Simple Algorithm
Quantum Computing · Rydberg State · 26 min

Scaling Ions: QCCD & Photonic Links

Shuttling ions and networking traps.

A single chain of ions gives you all-to-all connectivity, but not a million qubits. The two leading answers are to move the ions, shuttling them between zones of a chip-scale trap like charge packets in a camera sensor, and to link separate traps with photons. Each trades one hard problem for another.

What you’ll learn

  • Quantify why gates slow down and errors grow as a single chain gets longer.
  • Describe the QCCD architecture: transport, splitting and merging, reordering, junctions and sympathetic cooling.
  • Derive the motional excitation caused by moving a harmonic well, and the condition for excitation-free transport.
  • Compute heralded remote-entanglement rates and explain the 50% limit of linear-optics Bell measurements.
  • Show that one shared Bell pair, local operations and two classical bits implement a remote CNOT.

Why one chain does not scale

Three effects bite as a chain of NN ions grows.

  1. Weaker coupling. In the center-of-mass (COM) mode every ion moves with amplitude 1/N1/\sqrt N of the mode coordinate, so its Lamb–Dicke parameter for that mode is η/N\eta/\sqrt N. At fixed laser power, gates slow down.
  2. Crowded spectra. NN ions have NN modes per direction. The transverse modes most often used for gates crowd into a narrow band below the COM frequency, so isolating one mode needs long gates, and using all of them needs pulse shaping that closes ever more loops.
  3. Heating. Electric-field noise at the ions heats the modes, and the COM mode heats fastest.
DerivationThe COM heating rate grows with N
  1. nˉ˙=SF(ν)4Mℏν\dot{\bar n} = \frac{S_F(\nu)}{4M\hbar\nu}

    Heating rate of an oscillator of mass MM and frequency ν\nu driven by a fluctuating force with spectral density SFS_F at the mode frequency (in the convention common in the ion-trap literature).

  2. one ion: SF=e2SE  ⇒  nˉ˙1=e2SE(ν)4mℏν\text{one ion: } S_F = e^2S_E\;\Rightarrow\;\dot{\bar n}_1 = \frac{e^2S_E(\nu)}{4m\hbar\nu}

    A single ion of mass mm in field noise of spectral density SES_E.

  3. COM of N ions: F=NeE,  M=Nm  ⇒  nˉ˙COM=N2e2SE4Nmℏν=N nˉ˙1\text{COM of } N \text{ ions: } F = NeE,\ \ M = Nm\;\Rightarrow\;\dot{\bar n}_{\rm COM} = \frac{N^2e^2S_E}{4Nm\hbar\nu} = N\,\dot{\bar n}_1

    Spatially uniform noise pushes all ions together, so it couples only to the COM mode, whose heating rate grows linearly with NN. Modes that leave the center of mass fixed are insensitive to uniform noise and heat far more slowly.

  4. Empirically, SES_E grows steeply as the ion approaches the electrodes, roughly as d−4d^{-4} in many experiments, and cryogenic operation and surface cleaning reduce it by orders of magnitude (Brownnutt et al., 2015). Small chip traps pay for their density with heating.

Check your understanding

A 16-ion chain sits in spatially uniform electric-field noise. Compared with a single ion in the same trap, its COM mode heats

  1. 16 times more slowly
  2. at the same rate
  3. 4 times faster
  4. 16 times faster
Show the answer

D. 16 times faster

The force on the center of mass is 16eE16eE, so its noise power grows by 16216^2, while the mass grows by 16: nˉ˙COM=16 nˉ˙1\dot{\bar n}_{\rm COM} = 16\,\dot{\bar n}_1.

The quantum CCD

Kielpinski, Monroe and Wineland (2002) proposed keeping every chain small and moving the ions instead. A chip carries many trapping zones defined by segmented DC electrodes on top of the RF pseudopotential. Changing the electrode voltages in time transports potential wells, with their ions, between memory, interaction and readout zones. Two-qubit gates only ever act on a few ions sharing a well, so the gate physics stays that of a short chain.

