Fujitsu Built a Diamond-Spin Quantum Prototype. The Numbers That Would Prove It Scales Are Missing.

8 Sep 2026 – Fujitsu announced it has built a working prototype of a diamond-spin quantum computer that incorporates tin-vacancy (SnV) centers into photonic integrated circuits. The company described the prototype as the first to use high-brightness SnV centers integrated on a chip, based on its own research criteria.

The device is a single module built around a nanoscale diamond piece containing an SnV center coupled to an alumina waveguide – not a multi-qubit processor. It operates at −271.6°C (1.55 kelvin), warmer than the approximately 20 millikelvin (−273.13°C) typically required by superconducting quantum computers. That temperature difference changes the cryostat class: 1.55 K is reachable with a closed-cycle cryocooler and does not require the dilution refrigerator stage that superconducting and SiV-center systems need, which reduces the cooling infrastructure and raises the available cooling power. Fujitsu said the prototype was accessible through its Hybrid Quantum Computing Platform in a test environment without requiring additional specialist knowledge from users, according to the company’s announcement.

The work is based on a collaboration started in 2020 between Fujitsu, Delft University of Technology, and QuTech, the quantum technology research institute at TU Delft. Fujitsu developed three enabling technologies for the prototype: a heterogeneous material bonding process that attaches ion-implanted diamond substrates to alumina/silicon dioxide substrates and thins them from several hundred micrometers to several hundred nanometers; a photonic integrated circuit fabrication process that integrates nanometer-sized SnV-containing diamond crystals with alumina optical waveguides transparent in the visible light range; and a quantum circuit conversion mechanism that translates quantum gate descriptions into the combined light, microwave, and radio frequency control sequences required by diamond-spin qubits. Diamond processing drew on joint research with the University of Tokyo.

“The diamond-spin approach we have applied in this prototype not only offers exceptional scalability in its own right, but also has the potential to be integrated with superconducting quantum computers to further extend their capabilities,” said Vivek Mahajan, Corporate Executive Officer, Corporate Vice President and CTO at Fujitsu.

Kees Eijkel, General Director of QuTech, said the prototype represented “a major milestone in our strong collaboration” but noted that “demonstrating the scalability expected of diamond spin quantum computing remains a long and challenging journey.”

Fujitsu said it plans to develop a multi-module diamond-spin prototype by 2027. The company’s broader quantum roadmap targets a 250-logical-qubit system by fiscal 2030 – built on a 10,000-plus physical qubit superconducting architecture using its STAR early-FTQC architecture developed with Osaka University – and a 1,000-logical-qubit system by fiscal 2035 through the integration of superconducting and diamond-spin approaches.

My Analysis

Fujitsu’s prototype validates a fabrication pathway that diamond-spin quantum computing has needed for years. Integrating an SnV center into a photonic circuit on a chip, packaging it in a cryostat, and making it accessible through a software platform is a genuine engineering step – but “platform-accessible” is not the same as “benchmarked.” The announcement publishes no qubit count, no gate fidelity for the SnV qubits, and no inter-module entanglement rate. Those are the three numbers that would let anyone evaluate whether this modular architecture can actually scale, and all three are absent.

That matters because modularity is the central promise. Diamond-spin qubits are not going to compete with superconducting or trapped-ion systems on raw qubit count within a single processor. Their advantage is supposed to be photonic interconnection: you build small, high-fidelity modules and link them with light. The physics supporting that claim is strong. Whether the engineering is there is a different question, and this announcement does not answer it.

Why Tin-Vacancy Centers Exist as a Research Direction

The diamond-spin field has historically centered on nitrogen-vacancy (NV) centers, and for good reason. NV centers have excellent spin coherence – electron spin coherence times exceeding 50 milliseconds at cryogenic temperatures in isotopically purified diamond, and nuclear spin memories with minute-scale lifetimes. QuTech’s own group, led by Ronald Hanson, built the field’s landmark demonstrations on NV centers: the first loophole-free Bell test in 2015, a three-node quantum network in 2021, metropolitan-scale entanglement over 25 km of deployed fiber in 2024, and unconditional quantum gate teleportation between remote registers published in Nature Communications in May 2026.

