Superconducting qubits get most of the headlines, largely because IBM and Google are loud about their roadmaps. But if you talk to people inside the field about which hardware platform currently has the cleanest qubits — the lowest error rates, the longest coherence times, the most uniform performance across the chip — a lot of them will point to trapped ions. Companies like IonQ, Quantinuum (the merger of Honeywell Quantum Solutions and Cambridge Quantum), and Alpine Quantum Technologies have built their entire hardware strategy around this approach, betting that quality beats raw qubit count in the race toward useful quantum computation.
This article covers how trapped ion qubits actually work, the physics of ion traps and laser control, why the gates are so precise, and why — despite that precision — scaling this approach up to thousands of qubits is a genuinely different kind of hard problem than the one superconducting qubits face.
The Basic Idea: Use an Actual Atom
Where superconducting qubits are artificial atoms engineered out of circuit elements, trapped ion qubits use real atoms — typically ionized alkaline earth or lanthanide elements like ytterbium-171 (¹⁷¹Yb⁺), calcium-40 (⁴⁰Ca⁺), barium-137, or strontium-88. These atoms are ionized (stripped of an electron so they carry a net positive charge), which makes them controllable with electric and magnetic fields, and then trapped and manipulated using a device called a Paul trap or an RF (radio-frequency) ion trap.
The core advantage of using a real atom rather than an engineered circuit is that atoms of the same isotope are, by the laws of physics, absolutely identical. Every ¹⁷¹Yb⁺ ion in the universe has exactly the same energy levels as every other ¹⁷¹Yb⁺ ion. This eliminates a class of problems that plagues superconducting qubits, where fabrication variability means every chip has slightly different qubit frequencies that need individual calibration. With trapped ions, qubit-to-qubit variation from fabrication simply doesn’t exist — the qubits are, quite literally, natural.
How an Ion Trap Works
You can’t trap a charged particle in a stable equilibrium using only static electric fields — this is a consequence of Earnshaw’s theorem, which says that a charged particle cannot be held in stable equilibrium by electrostatic forces alone. So ion traps instead use oscillating radio-frequency electric fields, typically in a configuration called a linear Paul trap.
In a linear Paul trap, a set of electrodes generates a rapidly oscillating quadrupole electric field. The ion experiences an effective time-averaged potential, called a pseudopotential, that does create stable confinement in the plane perpendicular to the trap axis. Along the trap axis itself, static (DC) electrodes provide additional confinement. The net result is that ions get held in place in a linear chain, floating in ultra-high vacuum, isolated from the outside world except for the laser beams and fields the operator intentionally applies.
Because ions carry the same charge, they repel each other via Coulomb interaction, which naturally spaces them out along the trap into an evenly separated linear chain — this is actually useful, because it gives you a natural, physically enforced qubit register with well-defined spacing that a laser beam can be aimed at, ion by ion.
Encoding the Qubit
There are two dominant schemes for encoding quantum information in a trapped ion, and it matters which one a given system uses because it changes how you drive gates and how long coherence lasts.
Hyperfine qubits encode the $|0\rangle$ and $|1\rangle$ states in two hyperfine ground-state sublevels of the ion — energy differences that arise from the interaction between the nuclear spin and the electron’s magnetic moment. These transitions typically sit in the microwave or radio-frequency range (a few GHz), and because both states are ground states, they’re essentially immune to spontaneous decay. Hyperfine qubits in ions like ¹⁷¹Yb⁺ have demonstrated coherence times measured in minutes, which is extraordinarily long compared to the hundreds of microseconds typical of superconducting transmons.
Optical qubits instead encode information between a ground state and a long-lived metastable excited state, addressed directly with a laser at optical frequencies. These are used in ions like ⁴⁰Ca⁺ and ⁸⁸Sr⁺. Optical qubits have somewhat shorter natural lifetimes than hyperfine qubits (since the excited state can eventually decay), but they offer certain gate-speed and addressing advantages.
