Simple Algorithm
Quantum Computing · Excited State · 24 min

Superconducting Qubits

Artificial atoms made of Josephson junctions at 10 millikelvin.

A superconducting qubit is an electrical circuit you could see under a magnifying glass, yet it has discrete energy levels like an atom. Built from capacitors, wires and a Josephson junction, and cooled to about 10 millikelvin, it is the platform behind many of the field's landmark experiments. This lesson builds one from circuit theory up.

What you’ll learn

  • Quantize an LC circuit and explain why a harmonic oscillator cannot serve as a qubit.
  • Derive from the Josephson relations that a junction is a lossless, nonlinear inductor with energy −EJcos⁡φ-E_J\cos\varphi.
  • Write the transmon Hamiltonian, explain the EJ/EC≫1E_J/E_C\gg1 regime, and derive the anharmonicity α≈−EC\alpha\approx-E_C.
  • Describe microwave control, dispersive readout and two-qubit gates, with typical parameters as of the mid-2020s.

An electrical circuit as a quantum oscillator

Start with an LC circuit, a capacitor CC and an inductor LL. Its energy, written with the flux Φ\Phi through the inductor and the charge QQ on the capacitor, is that of a harmonic oscillator, with Φ\Phi playing position and QQ momentum:

H=Q22C+Φ22L,[Φ,Q]=iℏ,ωr=1LCH = \frac{Q^2}{2C}+\frac{\Phi^2}{2L},\qquad[\Phi,Q] = i\hbar,\qquad\omega_r = \frac{1}{\sqrt{LC}}
Canonical quantization gives H=ℏωr(a†a+12)H = \hbar\omega_r\big(a^\dagger a+\tfrac12\big), with Φ=ℏZ/2 (a+a†)\Phi = \sqrt{\hbar Z/2}\,(a+a^\dagger) and Z=L/CZ = \sqrt{L/C}.

For this to behave as a quantum object rather than a lossy classical circuit, two conditions must hold. No dissipation: superconductors such as aluminum, niobium and tantalum carry current without resistance below their critical temperature. No thermal excitation: the level spacing ℏωr\hbar\omega_r must be much larger than kBTk_BT.

Predict before you look

A qubit transition is at ω/2π=5\omega/2\pi = 5 GHz. At what temperature TT does kBT=ℏωk_BT = \hbar\omega? Answer in millikelvin.

Show the answer

240 mK

T=hf/kB=(6.63×10−34×5×109)/(1.38×10−23)≈0.24T = hf/k_B = (6.63\times10^{-34}\times5\times10^9)/(1.38\times10^{-23})\approx0.24 K; a handy rule is 1 GHz per 48 mK. At 10 mK the thermal excited population would be e−24≈4×10−11e^{-24}\approx4\times10^{-11}. In practice residual populations of order 0.1–1% are seen, because the qubit's effective temperature is set by stray radiation and non-equilibrium effects, not by the refrigerator alone.

The Josephson junction

A Josephson junction is two superconductors separated by a thin insulator, typically aluminum oxide about a nanometer thick, through which Cooper pairs tunnel coherently. It is described by the phase difference φ\varphi between the superconducting condensates on either side, which obeys the two Josephson relations:

I=Icsin⁡φ,V=Φ02πdφdt,Φ0=h2e≈2.07×10−15 WbI = I_c\sin\varphi,\qquad V = \frac{\Phi_0}{2\pi}\frac{d\varphi}{dt},\qquad\Phi_0 = \frac{h}{2e}\approx2.07\times10^{-15}\ \text{Wb}
IcI_c is the junction's critical current and Φ0\Phi_0 the superconducting flux quantum.
DerivationA nonlinear, lossless inductor
  1. dIdt=Iccos⁡φ dφdt=Iccos⁡φ 2πΦ0 V\frac{dI}{dt} = I_c\cos\varphi\,\frac{d\varphi}{dt} = I_c\cos\varphi\,\frac{2\pi}{\Phi_0}\,V

    Differentiate the first relation and use the second.

  2. V=LJ(φ) dIdt,LJ(φ)=Φ02πIccos⁡φV = L_J(\varphi)\,\frac{dI}{dt},\qquad L_J(\varphi) = \frac{\Phi_0}{2\pi I_c\cos\varphi}

    The junction acts as an inductor whose inductance depends on the phase, and hence on the current through it: it is nonlinear.

