양자 뉴스: Monday, July 13, 2026

Monday, July 13, 2026

Three stories this week, and none of them is a new qubit-count record — they're all about the unglamorous engineering that decides whether quantum computing scales and whether anyone outside a national lab ever gets to touch it. Start with access. QuTech, the Delft outfit that's been building superconducting hardware for a decade, put a 17-qubit processor called Tuna-17 online through its Quantum Inspire cloud — free, worldwide, no usage caps, over a hundred thousand shots per job. Seventeen qubits won't factor anything, and QuTech didn't publish gate fidelities, which is the number you'd actually want to see. But the point isn't the spec; it's that a student in Lagos or Lima can run circuits with mid-circuit measurement (peeking at some qubits partway through without collapsing the rest) on real hardware without a grant or an NDA — and the whole thing runs on a European supply chain (TNO, Qblox, QuantWare, Delft Circuits) instead of being rented from IBM. Second, manufacturability. Diraq, working with the Belgian foundry imec, ran eight silicon spin qubits as a linear array on a device made entirely on a standard 300-millimeter CMOS line — the same industrial process that stamps out the chip in your laptop — with one- and two-qubit gate fidelities above 99% and readout above 99.9% (published in Nature Communications). Eight is small. The claim that matters is that scaling up from their earlier two-qubit cell cost them no coherence, because 'we can make it in a normal fab' is the difference between a physics result and an industry. Third, the wiring problem — the least glamorous and most load-bearing constraint in the field. Every qubit normally needs its own control lines, which is how you get an architecture that works fine at 50 qubits and physically cannot work at a million. EeroQ, a company that floats electrons on superfluid helium, published in Physical Review Applied a control chip that shuttled electron packets a collective tens of kilometers with zero detectable charge loss over a billion cycles — using a scheme they argue could steer a million qubits with fewer than fifty control lines. It's a transport demo, not a computer; there are no qubits being computed on yet, and they say so plainly. But it's the exact bottleneck everyone else is quietly terrified of. The through-line: open access, ordinary fabs, and fewer wires. Records make the headlines, but this is the boring plumbing that decides whether quantum ever becomes something you learn to use instead of something you only read about.