Trapped-Ion Qubits

A trapped-ion quantum processor is the most literal realization of the dream of computing with single atoms. Each qubit is encoded in two long-lived internal states of a singly charged ion---such as $^{40}\mathrm{Ca}^+$ , $^{171}\mathrm{Yb}^+$ , $^9\mathrm{Be}^+$ , or $^{88}\mathrm{Sr}^+$ ---suspended in vacuum, cooled to low temperatures, and addressed by laser or microwave fields. Because Earnshaw's theorem forbids a static three-dimensional trap for a charge, confinement is dynamic: a radio-frequency quadrupole (the Paul trap) whips a saddle-shaped potential around fast enough that the time-average holds the ion in an effective harmonic well. The ions repel one another and freeze into a Coulomb crystal whose collective vibrations are quantized as phonons. That crystal is the secret of the platform. The qubit lives in atomic structure that nature reproduces identically for every ion and shields almost perfectly from its environment — which is why trapped ions hold the records for gate fidelity, measurement fidelity, and coherence-to-gate-time ratio. But that same isolation means the atoms have no direct way to talk to each other. The only wire available is the shared motion: a laser tuned to a motional sideband converts a spin flip into a phonon and back, so a single vibrational mode becomes a quantum bus entangling any pair in the chain. The full cycle — Doppler and sideband cooling, optical pumping to initialize, sideband- or geometric-phase gates, and state-dependent fluorescence readout — meets every DiVincenzo criterion. The frontier is not fidelity but scale: a single trap comfortably controls tens of ions, and going further means shuttling ions between zones on a chip or wiring separate traps together with photons.

Key takeaways

Trapped-ion qubits store quantum information in two long-lived internal states of atomic ions confined by oscillating (Paul) or static-plus-magnetic (Penning) fields under ultra-high vacuum.
The encodings are optical qubits (a ground and a metastable level split by an optical frequency, e.g. $S_{1/2}\leftrightarrow D_{5/2}$ in $^{40}\mathrm{Ca}^+$ , lifetime $\approx1.1\,\mathrm{s}$ ) and hyperfine qubits (two ground sublevels split by GHz, coherent for seconds to minutes).
Confinement is dynamic — Earnshaw forbids a static 3D minimum, so an RF saddle is rotated fast and its time-average, the pseudopotential $\psi\propto q^2|\nabla\Phi_{\mathrm{rf}}|^2/(4m\Omega_{\mathrm{rf}}^2)$ , does the trapping.
The shared phonon modes are the quantum bus: laser-driven sidebands with Rabi frequencies $\Omega\,\eta\sqrt{n}$ (red) and $\Omega\,\eta\sqrt{n+1}$ (blue) couple spin to motion and drive the Cirac–Zoller and Mølmer–Sørensen entangling gates.
Ions satisfy DiVincenzo with the best demonstrated fidelities (two-qubit $>99.9\%$ , single-qubit $>99.99\%$ , readout $>99.9\%$ ), at the cost of slower gates (μs–ms) and harder scaling than superconducting circuits.
Key idea: the qubit lives in internal states, while the shared, quantized vibration of the ion crystal provides the quantum bus within a trapped-ion chain---laser-driven sidebands couple the internal states to motion, enabling entangling gates between ions.

The core insight

The qubit lives in internal electronic states, but the only available quantum bus is the shared, quantized vibrational motion of the ion crystal: laser-driven motional sidebands convert a spin flip into a phonon and back, so that mediating gates through phonons turns identical, well-isolated atoms into a fully connected register.

Where this lives today

Trapped ions are commercial quantum processors. Quantinuum's H-series and IonQ's machines run the quantum-charge-coupled-device (QCCD) architecture, shuttling ions between dedicated storage, gate, and readout zones on microfabricated surface traps so a modest chain in any one zone stays well-controlled while the register grows to tens of qubits. These devices hold the field's fidelity records — two-qubit above 99.9%, single-qubit above 99.99%, readout above 99.9% — and the highest quantum-volume and logical-qubit benchmarks of any platform. The same toolbox powers the world's most accurate clocks ( $^{27}\mathrm{Al}^+$ quantum-logic clocks, fractional uncertainty below $10^{-18}$ ) and analog simulators of long-range Ising magnetism with 50 or more ions. The live open problem is interconnection — photonic links that entangle ions in separate traps by interfering their emitted photons on a beamsplitter, networking many small QCCD modules into a large fault-tolerant machine.

