客座文章：量子计算的“晶体管时刻”

A new paper demonstrating Grover’s search on a room-temperature diamond processor may echo the transistor’s takeover from vacuum tubes – and raises questions about how much vacuum-dependent quantum hardware survives the same transition.

Guest Post By Prof. Dr. Marius Grundmann, CEO, SAXON Q

Every disruptive technology has a quiet laboratory result that outsiders miss and insiders remember for decades. A newly published demonstration of Grover’s search algorithm on a room-temperature diamond quantum processor is a strong candidate for exactly that kind of moment.

The result is narrow in scope – a three-qubit search on a nitrogen-vacancy (NV) center system – but carries a broad implication. It suggests that quantum information processing is echoing the move classical computing made, out of vacuum-based hardware and into the semiconductor lattice. The physics is different, but the direction is the same.

The real question is how fast it happens – and what it means for the tens of billions of dollars committed to vacuum-dependent modalities.

How Silicon Buried the Vacuum Tube

The precedent is instructive. The first electronic computers, like ENIAC, ran on thousands of vacuum tubes – power-hungry, hot and unreliable enough that keeping one running was a full-time engineering discipline. The transistor, invented at Bell Labs in 1947, did not immediately outperform the vacuum tube. What changed the outcome was not a single leap in performance, but a trajectory.

Transistors miniaturized, integrated and improved on a curve that vacuum tubes structurally could not follow, because a vacuum tube cannot be fabricated by the billions onto a wafer. By 1958, the first integrated circuits demonstrated that semiconductor logic could be manufactured, not just assembled, and the vacuum tube’s fate was sealed. The lesson is that manufacturability and a favorable cost-and-performance curve beat raw early performance every time the two diverge.

Quantum Computing’s Vacuum-Tube Problem

Quantum computing today looks much like classical computing in its vacuum-tube phase – in more ways than one. Trapped-ion systems confine charged atoms using electromagnetic fields inside an ultra-high vacuum chamber. Neutral-atom arrays hold uncharged atoms in a lattice of laser tweezers, also inside ultra-high vacuum, and have not yet produced a multi-qubit Grover search. Superconducting qubits, the basis of most large announced processors, use dilution refrigerators to reach tens of millikelvin – and those refrigerators are themselves vacuum systems, built from nested vacuum-sealed stages.

All three face the same structural irony that doomed the vacuum tube. A wafer-scale fabrication process is only as useful as the roomful of cryogenic or vacuum infrastructure built around every chip it produces. Fabricating the qubit is not the bottleneck. Housing it is.

A Different Path in Diamond

Researchers in Germany ran a three-qubit Grover search – an eight-element unstructured search problem – on a commercial NV-center diamond processor at ambient temperature and pressure, with no vacuum chamber and no cryostat. Grover’s algorithm delivers a proven quadratic speedup over classical brute-force search, and because running it end-to-end exercises a full stack of gates rather than a single operation, it has become a standard way of stress-testing what a processor can actually do.

The calculation qubits were not the NV center’s electron spins but three long-lived nuclear spins – the intrinsic nitrogen-14 nucleus and two nearby carbon-13 nuclei. Shielded by the carbon lattice, they hold their quantum state for several milliseconds, a thousand times longer than the electron spin, which serves only as an optical interface.

The numbers matter. The nuclear qubits reached an average single-qubit Clifford fidelity of 99.90%, with two-qubit subspace gate fidelities averaging 95.7%. The system found a single marked state among eight with an average success probability of 77.3%, and either of two marked states with 87.0% – both well above the best possible classical result for the same number of oracle queries, 37.5% and 46.4%. Superconducting processors at millikelvin temperatures have reported comparable or lower success probabilities. Trapped-ion demonstrations of the same search range from 44 to 69%. The diamond result came at an average temperature of 296.3 kelvin – room temperature – on about 600 watts, rather than the tens of kilowatts a dilution refrigerator and its supporting plant require.

None of this means diamond has “solved” quantum computing. Ion traps still hold the best raw fidelities across all modalities. But the result reframes the comparison. The system behaves like a solid-state ion trap whose “ions” are permanently fixed in a crystal lattice rather than held in place by an apparatus.

Three Quiet Milestones

History rarely announces itself. The parallel to the semiconductor transition has been unfolding for years without much notice, and three moments stand out.

The first is a transistor moment. In 2020, researchers in Leipzig, Germany, demonstrated the first working one-qubit diamond-based quantum computer – proof that the device physics worked outside a research setting.

The second is an integration moment. In 2023, fabrication of NV centers was demonstrated at production scale under Germany’s national quantum computing initiative – the “transistor” manufactured repeatably rather than hand-assembled one defect at a time. A sulfur-doping and activation step during NV center creation lifts the yield of usable and entangled centers to roughly 85%, compared with roughly 1% for earlier, undoped approaches. That figure deserves attention because almost no competing spin-defect platform has published a fabrication yield at all, and yield is the industrial question – how reliably usable qubits can be placed where they are wanted, at scale.

The third is a performance moment. The recent Grover result shows that a commercial, room-temperature semiconductor platform can match the algorithmic performance of vacuum-dependent competitors.

Diamond – The Silicon of Quantum Computing?

Whether NV-center diamond is quantum computing’s silicon – the material system that wins the platform war – is an open question.

Early semiconductor electronics ran on germanium before the industry standardized on silicon, and a comparable shift could happen here. Silicon-vacancy and tin-vacancy centers in diamond have so far worked only at cryogenic temperatures, and other spin centers could eventually prove better. The point is not that any single defect center wins. It is that the category – solid-state spin qubits operating without vacuum or cryogenic infrastructure – now for the first time has a credible performance claim behind it.

What’s at Stake

If semiconductor quantum computing follows even a modest version of that improvement curve, the consequences run in two directions. The first is additive. An ambient-condition, low-power qubit platform can go places dilution refrigerators and vacuum chambers structurally cannot – including inside a vehicle, a mobile robot, a network-edge box or a hospital point-of-care device. That is not a modest expansion of quantum computing’s addressable market, it is a different market. It echoes how the microprocessor created categories, from the personal computer to the smartphone, that vacuum-tube computing was never going to reach.

The second consequence is subtractive, and less comfortable for incumbents. Tens of billions of dollars have been invested in superconducting, trapped-ion and neutral-atom quantum computing on the premise that these vacuum and cryogenics dependent approaches are the winners. If a manufacturable, room-temperature alternative closes the performance gap the way transistors eventually closed it on vacuum tubes, a meaningful share of that capital base is at risk of following vacuum-tube technology into a specialty niche. The same thing happened to CRT displays after LEDs matured. The incumbent did not lose because it was inferior at the moment of comparison. It lost because it could not get onto the semiconductor improvement curve, and the newer technology could.

The Grover search result out of Germany is a small experiment. The pattern it may be repeating is not.