Simple Algorithm
Quantum Computing · Ground State · 16 min

Quantum Hardware

Superconducting circuits, trapped ions, atoms and light: how qubits are built.

A qubit is an idea: two amplitudes on a sphere. To compute with one, someone has to build it, out of a circuit colder than outer space, a single atom floating in vacuum, or a lone photon. Every way of doing it strikes a different bargain between protecting quantum information and being able to steer it.

What you’ll learn

  • Explain what a physical system needs in order to serve as a qubit, and why isolation and control pull in opposite directions.
  • Drive a qubit with a resonant pulse and explain Rabi oscillations, π\pi-pulses and detuning.
  • Interpret T1T_1 and T2T_2, and estimate how many gates fit inside a coherence time.
  • Compare superconducting, trapped-ion, neutral-atom, photonic, spin and topological qubits by their strengths and weaknesses.
  • Describe honestly where the field stands as of the mid-2020s, and how quantum computers are graded.

Building a qubit: the wish list

Every qubit you have met so far was a ket on a page. A real one is a physical object with two distinguishable states that behave quantum mechanically: two energy levels of an atom, the spin of an electron, the polarization of a photon, or two oscillation states of a superconducting circuit.

Here is the catch. Almost every two-level system in the world is useless as a qubit. A coin has two states, but any superposition of heads and tails would be destroyed instantly, because air molecules and light are constantly "measuring" it. A good qubit must be quantum, and stay quantum long enough to compute.

Think first Qubit A stays coherent for 1 second but needs 1 millisecond per gate. Qubit B stays coherent for only 100 μs but needs just 50 ns per gate. Which can run more operations before its quantum information fades?

Qubit B: about 100 μs/50 ns=2000100\ \mu\text{s}/50\ \text{ns} = 2000 gates, against 1 s/1 ms=10001\ \text{s}/1\ \text{ms} = 1000 for qubit A. What counts is not coherence time on its own but the number of operations that fit inside it, and, as you will see, how accurate each operation is.

Steering a qubit: pulses and Rabi oscillations

How does a gate actually happen? A qubit's two levels differ in energy, and that energy difference corresponds to a frequency ff: about 5 GHz (microwaves) for a superconducting qubit, optical or microwave frequencies for atoms and ions.

Shine an oscillating field at exactly that frequency and the qubit starts to swing smoothly from ∣0⟩\ket{0} to ∣1⟩\ket{1} and back. This is a Rabi oscillation. The gate is the pulse: its duration sets the rotation angle and its phase sets the rotation axis.

rotation angle=Ω τ\text{rotation angle} = \Omega\,\tau
On resonance, a drive of strength Ω\Omega (the Rabi frequency, in radians per unit time) applied for a time τ\tau rotates the qubit by Ωτ\Omega\tau on the Bloch sphere. A full flip, ∣0⟩→∣1⟩\ket0\to\ket1, is a rotation by π\pi: a π\pi-pulse.
Predict before you look

In the simulator below the drive starts at Ω=2\Omega = 2 rad/μs. How long must a resonant pulse last to flip ∣0⟩\ket0 into ∣1⟩\ket1?

Show the answer

1.57 μs

A flip is a rotation by π\pi, so τ=π/Ω=π/2≈1.57\tau = \pi/\Omega = \pi/2\approx1.57 μs. Real superconducting qubits are driven far harder, so their π\pi-pulses take only about 10–50 ns.

This is why calibration matters so much. A pulse that is 1% too long, or slightly off frequency, leaves the qubit a little away from its target every time, and such small errors add up over thousands of gates.

The clock is ticking: decoherence

Left alone, a qubit is slowly corrupted by its environment in two ways, together called decoherence:

  • Energy relaxation, on a timescale T1T_1: an excited qubit gives its energy away and falls from ∣1⟩\ket1 to ∣0⟩\ket0.
  • Dephasing, on a timescale T2T_2: the relative phase of a superposition drifts randomly, so the interference that algorithms rely on washes out.
Predict before you look

A qubit has T2=100T_2 = 100 μs and each gate takes 50 ns. Roughly how many gates fit into one T2T_2?

Show the answer

2000 gates

100 μs/50 ns=2000100\ \mu\text{s}/50\ \text{ns} = 2000. In practice you want to finish well before T2T_2, because coherence decays continuously: by t=T2/10t = T_2/10 you have already lost about 10% of it.

Artificial atoms: superconducting circuits

Cool aluminum below about 1 K and it superconducts: current flows with zero resistance. Build a tiny circuit from a capacitor and a Josephson junction, an insulating barrier only a nanometer or two thick, and the circuit acquires discrete, unequally spaced energy levels, just like an atom. The most widely used design is the transmon.

The circuit is under a millimeter across and contains billions of electrons, yet it behaves as a single quantum object. It must sit in a dilution refrigerator at about 10 mK: at 5 GHz, the energy of one microwave photon corresponds to a temperature of only about 0.24 K, so any warmer environment would randomly excite the qubit.

