In short

A trapped ion sits still in a Paul trap. To compute with it, you need to rotate its qubit (single-qubit gate) and entangle it with a neighbour (two-qubit gate). Single-qubit gates on hyperfine ions: either a 12.6 GHz microwave horn driving the whole trap with magnetic-gradient addressing, or a pair of Raman laser beams focussed on one ion whose difference frequency equals the hyperfine splitting. Gate time 1–50 μs, fidelity 99.99% routine, 99.9999% at best. Two-qubit gates: the Mølmer-Sørensen (MS) gate drives two ions simultaneously with red- and blue-detuned tones relative to a shared motional mode; the ions' collective vibration acts as a bus that couples them, then disentangles itself at the end of the gate. Output is \exp[-i\theta/4 \cdot X_1 X_2]; at \theta = \pi/2 this turns |00\rangle into a Bell state. Fidelity on Quantinuum H2-56: 99.914%. Gate time: 100–300 μs. QCCD (quantum charge-coupled device) scales beyond a single chain: the trap has multiple zones, ions are shuttled between them by ramping DC voltages, giving effective all-to-all connectivity across dozens of ions. Quantinuum H2-56 implements this with 8 gate zones. State of the art 2024: Quantinuum H2-56 (56 qubits, 99.914% 2Q), IonQ Forte (64 qubits), Oxford Ionics (99.97% gate, photonic networking). Quantinuum's 2023 logical Bell state and 2024 logical-error-rate improvements put trapped ions at the front of the fault-tolerance race.

One chapter ago you caught a ytterbium ion in a Paul trap. It is hanging motionless in ultra-high vacuum, its two hyperfine ground states labelled |0\rangle and |1\rangle, its coherence time pushing past a second. A quantum computer wants more than a stationary qubit — it wants operations. It wants to rotate the ion around the Bloch sphere on command. It wants to entangle one ion with another. It wants to do both hundreds of times in a row, cleanly, before the qubit decoheres.

This chapter is the playbook for doing exactly that. Rotating an ion is done by shining the right electromagnetic wave at it — microwaves for hyperfine qubits, optical lasers for optical qubits — at the right frequency for the right number of nanoseconds. Entangling two ions is done by borrowing their motion: the ions in a chain are not isolated harmonic oscillators; they are Coulomb-coupled, and their collective vibration has well-defined quantum modes that you can use as a bus. Drive both ions with a carefully shaped bichromatic pulse, and the bus mode carries correlation between them, then lets go. What is left is entanglement.

These are the two moves. Master them, stitch them together with measurement, and you have a universal quantum computer — with the highest gate fidelity on any platform as of 2024.

Single-qubit gates — rotating one ion

A single-qubit gate is nothing more than a rotation of the ion's Bloch-sphere vector. To implement it, you apply an oscillating electromagnetic field at (or near) the qubit's transition frequency for a precisely controlled duration. Two standard techniques.

Microwave drive — for hyperfine qubits

The ^{171}Yb^+ hyperfine splitting is \omega_{hf}/2\pi = 12.642812 GHz — squarely in the microwave band. Point a microwave horn at the trap, tune it to 12.6 GHz with amplitude calibrated so the Rabi frequency \Omega_R/2\pi \approx 100 kHz, and every ion in the trap rotates at the same rate. A \pi/2 pulse (t = \pi / (2\Omega_R) \approx 2.5 μs) takes |0\rangle to (|0\rangle + i|1\rangle)/\sqrt 2. A \pi pulse (t \approx 5 μs) implements X.

Why microwaves are slow. The transition is a magnetic dipole (hyperfine levels differ by nuclear-spin alignment, not by electric-charge distribution), and magnetic dipoles couple to electromagnetic fields roughly \alpha^2 \sim 10^{-4} weaker than electric dipoles. Hence μs rather than ns gate times. The upside: microwaves are trivially cheap and stable compared to lasers. Commercial synthesisers hit 12 GHz with sub-millihertz line width.