PrimitiveWhat happensMain cost
Linear transportA well carries ions along the trap axis, over tens to hundreds of μmMotional excitation unless the velocity profile is well chosen (next step)
Split and mergeOne well holding two ions becomes two wells, or the reverseThe confinement passes through a weak, nearly quartic point, which tends to heat the ions
ReorderA two-ion crystal is rotated to swap the ions' positionsNeeded to bring arbitrary pairs together; adds time and heating
Junction transportIons are routed through X, Y or T junctions between trap armsSmall bumps in the RF pseudopotential near junctions can excite motion
Sympathetic coolingA coolant ion of another species, cooled next to the qubits, re-cools their shared modesTakes time; together with transport it often dominates the circuit duration

Moving an ion without shaking it

DerivationProof
  1. H=p22m+12mω2(x−x0(t))2H = \frac{p^2}{2m} + \frac12m\omega^2\big(x - x_0(t)\big)^2

    An ion in a well whose center moves.

  2. H′=T†HT−iℏ T†T˙=ℏω a†a−x˙0(t) p,T=e−ix0(t)p/ℏH' = T^\dagger HT - i\hbar\,T^\dagger\dot T = \hbar\omega\,a^\dagger a - \dot x_0(t)\,p,\qquad T = e^{-ix_0(t)p/\hbar}

    Move to the frame of the well (dropping the zero-point constant). The frame's motion appears as a coupling to the momentum.

  3. p=imℏω2 (a†−a)  ⇒  HI′(t)=−ix˙0(t)mℏω2(a†eiωt−a e−iωt)p = i\sqrt{\tfrac{m\hbar\omega}{2}}\,(a^\dagger - a)\;\Rightarrow\; H'_I(t) = -i\dot x_0(t)\sqrt{\tfrac{m\hbar\omega}{2}}\big(a^\dagger e^{i\omega t} - a\,e^{-i\omega t}\big)

    In the interaction picture of the oscillator this is a linearly driven oscillator.

  4. UI(T)=eiφ D^(α),α=−mω2ℏ∫0Tx˙0(t) eiωt dt■U_I(T) = e^{i\varphi}\,\hat D(\alpha),\qquad \alpha = -\sqrt{\frac{m\omega}{2\hbar}}\int_0^T\dot x_0(t)\,e^{i\omega t}\,dt\qquad\blacksquare

    As for the MS gate, the commutator of the drive at two times is a c-number, so the Magnus series stops at second order: a displacement times a global phase. Once the well is at rest again, the moving and lab frames agree on the motion relative to the well.

Real wells are not perfectly harmonic, their frequency changes along the path, and electrode voltages have finite bandwidth and noise, so practical waveforms are optimized numerically. The theorem still sets the logic: transport is a forced oscillator, and the excitation is the Fourier component of the forcing at the trap frequency.

Linking traps with light

A QCCD chip still lives in one vacuum system. To go further, separate traps (modules) can share entanglement through photons. Each ion is excited so that it emits a single photon whose polarization (or frequency, or time bin) is entangled with its qubit, 12(∣↑⟩∣H⟩+∣↓⟩∣V⟩)\tfrac{1}{\sqrt2}(\ket{\uparrow}\ket{H} + \ket{\downarrow}\ket{V}). Photons from two modules meet at a beam splitter in front of detectors, and particular two-detector coincidences project the two distant ions onto a Bell state. The ions never meet: this is entanglement swapping on the photons, first demonstrated between ions by Moehring et al. (2007).

DerivationThe heralding rate
  1. psucc=12 ηAηB,η=ηcollect ηfiber ηdetectp_{\rm succ} = \tfrac12\,\eta_A\eta_B,\qquad \eta = \eta_{\rm collect}\,\eta_{\rm fiber}\,\eta_{\rm detect}

    Both photons must arrive and be detected (probability ηAηB\eta_A\eta_B), and a linear-optics Bell analyzer recognizes at most half of the Bell states (theorem below). Without polarization-resolving detection only Ψ−\Psi^- is recognized and the factor becomes 14\tfrac14.

  2. R=psucc RattemptR = p_{\rm succ}\,R_{\rm attempt}

    The attempt rate is limited by the excitation pulse, the photon emission time and occasional re-cooling.

  3. Collection is usually the bottleneck: even a high-numerical-aperture lens captures only a modest fraction of a photon emitted into all directions, and every loss enters squared. Optical cavities, better optics and conversion to low-loss telecom wavelengths attack this.

Predict before you look

Each module delivers its photon to a detector with total efficiency η=2%\eta = 2\%. The two-photon Bell analyzer succeeds with probability 12ηAηB\tfrac12\eta_A\eta_B, and you make 10610^6 attempts per second. About how many Bell pairs per second do you herald?

