modularity being built directly into the physics of the platform. Photons are naturally suited to travelling long distances with relatively low loss, making communication between modules an intrinsic capability rather than a technical obstacle. Consequently, photonic architectures are often conceived as collections of interconnected components rather than single integrated processors. The result is an architectural vision that resembles modern data-centre infrastructure more closely than many other quantum-computing platforms. Whereas superconducting systems must carefully engineer cryogenic interconnects to bridge module boundaries, photonic systems can leverage networking concepts that already exist within classical communications infrastructure. The primary challenges shift from thermal engineering to optical loss, resource-state generation and fault-tolerant photonic protocols.

Silicon spin platforms: modularity inspired by semiconductors – Silicon spin quantum computing introduces yet another interpretation of modularity, one heavily influenced by the semiconductor industry's experience with advanced packaging and chip-level architectures. Developers in this field increasingly envision quantum processors that can take advantage of established CMOS manufacturing techniques and packaging ecosystems. If successful, silicon spin platforms could benefit from decades of accumulated expertise in wafer-scale fabrication, heterogeneous integration, and modular system design. In this context, modularity is not merely a route to technical scalability but potentially a route to industrial scalability as well.

What the future holds

Taken together, these developments suggest that modularity is becoming one of the defining themes of modern quantum-computing architecture. Yet the industry's growing commitment to modularity should not be mistaken for architectural convergence. Quite the opposite: the more companies embrace modularity, the more varied their implementations appear to become.

IBM's announcement provides an excellent illustration of this phenomenon. It demonstrates that modularity can extend beyond processors and into the cryogenic infrastructure. At the same time, developments in trapped ions, neutral atoms, photonics, and silicon spin systems reveal that there are many possible answers to the question of where a quantum computer should be divided into modules.

Perhaps the most important insight is that modularity is not a single technology but a design philosophy. Different platforms place module boundaries at different levels because they are attempting to solve different physical and engineering problems. The resulting architectures may look radically different, yet they are all responding to the same fundamental challenge: how to scale quantum computing more efficiently than a single monolithic device.

The debate is no longer whether quantum computers will be modular. It is about determining which forms of modularity will prove most effective, and whether the industry will eventually settle on common interfaces between them. That question may ultimately be every bit as important as the race to build better qubits.