Addressing one ion at a time. A microwave horn illuminates the whole trap. To make the gate ion-specific, you bathe the trap in a magnetic-field gradient: each ion along the chain sees a slightly different Zeeman shift, so its hyperfine splitting is slightly different. Tune the microwave to one ion's exact frequency and only that ion rotates.

Raman drive — for per-ion laser addressing

An alternative, used by Quantinuum, IonQ, and most academic groups: drive a two-photon Raman transition. Two laser beams at frequencies \omega_1 and \omega_2, both detuned by a large amount \Delta from an excited state, interfere to produce an effective drive at the difference frequency \omega_1 - \omega_2. Set \omega_1 - \omega_2 = \omega_{hf} and the hyperfine qubit rotates.

Raman gate level diagramA three-level lambda system diagram showing the two hyperfine ground states below an excited P state. Two laser beams drive the transitions from each ground state to the excited state, detuned by a large Delta. Their frequency difference equals the hyperfine splitting.Stimulated Raman transition — two lasers, one effective drive|0⟩F=0|1⟩F=1, m=0|e⟩ (²P₁/₂)Δω₁ω₂effective: ω₁ − ω₂ = ω_hftwo-photon stimulated Raman — excited state is never populated (large Δ)
A Raman gate drives both ground states to the same far-detuned excited state with two laser beams. The excited state $|e\rangle$ is virtual — never actually populated — because the detuning $\Delta$ is much larger than the Rabi frequencies. What survives is an effective $|0\rangle \leftrightarrow |1\rangle$ drive at frequency $\omega_1 - \omega_2$. Tune that to the hyperfine splitting and you rotate the qubit.

Why Raman wins for scale. Lasers can be focussed to ~1 μm — about the inter-ion spacing in a trap chain. Two tightly focussed Raman beams intersected on ion 4, say, drive only ion 4. Per-ion addressing without magnetic gradients. Gate time stays in the 1–10 μs range.

Effective Rabi frequency. The combined drive has effective Rabi frequency \Omega_{\text{eff}} = \Omega_1 \Omega_2 / (2\Delta) where \Omega_1, \Omega_2 are the single-beam Rabi frequencies on each transition and \Delta is the detuning. For typical Yb^+ numbers (\Omega_1 \sim \Omega_2 \sim 2\pi \times 10 MHz, \Delta \sim 2\pi \times 100 GHz), \Omega_{\text{eff}}/2\pi \sim 500 kHz — a \pi/2 pulse in 500 ns. Why: the two-photon process is second-order in the single-beam amplitudes and first-order in 1/\Delta, so detuning sets the speed.

For optical qubits

When the qubit is a narrow electric-quadrupole transition — as in ^{40}Ca^+ at 729 nm or ^{88}Sr^+ at 674 nm — you drive it directly with a single ultra-stable laser. No Raman trick needed. Gate time 10–100 μs depending on laser power; fidelity 99.99% routine.

Fidelity numbers

Gate type Platform Time Fidelity
Microwave (global) Yb^+ hyperfine 5–50 μs 99.99%
Raman (addressed) Yb^+, Ba^+ hyperfine 0.5–10 μs 99.99%
Optical quadrupole Ca^+, Sr^+ 10–100 μs 99.99%
Best reported Oxford Ca^+ (2023) 10 μs 99.9999%

The 99.9999% single-qubit fidelity from Oxford (Ballance, Lucas, and collaborators) corresponds to one error per million gates — the highest control fidelity ever demonstrated on any physical qubit.

Example 1: Raman gate Hamiltonian and pulse area

Design a \pi/2 pulse on a single Yb^+ ion using Raman beams at typical parameters.

Step 1. Write the two-photon Hamiltonian. After adiabatic elimination of the excited state |e\rangle, the effective Hamiltonian on the \{|0\rangle, |1\rangle\} subspace is

H_{\text{eff}} = \frac{\hbar \Omega_{\text{eff}}}{2} \big( |1\rangle\langle 0| + |0\rangle\langle 1| \big) + \text{light shifts}

with \Omega_{\text{eff}} = \Omega_1 \Omega_2 / (2\Delta). Why: when the detuning \Delta is much larger than both individual Rabi frequencies, the excited state population stays \sim (\Omega/\Delta)^2 \ll 1 and can be integrated out. What survives is a clean two-level drive.