The physical picture

Picture each ion as a tiny tuning fork — its electronic two-level system — suspended on a shared mattress, the crystal's vibrational modes. Plucking one ion's spin with a laser tuned slightly off the bare transition also kicks the mattress; the kick propagates and lets a distant ion feel it. That mechanical coupling is the only thing connecting otherwise perfectly isolated atoms, and engineering it is the whole art of the platform.

Figure

A linear RF Paul trap holds a chain of ions of spacing ≈ 5 μm in a shared harmonic well of secular frequency ν ≈ 2π × 1 MHz; the Coulomb-coupled chain has collective phonon modes — the in-phase centre-of-mass mode at ν₁ = ν_z and the breathing mode at ν₂ = √3 ν_z ≈ 1.73 ν_z — that act as the quantum bus

Confinement is a juggling act, not a bowl: the RF saddle points uphill along one axis and downhill along the other, swapping them tens of millions of times a second, so an ion that drifts off-center is on time-average always pushed back. The price of the qubit's isolation is that you must reach in through the shared motion to make two qubits interact — and the machinery is the sideband ladder in the second figure, whose relative rates are the load-bearing numbers of the platform.

Figure

The internal states g and e each carry a phonon ladder spaced by ħν: the carrier flips the spin at rate Ω leaving n unchanged, the blue sideband adds a phonon at Ω η √(n+1), the red removes one at Ω η √n — with η ≈ 0.1 the sidebands are about ten times weaker than the carrier, and the red sideband vanishes at n = 0

The core equations

Everything begins with the trap. A charge cannot be confined in all three directions by a static electric field alone---in charge-free space, Laplace's equation requires the curvatures of the electrostatic potential to sum to zero. Thus, a static quadrupole that confines along some directions must be anti-confining along another. The linear Paul trap sidesteps this by making the confining field oscillate.

\Phi(x,y,z,t)=\frac{U_{\mathrm{dc}}}{2}\big(\alpha x^2+\beta y^2+\gamma z^2\big)+\frac{V_{\mathrm{rf}}\cos(\Omega_{\mathrm{rf}}t)}{2}\big(\alpha' x^2+\beta' y^2+\gamma' z^2\big) \qquad (1)

where: $U_{\mathrm{dc}},V_{\mathrm{rf}}$ — static and RF voltage amplitudes; $\Omega_{\mathrm{rf}}$ — RF drive frequency (tens of MHz); $\alpha,\beta,\gamma$ — geometric coefficients constrained by $\alpha+\beta+\gamma=0$ (Laplace's equation). The first term is anti-confining somewhere; the second term, oscillating, is what rescues the trap. Averaging the fast RF-driven wiggle over one period leaves a smooth well — the pseudopotential. It is proportional to the square of the RF field, so it always points inward, and its curvature sets the secular frequency $\nu$ used everywhere below.

\psi_{\mathrm{pseudo}}(\mathbf r)=\frac{q^2}{4m\Omega_{\mathrm{rf}}^2}\,\big|\nabla\Phi_{\mathrm{rf}}(\mathbf r)\big|^2,\qquad \nu_r=\frac{qV_{\mathrm{rf}}}{\sqrt{2}\,m\Omega_{\mathrm{rf}}r_0^2} \qquad (2)

where: $q,m$ — ion charge and mass; $\Omega_{\mathrm{rf}}$ — RF drive frequency; $r_0$ — ion-to-electrode distance; $\nu_r$ — radial secular frequency of the resulting well. Fast driving of a saddle produces a genuine harmonic trap for the slow motion. With the trap set, write the bare Hamiltonian of one ion: an internal two-level qubit plus one quantized motional mode whose quanta are phonons.