  3. U(φ)=∫IV dt=∫Icsin⁡φ Φ02π dφ=−EJcos⁡φ,EJ=Φ0Ic2πU(\varphi) = \int IV\,dt = \int I_c\sin\varphi\,\frac{\Phi_0}{2\pi}\,d\varphi = -E_J\cos\varphi,\qquad E_J = \frac{\Phi_0I_c}{2\pi}

    The stored energy. No resistance appears anywhere, so the element is lossless. Replacing Φ2/2L\Phi^2/2L by −EJcos⁡(2πΦ/Φ0)-E_J\cos(2\pi\Phi/\Phi_0) makes the oscillator anharmonic.

Think first According to LJ=Φ0/(2πIccos⁡φ)L_J = \Phi_0/(2\pi I_c\cos\varphi), what happens to the junction's inductance as the current approaches IcI_c, and why does that matter?

As I→IcI\to I_c, φ→π/2\varphi\to\pi/2 and cos⁡φ→0\cos\varphi\to0, so LJL_J diverges: the junction becomes ever "softer" the harder it is driven. Because the effective inductance changes with the oscillation amplitude, the oscillation frequency depends on the energy, which is exactly what makes the energy levels unequally spaced.

The transmon

Shunt a junction with a capacitor and you get the Cooper-pair-box family of circuits. With nn the number of Cooper pairs that have tunneled across and φ\varphi the junction phase ([φ,n]=i[\varphi,n] = i), the Hamiltonian is

H=4EC (n−ng)2−EJcos⁡φ,EC=e22CΣH = 4E_C\,(n-n_g)^2-E_J\cos\varphi,\qquad E_C = \frac{e^2}{2C_\Sigma}
CΣC_\Sigma is the total capacitance, and ngn_g is an offset charge set by the electrostatic environment, which fluctuates (charge noise).
RegimeEJ/ECE_J/E_CBehavior
Cooper-pair box (1999)≲1\lesssim1strong anharmonicity, but the energy levels depend strongly on ngn_g: charge noise dephases it within nanoseconds
Transmon (2007)∼20\sim20–100100levels nearly independent of ngn_g; anharmonicity reduced but still a few hundred MHz

The trade-off is lopsided in the transmon's favor. Koch et al. (2007) showed that the sensitivity to ngn_g, the charge dispersion, falls off exponentially, as e−8EJ/ECe^{-\sqrt{8E_J/E_C}}, while the anharmonicity falls only algebraically. For EJ/EC≫1E_J/E_C\gg1 the phase stays near φ=0\varphi = 0, and you may expand the cosine:

DerivationAnharmonicity of the transmon
  1. −EJcos⁡φ≈−EJ+EJ2φ2−EJ24φ4-E_J\cos\varphi\approx-E_J+\frac{E_J}{2}\varphi^2-\frac{E_J}{24}\varphi^4

    Small oscillations around the minimum. In this regime ngn_g can be dropped, because its effect is exponentially small.

  2. H0=4ECn2+EJ2φ2=8EJEC (b†b+12),φ=(2ECEJ)1/4(b+b†)H_0 = 4E_Cn^2+\frac{E_J}{2}\varphi^2 = \sqrt{8E_JE_C}\,\big(b^\dagger b+\tfrac12\big),\qquad\varphi = \Big(\frac{2E_C}{E_J}\Big)^{1/4}(b+b^\dagger)

    A harmonic oscillator at the plasma frequency, ℏωp=8EJEC\hbar\omega_p = \sqrt{8E_JE_C}.

  3. −EJ24φ4=−EC12 (b+b†)4,⟨m|(b+b†)4|m⟩=6m2+6m+3-\frac{E_J}{24}\varphi^4 = -\frac{E_C}{12}\,(b+b^\dagger)^4,\qquad\mel{m}{(b+b^\dagger)^4}{m} = 6m^2+6m+3

    The quartic term is a perturbation of size about EC≪ℏωpE_C\ll\hbar\omega_p; to first order only its diagonal elements matter.

  4. Em≈−EJ+8EJEC (m+12)−EC12(6m2+6m+3)E_m\approx-E_J+\sqrt{8E_JE_C}\,\big(m+\tfrac12\big)-\frac{E_C}{12}\big(6m^2+6m+3\big)

    The levels are no longer equally spaced.