Step 2. Plug in numbers. \Omega_1 = \Omega_2 = 2\pi \times 10 MHz (readily achievable with 10 mW of 369 nm light focussed to 5 μm). \Delta = 2\pi \times 100 GHz (far below the ^2P_{1/2} state at 369.5 nm). Then

\Omega_{\text{eff}} = \frac{(2\pi \times 10^7)^2}{2 \cdot 2\pi \times 10^{11}} = 2\pi \times 500\,\text{kHz}.

Why: factors of 2π cancel cleanly when you work in angular-frequency units. The \Omega^2/\Delta scaling is what makes Raman gates slower than a resonant optical gate at the same intensity, but insensitive to laser frequency noise (which cancels in the difference).

Step 3. Compute the pulse duration. A \pi/2 rotation needs pulse area \Omega_{\text{eff}} \tau = \pi/2, so \tau = \pi / (2 \Omega_{\text{eff}}) = 1 / (4 \cdot 500\,\text{kHz}) = 500 ns. Why: on the Bloch sphere, a drive of angular frequency \Omega_{\text{eff}} applied for time \tau rotates the state by angle \Omega_{\text{eff}} \tau around the drive axis.

Step 4. Account for light shifts. Both beams also produce AC Stark shifts \delta_{\text{AC}} = \Omega^2/(4\Delta) \approx 2\pi \times 250 kHz on each level. If unbalanced, this adds an unwanted Z rotation during the pulse. Standard fix: balance the beam intensities so that the differential shift \delta_1 - \delta_2 \to 0. Why: equal Stark shifts on both qubit levels translate into a global phase, which is unobservable. Only the differential shift matters.

Step 5. Off-resonant scattering error. The probability of spontaneous emission during the gate is P_{\text{sc}} \approx (\Omega/\Delta)^2 \cdot \Gamma \tau where \Gamma = 2\pi \times 20 MHz is the ^2P_{1/2} decay rate. Plugging in: (10^{-4}) \cdot (2\pi \times 2 \times 10^7) \cdot (5 \times 10^{-7}) \approx 6 \times 10^{-5}. Why: scattering is the dominant error for Raman gates — you cannot push detuning arbitrarily large because you run out of laser power first. The trade-off \Omega^2/\Delta (gate speed) vs \Gamma/\Delta^2 (scattering) sets an optimal detuning for a given laser power.

Result. A 500-ns \pi/2 gate at 6 \times 10^{-5} scattering error per gate, corresponding to raw fidelity \approx 99.994\% per gate — before any further imperfections (pulse shaping, motional Debye-Waller factors). With pulse-shaping and composite sequences (BB1, SK1), published fidelities reach 99.9999%.

Rabi oscillation on a Raman gateA plot showing the probability of measuring the qubit in state one as a function of pulse duration, oscillating as a cosine with period equal to two pi over the effective Rabi frequency.Rabi oscillation — P(|1⟩) vs pulse durationt (μs)10π/2 (500 ns)π (1 μs)2π (2 μs)0.51.01.52.03.04.0
The population of the $|1\rangle$ state oscillates sinusoidally with the pulse time — a textbook Rabi flop. Stop the pulse at $t = 500$ ns for a $\pi/2$ gate (equal superposition); stop at $t = 1$ μs for a full bit flip. This is the diagnostic physicists run to calibrate every single-qubit pulse.

What this shows. Single-qubit control on a trapped ion reduces to two settings: pulse area (set by \Omega_{\text{eff}} \tau) and phase of the drive (which selects the rotation axis in the Bloch-sphere equator). Everything else is engineering: balancing light shifts, stabilising the laser, shaping the pulse to suppress motional coupling.

Two-qubit gates — the Mølmer-Sørensen gate

Single-qubit gates do not entangle. For a universal gate set, you also need one two-qubit gate — something like CNOT, or anything locally equivalent to it. On trapped ions, that gate is almost always the Mølmer-Sørensen (MS) gate, invented in 2000 by Anders Sørensen and Klaus Mølmer.