H_0=\hbar\nu\Big(a^\dagger a+\tfrac12\Big)+\frac{\hbar\omega_0}{2}\sigma_z \qquad (3)

where: $\nu$ — secular angular frequency of the mode; $a^\dagger,a$ — phonon creation/annihilation operators; $\omega_0$ — qubit transition angular frequency; $\sigma_z$ — Pauli operator on the qubit. The two terms are independent until a laser links them. The strength of that link is set by one dimensionless number, the Lamb–Dicke parameter — the ratio of the ground-state motional spread $x_0$ to the laser wavelength, i.e. how much a single photon's recoil disturbs the motion.

\eta=k\sqrt{\frac{\hbar}{2m\nu}}=k\,x_0 \qquad (4)

where: $k$ — projection of the laser wavevector on the motional axis; $x_0=\sqrt{\hbar/2m\nu}$ — RMS zero-point extent of the mode. For an optical transition $\eta\sim0.1$ ; the Lamb–Dicke regime $\eta\sqrt{2\bar n+1}\ll1$ holds once cooled near the ground state. Now the laser–ion coupling in the interaction picture. The spin-flip operator is dressed by the displacement operator of the motion: flipping the spin can simultaneously push the oscillator — the entire mechanism by which spin and motion communicate.

H_I=\frac{\hbar\Omega}{2}\,\sigma_+\,e^{\,i\eta(a e^{-i\nu t}+a^\dagger e^{\,i\nu t})}e^{-i\delta t}+\text{h.c.} \qquad (5)

where: $\Omega$ — bare (carrier) Rabi frequency; $\sigma_+=|e\rangle\langle g|$ — qubit raising operator; $\delta=\omega_L-\omega_0$ — laser detuning from the bare transition; $\eta$ — Lamb–Dicke parameter. Tuning $\delta$ to $0$ , $-\nu$ , or $+\nu$ selects the carrier, red, or blue transition. Expanding the displacement operator to first order in $\eta$ gives the three fundamental transitions and their rates — the numbers drawn in the sideband-ladder figure.

\Omega_{n,n}=\Omega,\qquad \Omega_{n,n-1}=\Omega\,\eta\sqrt{n},\qquad \Omega_{n,n+1}=\Omega\,\eta\sqrt{n+1} \qquad (6)

where: $\Omega_{n,n}$ — carrier ( $\delta=0$ ), flips spin, leaves $n$ ; $\Omega_{n,n-1}$ — red sideband ( $\delta=-\nu$ ), removes a phonon; $\Omega_{n,n+1}$ — blue sideband ( $\delta=+\nu$ ), adds one. The $\sqrt{n}$ scaling is bosonic enhancement, and $\sqrt{n}=0$ at $n=0$ terminates sideband cooling. A chain of $N$ ions has $N$ axial normal modes; small oscillations about equilibrium can be decomposed into independent harmonic oscillators, and the modes can be used as quantum buses for mediating interactions between ions.

H_{\mathrm{mot}}=\sum_{p=1}^{N}\hbar\nu_p\Big(a_p^\dagger a_p+\tfrac12\Big),\qquad \nu_1=\nu_z,\quad \nu_2=\sqrt{3}\,\nu_z \qquad (7)

where: $\nu_p$ — frequency of mode $p$ ; $\nu_1$ — centre-of-mass (all ions in phase, at the single-ion frequency $\nu_z$ ); $\nu_2$ — breathing/stretch mode. Any ion's laser can excite a shared mode, so the phonon couples non-neighbouring ions — full connectivity. The workhorse entangling operation is the Mølmer–Sørensen gate, a bichromatic drive whose net effect after a closed phase-space loop is a pure spin–spin interaction, with no leftover phonon entanglement.

U_{\mathrm{MS}}=\exp\!\big[-i\,\chi\,J_x^2\big],\qquad J_x=\tfrac12\textstyle\sum_i \sigma_x^{(i)},\qquad \chi=\Big(\frac{\eta\Omega}{\delta}\Big)^2\pi \qquad (8)

where: $\chi$ — accumulated geometric phase ( $\chi=\pi/2$ is maximally entangling); $J_x$ — collective spin operator; $\delta$ — detuning of the two tones from the sidebands. The gate builds $\sigma_x\sigma_x$ coupling directly, the CNOT-class primitive. Finally, readout is a photon-counting test: one state scatters on a closed cycling transition (bright), the other stays dark, and over a few hundred microseconds the two Poissonian count distributions barely overlap.

