In short: No single qubit wins — superconducting and trapped-ion lead today on scale and fidelity respectively, photonic bets on room-temperature networking, spin qubits chase CMOS-foundry density, and topological qubits promise error resistance that is still being proven in the lab.
Ask ten quantum hardware engineers which qubit is "best" and you will get eleven answers. The reason is that a qubit is a bundle of tradeoffs across five axes that rarely move together: coherence time (how long the quantum state survives), gate fidelity (how clean each operation is), gate speed, connectivity (which qubits can talk directly), and manufacturability. Optimise one and you usually pay somewhere else. This article walks through the five mainstream modalities, the physics behind each tradeoff, and where the real engineering pain lives.
The five contenders at a glance
Below are representative, order-of-magnitude figures as of the mid-2020s. Treat them as a map, not a datasheet — every lab reports slightly different numbers and they improve each year.
| Modality | Coherence (T2) | 2-qubit gate | Gate fidelity | Operating temp | Headline strength |
|---|---|---|---|---|---|
| Superconducting | ~100 µs | ~20–60 ns | ~99.5–99.9% | ~15 mK | Fast gates, chip fab |
| Trapped ion | seconds–minutes | ~10–100 µs | ~99.9%+ | near room (ions in vacuum) | Fidelity, all-to-all |
| Photonic | no decoherence in flight | probabilistic | loss-limited | room temperature | Networking, no cryostat |
| Spin (silicon) | ~ms (isotopic Si-28) | ~10–100 ns | ~99%+ | ~100 mK–1 K | CMOS density |
| Topological | intrinsically protected (goal) | braiding | unproven | ~tens of mK | Hardware error resistance |
Key fact: A useful figure of merit is gate-operations-per-coherence-time. A superconducting qubit with 100 µs T2 and 40 ns gates fits ~2,500 operations; a trapped ion with 1 s coherence and 50 µs gates fits ~20,000. Fidelity, not raw coherence, is what decides whether those operations are usable.
Superconducting qubits: the industrial frontrunner
Superconducting transmons are lithographed aluminium/niobium circuits behaving as nonlinear LC oscillators, where a Josephson junction provides the anharmonicity that lets you address just two energy levels. They are what IBM, Google, and Rigetti have scaled to hundreds of qubits. The attraction is that they are fabricated with the same deposition and etch tooling used in conventional chip making, and gates run in tens of nanoseconds — microwave pulses, not laser systems.
The cost is the dilution refrigerator: transmons live at ~15 millikelvin, colder than deep space, because their energy splitting (~5 GHz) is only distinguishable from thermal noise near absolute zero. Coherence sits around 100 µs and connectivity is nearest-neighbour on a 2D lattice, so a logical algorithm needs many SWAP gates to shuffle information across the chip. If you want the deep dive, see our companion piece on how superconducting qubits work.
Trapped-ion qubits: the fidelity champion
Here the qubit is a single atomic ion (ytterbium or calcium are common) suspended in vacuum by oscillating electric fields in a Paul trap. The two states are long-lived electronic or hyperfine levels, so coherence stretches to seconds and gate fidelities routinely exceed 99.9% — the best demonstrated two-qubit numbers of any platform. Crucially, ions in a shared trap couple through their collective motion, giving all-to-all connectivity: any ion can entangle with any other without routing overhead.
The drawback is speed and scale. Gates use precisely tuned lasers and take microseconds to tens of microseconds — a thousand times slower than transmons. Packing more than a few dozen ions into one trap gets unwieldy, so companies like IonQ and Quantinuum pursue QCCD architectures that physically shuttle ions between trap zones, trading engineering complexity for modular scaling.
Key fact: Connectivity matters as much as fidelity. On a nearest-neighbour superconducting chip a single non-local gate can expand into a dozen physical operations; trapped-ion all-to-all coupling can make a shallow-looking algorithm run markedly shorter in practice.
