Strip one electron off an atom, hang it in nothing on an electric field oscillating millions of times a second, and you get the most accurate qubit in the business — and a single speck of matter that is, at once, a quantum memory good for minutes, a processor with all-to-all wiring, and a node that can fire its state down a fiber.
Part of the How to Build a Quantum Computer series.
The previous post in this series ended on a confession: a superconducting qubit is a circuit pretending to be an atom. This post is about the platform that refuses the pretense and catches a real one. Take an ordinary atom, strip a single electron so it carries a net charge, and hold that ion suspended in vacuum on nothing but electric fields. What you get is, by several measures, the most precise qubit anyone has built — and a quietly remarkable economy of function. The same lonely, levitating atom can store a quantum state for minutes, compute on it with any other ion in the trap, and hand it to a photon you fire down an optical fiber to a node across the city. Most platforms make you pick one of those jobs. The trapped ion just shrugs and does all three. Let us take the shrug seriously, because the physics that makes it possible is genuinely beautiful — and the bill it hands you is genuinely steep.
Start where every platform must: what is the two-level system? For a superconducting qubit it was a fabricated circuit; here it is the genuine article. The qubit is a single atomic ion — one atom of, say, ytterbium or calcium or barium, with one electron removed so it carries charge $+e$ — and the two states $\vert 0\rangle$ and $\vert 1\rangle$ are two of that atom’s own internal electronic levels. Nothing is being faked. The discreteness that a transmon had to engineer with a Josephson junction comes free, because an atom’s electrons genuinely can only sit on discrete energy levels. Pick two of them and you have your bit.
Which two? There are two standard choices, and the distinction matters later.
Hold onto that one-second number; it quietly draws a line we will return to when the coherence records arrive. Here is the part that fabricated qubits openly envy. Every $^{171}\mathrm{Yb}^+$ ion is fundamentally identical to every other — not “matched to within manufacturing tolerance,” but the same, because atoms are not manufactured. Two transmons coming off the same wafer have slightly different frequencies and need individual calibration; the field calls them snowflakes. Two ions of the same isotope have exactly the same transition frequencies, set by nature to a precision no foundry will ever touch. (The honest footnote: the atoms are identical, but their environments are not. Stray fields, crosstalk, micromotion, and position-dependent heating still vary ion to ion, so nobody actually skips calibration — they just start from a far better place.)
So the qubit itself is the easy part: nature supplies it, pre-matched, for free. Everything hard is about holding the thing still and talking to it without knocking it loose.
Why you cannot just build a bowl. Your instinct, holding a charged particle, is to surround it with electrodes shaped like a bowl and let it settle at the bottom. Nature forbids it. In the empty space between your electrodes the electric potential obeys Laplace’s equation $\nabla^2\Phi = 0$, and a function with no sources has no local minimum — every “bowl” is secretly a saddle, curving up along one axis and down along another. This is Earnshaw’s theorem: no arrangement of static charges can trap another charge in stable equilibrium. Push down in two directions and the potential springs a leak in the third, and your ion rolls out.
The trick: spin the saddle. Wolfgang Paul’s insight, which shared the 1989 Nobel Prize in Physics
quadratic near the center, so the ion oscillates at a slow secular frequency $\omega_{\text{sec}}\ll\Omega_{\text{RF}}$ with a small fast jitter (micromotion) riding on top. A linear Paul trap builds this from four parallel rod electrodes that form a radial quadrupole to squeeze the ion sideways — the RF drive sits on one opposing pair, the other held at RF ground — plus a pair of DC endcaps to push it inward along the axis. Several ions, all positive, repel one another and string out along the trap axis into a linear chain whose shared vibrations every ion feels at once — the canonical picture of trapped-ion computing, and the subject of Figure 2.