P_{\mathrm{bright}}=\frac{1}{2}\Big(1-\mathrm{erf}\!\Big[\frac{n_{\mathrm{th}}-\bar n_{\mathrm{dark}}}{\sqrt{2}\,\sigma}\Big]\Big),\qquad \bar n_{\mathrm{min}}\approx\Big(\frac{\Gamma}{2\nu}\Big)^2\ll1 \qquad (9)

where: $n_{\mathrm{th}}$ — count threshold; $\bar n_{\mathrm{dark}}$ — mean dark counts; $\sigma$ — counting noise; $\Gamma$ — cooling-transition linewidth; $\bar n_{\mathrm{min}}$ — residual phonon number after cooling. The right-hand relation is the cooling floor: for $\nu\gg\Gamma$ the motion sits essentially in its ground state.

Derivation AWhy the trap must oscillate and how the pseudopotential appears

The goal is to prove that no static field can trap an ion in three dimensions, then show how oscillating that same forbidden field builds a real well. We start from Laplace's equation for the geometric constraint, split the motion into a slow secular drift plus fast micromotion, solve for the micromotion, and average the force over one RF period. The residue is the pseudopotential — a harmonic trap built from a field that is confining nowhere at any instant.

Step 1. In charge-free space $\nabla^2\Phi=0$ , so for a quadrupole $\alpha+\beta+\gamma=0$ : at least one coefficient is negative and the static field expels the ion along that axis. A three-dimensional electrostatic minimum simply does not exist. Why: this is Earnshaw's theorem made concrete — the reason the platform runs on time-dependent fields.

Step 2. Write the RF equation of motion $m\ddot{\mathbf r}=qE_{\mathrm{rf}}(\mathbf r)\cos(\Omega_{\mathrm{rf}}t)$ with $E_{\mathrm{rf}}=-\nabla\Phi_{\mathrm{rf}}$ , and split the trajectory into a slow drift and a small fast wiggle, $\mathbf r=\mathbf R+\boldsymbol\xi$ . Why: the drive is far faster than the secular motion, so the two timescales separate cleanly.

Step 3. To leading order $m\ddot{\boldsymbol\xi}\approx qE_{\mathrm{rf}}(\mathbf R)\cos(\Omega_{\mathrm{rf}}t)$ , giving $\boldsymbol\xi\approx -qE_{\mathrm{rf}}(\mathbf R)\cos(\Omega_{\mathrm{rf}}t)/(m\Omega_{\mathrm{rf}}^2)$ — the micromotion, in phase with the drive and shrinking as $\Omega_{\mathrm{rf}}$ grows. Why: a fast drive gives the ion no time to run away before the field reverses.

Step 4. Average the force on the drift $\mathbf R$ over one RF period. The cross term $\langle qE_{\mathrm{rf}}(\mathbf R+\boldsymbol\xi)\cos(\Omega_{\mathrm{rf}}t)\rangle$ survives because $\boldsymbol\xi$ itself oscillates as $\cos(\Omega_{\mathrm{rf}}t)$ , leaving $-\nabla\psi_{\mathrm{pseudo}}$ with $\psi_{\mathrm{pseudo}}=q^2|E_{\mathrm{rf}}|^2/(4m\Omega_{\mathrm{rf}}^2)$ . Why: averaging a driven oscillator's response against its own drive is the standard ponderomotive potential, confining wherever the RF field grows outward. *Note: $\alpha+\beta+\gamma=0$ follows directly from $\nabla^2\Phi=0$ , and the pseudopotential $\psi=q^2|\nabla\Phi_{\mathrm{rf}}|^2/(4m\Omega_{\mathrm{rf}}^2)\ge0$ is manifestly non-negative — confirming that the time-averaged well always confines.*

Go deeper · Derivation B · 1 minCarrier and sideband Rabi frequencies from the Lamb–Dicke expansion

Derivation BCarrier and sideband Rabi frequencies from the Lamb–Dicke expansion

The aim is to extract, from the full laser–ion coupling, the three transition rates that make trapped-ion control possible: the carrier that flips the spin alone, and the red and blue sidebands that exchange one phonon with the spin. We expand the motional displacement operator in the small parameter $\eta$ and identify which term becomes resonant for each detuning $\delta$ ; the bosonic $\sqrt{n}$ factors that fall out set the speed of every gate and give ground-state cooling a natural stopping point.