Photonic qubits: no cryostat, but probabilistic gates
Photonic qubits encode information in light — polarization, path, or time-bin of single photons. Photons barely interact with the environment, so there is effectively no decoherence "in flight," and the hardware runs at room temperature. This makes photons the natural carrier for quantum networking and the basis of measurement-based schemes pursued by PsiQuantum and Xanadu.
The hard part is that photons also barely interact with each other, so deterministic two-qubit gates are difficult. Linear-optical approaches rely on probabilistic entangling operations plus heavy multiplexing, and photon loss in waveguides and detectors is the dominant error channel. The engineering race is really about near-unity-efficiency single-photon sources and detectors.
Spin qubits: betting on the foundry
Spin qubits store a bit in the spin of a single electron (or hole) confined in a silicon quantum dot, controlled by magnetic or electric fields. Their headline appeal is manufacturing: a spin qubit is roughly the size of a transistor, so in principle you ride the entire investment of the CMOS industry — the same lithography and process control that Synopsys and Siemens EDA tools already model for logic chips. Intel's work on isotopically purified silicon-28 (which removes nuclear-spin noise) has pushed coherence into the millisecond range.
The challenges are uniformity and control wiring. At the atomic scale, tiny fabrication variations shift each dot's resonance, so calibration is per-qubit, and fanning out control lines to millions of densely packed dots is an unsolved interconnect problem. Spin qubits also need sub-kelvin operation, though "hot" variants targeting ~1 K would dramatically ease the cryogenic budget. Teams building these devices often lean on the same mixed-signal verification discipline used for commercial silicon — the kind of rigor our partners at Avecas VLSI design services apply to analog/mixed-signal tapeouts.
Topological qubits: the long shot with the biggest payoff
Topological qubits are the outlier: they do not yet exist as a scalable product. The idea is to store quantum information non-locally in exotic quasiparticles (Majorana zero modes), so that local noise — the thing that plagues every other platform — cannot corrupt the state. Logic gates would be performed by "braiding" these quasiparticles, with error suppression built into the hardware itself rather than bolted on with error-correcting codes.
If it works, the saving is enormous: current estimates suggest fault-tolerant machines on conventional qubits may need ~1,000 physical qubits per logical qubit, a ratio topological hardware could slash. The catch is that unambiguous, controllable Majorana modes have been extraordinarily hard to demonstrate, and claims have been contested. For a deeper look at why error correction dominates the resource budget, see our explainer on quantum error correction.
So which should you learn?
For a career in quantum hardware, superconducting and trapped-ion are where most commercial jobs are today, spin qubits are the bet if you come from a semiconductor/VLSI background, and photonics suits those drawn to optics and networking. The honest engineering answer is that the field is modality-agnostic at the software and error-correction layer, so understanding all five makes you far more valuable than specialising too early.
Want to go from "I can name the qubits" to "I can reason about coherence, fidelity, and error budgets like an engineer"? Explore the CourseTron Quantum Computing track — hands-on, interview-focused, and built by practising semiconductor engineers.
Frequently asked questions
Which qubit type has the longest coherence time?
Trapped ions, by a wide margin — their atomic states can hold coherence for seconds to minutes, versus roughly 100 microseconds for superconducting transmons. Photons are a special case: they do not decohere while travelling, but they are instead limited by loss.
Why do superconducting qubits need to be so cold?
Their energy-level splitting corresponds to microwave frequencies (~5 GHz). At temperatures much above ~15 mK, thermal energy would randomly excite the qubit, destroying the quantum state. The dilution refrigerator keeps thermal noise well below that splitting.
Are topological qubits real yet?
Not as scalable, controllable devices. The underlying Majorana physics is actively researched and some signatures have been reported, but a clearly demonstrated, braidable topological qubit performing logic gates has not been convincingly shown at the time of writing.
Do spin qubits really use the same fabs as normal chips?
Largely yes — silicon spin qubits are made with CMOS-compatible processes, which is their central advantage. The open problems are device-to-device uniformity and routing control lines to huge arrays, not the basic fabrication steps.
References & Further Reading
Training and workshops for teams are also available via our sister brand ChipXpert.