The key idea: the ions in a chain are not independent. Coulomb repulsion couples their positions, so the chain as a whole has collective motional modes — normal modes of vibration, each a quantum harmonic oscillator. For a two-ion chain, the two modes are:

These phonon modes are your bus. A clever pulse couples the two ions' internal states to the bus, lets the bus mediate an interaction, then decouples the bus cleanly at the end of the gate. The internal states come out entangled; the motional mode returns to whatever it was doing before.

The pulse — bichromatic sideband drive

The MS protocol applies two laser tones simultaneously to both ions:

Here \omega_0 is the qubit transition frequency, \omega_m is the bus-mode frequency, and \epsilon is a small gate-closing offset.

Mølmer-Sørensen pulse structureA laser frequency spectrum diagram with a carrier frequency in the centre, and two symmetric sidebands — one red-detuned and one blue-detuned by the motional frequency. A phase-space spiral loop is shown below that starts and ends at the origin.MS gate — laser spectrum and phase-space loopω₀ (carrier)ω₀ − ω_m − εred sidebandω₀ + ω_m + εblue sidebandboth tones drive both ions simultaneouslymotional phase-space trajectoryRe αIm αclosed loopstart=end
Top: the laser spectrum for the MS gate. Two tones flank the carrier, equidistant from $\omega_0 \pm \omega_m$ by the small offset $\epsilon$. Each tone individually would excite motion and entangle ions with phonons; together they interfere, driving a closed loop in motional phase space. Bottom: the trajectory of the motional coherent-state amplitude $\alpha(t)$ as the gate unfolds — the loop closes at the gate end, leaving motion disentangled from the qubits.

Why it entangles

Expanding the bichromatic drive in the interaction picture (and moving to the Lamb-Dicke regime where the ion's motion is small compared to the laser wavelength), the effective Hamiltonian is

H_{\text{MS}} = \hbar \eta \Omega \, (X_1 + X_2) \, \big( a \, e^{-i\epsilon t} + a^\dagger e^{i\epsilon t} \big)

where \eta is the Lamb-Dicke parameter (typically 0.05–0.2), \Omega is the single-beam Rabi frequency, and a, a^\dagger are the bus-mode annihilation and creation operators. Notice the structure: the collective X operator S_X = X_1 + X_2 is what couples to the bus.

Integrate this Hamiltonian for time \tau = 2\pi/\epsilon (one full phase-space loop) and the motional degree of freedom disentangles, leaving the pure spin-spin unitary

U_{\text{MS}}(\theta) = \exp\!\left[-i \frac{\theta}{4} (X_1 \otimes X_2) \right], \qquad \theta = \frac{2\pi \eta^2 \Omega^2}{\epsilon^2}.

Set \theta = \pi/2. Apply to |00\rangle: the output is (|00\rangle - i|11\rangle)/\sqrt 2 — a Bell state, up to local single-qubit phases.

Why it's robust

The MS gate has a remarkable property: the phase \theta picked up during a closed phase-space loop is geometric — it is the enclosed area in phase space — and is therefore independent of which Fock state the motion started in. You do not need the ground state of motion; you just need a small thermal occupation \bar n with \bar n \eta^2 \ll 1 (the Lamb-Dicke regime). This makes MS tolerant of imperfect cooling, which is a big practical win.

Fidelity today

Platform Gate Time Fidelity
Quantinuum H2-56 MS (Raman) 150 μs 99.914%
IonQ Forte MS (Raman) 200–400 μs 99.8%
Oxford Ca^+ (2022) MS (optical) 100 μs 99.97%
Innsbruck (Blatt) MS (optical, small chain) 100 μs 99.9%

The 99.914% number from Quantinuum's H2-56 (measured 2024) is the record two-qubit fidelity on any commercial device. The 99.97% from Oxford is even higher but measured on a small academic setup.

Example 2: Designing an MS pulse for a two-ion chain

Given a two-ion ^{171}Yb^+ chain with the centre-of-mass mode at \omega_{\text{COM}}/2\pi = 2 MHz, design a Bell-state-generating MS pulse with a 100 μs gate time.