Step 1. In the Lamb–Dicke regime $\eta\sqrt{2\bar n+1}\ll1$ , expand $e^{i\eta(ae^{-i\nu t}+a^\dagger e^{i\nu t})}\approx 1+i\eta(ae^{-i\nu t}+a^\dagger e^{i\nu t})$ inside $H_I$ . Why: the wavepacket is far smaller than the wavelength, so first order in $\eta$ captures single-phonon processes and higher terms are negligible.

Step 2. The " $1$ " term multiplies $e^{-i\delta t}$ ; choosing $\delta=0$ makes it stationary, a resonant spin flip with $\Omega_{n,n}=\Omega$ , independent of $n$ . Why: this is the carrier — pure single-qubit rotation, insensitive to the motional state.

Step 3. The $a$ term carries $e^{-i(\delta+\nu)t}$ ; setting $\delta=-\nu$ freezes it, coupling $|g,n\rangle\leftrightarrow|e,n-1\rangle$ at rate $\Omega\,\eta\sqrt{n}$ from $a|n\rangle=\sqrt{n}\,|n-1\rangle$ . Why: the red sideband removes a phonon while exciting the ion — the coherent half of every cooling cycle.

Step 4. The $a^\dagger$ term carries $e^{-i(\delta-\nu)t}$ ; setting $\delta=+\nu$ couples $|g,n\rangle\leftrightarrow|e,n+1\rangle$ at rate $\Omega\,\eta\sqrt{n+1}$ . At $n=0$ the red rate vanishes but the blue does not, so $|g,0\rangle$ is the dark state that ends cooling. Why: the $\sqrt{n}$ versus $\sqrt{n+1}$ asymmetry makes the ground state a trap for the cooling dynamics. *The three rates $\Omega,\ \Omega\eta\sqrt{n},\ \Omega\eta\sqrt{n+1}$ are confirmed numerically for $^{40}\mathrm{Ca}^+$ at 729 nm with $\nu=2\pi\times1\,$ MHz, giving $\eta=0.097\ll1$ .*

Go deeper · Derivation C · 1 minThe Mølmer–Sørensen geometric phase from a closed phase-space loop

Derivation CThe Mølmer–Sørensen geometric phase from a closed phase-space loop

The goal is to explain why a two-tone laser field entangles two ions with a phase that does not care about the temperature of the motion — the property that turned trapped ions into a workhorse technology. We drive symmetrically about the red and blue sidebands so no single-phonon transition is resonant, and follow the motional mode's trajectory in phase space as it closes on itself at the right gate time. The enclosed area is a geometric phase in $J_x^2$ ; because the loop closes for every starting point, the phonon disentangles and only the spin–spin phase survives.

Step 1. Apply two tones at $\omega_0\pm(\nu-\delta)$ . Neither drives a real sideband (each is detuned by $\delta$ ), but together they produce a force $\propto J_x=\tfrac12\sum_i\sigma_x^{(i)}$ that displaces the motional mode in a $J_x$ -dependent direction. Why: symmetric detuning cancels the resonant single-phonon terms, leaving a second-order spin-dependent push — the essence of a geometric gate.

Step 2. The mode traces $\hat A(\tau)=\frac{\eta\Omega}{2\delta}\big(1-e^{i\delta\tau}\big)J_x$ , a circle of radius $\propto\eta\Omega/\delta$ . At the gate time $\tau=2\pi/\delta$ the loop closes, $\hat A=0$ , and the motion returns to its start regardless of the initial phonon number. Why: closure for all $n$ is the whole robustness argument — a warm ion works as well as a cold one.