Cooling. A trapped ion still jiggles thermally, and that motion is the very thing the gates will use, so it must be cold and well-defined. Lasers do the cooling. Doppler cooling — a beam tuned just below an atomic transition, so the ion preferentially absorbs photons when moving toward the beam and gets nudged to a stop — brings it to millikelvin scales; resolved-sideband cooling then removes motional quanta one at a time until the ion sits essentially in its motional ground state, an effective kinetic temperature of tens of microkelvin. This is a temperature of the ion’s motion, not of a surrounding bath, and it is reached with lasers, not a dilution fridge. The trap sits in ultra-high vacuum, around $10^{-11}$ mbar, but the apparatus around it runs at room temperature. Vacuum emptier than low orbit, lasers fussing over it around the clock, chilled in its motion to within a whisker of absolute zero: comfortably the most pampered atom in the universe.
Single-qubit gates are the easy half. Flip $\vert 0\rangle\leftrightarrow\vert 1\rangle$ on one ion with a resonant pulse: microwaves (or a pair of laser beams driving a stimulated Raman transition) for a hyperfine qubit, a narrow laser for an optical qubit. Tightly focused beams, or distinct microwave near-fields, address one ion at a time.
Two-qubit gates are where the genius lives — and where the shared vibration of Figure 2 pays off. You cannot entangle two ions by pushing them at each other directly; they are micrometers apart and you must not heat them. Instead you use their collective motion as an intermediary. The chain’s normal modes — the whole string sloshing in lockstep (the center-of-mass mode), or stretching and breathing — are each a quantized harmonic oscillator whose excitations are the phonons of Figure 2. The trick, due first to Cirac and Zoller in 1995
Here is the mechanism, and it is worth slowing down for. Illuminate the two target ions with a bichromatic field — two laser tones straddling the qubit frequency, one sitting near the red motional sideband and one near the blue, each detuned by a small amount $\delta$ from its sideband. (The detuning $\delta$ from each sideband is tiny — tens of kHz against the MHz-scale sideband offset $\nu$; Figure 3 has the geometry.) Together they produce a force on each ion proportional to its qubit operator $\hat\sigma_x$. Conditioned on the joint spin state, this force pushes the shared motional mode in different directions, sending it on a loop through phase space — the plane of the mode’s position and momentum. The detuning is chosen so that after one gate time $t_{\text{gate}} = 2\pi/\delta$ the loop closes: the motion returns exactly to where it started and disentangles itself from the qubits, but the loop has swept out an area, and that enclosed area becomes a geometric phase stamped on the joint spin state. The collective effect is captured by one almost suspiciously tidy operator,
\[H_{\text{MS}} \;\propto\; \hat S_x^{\,2}, \qquad \hat S_x \equiv \sum_i \hat\sigma_x^{(i)},\]the square of a total spin. Square a sum and out falls a cross term — one ion’s $\hat\sigma_x$ times the other’s — and that is the entangler, turning a separable product state into a Bell state. Pause on what just happened: two laser tones and a single shared wobble of the chain have welded the fates of two atoms that never touched. The deep virtue, and the reason the Mølmer–Sørensen gate displaced its predecessor, is that, because the motion returns to its starting point, the final entanglement barely depends on how many phonons were in the mode to begin with. The bus can be a little warm and the gate still works. Figure 3 sketches it.
The payoff of routing entanglement through a shared bus is all-to-all connectivity: any pair of ions entangles directly, wherever they sit in the chain — a structural advantage we will weigh against the platform’s structural curse, slowness, once we have counted its strengths.
Four things set trapped ions apart — record gate fidelity, absurd coherence, clean single-shot readout, and all-to-all wiring — and in each one the precision of the numbers is the whole point. Weigh them exactly, and resist the urge to round up.