Step 1. Pick the gate-closing offset. \tau = 2\pi/\epsilon and we want \tau = 100 μs, so \epsilon = 2\pi / (10^{-4}\,\text{s}) = 2\pi \times 10 kHz. Why: the motional-phase-space trajectory spirals with angular frequency \epsilon; one full orbit closes the loop. Smaller \epsilon means longer, gentler gates.

Step 2. Set the Lamb-Dicke parameter. For a Raman gate at 355 nm with \omega_{\text{COM}}/2\pi = 2 MHz on a Yb^+ ion, \eta \approx 0.10. Why: \eta = k \sqrt{\hbar / (2 m \omega_m)} where k is the laser wavenumber and m the ion mass. For our numbers: k = 2\pi/(355\,\text{nm}) = 1.77 \times 10^7 rad/m, so \eta \approx 0.10 and \eta^2 \approx 0.01.

Step 3. Pick the Rabi frequency. We want \theta = \pi/2, i.e. 2\pi \eta^2 \Omega^2 / \epsilon^2 = \pi/2. Solving, \Omega^2 = \epsilon^2 / (4 \eta^2), so \Omega = \epsilon / (2\eta) = (2\pi \times 10\,\text{kHz}) / (0.2) = 2\pi \times 50 kHz. Why: the entangling angle scales as (\eta\Omega/\epsilon)^2 — you set three dials (\eta, \Omega, \epsilon) such that one closed loop encloses the right area.

Step 4. Verify by the gate angle formula.

\theta = \frac{2\pi \cdot 0.01 \cdot (2\pi \times 50000)^2}{(2\pi \times 10000)^2} = 2\pi \cdot 0.01 \cdot 25 = \pi/2.

Why: algebra check. The ratio \Omega^2/\epsilon^2 = 25 and 2\pi \eta^2 \cdot 25 = 2\pi \cdot 0.25 = \pi/2. Every factor of 2\pi tracks.

Step 5. Apply to |00\rangle. The output is

U_{\text{MS}}(\pi/2) |00\rangle = \cos(\pi/8) |00\rangle - i \sin(\pi/8) |11\rangle \cdot \sqrt 2 \cdot \text{normalisation} = \frac{1}{\sqrt 2}(|00\rangle - i|11\rangle)

— a Bell state (specifically |\Phi^-\rangle up to single-qubit phases). Why: expanding \exp[-i \pi/8 \cdot X_1 X_2] on |00\rangle and using X_1 X_2 |00\rangle = |11\rangle and (X_1 X_2)^2 = I, we get \cos(\pi/8) |00\rangle - i \sin(\pi/8) |11\rangle. At \theta = \pi/2, \pi/8 is the correct argument and this is (after a sign rearrangement) a Bell state.

Result. \Omega/2\pi = 50 kHz, \epsilon/2\pi = 10 kHz, \tau = 100 μs. Bell state fidelity on calibrated Quantinuum H2 hardware: 99.9%+.

Two-ion chain with collective modesA schematic of two ions aligned along a horizontal axis connected by a coupling spring representing Coulomb repulsion, with arrows showing the two normal modes — centre of mass (both ions move together) and stretch (ions move oppositely).Two-ion normal modescentre-of-massω_COM = ω_zstretchω_stretch = √3 · ω_zthe MS bichromatic drive picks one mode as the bus; the other mode is far enough off-resonance to ignore
The two-ion normal modes. Centre-of-mass: both ions move together. Stretch: opposite phases. An MS gate driven near $\omega_{\text{COM}}$ uses COM as the bus; the stretch mode sits $(\sqrt 3 - 1) \omega_z \approx 0.7 \omega_z$ higher and is off-resonant for the MS tones, so it plays no role. Choosing which mode to use is a free parameter in MS gate design.

What this shows. The MS gate is the canonical example of using an auxiliary degree of freedom (motion) as a temporary bus — excited, used to mediate interaction, then returned to its initial state. This bus-mode pattern recurs throughout physics: it is how the electron-phonon interaction in superconductors mediates Cooper pairing, how cavity QED couples atoms, how circuit QED couples transmons. Trapped-ion MS is arguably the cleanest implementation of the idea.