Step 3. By the displacement-operator algebra, a closed loop of area $\mathcal A$ imprints $e^{-i\Phi J_x^2}$ with $\Phi=(\eta\Omega/\delta)^2\pi$ equal to (twice) the enclosed area. Why: this is Berry's geometric phase realized mechanically — it depends only on the area swept, not the path.

Step 4. Choose the intensity so $\Phi=\pi/2$ ; then $U=\exp[-i(\pi/2)J_x^2]$ , locally equivalent to a maximally entangling CNOT-class gate. Why: a quarter-turn of enclosed area turns the geometric phase into a full two-qubit gate, the primitive of universal logic. *The loop closure $\hat A(2\pi/\delta)=0$ and the geometric phase $\Phi=(\eta\Omega/\delta)^2\pi$ together confirm that the motion fully disentangles at the gate time.*

Limiting cases

The physics collapses into clean regimes set by $\eta$ and by the ratio of the trap frequency $\nu$ to the linewidth and coupling.

Case	Result	Meaning
Lamb–Dicke regime	$\eta\sqrt{2\bar n+1}\ll1$	Ion wavepacket much smaller than the wavelength; coupling expands to first order in $\eta$ , giving clean carrier and first-order sidebands.
Resolved-sideband regime	$\nu\gg\Gamma,\ \Omega$	Trap frequency exceeds linewidth and Rabi rate, so sidebands at $\omega_0\pm\nu$ are distinct. Enables ground-state cooling to $\bar n\sim(\Gamma/2\nu)^2$ .
Carrier / free-atom limit	$\eta\to0$	Recoil decouples spin from motion; the laser drives only the bare transition. Single-qubit rotations live here.
Many-ion / spectral crowding	$N$ large $\Rightarrow\Delta\nu_{\text{modes}}\to0$	The $3N$ modes crowd together, making single-mode addressing harder and gates slower; motivates QCCD and photonic modularity.

Worked examples

1 · The Lamb–Dicke parameter of a calcium qubit (fully worked)

A $^{40}\mathrm{Ca}^+$ ion is driven on its $S_{1/2}\leftrightarrow D_{5/2}$ optical qubit transition at $\lambda=729\,\mathrm{nm}$ in a trap of axial secular frequency $\nu=2\pi\times1\,\mathrm{MHz}$ . Find the zero-point extent $x_0$ of the motion, the Lamb–Dicke parameter $\eta$ , and confirm the Lamb–Dicke regime holds after ground-state cooling.

Step 1. Zero-point extent. With $m=40\times1.6605\times10^{-27}=6.642\times10^{-26}\,\mathrm{kg}$ and $\nu=2\pi\times10^6\,\mathrm{s^{-1}}$ ,

x_0=\sqrt{\frac{\hbar}{2m\nu}}=\sqrt{\frac{1.0546\times10^{-34}}{2(6.642\times10^{-26})(6.283\times10^{6})}}=\mathbf{1.12\times10^{-8}\ m}=11.2\ \mathrm{nm}.

The ground-state wavepacket is about 11 nm — tiny compared with the 729 nm wavelength.

Step 2. Lamb–Dicke parameter. The laser wavenumber is $k=2\pi/\lambda=8.62\times10^{6}\,\mathrm{m^{-1}}$ , so

\eta=k\,x_0=(8.62\times10^{6})(1.12\times10^{-8})=\mathbf{0.097}.

This is the canonical " $\eta\sim0.1$ " for an optical qubit — the sidebands are about ten times weaker than the carrier.

Step 3. Regime check. After resolved-sideband cooling to $\bar n\approx0.05$ , $\eta\sqrt{2\bar n+1}=0.097\times\sqrt{1.1}=\mathbf{0.10}\ll1$ , so the first-order expansion behind equation (6) is well justified and single-phonon sidebands dominate. η = 0.097, x₀ = 11.2 nm, Lamb–Dicke regime confirmed.

2 · Timing a Mølmer–Sørensen gate

Using $\eta=0.10$ , choose parameters for a maximally entangling MS gate ( $\Phi=\pi/2$ ) and find the gate time and Rabi frequency for a detuning $\delta=2\pi\times25\,\mathrm{kHz}$ .