  5. ℏω01=E1−E0≈8EJEC−EC,α=(E2−E1)−(E1−E0)≈−EC\hbar\omega_{01} = E_1-E_0\approx\sqrt{8E_JE_C}-E_C,\qquad\alpha = (E_2-E_1)-(E_1-E_0)\approx-E_C

    The anharmonicity is about −EC-E_C: typically −200-200 to −300-300 MHz, against qubit frequencies of 4–6 GHz. Exact diagonalization gives a somewhat larger magnitude, about 1.15 EC1.15\,E_C at EJ/EC=50E_J/E_C = 50.

Predict before you look

To leading order the relative anharmonicity is ∣α∣/ω01≈EC/8EJEC=1/8EJ/EC\abs\alpha/\omega_{01}\approx E_C/\sqrt{8E_JE_C} = 1/\sqrt{8E_J/E_C}. For a transmon with EJ/EC=50E_J/E_C = 50, what is ∣α∣/ω01\abs\alpha/\omega_{01}, in percent?

Show the answer

5 %

1/400=1/20=5%1/\sqrt{400} = 1/20 = 5\%; exact diagonalization gives about 6%. Only a few percent separates the 1→21\to2 transition from the qubit transition, which is why gates cannot be made arbitrarily fast without leaking into ∣2⟩\ket2.

Microwave control

A transmon is driven by a microwave voltage coupled capacitively to its island, which adds a term proportional to nn to the Hamiltonian. Restricted to the two lowest levels, in a frame rotating at the drive frequency ωd\omega_d, and after the rotating-wave approximation, this becomes

Hrot=ℏΔ2 σz+ℏΩ(t)2(cos⁡ϕ σx+sin⁡ϕ σy),Δ=ω01−ωdH_{\text{rot}} = \frac{\hbar\Delta}{2}\,\sigma_z+\frac{\hbar\Omega(t)}{2}\big(\cos\phi\,\sigma_x+\sin\phi\,\sigma_y\big),\qquad\Delta = \omega_{01}-\omega_d
The envelope Ω(t)\Omega(t) sets the rotation angle ∫Ω dt\int\Omega\,dt, the phase ϕ\phi sets the rotation axis in the xyxy-plane, and a detuning Δ\Delta tilts the axis toward zz.
Think first Why not simply make every pulse 1 ns long to beat decoherence?

A short pulse has a broad spectrum, of width roughly 1/t1/t. With ∣α∣/2π≈200\abs\alpha/2\pi\approx200–300300 MHz, a 1 ns pulse would strongly drive the 1→21\to2 transition and leak population out of the qubit. Gate times must stay well above 1/∣α∣1/\abs\alpha (here about 0.5–0.8 ns), and DRAG shaping lets you approach that limit; in practice single-qubit gates take roughly 10–50 ns.

Reading out and keeping coherence

The qubit is read out through a superconducting microwave resonator (frequency ωr\omega_r, typically 6–8 GHz) coupled to it with strength gg. When the two are far detuned, ∣ω01−ωr∣≫g\abs{\omega_{01}-\omega_r}\gg g, they cannot exchange energy, but the resonator's frequency shifts according to the qubit's state:

Hdisp≈ℏ(ωr+χσz)a†a+ℏω~012 σz,χ≈g2Δ  (two-level approximation)H_{\text{disp}}\approx\hbar\big(\omega_r+\chi\sigma_z\big)a^\dagger a+\frac{\hbar\tilde\omega_{01}}{2}\,\sigma_z,\qquad\chi\approx\frac{g^2}{\Delta}\ \ \text{(two-level approximation)}
Here Δ=ω01−ωr\Delta = \omega_{01}-\omega_r. The resonator sits at ωr±χ\omega_r\pm\chi depending on the qubit state; for a transmon the ∣2⟩\ket2 level modifies χ\chi, as derived in the Rydberg State lesson on readout.

Send a weak microwave tone near ωr\omega_r and look at the phase and amplitude of the transmitted or reflected signal: they differ for ∣0⟩\ket0 and ∣1⟩\ket1. The measurement is quantum non-demolition: it does not change σz\sigma_z, so repeating it gives the same answer. The signal is only a few photons, so it is amplified first by a near-quantum-limited Josephson parametric amplifier at millikelvin, then by a cryogenic HEMT amplifier at 4 K. As of the mid-2020s, readout typically takes a few hundred nanoseconds, with assignment fidelities of order 99%.

Check your understanding

In dispersive readout, what physical quantity carries the information about the qubit state?