QCCD — scaling beyond one chain

A single Paul trap chain starts to struggle past ~30 ions. Transverse modes soften, individual-ion addressing becomes harder, and the dense phonon spectrum makes single-mode MS gates impossible.

The fix is architectural: don't hold all your ions in one chain. Split the trap into multiple zones, connected by corridors along which ions can be shuttled. Perform MS gates in a "gate zone" on just a few ions at a time. Park the others in "memory zones" where they sit undisturbed. Load fresh ions from an atomic oven into a "loading zone". This is the quantum charge-coupled device (QCCD) architecture, proposed by Kielpinski, Monroe, and Wineland in 2002 and commercialised by Honeywell/Quantinuum.

QCCD shuttle architectureA top-down schematic of a quantum charge coupled device trap showing multiple zones — two gate zones, two memory zones, a loading zone — connected by narrow shuttling corridors. Arrows show ions moving between zones.QCCD — segmented trap with shuttle corridorsGate zone AMS gates hereMemoryidle qubitsGate zone BMS gates hereLoadingion oven + trapions shuttle between zones by ramping DC electrode voltageseach transport step: 10–100 μs; recooling: ~1 ms
Top-down view of a QCCD trap. Gate zones (red border) host MS gates on small chains. Memory zones store idle qubits. A loading zone can insert fresh ions. Shuttling corridors connect all zones; by ramping DC voltages on segmented electrodes, ions move between zones on demand. Effective connectivity becomes *all-to-all* across the full register.

The cost of shuttling

Every transport step heats the ion a little — moving a charged particle through a varying potential inevitably kicks its motion. Before the next MS gate, the ion must be re-cooled to within the Lamb-Dicke regime. Quantinuum budgets about 1 ms per gate cycle for this: the gate itself is 150 μs, but another ~850 μs is shuttling and re-cooling around it. This is the dominant overhead of QCCD.

The payoff: arbitrary pairs of qubits can be brought together for a gate. The connectivity is not fixed by a 2D superconducting layout — it is dynamic. For algorithms with heavy non-local interaction (quantum error correction with non-local codes, quantum simulations of long-range Hamiltonians), this is a huge advantage.

Quantinuum H2-56 — 2024 flagship

The Quantinuum H2-56 system, announced in 2024, is the state of the art.

These specs put H2-56 at the frontier of logical-qubit demonstrations — which is the next step.

IonQ Forte — a different 2024 design

IonQ's Forte machine pushes the qubit count higher (64 announced) but uses a single-chain architecture rather than QCCD. The bet: algorithmic compression via all-to-all-in-the-chain connectivity beats the shuttle overhead. Gate fidelity on Forte is lower than Quantinuum (99.8% 2Q) but total circuit throughput is competitive for certain workloads.

IonQ's next-generation system, Tempo, uses integrated-photonics laser delivery — waveguide-based laser addressing on a chip-scale trap — aimed at eliminating the bulky free-space optics that limit current trap systems.

Oxford Ionics — photonic networking

Oxford Ionics (spun out of the University of Oxford, 2023) takes yet another angle: small high-fidelity traps connected by photonic fibre links. Each trap has ~10 ions; ions emit photons entangled with their internal state; coincidence detection on the photons at a beamsplitter heralds entanglement between ions in different traps. Oxford Ionics reports 99.97% single-trap gate fidelity and is demonstrating inter-trap entanglement rates increasing year over year.

Fault tolerance on trapped ions — the leading race

The combination of high fidelity and all-to-all connectivity (within a QCCD or short chain) has made trapped ions the leading platform for early fault-tolerant quantum computing — running circuits where the logical error rate is lower than the physical error rate.

Quantinuum's milestones

These results use roughly 20–40 physical qubits to protect a few logical qubits. The regime is not yet useful (you can't break RSA with four logical qubits), but it is the first clear demonstration that error correction works end-to-end. The race is now to scale: 30 logical qubits would let you run Shor on small integers; 1000 logical qubits would let you do useful quantum chemistry.