NV centers have a severe optical limitation. Their Debye-Waller factor – the fraction of photon emission into the zero-phonon line, which is the spectrally narrow emission needed for high-fidelity photonic operations – is approximately 3%. That means 97% of emitted photons are spectrally unsuitable for entanglement generation. Every photonic interconnect protocol linking NV-center modules is constrained by this: entanglement generation rates remain low because the system must wait for the rare zero-phonon-line photon.

SnV centers address this directly. Their inversion-symmetric crystal structure, shared with other group-IV color centers (SiV, GeV), gives them a Debye-Waller factor around 60% – roughly 20 times higher than NV. Their quantum efficiency is approximately 80%. And the heavier tin atom produces a large ground-state splitting of about 850 GHz, which means phonon-induced decoherence freezes out at temperatures above 1 kelvin rather than requiring the millikelvin conditions that silicon-vacancy (SiV) centers demand. A 2024 study demonstrated SnV electron spin coherence times of 433 ± 23 microseconds under Hahn echo and 10 ± 1 milliseconds under dynamical decoupling, using a superconducting waveguide for microwave control (Karapatzakis et al., Phys. Rev. X 14, 031036). Those measurements were taken at 50 mK – below the Fujitsu prototype’s 1.55 K operating point – and coherence at the higher temperature may differ.

Fujitsu’s Japanese-language press materials describe SnV emission efficiency as roughly 10 times that of NV, and Shintaro Sato’s briefing ties this to “about 10× computing efficiency” through more entanglement attempts per unit time. The Debye-Waller ratio alone (~60% vs ~3%) implies up to ~20× more zero-phonon-line photons per emission event; the 10× figure likely reflects collection efficiency losses in the specific waveguide geometry. Either way, the optical advantage is real. What the 10× or 20× improvement does not automatically deliver is a 10× or 20× improvement in useful computation, because SnV coherence times are one to two orders of magnitude shorter than NV. The trade-off is intentional: SnV gives up some coherence for dramatically better optical properties, betting that faster entanglement generation will more than compensate. Whether that bet pays off depends on measurements this prototype does not report.

The Coherent Cooperativity Result That Underpins This

The most important recent result for the SnV-photonic program is not the Fujitsu prototype itself but a QuTech paper published in Physical Review X in spring 2026. The Hanson group demonstrated above-unity coherent cooperativity of SnV centers embedded in diamond photonic crystal cavities – the first time this threshold has been crossed for a tin-vacancy center in a nanophotonic device (Codreanu et al., Phys. Rev. X 16, 021060). Other diamond color centers – notably Harvard’s SiV work – had previously achieved cooperativity above one, but SnV had not, and tin-vacancy is the emitter Fujitsu is building its modular architecture around.

Coherent cooperativity above one means the useful quantum interaction between the emitter and the cavity dominates over dephasing noise. Below that threshold, the cavity makes the emitter brighter but not more quantum-mechanically useful. Above it, high-fidelity entanglement generation protocols become viable. The team reported a scalable fabrication outcome: across two separate chips, they measured 327 devices with high average quality and yield, which matters because a modular architecture needs many such devices, not one.

Hanson stated in the QuTech press release that the result is “important for efficiently linking qubit modules into one large computer” in the context of the Fujitsu collaboration. That comment connects the lab physics to the engineering announcement less than three months later. The prototype integrates SnV centers into photonic circuits; the spring paper shows those circuits can, in principle, support the entanglement generation rates a modular architecture requires.

In principle. The spring paper demonstrated cooperativity in individual cavity-emitter devices. The prototype packages those devices into something accessible through Fujitsu’s hybrid platform. Neither publication reports an entanglement rate between two modules, which is the number that separates a modular quantum computer from a collection of isolated qubits.

What the Announcement Does Not Contain

Three categories of data are absent, and each one is necessary to evaluate the scalability claim.