Either way, the essential point is that the qubit’s “clock frequency” — the frequency of the transition between its two logical states — is fixed by nature, and known to extraordinary precision from atomic physics and spectroscopy. This is part of why trapped ion qubits are so clean: you’re not fighting fabrication-induced randomness, you’re working against a well-characterized, highly stable atomic reference.
Single-Qubit Gates: Extremely High Fidelity
Single-qubit gates on trapped ions are performed by applying a precisely tuned laser pulse (or, for hyperfine qubits, a microwave or two-photon Raman laser pulse) resonant with the qubit’s transition frequency. The pulse’s duration and phase determine the rotation performed on the qubit’s state, described using the same rotation formalism as any two-level quantum system:
$$|\psi\rangle = \cos\left(\frac{\theta}{2}\right)|0\rangle + e^{i\phi}\sin\left(\frac{\theta}{2}\right)|1\rangle$$
Because the transition frequency is extremely well-defined (atomic clock transitions are, after all, the basis for the world’s most precise timekeeping devices), single-qubit gate fidelities in trapped ion systems routinely exceed 99.9%, and some demonstrations have pushed into the “five nines” range (99.999%) — numbers that are difficult to match with any other current qubit technology.
Two-Qubit Gates: The Mølmer–Sørensen Gate
This is where trapped ion physics gets genuinely elegant, and it’s worth understanding because it’s conceptually different from how superconducting qubits couple.
Trapped ions in a linear chain aren’t just individually confined — because they repel each other electrostatically, the whole chain has collective vibrational modes, much like a set of masses connected by springs. These vibrational modes are quantized, meaning the chain’s motion can be described in terms of quantized units of vibrational energy called phonons.
The trick behind the most widely used two-qubit gate for trapped ions — the Mølmer–Sørensen gate — is to use this shared vibrational mode as a communication bus between two ions that otherwise have no direct interaction. By applying bichromatic laser pulses (two closely spaced frequencies) tuned near the ion’s transition frequency but symmetrically detuned from a particular vibrational mode, you can generate an effective spin-spin interaction between two ions, mediated by virtual excitation of the shared phonon mode. Done correctly, this entangles the two ions’ internal (qubit) states without leaving any residual entanglement with the vibrational motion itself, which is essential — any lingering ion-phonon entanglement would act as a decoherence channel.
The gate is elegant because, in principle, it doesn’t require the two ions to be physically adjacent — any two ions in the same trap share the same set of collective vibrational modes, so any ion can, in principle, be entangled with any other ion in the same chain. This gives trapped ion systems something superconducting chips fundamentally cannot offer on a 2D lattice: all-to-all connectivity within a single trap, without needing intermediate SWAP operations to move information across the register.
Two-qubit gate fidelities on trapped ion systems have reached above 99.9% in leading demonstrations, generally regarded as the highest two-qubit gate fidelities of any qubit modality currently in operation.
Readout: State-Dependent Fluorescence
Measuring a trapped ion qubit uses a technique called state-dependent fluorescence detection. A laser tuned to a cycling transition (one where the excited state can only decay back to the same ground state it came from) is shone on the ion. If the ion is in the state that couples to this transition, it will scatter thousands of photons, which can be collected by a camera or photomultiplier and clearly registered as “bright.” If the ion is in the other qubit state, which doesn’t couple to the readout laser, it stays “dark” — no photons scattered. This bright/dark contrast is so strong that readout fidelities routinely exceed 99.9%, another area where trapped ions currently lead other platforms.
The Scalability Problem
Given all these advantages — long coherence, near-perfect gate fidelities, all-to-all connectivity — you might reasonably ask why trapped ion companies aren’t leading the qubit-count race the way superconducting companies are. The answer is that trapped ions face a genuinely different and, in some ways, harder scaling problem.
Gate speed. Trapped ion gates, whether single- or two-qubit, typically take on the order of microseconds to tens of microseconds to execute — roughly 100 to 1,000 times slower than superconducting gates, which run in tens of nanoseconds. For circuits requiring many sequential operations, this speed disadvantage compounds.
