Quantum Hardware & Control Engineer

A qubit is a physical object with two addressable levels

Every quantum computer, whatever the brochure says, is a collection of physical systems that each have two energy levels we can isolate, drive, couple and read out. Pick two levels of anything — an atom, an electron spin, a superconducting circuit, a photon's path — and you have a candidate qubit. The entire engineering discipline of quantum hardware is about making those two levels (a) stay what you set them to for long enough, (b) change only when you tell them to, and (c) interact with a neighbour on demand.

DiVincenzo wrote this down in 2000 as five criteria, and they are still the checklist a hardware engineer uses to judge a platform: well-defined qubits that scale; the ability to initialise them; coherence times long compared with gate times; a universal set of gates; and qubit-specific measurement. Two more cover communication: converting stationary qubits to flying ones and transmitting them. Notice that the criteria ask for a system that can scale but set no qubit count and say nothing about speed. Those come later, and they are where the modalities diverge.

The five modalities, by what the qubit physically is

Superconducting circuits. The qubit is the lowest two levels of a nonlinear LC oscillator, where the nonlinearity comes from a Josephson junction. The dominant design, the transmon, resonates at a few gigahertz and is driven with shaped microwave pulses. It lives at about 10–20 millikelvin at the bottom of a dilution refrigerator. Gates are fast — tens of nanoseconds for single-qubit, a few hundred nanoseconds or less for two-qubit — but coherence is short, typically tens to hundreds of microseconds, with the best reported devices approaching a millisecond. Qubits are lithographically fabricated, which is both the attraction (chips, foundries) and the problem (every qubit is slightly different). Kjaergaard et al. is the standard tour of the design space.

Trapped ions. The qubit is two internal levels of a single atomic ion — hyperfine or optical states — held in vacuum by radio-frequency electric fields and manipulated with lasers (or, for some species, microwaves). Every ion of a species is identical, so there is no fabrication spread. Coherence times are seconds or longer, and two-qubit gate fidelities above 99.9% have been reported, the best of any modality. The price is speed: gates take microseconds to hundreds of microseconds, and the shared motional modes that mediate entanglement get crowded beyond a few tens of ions in one chain. Scaling means shuttling ions between zones or linking traps with photons.

Neutral atoms. The qubit is again two atomic levels, but the atom is uncharged and held in an optical tweezer — a focused laser beam. Arrays of hundreds to over a thousand atoms have been assembled, and because tweezers can be moved, connectivity is reconfigurable mid-circuit. Entanglement comes from the Rydberg blockade (Module 5). Coherence is long, gates are sub-microsecond to microsecond, but loading and rearranging atoms is slow, atoms are sometimes lost, and the whole array is typically run as one slow cycle.

Photons. The qubit is a property of a single photon — which of two waveguides it travels in, or which time slot. Photons barely interact with anything, which gives room-temperature transport and no decoherence from the environment, but also makes deterministic two-photon gates impossible with linear optics. Photonic schemes therefore use probabilistic gates heralded by detectors, or measurement-based computing on large entangled resource states. The dominant error is loss: a lost photon is an erased qubit. Single-photon detectors usually need cryogenic cooling even if the chip does not.

Spins in semiconductors. The qubit is the spin of an electron (or hole, or nucleus) confined in a silicon or germanium quantum dot, controlled with microwave magnetic or electric fields and coupled to neighbours through the exchange interaction. The footprint is tiny — tens of nanometres — and the fabrication is close to CMOS, which is the whole bet: if it works, it inherits the semiconductor industry. Two-qubit fidelities above 99% were reported by several groups in 2022, and some devices operate above 1 kelvin, which relaxes the cryogenic budget enormously. The unsettled questions are uniformity across millions of dots and the control-wiring fan-out.

Topological qubits deserve a note because they are in the news: the proposal is to store information non-locally in Majorana zero modes so that local noise cannot reach it. As of today the existence of a working topological qubit is not established to the field's satisfaction; treat claims with the scepticism you would apply to any single-group result that has not yet been reproduced.

How to compare them honestly

A useful way to compare is a ratio, not a single number: how many gates fit inside one coherence time. A superconducting qubit with 100 µs coherence and 30 ns gates fits about 3,000 single-qubit operations; an ion with 1 s coherence and 10 µs gates fits 100,000. By that ratio ions win — but a superconducting processor completes a short circuit, readout included, in roughly the time an ion trap spends on a few gates, so for workloads that need millions of shots the arithmetic flips. Connectivity is the third axis: ions and atoms offer all-to-all or reconfigurable coupling; superconducting and spin devices are fixed lattices and pay SWAP overhead for everything non-local.

No modality leads on all three, which is why the field has not converged and why your mission table needs a column for the obstacle.

What to remember

• A qubit is any physical two-level system you can isolate, drive, couple and read; DiVincenzo's criteria are the checklist.
• Superconducting: fast, fabricated, short-lived. Ions: slow, identical, long-lived, best fidelities. Atoms: large reconfigurable arrays, slow cycle. Photons: loss-limited, probabilistic gates. Spins: tiny, CMOS-like, uniformity unproven.
• Compare gates-per-coherence-time and wall-clock throughput separately; they rank the platforms differently.
• Topological qubits are a proposal with contested evidence, not a shipping modality.