Think first An ordinary electrical LC circuit also oscillates at a few GHz. Why can't it serve as a qubit, and how does the Josephson junction fix this?

A plain LC circuit is a harmonic oscillator: its energy levels are equally spaced, so a pulse tuned to ∣0⟩→∣1⟩\ket0\to\ket1 also drives ∣1⟩→∣2⟩\ket1\to\ket2, ∣2⟩→∣3⟩\ket2\to\ket3, and so on up the ladder. The junction is a nonlinear inductor that makes the spacing uneven (the ∣1⟩→∣2⟩\ket1\to\ket2 transition sits a few hundred MHz below ∣0⟩→∣1⟩\ket0\to\ket1), so a carefully shaped pulse addresses only the two lowest levels.

Chips also have a limited native gate set. There is usually no CNOT "button": many superconducting processors (and neutral-atom machines) natively perform the controlled-Z (CZ) gate, which flips the sign of ∣11⟩\ket{11} only, and a compiler rewrites every circuit in terms of native gates. Try being the compiler.

Natural atoms: trapped ions and neutral atoms

Instead of engineering an artificial atom, why not use a real one? Every atom of a given isotope is identical, so the qubits come out the same, straight from nature.

Trapped ions. Ions, atoms with an electron removed, are held in ultra-high vacuum by oscillating electric fields and line up in a chain. Laser or microwave pulses drive single-qubit gates. Because the ions repel each other, they share collective vibrations, and this shared motion lets any pair in the chain be entangled.

Neutral atoms. Hundreds to thousands of uncharged atoms are held in arrays of tightly focused laser beams called optical tweezers. To entangle two neighbors, lasers briefly excite them to huge, strongly interacting Rydberg states. Atoms can even be moved around in the middle of a computation.

Typical figure (mid-2020s)SuperconductingTrapped ionsNeutral atoms
Two-qubit gate time~20–500 ns~10–200 μs~0.1–1 μs
Coherencetens to hundreds of μsseconds or longer~1 s (idle)
Connectivitynearest neighbors on a fixed chipall-to-all within a chainreconfigurable by moving atoms
Main challengewiring, crosstalk, material defectsslow gates; scaling beyond one chainatom loss; slower readout
Orders of magnitude only; individual devices vary widely, and the leading numbers move every year.
Check your understanding

Why can any two ions in a chain be entangled directly, even if they are not neighbors?

  1. Because all ions of the same isotope are identical
  2. Because the laser beam is wide enough to hit every ion at once
  3. Because the ions share collective vibrational modes, which act as a "bus" connecting every ion
  4. Because ions in vacuum never decohere
Show the answer

C. Because the ions share collective vibrational modes, which act as a "bus" connecting every ion

The ions repel one another, so they move together like masses on springs. A laser can make an ion's internal state push on a shared mode of motion, and that motion in turn acts on any other ion in the chain: an entangling gate between arbitrary pairs.

Light, spins and topology

Photons barely interact with anything, so they keep their quantum state well and travel far through optical fiber. That makes them the natural carriers for quantum networks. The same aloofness makes two-qubit gates hard: photonic schemes rely on measurements and feed-forward instead, and their main enemy is photon loss.

Spin qubits store information in the spin of single electrons trapped in silicon or germanium "quantum dots". They are tiny (around 100 nm), can be made with chip-industry techniques, and run at roughly 0.1–1 K, but no two dots are quite alike.

Topological qubits would store information non-locally, in exotic excitations called Majorana zero modes, so that local noise could not corrupt it. As of the mid-2020s this approach remains experimental: claims have been made, and the key signatures are still debated.

Think first Photons hardly decohere at all. Why aren't all quantum computers photonic?

Because the property that protects them, not interacting with their surroundings, also stops them from interacting with each other. Entangling gates between photons must be built from measurements and are probabilistic, every component loses some photons, and making single photons on demand is hard. Photonic architectures must design around all three.

Grading machines, and the road ahead

How do you compare a 30-qubit ion trap with a 1000-qubit chip? Not by counting qubits. Researchers report:

  • Fidelities: how close each gate, state preparation and readout comes to ideal (the fidelity).
  • Randomized benchmarking: run long random gate sequences that should undo themselves, and watch how fast the success rate decays. The decay rate gives the average error per gate, untainted by readout errors.
  • Holistic benchmarks such as quantum volume: the largest "square" random circuit (as many layers as qubits) that the whole machine runs reliably.

The Excited State level of this topic turns each part of the story into real physics:

Excited State lessonWhat you will do there
DiVincenzo Criteria & BenchmarksMake the wish list quantitative: gate fidelities, T1T_1 and T2T_2, readout and SPAM errors, and why randomized benchmarking works.
Superconducting QubitsQuantize a circuit, derive the transmon's anharmonicity, drive it with microwaves and read it out through a resonator.
Trapped IonsTrap ions with oscillating fields and entangle them through their shared motion.
Neutral Atoms & Rydberg ArraysUse the Rydberg blockade to entangle atoms, and rearrange arrays mid-circuit.
Photons, Spins & Topological QubitsLinear optics and fusion, spins in quantum dots, and the bet on Majorana modes.
At the Rydberg State level these branch into randomized-benchmarking theory, quantum volume and cross-entropy benchmarks, gate set tomography, transmon physics, two-qubit gates and couplers, dispersive readout and cryogenics, and more.