Why ions lead here

Superconducting platforms are ahead on physical-qubit count (IBM Condor has 1121 qubits as of 2023) but behind on fidelity. Error correction pays a steep overhead — typically 1000 physical per 1 logical at current fidelities. On ions, the overhead is closer to 30:1 because the physical fidelity is so much higher. As of 2026, ions have the logical lead, and superconducting has the physical lead. Which wins the useful-computation race depends on whose scaling curve bends first.

Indian context

IIT Delhi (Prof. Rajdeep Chatterjee and colleagues) has operated a ^{171}Yb^+ ion-trap experiment since 2020 under India's National Quantum Mission pillar for trapped ions. As of 2026, the IIT Delhi group targets a 10-qubit demonstrator; the 8-year NQM roadmap calls for a 30-qubit QCCD by 2028. IISER Pune and the Raman Research Institute, Bangalore also have ion-trapping groups, with RRI's effort focused on single-ion quantum optics. TIFR Mumbai runs a neutral-atom programme that shares many techniques with trapped ions.

The ion-trap community in India is small (perhaps 20 active researchers as of 2026) but growing rapidly. NQM funding plus industry interest (TCS, Infosys, HCL, Wipro all have quantum efforts that are aware of ion-trap roadmaps) is accelerating it.

Common confusions

Going deeper

If you understand that single-qubit gates on an ion are microwave or Raman pulses that rotate its Bloch vector, that the Mølmer-Sørensen gate uses the ion chain's collective motional mode as a temporary bus to generate an X_1 X_2 interaction that becomes a Bell-state-creating gate at \theta = \pi/2, that the MS gate tolerates thermal motion because the entangling phase is geometric, that QCCD architectures scale beyond one chain by shuttling ions between zones, and that Quantinuum H2-56 holds the current two-qubit fidelity record at 99.914% while running the first logical-qubit demonstrations with sub-threshold error rates — you have chapter 167. What follows is the full Raman-gate derivation, the MS magnus-expansion geometric-phase calculation, QCCD design details, current fault-tolerance milestones, and the main error sources that limit today's hardware.

Raman gate — adiabatic elimination in detail

The full three-level lambda-system Hamiltonian for a Raman gate on \{|0\rangle, |1\rangle, |e\rangle\} with two lasers is

H = \frac{\hbar \Omega_1}{2} (|e\rangle\langle 0| e^{-i\Delta_1 t} + \text{h.c.}) + \frac{\hbar \Omega_2}{2} (|e\rangle\langle 1| e^{-i\Delta_2 t} + \text{h.c.})

where \Delta_1, \Delta_2 are the detunings of each laser from its respective transition. Under two-photon resonance (\Delta_1 \approx \Delta_2 \equiv \Delta, with small two-photon detuning \delta = \Delta_1 - \Delta_2 + \omega_{hf}), the excited-state amplitude in the rotating frame is approximately c_e \approx -(\Omega_1 c_0 + \Omega_2 c_1)/(2\Delta) in the adiabatic (slowly varying c_0, c_1) limit. Substituting back into the Schrödinger equations for c_0 and c_1 gives the effective two-level Hamiltonian

H_{\text{eff}} = \frac{\hbar}{2} \begin{pmatrix} -\Omega_1^2/(2\Delta) & \Omega_1 \Omega_2^*/(2\Delta) \\ \Omega_1^* \Omega_2 /(2\Delta) & -\Omega_2^2/(2\Delta) - 2\delta \end{pmatrix}

— with off-diagonal elements \Omega_{\text{eff}}/2 = \Omega_1 \Omega_2 / (4\Delta) driving Rabi oscillations, and diagonal AC Stark shifts -\Omega_{1,2}^2/(4\Delta) that must be cancelled. The residual excited-state population |c_e|^2 \sim (\Omega/\Delta)^2 times the spontaneous-emission rate sets the scattering floor.