Qubit count and gate fidelities. The announcement does not state how many qubits the prototype operates – or whether the SnV spin was coherently controlled in this specific package at all. “Working prototype” here means the device is cryogenically operational and platform-accessible; it does not mean a randomized-benchmarking or gate-fidelity number has been published. The Fujitsu-QuTech collaboration has separately published NV-center two-qubit gate fidelities of 99.93% (Phys. Rev. Applied 23, 034052, 2025) and unconditional quantum gate teleportation between remote NV-center registers in separate cryostats (Nature Communications 17, 4694, 2026). Neither of those results was achieved with SnV centers. Transferring NV performance numbers to an SnV prototype would be a category error – the gate mechanisms, control sequences, and noise profiles differ. Until Fujitsu publishes SnV-specific gate fidelities, the prototype’s computational capability is undefined.

Inter-module entanglement rate. This is the metric that determines whether photonic modularity works at scale. As I noted in my analysis of IonQ’s photonic interconnect announcement in April 2026, the gap between academic photonic interconnect demonstrations and the throughput needed for practical modular quantum computing remains one of the most significant engineering challenges facing any modular approach. For scale: Harvard’s 35 km SiV entanglement demonstration achieved roughly 1 Bell pair per second. Useful modular quantum computing likely requires thousands of high-fidelity Bell pairs per second, depending on gate times and error correction overhead. Fujitsu’s announcement describes a single-module prototype; the multi-module system is a 2027 target. Without an entanglement rate, the modular architecture remains a design intent, not an engineering result.

Error correction performance. Diamond-spin proponents argue that high native gate fidelities could reduce the physical-to-logical qubit overhead compared to superconducting systems, where surface codes typically require on the order of 1,000 physical qubits per logical qubit. (Fujitsu’s own STAR architecture for its 2030 superconducting system is designed specifically to reduce that overhead, so the 1,000:1 figure is a baseline, not the state of the art on every platform.) The overhead-reduction claim for diamond-spin is plausible in physics but undemonstrated in SnV hardware. No error correction experiment has been performed on an SnV-center quantum processor.

How This Compares to Other Modular Approaches

Fujitsu is not alone in pursuing photonically linked quantum modules. At least three other programs are working the same problem with different qubit platforms, and each has demonstrated capabilities the Fujitsu prototype has not.

IonQ announced in April 2026 that it had photonically interconnected two independent trapped-ion quantum computers, generating entanglement between commercial systems at a distance – described as the first demonstration of networked commercial quantum computers. The same day, IonQ was selected for DARPA’s Heterogeneous Architectures for Quantum (HARQ) program, which targets networked quantum computers combining different qubit types. IonQ’s 2025 acquisition of Lightsynq brought diamond-based photonic quantum memory technology that promises up to 50× improvement in entanglement rates, though those gains have not been publicly demonstrated at system level.

Harvard’s Lukin group, using SiV centers in nanophotonic diamond cavities, demonstrated entanglement of quantum memory nodes over 35 kilometers of deployed telecom fiber in the Boston area (Knaut et al., Nature, 2024) and universal distributed blind quantum computing with solid-state qubits (Wei et al., Science 388, 509, 2025). The Harvard program uses silicon-vacancy rather than tin-vacancy centers, requiring millikelvin operation, but it has demonstrated multi-node functionality and telecom-wavelength compatibility that the Fujitsu prototype has not addressed. Telecom compatibility matters: SnV emits at approximately 619 nm in the visible range, and connecting SnV modules over fiber distances would require quantum frequency conversion to the telecom band – a step Harvard’s SiV program has already demonstrated but Fujitsu’s SnV prototype has not.

QuTech’s own NV-center program – running in the same building as the SnV work – has demonstrated three-node quantum networking, metropolitan entanglement over 25 km of deployed fiber, and the unconditional gate teleportation result from May 2026. Fujitsu’s SnV prototype is a parallel bet inside the same collaboration: it wagers that SnV’s optical advantages will eventually overtake NV’s more mature multi-node demonstrations.