Step 1. Entangling condition. Setting $\Phi=(\eta\Omega/\delta)^2\pi=\pi/2$ gives $\eta\Omega/\delta=1/\sqrt2$ , so the carrier Rabi frequency must be $\Omega=\delta/(\eta\sqrt2)=(2\pi\times25\,\mathrm{kHz})/(0.10\times1.414)=\mathbf{2\pi\times177\ kHz}$ .

Step 2. Gate time. The loop closes at $\tau=2\pi/\delta=1/(25\times10^3)=40\,\mu\mathrm{s}$ , setting the two-qubit gate duration in this example. A hyperfine clock qubit with coherence lasting several seconds could still accommodate $\sim10^5$ such gates within one coherence time. A 40 μs gate at Ω = 2π × 177 kHz, comfortably within coherence.

Physics of the portrait

Trapped-ion quantum computing is inseparable from one experimentalist who turned single atoms into the most precisely controlled objects in physics.

Origins · 1 minDavid J. Wineland and how this was discovered

David J. Wineland

1944– · Nobel Prize in Physics 2012 · laser cooling and quantum control of trapped ions

David Wineland built the experimental foundation on which trapped-ion quantum computing stands. Working at NIST in Boulder from the late 1970s, he and his group demonstrated laser cooling of trapped ions in 1978 — among the first cooling of matter with light. In 1995, months after Cirac and Zoller proposed the trapped-ion quantum computer, his group demonstrated the first controlled-NOT gate between an ion's internal and motional states — the first experimental quantum logic gate in any system. His laboratory pioneered resolved-sideband cooling to the motional ground state, near-unit-fidelity quantum-jump detection, Schrödinger-cat states of a single ion's motion, and quantum-logic spectroscopy — the trick that made $^{27}\mathrm{Al}^+$ optical clocks the most accurate ever built. He shared the 2012 Nobel Prize in Physics with Serge Haroche "for ground-breaking experimental methods that enable measuring and manipulation of individual quantum systems." Nearly every technique in this article carries his fingerprints.

History

1953

— Wolfgang Paul introduces the radio-frequency quadrupole ion trap

1978

— Wineland and Neuhauser demonstrate laser cooling of trapped ions

1995

— Cirac and Zoller propose the ion-trap quantum computer; the first CNOT is demonstrated

1999

— Sørensen and Mølmer introduce the thermally robust geometric-phase gate

2012

— Wineland shares the Nobel Prize for control of individual quantum systems

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information, §7.6 (the ion-trap quantum computer) · Preskill, Lecture Notes for Physics 219, physical-implementations chapter · Wineland et al., "Experimental Issues in Coherent Quantum-State Manipulation of Trapped Atomic Ions," J. Res. NIST 103, 259 (1998) · Leibfried, Blatt, Monroe, Wineland, "Quantum dynamics of single trapped ions," Rev. Mod. Phys. 75, 281 (2003) · Häffner, Roos, Blatt, "Quantum computing with trapped ions," Phys. Rep. 469, 155 (2008) · Bruzewicz, Chiaverini, McConnell, Sage, "Trapped-ion quantum computing: Progress and challenges," Appl. Phys. Rev. 6, 021314 (2019). The key results in this article are: $\alpha+\beta+\gamma=0$ from $\nabla^2\Phi=0$ ; pseudopotential $\psi_{\mathrm{pseudo}}=q^2|\nabla\Phi_{\mathrm{rf}}|^2/(4m\Omega_{\mathrm{rf}}^2)$ via secular/micromotion separation; Lamb–Dicke rates $\Omega,\ \Omega\eta\sqrt{n},\ \Omega\eta\sqrt{n+1}$ ; normal-mode frequencies $\nu_1=\nu_z$ , $\nu_2=\sqrt{3}\,\nu_z$ ; MS phase $\Phi=(\eta\Omega/\delta)^2\pi$ with loop closure at $\tau=2\pi/\delta$ ; and $x_0=11.2\,$ nm, $\eta=0.097$ for $^{40}\mathrm{Ca}^+$ at 729 nm, $\nu=2\pi\times1\,$ MHz. Quoted fidelity and clock figures are representative published values.

