40. Trapped-Ion Quantum Computing
40.1Trapped ions
The qubit is two internal states of a positively charged atom: hyperfine ground states (¹⁷¹Yb⁺, ⁴⁰Ca⁺) or optical excited states. Atoms are identical by construction — no fabrication variance, no dielectric defects — so every ion in every laboratory is the same qubit. State of the art: two-qubit gate fidelities at 99.9% and above (Quantinuum's H-series), the highest published two-qubit fidelities of any commercial platform, with coherence limited only by ambient field noise and reaching seconds to minutes. Chains of tens of ions sit in a linear trap; today's machines run tens of qubits with full connectivity inside the chain. The cost of all this quality is speed — everything is slower (40.6, 40.7).
40.2Electromagnetic traps
Earnshaw's theorem forbids static fields from confining a charged particle, so the Paul trap uses an oscillating RF quadrupole field whose time-averaged potential is confining. Ions repel each other via Coulomb force and crystallize into a chain spaced ~5 µm apart. The chain's collective vibrations — quantized normal modes, or phonons — are shared by all ions; they are not noise to be eliminated but the data bus that entangling gates ride (40.6). Surface traps, with electrodes lithographed onto a chip, are the scaling path: they allow ions to be shuttled between storage, gate, and readout zones. Trap design is micromachining plus RF engineering, not cleanroom quantum fabrication.
40.3Laser cooling
Gates need the shared motion nearly frozen. Doppler cooling: a red-detuned laser makes the ion absorb preferentially when moving toward the beam, dumping a momentum quantum each cycle — down to millikelvin. Sideband cooling then resolves individual motional modes and removes phonons one by one, reaching the motional ground state required for the highest-fidelity gates. Cooling runs between computations and refreshes the bus that gates heat up. The laser system is a serious engineering stack — phase-locked, sub-kHz-linewidth, frequency-stabilized lasers plus classical optics racks — and it is the platform's equivalent of control electronics: expensive, software-defined, and central to performance.
40.4State preparation
Initialization by optical pumping: a polarized laser drives the ion until population funnels into one ground state and stays there — a dark state it cannot absorb from. Fidelity exceeds 99.9% in microseconds, and combined with sideband cooling (40.3) it prepares the register's all-zeros state. DiVincenzo criterion 2 is satisfied almost trivially — in contrast with superconducting chips, where initialization is either "wait for T1" or an actively engineered reset. The general pattern is worth naming: ion platforms spend time to obtain near-perfect primitives; superconducting platforms spend hardware to obtain speed. Neither is free, and the choice shapes everything downstream, including error correction.
40.5Laser control
Single-qubit gates are Rabi rotations on the qubit transition, driven by a focused laser beam (optical qubits) or a microwave horn (hyperfine qubits); durations ~1–100 µs. Addressing individual ions in a 5 µm-spaced chain is an optics problem: tightly focused beams, acousto-optic deflectors for fast beam switching, and careful polarization control at every site. Gate phase is set by optical path length, so phase noise in the beam paths translates directly into dephasing errors — active path stabilization is a genuine engineering discipline here. Control complexity per qubit is higher than superconducting platforms', but it is optical and room-temperature rather than cryogenic microwave.
40.6Entangling gates
Ions do not interact directly; entangling gates ride the shared phonons of 40.2. Cirac–Zoller (1995) mapped qubit state onto motion and back; Mølmer–Sørensen applies a bichromatic field that couples the whole chain to an effective XX-type interaction independent of the motion's exact state — this scheme and derivatives power modern machines. Durations ~10–600 µs; fidelities 99.9%+. The prize is connectivity: any pair of ions in the chain can gate directly — all-to-all — which eliminates most of the SWAP overhead that lattice platforms pay (44.4) and makes small circuits dramatically more efficient. The price: gates and heating disturb the bus, limiting practical chain length to tens of ions.
40.7Measurement
State-dependent fluorescence: drive a cycling transition that only |1⟩ scatters, and the ion either glows (thousands of photons) or stays dark, collected by a photomultiplier or camera. Fidelity exceeds 99.9%, each ion in the chain is individually resolved, and crosstalk is negligible. This is the most ideal projective measurement of any platform — nearly quantum non-demolition. The cost is speed: detection takes ~100 µs to 1 ms, plus recooling afterward. For algorithms this is fine; for error correction (Part X) it is the bottleneck, because QEC demands fast repeated cycles of reset, gate, and measure — so the ion platform's fault-tolerance rate is limited by its best feature, its measurement.
40.8Scaling
One chain shares one motional bus, and modes crowd together as ions are added — tens of ions is the practical chain limit. Three scaling paths exist. Shuttling: physically move ions between storage and gate zones through junction electrodes — Quantinuum's QCCD architecture works this way, and its shuttling-based machines with 50+ qubits are cloud-accessible today. Photonic links: entangle separate chains remotely via photons emitted and collected through cavities — demonstrated at small scale, currently slow, potentially unlimited. Larger two-dimensional traps: still exploratory. The open question is arithmetic: whether gate speed plus shuttling overhead can deliver the fast cycle times fault tolerance wants, or whether fidelity headroom must buy back the time.