The comparative picture is specific. IonQ and Harvard have demonstrated inter-module entanglement. QuTech’s NV program has demonstrated multi-node operations. Fujitsu’s SnV prototype has demonstrated chip-level fabrication and platform integration. Fabrication is the earliest of these milestones; entanglement generation and multi-node operation are later milestones that Fujitsu has not yet reached.

The Roadmap Arithmetic

Fujitsu’s stated roadmap – 250 logical qubits by fiscal 2030 and 1,000 logical qubits by fiscal 2035 – deserves careful reading. The 2030 target is explicitly a superconducting system: a 10,000-plus physical qubit machine using Fujitsu’s STAR early-FTQC architecture, developed under a NEDO-led project with RIKEN and AIST. Diamond-spin is not the path to 250 logical qubits. Fujitsu’s own press materials position diamond-spin as a sixth hardware approach alongside superconducting – a portfolio bet, not a pivot.

The 1,000-logical-qubit target for 2035 is where diamond-spin enters, through integration with superconducting systems. Fujitsu’s CTO framed the diamond-spin approach as having “the potential to be integrated with superconducting quantum computers to further extend their capabilities.” That language describes a research ambition, not an engineering plan. No published demonstration of a hybrid superconducting-diamond-spin interface exists. The physics is not obviously impossible – but the engineering is substantial: converting microwave photons (4–8 GHz) from superconducting circuits into optical photons (~619 nm for SnV) with high efficiency and low noise, without dissipating thermal energy into the dilution refrigerator, is an unsolved transduction problem. No component of that interface has been prototyped.

To progress from this single-module prototype to its planned 2027 multi-module system, Fujitsu must first demonstrate SnV gate fidelities competitive with NV benchmarks. It will then need to fabricate multiple uniform modules and achieve inter-module entanglement rates fast enough to support quantum error correction. Between 2027 and 2035, the program would need to scale from multi-module to a system contributing meaningfully to 1,000 logical qubits – which, even with aggressive overhead assumptions, implies hundreds of high-fidelity modules linked at thousands of Bell pairs per second.

Each step in that sequence has been achieved individually in different labs, on different platforms, at different scales. No one has assembled them into a single system.

What This Changes for the Path to a CRQC

Does this announcement alter the timeline to a cryptographically relevant quantum computer? Not measurably.

Diamond-spin quantum computing is not on the near-term path to a CRQC. The leading contenders for fault-tolerant quantum computation at cryptographically relevant scale remain superconducting (Google, IBM, Fujitsu’s own STAR program), trapped-ion (Quantinuum), and neutral-atom (QuEra) platforms, all of which have larger qubit counts, more mature error correction demonstrations, and deeper engineering pipelines. Diamond-spin’s strategic relevance is in quantum networking – linking processors, enabling distributed computation, building repeater infrastructure – and in the long-term possibility of hybrid architectures where high-fidelity diamond-spin modules supplement other platforms. It is not a CRQC candidate itself, but it may become part of a CRQC architecture through networking.

Fujitsu’s prototype is a step toward that long-term architecture. It confirms that an SnV center can be fabricated into a photonic circuit on a chip, packaged into a cryogenic system, and accessed through a software platform. Those are necessary but not sufficient conditions for a modular diamond-spin quantum computer, and the distance between the two is where the engineering work remains.

The SnV physics is promising. The QuTech cooperativity result was a genuine advance. The fabrication demonstrated in this prototype is real. And the scalability case remains, for now, a projection built on laboratory results that have not been assembled into a working modular system. When Fujitsu publishes qubit counts, gate fidelities, and an inter-module entanglement rate, this prototype will be possible to evaluate. Until then, it is a fabrication milestone – and fabrication milestones are worth reporting precisely because they are prerequisites for everything that follows.

Organizations planning PQC migration should not adjust their timelines based on this announcement. The deadlines that drive action – regulatory, contractual, insurance – are set independently of any single hardware result, and that independence is exactly their value.