Key takeaways

  • A useful qubit needs DiVincenzo's ingredients: scalable, resettable, long-lived, universally controllable, and individually measurable.
  • Gates are pulses: a resonant drive rotates the qubit (a Rabi oscillation), the duration sets the angle, and detuning spoils the rotation.
  • T1T_1 (energy loss) and T2T_2 (phase loss) set the clock; what counts is how many accurate gates fit inside it.
  • Superconducting circuits are fast but need ~10 mK; ions and atoms are identical and long-lived but slower; photons travel well but barely interact; spins are tiny; topological qubits are still unproven.
  • As of the mid-2020s machines have up to thousands of noisy physical qubits and the first logical qubits; useful fault-tolerant computing still needs much more scale and lower error rates.

Check yourself

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

Question 1

A resonant pulse of duration τ\tau takes a qubit from ∣0⟩\ket0 to ∣1⟩\ket1. What does a resonant pulse of duration 2τ2\tau do to a qubit starting in ∣0⟩\ket0?

  1. Leaves it in ∣1⟩\ket1, more firmly
  2. Puts it in an equal superposition of ∣0⟩\ket0 and ∣1⟩\ket1
  3. Excites it further, to ∣2⟩\ket2
  4. Brings it back to ∣0⟩\ket0
Show the answer

D. Brings it back to ∣0⟩\ket0

The rotation angle grows linearly with duration: τ\tau gives π\pi (a flip), so 2τ2\tau gives 2π2\pi, a full turn back to ∣0⟩\ket0 (up to an unobservable global phase). That is the oscillation in "Rabi oscillation".

Question 2

In one qubit, a superposition loses its phase coherence in about 20 μs, while an excited state takes about 100 μs to relax. Which description is right?

  1. T2≈20T_2\approx20 μs and T1≈100T_1\approx100 μs: dephasing is the faster process here
  2. T1≈20T_1\approx20 μs and T2≈100T_2\approx100 μs
  3. This is impossible, because T2T_2 can never be shorter than T1T_1
  4. Only T1T_1 matters for computing, so this qubit is effectively perfect for 100 μs
Show the answer

A. T2≈20T_2\approx20 μs and T1≈100T_1\approx100 μs: dephasing is the faster process here

T1T_1 measures energy relaxation and T2T_2 the loss of phase. Relaxation alone would allow T2T_2 up to 2T12T_1, but extra noise that only scrambles the phase can make T2T_2 much shorter than T1T_1. Algorithms need phase coherence, so here the 20 μs clock is the one that counts.

Question 3

Why does a superconducting qubit need a Josephson junction instead of an ordinary coil (inductor)?

  1. Without the junction the circuit would not superconduct
  2. The junction's nonlinearity makes the energy levels unequally spaced, so pulses can address just ∣0⟩↔∣1⟩\ket0\leftrightarrow\ket1
  3. The junction cools the circuit to 10 mK
  4. The junction lets one circuit hold many qubits
Show the answer

B. The junction's nonlinearity makes the energy levels unequally spaced, so pulses can address just ∣0⟩↔∣1⟩\ket0\leftrightarrow\ket1

With an ordinary inductor the circuit is a harmonic oscillator with equally spaced levels, and a pulse would climb the whole ladder. The junction acts as a lossless nonlinear inductor that makes the ladder uneven.

Question 4

Which pairing of platform and principal challenge is correct?

  1. Trapped ions: coherence times too short for useful gates
  2. Superconducting circuits: gates too slow
  3. Photonic qubits: photon loss and weak photon–photon interactions
  4. Neutral atoms: no two atoms are identical
Show the answer

C. Photonic qubits: photon loss and weak photon–photon interactions

Photons decohere very little but are easily lost, and getting two photons to interact is hard. The other pairings invert real strengths: ions have very long coherence, superconducting gates are the fastest, and atoms of one isotope are identical.

Question 5

Why don't experts rank quantum computers by their number of qubits?

  1. Qubit counts are kept secret, so they cannot be compared
  2. Every platform has roughly the same number of qubits
  3. Qubit number is irrelevant, because only a few qubits can ever be entangled at once
  4. Usefulness depends on error rates, connectivity and coherence too, so many noisy qubits can do less than fewer excellent ones
Show the answer

D. Usefulness depends on error rates, connectivity and coherence too, so many noisy qubits can do less than fewer excellent ones

A computation fails if errors pile up faster than they can be handled. Benchmarks such as randomized benchmarking and quantum volume therefore measure quality, not just quantity.

Go deeper

Further reading

  • D. P. DiVincenzo, "The physical implementation of quantum computation", Fortschritte der Physik 48, 771 (2000); arXiv:quant-ph/0002077.
  • M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, ch. 7, "Quantum computers: physical realization" (Cambridge, 10th anniv. ed., 2010).