The MS gate — Magnus expansion to derive geometric phase

The MS Hamiltonian in the interaction picture (after rotating-wave approximation) is

H_I(t) = \hbar \eta \Omega \, S_X \, (a e^{-i\epsilon t} + a^\dagger e^{i\epsilon t})

where S_X = \sigma_x^{(1)} + \sigma_x^{(2)}. The time-ordered evolution operator U(\tau) = \mathcal{T} \exp[-i \int_0^\tau H_I dt/\hbar] is not expressible as a simple exponential because H_I at different times does not commute. The Magnus expansion gives

U(\tau) = \exp\!\left[\Omega_1 + \Omega_2 + \Omega_3 + \cdots\right]

with \Omega_1 = -i/\hbar \int H_I dt, \Omega_2 = -1/(2\hbar^2) \int\int [H_I(t_1), H_I(t_2)] dt_1 dt_2, and higher nested-commutator terms. For the MS Hamiltonian with its specific S_X \otimes \text{(motional)} structure:

So after the loop closes, only the S_X^2 phase survives:

U_{\text{MS}}(\tau) = \exp\!\left[-i \frac{\pi \eta^2 \Omega^2}{2 \epsilon^2} S_X^2\right] = \exp\!\left[-i \frac{\pi \eta^2 \Omega^2}{2 \epsilon^2} (2 + 2 X_1 X_2)\right]

with the identity absorbed into a global phase. Setting \theta = 2\pi \eta^2 \Omega^2 /\epsilon^2, we get U_{\text{MS}}(\theta) = \exp[-i \theta/4 \cdot X_1 X_2]. Thermal insensitivity follows because \Omega_2 has no a, a^\dagger left — the geometric phase is independent of initial phonon number.

QCCD design considerations

Modern QCCD chips (Quantinuum H2, Sandia's HOA-2 prototype) have electrode structures with 30–100 DC segments per zone, plus RF blades for radial confinement. Shuttling requires smooth time-dependent voltage profiles on adjacent segments: too fast and the ion gets kicked (heating); too slow and the chip throughput collapses. Quantinuum uses "bucket brigade" protocols optimised to move ions between zones in 50 μs with \Delta \bar n < 1 quanta of heating.

Junction traps — where shuttle corridors branch — are the hardest fabrication challenge. An ion must navigate a Y-junction without getting lost at the branch point. Current designs use gradient-index transitions and achieve >99.99% shuttle fidelity per junction.

Error sources — what limits today's fidelity

Error Magnitude per gate Mitigation
Spontaneous scattering (Raman) 10^{-4}–10^{-5} Detuning \Delta; composite pulses
Motional heating 10^{-4} Sympathetic cooling with Ba^+
Motional-mode crosstalk 10^{-4} Pulse shaping to null other modes
Beam intensity noise 10^{-4} Active intensity stabilisation
Laser phase noise 10^{-4} Cavity-locked lasers, phase cancellation
Magnetic-field noise 10^{-5} Clock states; mu-metal shielding
Addressing crosstalk 10^{-4} Tight beam focus; pulse shaping

Total MS fidelity ~99.9% corresponds to a budget of roughly 10^{-3} error per gate, built up from these contributions. The current frontier (99.97% Oxford, 99.914% Quantinuum) is a matter of chipping down each line item by a factor of 2–3. None of the lines has a fundamental floor — every one is in principle improvable.

The fault-tolerance demos — 2022–2026

These are milestones on the path to useful fault-tolerant computation: the physics works, the thresholds are crossed, the engineering is now about scale.

Where this leads next

References

  1. Anders Sørensen and Klaus Mølmer, Entanglement and quantum computation with ions in thermal motion (2000) — arXiv:quant-ph/0002024.
  2. Colin D. Bruzewicz et al., Trapped-Ion Quantum Computing: Progress and Challenges (2019), Applied Physics Reviews — arXiv:1904.04178.
  3. Quantinuum, H2-1 Quantum Computer Performance Characteristics (2024) — quantinuum.com/products-solutions/quantinuum-systems.
  4. Wikipedia, Trapped ion quantum computer.
  5. IonQ, Forte System Documentation — ionq.com/quantum-systems/forte.
  6. John Preskill, Lecture Notes on Quantum Computation, Chapter 7 — theory.caltech.edu/~preskill/ph229.