Laser control complexity. Superconducting qubits are controlled with electrical signals down coaxial cables — a relatively mature engineering discipline. Trapped ions require precisely aimed, phase-stable, individually addressable laser beams for every ion in the register. As chains grow longer, keeping every ion individually addressable with tightly focused laser spots, without crosstalk onto neighboring ions, becomes a serious optical engineering challenge.
Vibrational mode crowding. The Mølmer–Sørensen gate’s reliance on shared vibrational modes is a double-edged sword. As you add more ions to a single chain, you also add more vibrational modes, and these modes get closer together in frequency, making it harder to selectively address just the one mode you want for a given gate without accidentally exciting others. This is widely regarded as one of the fundamental limits on how long a single linear ion chain can practically be — most current systems operate with chains of a few tens of ions, not hundreds or thousands.
The QCCD architecture as a scaling answer. The industry’s leading proposed solution to this problem is the Quantum Charge-Coupled Device (QCCD) architecture, pioneered largely by groups that became Honeywell Quantum Solutions and then Quantinuum. Instead of trying to fit ever more ions into one long trap, QCCD uses a network of smaller trap “zones” connected by electrode-controlled pathways. Ions can be physically shuttled between zones — moved, split apart, recombined — using time-varying voltages on the trap electrodes. Gates are performed within small local zones, and quantum information is transported around the chip by physically moving the ions themselves rather than by any electromagnetic coupling. This is genuinely one of the more creative engineering solutions in quantum computing hardware: instead of building the connectivity into fixed wiring, you make the qubits mobile.
Photonic interconnects for modularity. For truly large-scale systems, several groups (including IonQ) are pursuing photonic interconnects — using entangled photons to link separate ion traps together, allowing a network of smaller trap modules to behave as one larger logical quantum computer. This remains an active area of research rather than a fully mature production technology.
Real-World Standing and Recent Milestones
Quantinuum’s H-series systems, built on the QCCD architecture, have consistently posted some of the highest quantum volume and two-qubit gate fidelity numbers publicly reported across the industry. IonQ has pursued a somewhat different commercial strategy, emphasizing algorithmic qubit counts and cloud accessibility through partnerships with AWS, Azure, and Google Cloud. Both companies, along with academic groups, have used trapped ion systems for some of the most convincing demonstrations of quantum error correction to date — precisely because the underlying physical error rates are low enough that error correction can actually show a net benefit rather than just adding overhead.
Advantages, Summarized
- Near-identical, fabrication-free qubits — no chip-to-chip variability
- The longest demonstrated coherence times of any qubit modality (minutes, for hyperfine qubits)
- The highest demonstrated single- and two-qubit gate fidelities currently reported
- All-to-all connectivity within a trap, without SWAP-gate overhead
- Room-temperature (or near-room-temperature) vacuum chamber operation for the trap itself, avoiding the need for a dilution refrigerator (though some laser and control systems still require careful thermal management)
Limitations, Summarized
- Gate speeds one to three orders of magnitude slower than superconducting qubits
- Laser control systems are optically and mechanically complex, and harder to miniaturize than microwave electronics
- Vibrational mode crowding limits practical single-chain qubit counts
- Scaling requires either QCCD-style ion shuttling or photonic interconnects, both of which are active engineering frontiers rather than fully solved problems
- Current systems, while high-fidelity, still have far fewer total qubits than the largest superconducting chips
Established Technology vs. What’s Still Developing
Trapped ion physics itself — Paul traps, hyperfine and optical qubit encoding, Mølmer–Sørensen gates, fluorescence readout — is decades-old, extremely well-understood atomic physics, and the commercial systems built on it are real, cloud-accessible hardware today. What’s still an open engineering frontier is scaling: QCCD ion-shuttling architectures are operating at meaningful scale in production systems, but extending them to the thousands-of-qubits regime needed for large-scale fault-tolerant computation, and doing so while keeping gate speeds and fidelities competitive, remains unproven at scale. Photonic interconnection between separate trap modules is even earlier-stage — promising in laboratory demonstrations, but not yet a mature commercial technology.