Builds on: DiVincenzo Criteria · Quantum Harmonic Oscillator · Two-Level Systems and Rabi Oscillations. Leads to: Cirac–Zoller Gate · Mølmer–Sørensen Gate · Sideband Cooling · Superconducting Qubits.

Ask the tutor: Want to see how the phase-space loop of a Mølmer–Sørensen gate closes step by step, or how sideband cooling drives the ion into its motional ground state?

Why are trapped ions so much more coherent than most other qubits?

The qubit lives in the internal electronic structure of an atom, which nature makes identical for every ion and which couples only weakly to the outside world. A hyperfine "clock" transition is, to first order, insensitive to magnetic field, so ambient noise barely dephases it — coherence reaches seconds to minutes — and there are no fabrication imperfections making one qubit differ from another. The cost of this isolation is speed: because the atoms ignore their environment they also ignore each other, and the only way to make two interact is to reach in through the shared motion with a laser, which is slow.

What exactly is a phonon here, and why is it called a bus?

A phonon is one quantum of a collective vibrational mode of the ion crystal. Because the ions repel through the Coulomb force they cannot move independently; their oscillations organize into normal modes, each a quantum harmonic oscillator whose excitations are phonons. The lowest — centre-of-mass — has every ion swinging in phase, so a laser on any single ion can add or remove a shared phonon that all ions feel. That shared quantum is the "bus": you write a spin state into motion on one ion and read it out on another, entangling ions that are not neighbours — the analogue of a wire between distant components.

How is a red sideband different from a blue sideband?

Both flip the qubit's spin, but they do opposite things to the motion. The red sideband, tuned one trap frequency below the bare transition ( $\delta=-\nu$ ), removes one phonon as it excites the ion, at rate $\Omega\eta\sqrt{n}$ . The blue sideband, tuned one trap frequency above ( $\delta=+\nu$ ), adds one phonon, at rate $\Omega\eta\sqrt{n+1}$ . The asymmetry matters at the bottom of the ladder: at $n=0$ the red rate is $\Omega\eta\sqrt0=0$ while the blue is nonzero — which is exactly why repeatedly driving the red sideband cools the ion and then stops cleanly in the motional ground state.

Why is the Mølmer–Sørensen gate preferred over the original Cirac–Zoller gate?

The Cirac–Zoller gate uses a single, specific phonon as an intermediary and so demands that the motion start in the ground state — any residual thermal phonon corrupts it. The Mølmer–Sørensen gate instead applies a bichromatic drive that never resonantly excites a phonon, producing a spin-dependent force that sends the motional mode on a closed loop in phase space. Because the loop closes for every initial phonon number, the motion fully disentangles at the end and the geometric phase depends only on the enclosed area, not on temperature. That robustness to warm, Doppler-cooled motion made it the experimental workhorse.

What is micromotion and why do experimenters fight to minimize it?

Micromotion is the fast, driven wiggle of the ion at the RF trap frequency, on top of its slow secular motion. It is unavoidable in principle — it is what averages to produce the confining pseudopotential — but excess micromotion appears when stray static fields push the ion off the RF null, where the oscillating field is nonzero. That excess Doppler-shifts and amplitude-modulates the laser interaction, weakening the sidebands, shifting clock transitions, and adding heating. Experimenters null it with compensation electrodes — a routine daily calibration of any trapped-ion apparatus.

How are trapped-ion systems supposed to scale beyond tens of qubits?

A single trap cannot host arbitrarily many well-controlled ions: as the chain grows its motional modes crowd together, addressing one mode gets harder, and gates slow down. The quantum-charge-coupled-device (QCCD) architecture keeps chains short and shuttles ions between dedicated memory, gate, and readout zones on a chip, so any gate acts on a small, clean subregister — how today's commercial machines scale. Beyond a single chip, photonic interconnects entangle ions in separate traps by interfering their emitted photons on a beamsplitter, a coincidence heralding a remote pair. Networking many small QCCD modules this way is the leading route to a large fault-tolerant machine.