Silicon Qubits Aren’t Winning Yet. Manufacturing Could Change That

The narrative often suggests an inevitable progression: silicon, the bedrock of the digital age, will seamlessly transition into the quantum era, transforming sand into transistors, and then into spin qubits. This clean, linear arc, however, misrepresents silicon’s actual standing in the quantum landscape. Silicon spin qubits aren’t vying for a “physics contest” they never claimed to enter. Instead, their appeal lies in a manufacturing proposition, presented as a materials story—a distinction critical to whether this gamble pays off.

As detailed in Ajay P. Manuel’s recent feature for The Quantum Insider, the foundational ideas emerged in 1998. Daniel Loss and David DiVincenzo proposed using the spin of a trapped electron as a qubit, while Bruce Kane, then a postdoc at the University of New South Wales, suggested implanting phosphorus atoms into silicon to read the donor’s spin. Both proposals hinged on the semiconductor industry’s manufacturing prowess being adaptable to quantum hardware. This wager remains active.

Bridging the fidelity gap, awaiting the count leap

Silicon proponents deserve credit; recent figures are concrete, not speculative. Silicon Quantum Computing reported 99.99% two-qubit fidelity in a 2025 Nature paper, matching IonQ’s trapped-ion record. Separately, Diraq and imec, in a September 2025 Nature publication, achieved over 99% two-qubit fidelity using randomly selected devices from a standard 300mm industrial wafer and imec’s existing process flow. This latter detail is significant: random selection from a production wafer implies a yield claim, a metric deeply understood in fabrication facilities.

Thus, the fidelity gap that hindered silicon for a decade has largely closed. However, the count gap persists. Silicon spin platforms have demonstrated only about a dozen qubits at a meaningful scale. In contrast, IBM’s Condor featured over a thousand physical qubits, whereas neutral-atom and trapped-ion systems operate in the hundreds to thousands. A disparity of this magnitude is not a trivial error; framing silicon as a current contender for raw scale overlooks present realities in favor of PR cycles.

The overlooked nuances of the silicon story

Isotopic purification is often presented as a resolved issue, which it largely is. Still, the necessity of this step reveals complexities often omitted from simplified narratives. Natural silicon contains approximately 4.7% silicon-29, an isotope with a non-zero nuclear spin that generates fluctuating magnetic fields, thereby scrambling electron spin coherence. Removing silicon-29 dramatically improves coherence times.

This requirement complicates the manufacturing argument: purified silicon-28 substrate is not a volume product of the trillion-dollar semiconductor industry, which optimizes for natural-abundance silicon. Consequently, the claim that quantum technology can simply inherit existing infrastructure comes with a significant asterisk, as isotopically enriched feedstock is a specialized, not commodity, input.

The true competitive landscape

For fairness, the strengths of other modalities must be acknowledged, as each surpasses silicon in areas where silicon currently falls short. Trapped ions from Quantinuum and IonQ maintain two-qubit fidelities of 99.97% to 99.99% with physical-to-logical overheads as low as 2 to 1, representing the field’s highest efficiency.

Superconducting platforms lead in raw qubit count and mature control stacks, albeit with a demanding 105 to 1 overhead per logical qubit, prompting IBM’s shift towards modular, error-corrected designs like Heron. Neutral atoms have recently garnered significant institutional backing, with Google launching its own program alongside its QuEra stake, prioritizing scaling potential over current fidelity.

Against this backdrop, silicon’s honest position is one of trajectory, not leadership. The first logical-qubit operations in silicon only appeared in early 2026, years behind trapped-ion and neutral-atom equivalents. A January 2026 PRX Quantum benchmarking study found that fidelity still degrades as circuit depth and qubit count increase simultaneously. This indicates that impressive single- and two-qubit numbers do not yet translate to scalable performance. Best-case fidelity on a handful of qubits is a data point, not a functional machine.

The bottleneck: An engineering challenge, the core of the bet

Silicon’s compelling argument is this: the remaining hurdles are no longer physics problems. Challenges include routing control and readout to thousands of qubits without exceeding the wiring budget of a dilution refrigerator, integrating cryogenic control electronics directly with qubits, and ensuring wafer-sample uniformity across full production runs. These are manufacturing problems. As the feature concludes:

“None of these are physics problems in the way isotopic purification or single-electron control once were. Rather, they are engineering and manufacturing problems, which is exactly the kind of problem silicon’s inherited industry has spent seventy years getting good at.”

This represents the strongest case for silicon, and it is genuinely robust. Diraq’s “hot-qubit” results demonstrated 98.92% two-qubit fidelity at 1 Kelvin, which is ten times warmer than superconducting qubits tolerate, thereby eliminating one of the field’s most costly cooling requirements. Silicon qubits are also approximately a thousand times smaller than superconducting transmons. The combination of small size, warm tolerance, and fab compatibility is a unique advantage among competing modalities.

The issue at hand is straightforward: a shorter path from lab to fab is only victorious if the destination is reached before a competitor’s longer path. Currently, silicon remains at a dozen qubits while rivals field thousands. The infrastructure argument is sound, but it remains a promise yet to be fully realized.