Cooling and Vacuum Requirements: A Different Kind of “Cold”
It’s worth clarifying a point that often confuses people coming from the superconducting side of the field: trapped ion systems don’t need dilution refrigeration to reach millikelvin temperatures the way superconducting chips do, since the qubit itself (an individual atomic ion) doesn’t rely on a superconducting phase transition to function. What trapped ion systems need instead is an extremely good vacuum — typically ultra-high vacuum chambers at pressures around $10^{-11}$ torr or lower — to prevent background gas molecules from colliding with the trapped ions and knocking them out of the trap or introducing motional heating. Many systems additionally use laser cooling techniques, such as Doppler cooling and resolved-sideband cooling, to bring the ions’ motional (vibrational) state down to or near its quantum ground state before performing gates, since residual thermal motion in the ion chain degrades two-qubit gate fidelity by introducing noise into the shared vibrational mode the Mølmer–Sørensen gate depends on. This is a genuinely different engineering discipline from cryogenic dilution refrigeration — it’s built on decades of atomic, molecular, and optical (AMO) physics rather than solid-state cryogenics — and it’s part of why trapped ion labs tend to look different from superconducting labs: less about giant cylindrical fridges, more about optical tables, laser systems, and vacuum chambers.
Barium and the Push for Better Optical Qubits
It’s worth mentioning a relatively recent and instructive development in ion species selection. Quantinuum’s more recent hardware generations have moved toward using barium ions (¹³⁷Ba⁺) rather than the ytterbium used in earlier systems, partly because barium’s relevant optical transitions sit at wavelengths that are more compatible with standard telecom-grade fiber optics and diode laser technology, which simplifies the laser engineering considerably and opens a more direct path toward photonic interconnects between separate trap modules. This kind of species selection decision illustrates something important about trapped ion engineering that doesn’t come up in the idealized physics discussion: the “periodic table” of usable ion species is a genuine design space, with real trade-offs between coherence properties, laser wavelength convenience, and manufacturability of the surrounding photonic and electronic infrastructure, and different companies have made different bets within that space.
Benchmarking: Quantum Volume and Beyond
Comparing hardware platforms fairly is harder than it looks, and trapped ion companies have generally leaned into a metric called quantum volume, introduced by IBM but adopted across the industry as a rough single-number benchmark, which captures a combination of qubit count, connectivity, gate fidelity, and circuit depth into one figure rather than relying on qubit count alone. Quantum volume is defined, loosely, as the largest square circuit (equal number of qubits and circuit depth) that a device can run successfully above a defined fidelity threshold, expressed as $\log_2(\text{QV})$ equal to that largest successful square dimension. Because trapped ion systems benefit from all-to-all connectivity and high gate fidelities, they have historically posted strong quantum volume numbers relative to their comparatively modest total qubit counts — a useful illustration of the article’s broader point that raw qubit count and genuine computational capability are not the same thing. More recently, the field has increasingly supplemented or replaced quantum volume with application-oriented benchmarks that measure performance on specific, more representative algorithmic workloads, since a single abstract number, however useful, can’t fully capture how a real algorithm will perform on real hardware.
The Trade-Off in One Sentence
If superconducting qubits are betting that fabrication scale and gate speed will eventually outrun their connectivity and coherence limitations, trapped ion systems are betting the opposite: that near-perfect gate fidelity and natural qubit uniformity matter more in the long run than raw qubit count, and that clever engineering (QCCD shuttling, photonic links) will eventually solve the scaling problem without sacrificing that precision. Which bet wins is genuinely an open question in the field right now, and it’s a big part of why serious observers of quantum computing tend to track multiple hardware platforms rather than assuming any one approach has already won.