Show the answer

200 pairs/s

p=12(0.02)2=2×10−4p = \tfrac12(0.02)^2 = 2\times10^{-4}; times 10610^6 attempts per second gives about 200 pairs per second. Rates of order 10210^2 per second were reported in the early 2020s, still well below local gate rates, so modules must store link entanglement in memory qubits while computing.

A Bell pair between modules is not yet a gate. It becomes one through gate teleportation: one shared Bell pair, local operations and one classical bit sent each way implement a CNOT between qubits in different modules, and Eisert, Jacobs, Papadopoulos and Plenio (2000) showed these resources are optimal. Verify it below. The composer has no classical feed-forward, so each "measure, send the bit, correct" step is replaced by the equivalent quantum-controlled correction (the deferred-measurement principle), which gives identical statistics.

Key takeaways

  • Long chains stall: per-ion coupling to the COM mode falls as 1/N1/\sqrt N, transverse spectra crowd, and uniform field noise heats the COM mode NN times faster.
  • QCCD keeps chains small and moves ions between zones; sympathetic cooling with a second species removes the heat that transport adds.
  • Moving a harmonic well is a forced oscillator: the excitation is the Fourier component of the well's velocity at the trap frequency, so suitably timed fast transport can be excitation-free.
  • Photonic links herald remote Bell pairs with probability ≤12ηAηB\le\tfrac12\eta_A\eta_B per attempt; one Bell pair plus local operations and one bit each way give a remote CNOT.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

What is the central idea of the QCCD architecture?

  1. Use one very long chain and address its ions with more laser beams
  2. Couple every ion to a common microwave cavity
  3. Keep interacting chains small, and move ions between memory, interaction and readout zones by changing electrode voltages
  4. Replace the RF pseudopotential with static magnetic fields
Show the answer

C. Keep interacting chains small, and move ions between memory, interaction and readout zones by changing electrode voltages

QCCD turns connectivity into a routing problem: ions travel to wherever they need to interact, so every gate acts on a short, well-behaved chain.

Question 2

Why is sympathetic cooling in QCCD machines done with a different ion species?

  1. The coolant's cooling light is far from resonance with the qubit ions, so it removes motional energy without scattering on the qubits
  2. Different species carry different charges, so they cool faster
  3. A heavier coolant makes the chain longer and more stable
  4. The qubit species cannot be laser-cooled at all
Show the answer

A. The coolant's cooling light is far from resonance with the qubit ions, so it removes motional energy without scattering on the qubits

Cooling relies on scattering photons. A second species can scatter as many as needed while the qubit ions, whose transitions are far away, keep their quantum states.

Question 3

A harmonic well is moved with a velocity profile whose Fourier transform at the trap frequency ω\omega vanishes. Ideally, the ion arrives

  1. in a coherent state whose amplitude grows with the distance traveled
  2. heated by an amount proportional to the transport time
  3. in the motional ground state, whatever its initial state
  4. in the same motional state it started in, whatever the distance or speed
Show the answer

D. in the same motional state it started in, whatever the distance or speed

The excitation is the displacement α∝∫x˙0eiωtdt\alpha \propto \int\dot x_0e^{i\omega t}dt, which vanishes by assumption. Transport is unitary on the motion; it neither heats nor cools in the ideal harmonic case.

Question 4

In a two-photon heralding scheme, you improve photon collection by a factor of 3 in both modules. The remote-entanglement rate improves by

  1. 3×
  2. 9×
  3. 6×
  4. √3×
Show the answer

B. 9×

psucc=12ηAηBp_{\rm succ} = \tfrac12\eta_A\eta_B is linear in each efficiency, and both improve: 3×3=93\times3 = 9.

Further reading

  • D. Kielpinski, C. Monroe & D. J. Wineland, "Architecture for a large-scale ion-trap quantum computer", Nature 417, 709 (2002).
  • J. M. Pino et al., "Demonstration of the trapped-ion quantum CCD computer architecture", Nature (2021).
  • D. L. Moehring et al., "Entanglement of single-atom quantum bits at a distance", Nature 449, 68 (2007).
  • C. Monroe et al., "Large-scale modular quantum-computer architecture with atomic memory and photonic interconnects", Physical Review A 89, 022317 (2014).
  • J. Eisert, K. Jacobs, P. Papadopoulos & M. B. Plenio, "Optimal local implementation of nonlocal quantum gates", Physical Review A 62, 052317 (2000).