Fidelity that strains belief. A single-qubit gate is the operation ions do most flawlessly. Back in 2014 the Lucas group at Oxford demonstrated single-qubit gates on $^{43}\mathrm{Ca}^+$ with an error of about $1\times10^{-6}$ — “six nines,” one slip in a million
Keep the single-qubit and two-qubit records strictly apart, because the two-qubit gate is much harder. Its lab record has marched from $99.9\%$ in 2016
Coherence measured in minutes. Where a transmon forgets in a fraction of a millisecond, an ion remembers for an almost comic length of time. A single $^{171}\mathrm{Yb}^+$ ion at Tsinghua (Kihwan Kim’s group) held a quantum state coherent for more than ten minutes — about $660$ s — in 2017
One detail makes the feat fair rather than miraculous, and it pays off a promise from the very first section. These records are all hyperfine clock qubits, whose two states are both stable ground-state sublevels — so there is no spontaneous-decay floor to fall through. An optical qubit cannot do this: its upper state is the metastable level from Figure 1, and that $\sim 1$ s lifetime caps its coherence no matter how quiet the lab. The two encodings really are different beasts, and this is where the difference bites. (A 2026 preprint from the same Tsinghua group stretches a two-ion encoding to an extrapolated $\sim 10.5$ hours
Readout you can trust in a single shot. Measuring an ion is almost embarrassingly direct. Shine a detection laser tuned to a fast cycling transition: if the qubit sits in one state it scatters photons by the thousands and the ion glows visibly bright; if it sits in the other it stays off-resonant and dark. This state-dependent fluorescence is read in a single shot: point a sensitive camera or photomultiplier at it, count photons for a few hundred microseconds, and bright versus dark hands you the answer — no averaging over many runs, no ambiguous analog voltage to threshold — at fidelities around $99.9\%$. One atom, one glance, one bit.
All-to-all connectivity. Because every ion couples to the same shared modes, any pair can be entangled directly — no relaying a state down a line of neighbors. On a nearest-neighbor chip, entangling two far-apart qubits means a parade of SWAP gates to walk them together and back; an all-to-all machine skips that bookkeeping entirely, so the same algorithm compiles to a shallower circuit with fewer operations — and fewer operations mean fewer chances to fail. The advantages compound.
Taken together — record fidelities, minutes of coherence, single-shot readout, and full connectivity — these are exactly the properties you want if you care about quality of computation over raw speed and scale.
Now settle the bill, because the same physics that buys all that precision charges for it.
The gates are slow. A high-fidelity two-qubit ion gate takes microseconds — sometimes many. The careful 2016 record reached its $99.9\%$ fidelity at a $100\,\mu\text{s}$ gate time; the very same study, pushed to its fastest $3.8\,\mu\text{s}$ gate, shed almost all of that quality and fell to about $97\%$
which lands in microseconds. No clever pulse erases that scale entirely. None of which makes slow gates secretly a virtue — they are a genuine cost. But the honest counter is subtler and more interesting: what ultimately matters is not raw gate speed but how many useful operations you complete before an error creeps in, and very long coherence times multiplied by very high fidelities buy you an enormous number of those. It is the tortoise — slower per step, but far fewer missteps over the long haul that a real quantum algorithm demands.
Scaling one chain is hard. You cannot simply keep adding ions to a single string. The more ions share a trap, the denser and softer their motional spectrum becomes: modes crowd together, become harder to address individually, heat more easily, and slow the gates further. Today’s leading single-chain machines hold only on the order of a hundred ions. Two escape routes are under active construction, and they map neatly onto the field’s two leading companies. The first is the QCCD (“quantum charge-coupled device”) architecture: physically shuttle ions between dedicated storage and interaction zones on a segmented chip, so every gate happens in a small, well-behaved group of two or three ions — Quantinuum’s bet. There is no free lunch in this: every shuttle reheats the ions’ motion — the very thing the gates need kept cold — so these machines co-trap a second “coolant” species and sympathetically re-cool the qubits between gates, trading a connectivity win for a transport-and-cooling problem. The second is photonic interconnect: build many small, excellent traps and stitch them into one modular machine using photons — the same ion-emits-a-photon trick that, as we are about to see, also makes ions natural network nodes. This is IonQ’s bet.
Lasers and vacuum, everywhere. A trapped-ion system is an optics cathedral. Each species needs a small orchestra of frequency-stabilized lasers — for cooling, state preparation, gates, and readout — plus the ultra-high vacuum ($\sim 10^{-11}$ mbar) that keeps stray gas molecules from knocking the ion out of the trap, plus electrode surfaces clean enough to suppress “anomalous heating,” the stubborn, not-fully-understood noise that warms the motional modes the gates depend on. There is no dilution fridge, which is a real relief next to superconducting; but “room-temperature vacuum chamber wrapped in lasers” is not exactly a laptop either. Different cathedral, comparable upkeep.
