The Hardware Zoo: Six Ways To Build A Qubit

A qubit is an idea before it is a device. Anything that can hold a controllable superposition of two states, be entangled with its neighbours and be measured reliably can serve as one. That is why quantum computers are being built from very different physical systems, from tiny electrical circuits to individual atoms held by lasers.

No approach has won yet, and the trade-offs matter when you read the news. A record qubit count on one platform may mean less than a smaller count on another. This article explains the six main families, what each does well, what holds it back and which results are confirmed and which are not.

Superconducting Circuits

Superconducting qubits are small electrical circuits printed on a chip and cooled until their metal conducts without resistance. IBM describes the operating temperature as roughly a hundred times colder than one degree above absolute zero, which takes large specialist refrigerators.¹ The main advantage is speed. NIST lists quick processing and the ability to reuse established chip-making methods among their strengths, and notes that ion qubits hold their states for longer.²

It is one of the most developed approaches, and the one behind several of the best-known results. Google’s Willow chip, announced in December 2024, has 105 qubits and produced the first clear below-threshold surface code result.³ IBM announced its 120-qubit Nighthawk processor in November 2025, with 218 tunable couplers linking qubits to up to four neighbours in a square grid.⁴ The main obstacles are the large cooling systems the chips require and the fact that, in most current designs, each qubit talks directly only to its nearest neighbours. IBM is developing longer-range connections for its error correction plans.⁵

D-Wave also builds superconducting machines, but they are quantum annealers, which work on a different principle from the gate-based circuits described in this series.⁶ Annealer qubit counts should not be compared with gate-based qubit counts.

Trapped Ions

Trapped-ion machines hold individual charged atoms in place with electromagnetic fields and control them with lasers or microwaves. IBM notes that trapped ions are known for long coherence times and high-fidelity measurements, but run more slowly than superconducting qubits.¹ Because ions can be physically moved and regrouped, qubits that start far apart can still be brought together to interact.

Quantinuum’s Helios uses 98 barium ions that are shuttled around a ring-shaped trap, and its February 2026 preprint reported computations with up to 94 logical qubits, with important qualifiers explained in the error correction article.⁷ IonQ publishes a roadmap targeting 10,000 physical qubits in 2027 and 2 million physical qubits, or 80,000 logical qubits, in 2030.⁸ Those are IonQ’s own targets, not results. The main obstacles for trapped ions are speed and the difficulty of scaling the traps and their control systems.

Neutral Atoms

Neutral-atom machines trap uncharged atoms in grids of tightly focused laser beams, known as optical tweezers, and can rearrange them during a computation. This makes very large arrays possible. In September 2025 a Caltech team reported holding more than 6,100 atoms in a tweezer array with coherence times of about 12.6 seconds.⁹ That is an array of qubits, not a computer running full two-qubit logic at that size, but it shows how far the platform can scale.

Harvard, MIT and QuEra reported a fault-tolerant architecture on up to 448 atoms that operated below the error correction threshold.¹⁰ Infleqtion announced in September 2026 that it had entangled 30 logical qubits on 80 physical atoms, a claim made in a company press release.¹¹ One known obstacle is atoms that escape their traps during a computation; the Harvard team built the detection of lost atoms into its error correction.¹⁰ Scaling full two-qubit logic to the size of the largest arrays is another.

Photonics

Photonic approaches encode qubits in particles of light travelling through optical chips. Light is already the standard medium for carrying information between computers, which makes photonics a natural fit for linking machines. The central difficulty is loss: photons are easily absorbed or scattered, and a lost photon is a lost qubit.

PsiQuantum announced its Omega chipset in February 2025, manufactured at GlobalFoundries. It reported 99.98 percent fidelity for single-qubit state preparation and measurement and 99.72 percent fidelity for links between chips, results also published in Nature.¹² These are results for components, not for a working large-scale computer.

Silicon Spin Qubits

Spin qubits store information in the spin of single electrons held in tiny silicon structures, closely related to ordinary transistors. Their appeal is size and manufacturability. Intel describes its Tunnel Falls chip as built with CMOS (complementary metal-oxide-semiconductor) technology, the standard process for making computer chips, and its research as aimed at very small qubits that can operate at higher temperatures than other approaches.¹³ Intel also notes that commercial systems will need to scale to over a million qubits, which is the argument for reusing chip factories. This series treats the platform as earlier stage than the three above, because none of the error correction results covered in this series comes from spin qubits.

Topological Qubits: Claimed, Not Confirmed

Topological qubits would store information in exotic states of matter called Majorana zero modes, in a way that is naturally protected from many errors. The hope is that such qubits would need much less error correction than other types. NIST’s explainer, updated in May 2026, still describes this kind of qubit as hypothetical.² In February 2025 Microsoft announced a chip called Majorana 1 together with a Nature paper on measurements in its devices.¹⁴

The scientific case is disputed. The paper itself describes its results as consistent with Majorana zero modes, not as proof of them. In the peer review file Nature published with the paper, the journal’s editors added a note that the results do not represent evidence for Majorana zero modes in the reported devices, and that the paper was published for its device design.¹⁵ In March 2025 physicist Henry Legg posted a critique arguing that the method Microsoft uses to identify the topological phase is unreliable.¹⁶ In June 2026 Nature published a formal comment by Legg on that detection method, together with a reply from Microsoft’s team.¹⁷

Microsoft has pressed on. In June 2026 it announced Majorana 2, a chip it says has qubits 1,000 times more reliable than before, and brought forward its target for a scalable machine to 2029.¹⁸ An accompanying preprint reports that the parity its qubits would store survived for about 20 seconds in a new device.¹⁹ Those are Microsoft’s own results and goals, not yet peer reviewed. As of October 2026, no independent group has reported confirming a topological qubit.