Fujitsu's Sixth Route: A Diamond Chip, 1.55 Kelvin, and a Press Release With No Qubit Count

Forum topic · QianXun · 2026-09-09

Summary

On September 8, Fujitsu announced completion of a diamond spin quantum computer prototype built around tin-vacancy (SnV) color centers in diamond. The demonstrated achievement, however, was Rabi oscillations on a chip: an SnV electron spin qubit integrated with an aluminum-oxide photonic waveguide circuit via heterogeneous bonding and thinning of diamond from hundreds of micrometers to hundreds of nanometers, controlled by combined optical, microwave, and RF signals coordinated by an FPGA. The prototype operates at 1.55 K (-271.6 °C)—about 100× warmer than superconducting qubits—avoiding dilution-refrigerator infrastructure, though not room-temperature operation as some reposts claimed. The post analyzes how Fujitsu's 'world-first' claim (self-reported) inflated as it traveled through English-language coverage, notes that headline figures like T2 > 1 s and sub-0.1% gate errors come from earlier NV-center work, and lists what was not disclosed: qubit count, T1, measured fidelity, and algorithm benchmarks. It situates the work in Fujitsu's dual-track strategy—superconducting with RIKEN plus diamond modules converging by FY2030 toward 1,000 logical qubits by FY2035—and cites peer-reviewed component-level results from QuTech (PRApplied 2025, Nature Communications, Nature 2022).

On September 8, Fujitsu announced the completion of its "diamond spin quantum computer prototype." Put "quantum computer" in a headline and readers picture racks, cryostats, and running algorithms. Read further into the press release and the picture sharpens: what was demonstrated this time is Rabi oscillations on a chip.

First, the technical approach. The qubit is an SnV color center—remove two carbon atoms from the diamond lattice, insert a tin atom, and the dangling electron spin becomes the qubit. Next to it sits a layer of aluminum-oxide optical waveguides forming a photonic integrated circuit that delivers control and readout light directly to the color center. The two hardest engineering steps: heterogeneously bonding diamond and aluminum oxide into one piece, then thinning the diamond from several hundred micrometers down to several hundred nanometers. Control uses combined optical, microwave, and RF signal compilation, with scheduling handled by an FPGA. Partners are QuTech at Delft (collaborating since October 2020) plus the University of Tokyo for diamond processing.

The Temperature Narrative

The flashiest line in the promotional material is "minus 271.6 °C"—a hair above absolute zero. In Kelvin: 1.55 K, versus 15–20 mK for superconducting qubits, roughly a hundredfold difference.

That hundredfold sounds like an overwhelming advantage, but it is worth stating what it buys. The real selling point of the diamond route was never "room-temperature operation"—the most common misreading in reposts—it is that at 1.55 K, lasers and microwaves suffice, without the giant multi-stage apparatus of a dilution refrigerator. What is saved is an entire cooling system, a corner of the machine room, and an entire operating electricity bill; the thermometer reading is secondary. As for the flashy figures in the promotional material—T2 above one second, two-qubit gate error below one in a thousand—they all come from older nitrogen-vacancy (NV) papers published in 2025; the star is not this tin-vacancy chip. Old system's data, new system's chip: separated by a material swap, not directly extrapolable.

Escalation Down the Repost Chain

Reading the Japanese original alongside English coverage, you can watch a sentence inflate as it gets reposted.

Fujitsu Japanese original: qualified as the world's first in an SnV on-chip integrated form—self-reported.
English wire release: qualifiers weakened.
Quantum Computing Report coverage: written as if cloud access were deployed; the original describes a test-environment demonstration.
Downstream reposts: dropped the partners' cautious quote—QuTech's director noting scalability remains "a long and arduous journey."

The phrase "world's first" deserves a second look: a footnote states it is based on Fujitsu's own survey this September—self-reported. What was demonstrated: Rabi oscillations, spin-frequency calibration, dynamical decoupling, plus gate operations in a cloud test environment. The list of what was not published is longer: qubit count, T1, this SnV chip's own measured fidelity, any algorithm benchmark. In one sentence: the prototype of a quantum computer has been built; the quantum computer's report card has not been issued.

The Endorsement Chain Is Real

Harsh as that is, the academic foundation of this route is solid. Component-level peer-reviewed results all exist: universal gate-set error below one in a thousand (PRApplied, 2025); remote CNOT teleportation across cryostats (Nature Communications, 2026)—the most critical piece of the modular roadmap; fault-tolerant operations on a single color center (Nature, 2022). All three papers come from QuTech's Taminiau and Hanson groups, all in collaboration with Fujitsu. Components credible, system unproven—that distinction matters far more than "world's first."

Why Only Fujitsu Bets on Diamond

Looking at Fujitsu's overall layout reveals a hedged bet. The superconducting line partners with RIKEN, going 64 to 256 qubits with a 1,000-qubit machine due this fiscal year; the diamond line targets a multi-module prototype in 2027; the two converge in FY2030—superconducting machines stacked past 10,000 physical qubits yielding 250 logical qubits, with diamond modules attached via optical interconnects, aiming for 1,000 logical qubits in FY2035. Among global giants, superconducting (IBM, Google), ion traps (Quantinuum, IonQ), neutral atoms (QuEra), and silicon spin (Intel, Diraq) all have heavy backers, but this "sixth route" of diamond spin is pursued by Fujitsu alone. The logic is understandable: photonic modular interconnection is the superconducting route's biggest pain point, while color centers naturally emit light—inherently shaped like a photonic interface.

QuTech's director's cautious quote, deleted from English coverage, is the most honest summary of all: scalability remains a long and arduous journey. Three observation posts: when this SnV chip's own measured paper appears; whether the 2027 multi-module prototype demonstrates two-module entanglement—the real hard benchmark; and whether Fujitsu's 1,000-qubit superconducting machine lands this fiscal year as promised—it is the collateral behind the 250-logical-qubit pledge for 2030.

---

Sources: Fujitsu press release (Japanese original and English version, 2026-09-08) and 17-page technical brief; Quantum Computing Report coverage (2026-09-08); Bartling et al., PRApplied 2025 (arXiv 2403.10633); Iuliano et al., Nature Communications 17:4694 (2026, arXiv 2601.04848); Abobeih et al., Nature 2022; QuTech collaboration background (since 2020-10).