The trapped-ion scoreboard is less about raw qubit counts — the headline-grabbing flex of the superconducting world — and more about fidelity, error correction, and the first real steps toward networking. Read each result for what it is.
Quantinuum. Honeywell’s quantum unit merged with Cambridge Quantum to form Quantinuum, and its QCCD H-series has spent years quietly setting records — particularly in quantum volume, a whole-machine benchmark that rewards quality and connectivity rather than mere qubit count. In November 2025 it launched Helios: 98 physical qubits, its first machine built on $^{137}\mathrm{Ba}^+$ rather than ytterbium (barium’s visible $493$ nm light is kinder to optics and fiber than ytterbium’s ultraviolet), with all-to-all connectivity, a single-qubit fidelity of $99.9975\%$ and a two-qubit fidelity of $99.921\%$ across all pairs
Earlier, in 2024, a Microsoft–Quantinuum collaboration ran error correction on H2 hardware and reported logical error rates 11 to 800 times below the matching physical baselines, across experiments using up to 12 logical qubits — results since validated in a peer-reviewed 2026 paper
IonQ. The other commercial heavyweight, IonQ pursues the modular, photonically-networked path. It long ran on $^{171}\mathrm{Yb}^+$ and has since moved to barium, whose friendlier wavelengths ease both optics and the photonic links its architecture depends on; its barium “Tempo” system reached a milestone benchmark in 2025
The academic record-setters. Behind the companies sit the labs that set the fidelity ceilings — Oxford’s single-qubit results at the $10^{-7}$ level
A trapped ion is a slow, fussy, ferociously precise qubit — and, almost uniquely, a shape-shifter. Store a state and the ion is a quantum memory; entangle two and it is a processor; let one breathe out a photon and it is a network node. The same charged speck of matter, held in nothing, does all three jobs we normally build separate machines for.
That networking role is not a footnote here — it is a real strength, and the reason this series keeps one eye on whether a qubit can network as well as compute. A superconducting qubit speaks in microwave photons that cannot travel down an optical fiber, so linking two of them over distance demands a lossy microwave-to-optical translation step that nobody has yet made work well. An ion has no such problem: it natively emits an optical photon entangled with its own internal state. Catch that photon, send it down a fiber, interfere it with a photon from a distant ion, and a successful measurement leaves the two faraway ions entangled — having never touched. This is demonstrated physics, not a hope: two ions about a metre apart were heralded into entanglement by Monroe’s group back in 2007
| Superconducting | Trapped ion | Neutral atom | Photonic | |
|---|---|---|---|---|
| Qubit | printed circuit (transmon) | a real atom (ion) | a real atom | a photon |
| Gate speed | very fast (~10–70 ns) | slow (~µs) | slow (~µs) | n/a (measurement) |
| Coherence | short (~0.1–1.7 ms) | very long (s–min) | long (s) | loss-limited |
| Connectivity | mostly nearest-neighbor | all-to-all | reconfigurable | hard (no interaction) |
| Operating temp | ~10 mK (dilution fridge) | room-temp vacuum | room-temp vacuum | room temperature |
| As a network node | needs transduction | natural (emits photons) | promising | it is the photon |
No platform wins every row, which is exactly why all five families this series covers are still racing. Superconducting out-sprints decoherence; the trapped ion out-lasts and out-connects it, and pays in microseconds per gate and in the difficulty of herding more than a hundred atoms into one trap. If your wish list is a cloud-accessible engineering juggernaut today, superconducting still leads; if it is record fidelity, long memory, all-to-all wiring, and a photon you can fire at a distant node, the ion is the one to beat. Either way, there is a different question hiding in the table’s fourth column: if catching one real atom is this good but herding many is this hard, what if you did not strip the electron at all — and held a neutral atom in a tweezer of pure light, with thousands of traps conjured from a single laser? That is the next route in this series: neutral atoms.
More in this series — How to Build a Quantum Computer: Superconducting · Trapped ion · Neutral atom · Photonic · Other platforms