Europe’s quantum pilot lines face a challenge that extends beyond producing more chips. Useful scale depends on repeatable devices, test capacity, and the optical, vacuum, and cooling systems around them.
The factory must deliver an operating device
Manufacturing a quantum chip and manufacturing a quantum computer are different achievements. A fabrication line may produce a precisely patterned wafer while the eventual system still needs demanding optical alignment, extensive calibration, or lengthy cryogenic testing. Those steps determine how many usable instruments emerge, how long they take to build, and whether another team can reproduce the result.
The comparison between hardware platforms is consequently more complicated than asking which one can use a semiconductor factory. Superconducting circuits, electron spins, trapped ions, neutral atoms and photons place different demands on materials and packaging. Each can benefit from industrial process control, but the relevant defects and acceptance tests differ.
A productive European manufacturing strategy should measure progress at those interfaces. It should ask which operations have become reproducible, which still require individual adjustment, and which losses appear only after assembly. That approach gives component suppliers a concrete role without assuming that a larger wafer or a new pilot line establishes a complete computing capability.
Wafer statistics reveal what selected devices conceal
An imec-led study published in Nature in 2024 demonstrated superconducting transmon qubits fabricated on 300-millimetre wafers using optical lithography and reactive-ion etching. Of 400 qubits examined without pre-selection, 393 were functional and fully characterised. The authors also measured substantial variation across the wafer and over time. Their data therefore provide a manufacturing distribution, not just the performance of a particularly successful device.
The reported 98.25% functional yield must retain that definition. It does not mean that 98.25% of complete quantum processors would meet a chosen application specification. Nor do the best observed coherence values describe every qubit on the wafer. Coherence measures how long quantum information remains usable under specified conditions; variation affects how devices can be operated together.
This is a valuable shift in evidence. A production process needs a description of acceptable performance and the fraction of devices that meet it. Reporting the distribution makes process improvement possible and helps customers decide whether the output suits their architecture.
Testing capacity can become a manufacturing bottleneck
A separate Intel study used a cryogenic wafer prober to examine hundreds of silicon spin-qubit devices at an electron temperature of 1.6 kelvin. Automated measurements collected information across full 300-millimetre wafers and fed it back into process development. On one representative wafer, the authors reported 99.8% quantum-dot yield and 96% full-device yield for the tested structures.
Those yields concern the paper’s electrical and device criteria. They are not measurements showing that every quantum dot performs high-fidelity quantum logic. This distinction is central to manufacturing: an efficient screening measurement can identify promising devices without replacing every later functional test.
Quantum operation provides another level of evidence. A 2025 study of industrially fabricated silicon–silicon-germanium devices reported single-qubit gate fidelities above 99%, alongside measurements of charge noise, relaxation and coherence. The researchers used a 300-millimetre process from material growth through micromagnet integration.
Taken together, these studies support complementary questions. Can the factory make uniform structures, can screening find them efficiently, and can selected devices perform the intended operations? A procurement specification should identify which of those claims has actually been demonstrated.
Photonic integration still has a system around it
PsiQuantum’s 2025 Nature paper describes a photonic platform developed with GlobalFoundries using a commercial 300-millimetre manufacturing process. It integrates photon generation, manipulation and detection, adapting established silicon photonics with additional quantum-specific process modules. This is primary technical evidence from the developing company and its collaborators, rather than an independent assessment of the company’s complete computing roadmap.
The paper’s physical assembly is revealing: photonic dies connect to fibres and electronics, while superconducting photon detectors operate in a cryogenic arrangement. Photonic processing should therefore not automatically be described as requiring no cryogenics. Different parts of a photonic system can have different temperature requirements.
Optical interfaces also belong inside the performance account. A photon lost in coupling or routing cannot be recovered merely because the preceding component was fabricated reproducibly. For production engineers, the relevant question is how fabrication, packaging and assembly jointly affect the probability of delivering the required optical event.
A foundry-compatible process is therefore a route to scaling components, while a complete machine must also manage interconnection and operational overhead. That distinction leaves room for substantial manufacturing progress without turning a component demonstration into a claim of useful fault-tolerant computation.
Trapped particles move complexity into light delivery
Trapped-ion systems illustrate another role for integrated photonics. In research reported by MIT in January 2026, waveguides and precisely designed emitting structures delivered intersecting laser beams to cool a trapped ion. The work demonstrated polarisation-gradient cooling through integrated optics, a technique previously established using bulk optical components. It addressed one operation in an ion system; the team identified cooling multiple ions as future work.
Integration can change where alignment is established and how it is preserved. Instead of adjusting every beam path independently on an optical bench, part of the geometry is defined in fabrication. However, the chip still needs suitable input light, electrical connections and a functioning trap environment. The demonstration should not be interpreted as eliminating lasers or every external optical component.
Europe’s CHAMP-ION programme explicitly targets a pilot-line network for ion-trap chips, including manufacturing standards and supply-chain development. Its official project description presents these as objectives, not completed industrial outcomes. This provides a concrete context for evaluating whether future chips become easier to assemble and test.
Vacuum packaging has its own performance boundary
Neutral-atom systems and many atomic sensors require a controlled vacuum environment as well as optical components. A September 2026 preprint reports a compact strontium-trapping package combining a chip-based oven, planar grating, miniature pump and additively manufactured titanium chamber. The authors trapped up to 100,000 atoms in a 750-millilitre arrangement and reported oven heating power below one watt.
That power figure concerns the oven, not the complete quantum instrument. The result demonstrates an atom-source subsystem, not a finished clock, sensor or neutral-atom computer. As a preprint, it should also be distinguished from a peer-reviewed production reliability study.
For a manufacturer, miniaturisation poses questions about assembly tolerance, access for servicing and the retention of vacuum performance after transport. A compact package can reduce one constraint while making another less accessible. Qualification therefore needs to follow the package through the operations it is expected to experience, rather than stop at its smallest demonstrated dimensions.
Pilot lines need comparable acceptance criteria
The official Q-PLANET framework description sets out plans to develop and validate manufacturing processes, design automation and testing for neutral-atom quantum chips. This makes the programme a relevant place to establish shared interfaces and acceptance methods. Its existence does not establish that a standardised, high-yield supply chain is already operating.
The fairest comparison across platforms is not a single yield percentage. It is a clearly defined sequence: fabricated component, screened component, assembled module, and operating system. Each stage should state its denominator, acceptance criteria, and test conditions. Otherwise, two impressive percentages may refer to entirely different products.
Batch consistency deserves a place in that comparison. A wafer map describes spatial variation within the measured material; repeated production runs establish whether the process returns to the same operating range. Customers also need to know which adjustments remain their responsibility after delivery. A component that requires individual tuning can still be useful, but its price alone will not describe the work needed to integrate it. Pilot lines could make that work visible by delivering reference test procedures, interface documentation and records of accepted deviations alongside the hardware. Such records would give smaller equipment makers a clearer basis for designing compatible products and investigating faults.
For Europe’s suppliers, that discipline can make specialisation commercially useful. A company supplying vacuum hardware, photonic packaging or cryogenic tests need not choose the eventual winning quantum computer architecture. It needs to demonstrate how its process improves a specified interface. Quantum manufacturing advances when those improvements survive assembly and can be repeated by the next production team.