  1. The photons the qubit emits when it decays
  2. The state-dependent frequency shift ±χ\pm\chi of the resonator, seen as a phase shift of the probe tone
  3. The voltage across the Josephson junction
  4. A temperature rise of the chip
Show the answer

B. The state-dependent frequency shift ±χ\pm\chi of the resonator, seen as a phase shift of the probe tone

The resonator frequency depends on the qubit state, so a probe tone near ωr\omega_r acquires a state-dependent phase and amplitude. The qubit does not have to decay for this, which is what makes the measurement quantum non-demolition.

Loss mechanismWhat it isMitigation
Dielectric loss (TLS)microscopic two-level defects in amorphous oxides at surfaces and interfaces absorb energybetter materials (e.g. tantalum, cleaner interfaces); geometries that store less energy near surfaces
Purcell decaythe qubit radiates through its readout resonator into the measurement linelarger detuning; Purcell filters
Quasiparticlesbroken Cooper pairs, created by stray infrared light or ionizing radiation, tunnel across the junctionshielding and filtering; gap engineering; phonon traps
Flux and charge noiseslow frequency fluctuations that dephase the qubitsweet spots; dynamical decoupling; the transmon's charge insensitivity
Thermal photonsresidual photons in the readout resonator dephase the qubitattenuation and filtering of the lines

Two-qubit gates and the frontier

Neighboring transmons are coupled capacitively, directly or through a coupler, with coupling strengths of a few to tens of MHz. Two families of entangling gates dominate:

  • Fixed-frequency qubits, all-microwave gates. The cross-resonance gate drives the control qubit at the target qubit's frequency, producing a conditional Z⊗XZ\otimes X rotation.
  • Frequency-tunable qubits or couplers. A flux pulse brings ∣11⟩\ket{11} into resonance with ∣20⟩\ket{20}; one full oscillation returns ∣11⟩\ket{11} with a minus sign, a CZ gate. Tunable couplers switch the interaction on for the gate and off otherwise.
CZ=diag⁡(1,1,1,−1)\text{CZ} = \diag(1,1,1,-1)
Two-qubit gates take roughly 20–500 ns, with fidelities around 99–99.9% on good pairs (mid-2020s). The main error sources are decoherence during the gate, leakage, and residual ZZZZ coupling between idle neighbors.

Key takeaways

  • An LC circuit quantizes to a harmonic oscillator with [Φ,Q]=iℏ[\Phi,Q] = i\hbar; its equal level spacing makes it useless as a qubit.
  • A Josephson junction is a lossless nonlinear inductor with energy −EJcos⁡φ-E_J\cos\varphi; the transmon, H=4EC(n−ng)2−EJcos⁡φH = 4E_C(n-n_g)^2-E_J\cos\varphi, operates near EJ/EC∼50E_J/E_C\sim50, where charge dispersion is exponentially suppressed.
  • Perturbation theory gives ℏω01≈8EJEC−EC\hbar\omega_{01}\approx\sqrt{8E_JE_C}-E_C and α≈−EC\alpha\approx-E_C, a few hundred MHz: enough to address the qubit with shaped 10–50 ns pulses.
  • Readout uses the dispersive shift χ≈g2/Δ\chi\approx g^2/\Delta of a resonator, amplified near the quantum limit; entangling gates use cross-resonance or flux-tuned CZ via ∣11⟩\ket{11}–∣20⟩\ket{20}.
  • Coherence is limited by materials (TLS defects), quasiparticles, Purcell decay and thermal photons, and requires operation near 10 mK.

Check yourself

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

Question 1

Why can't a superconducting LC resonator by itself serve as a qubit?

  1. Its levels are equally spaced, so a drive cannot isolate the ∣0⟩↔∣1⟩\ket0\leftrightarrow\ket1 transition
  2. It loses energy too quickly because of resistance
  3. It cannot be cooled below 1 K
  4. It has only one energy level
Show the answer

A. Its levels are equally spaced, so a drive cannot isolate the ∣0⟩↔∣1⟩\ket0\leftrightarrow\ket1 transition

All transitions of a harmonic oscillator have the same frequency, so driving it produces a coherent state rather than a two-level rotation. The Josephson junction's nonlinearity fixes this.

Question 2

A transmon has EC/h=250E_C/h = 250 MHz and EJ/EC=50E_J/E_C = 50. Which values are closest, using the leading-order formulas?

  1. ω01/2π≈5.0\omega_{01}/2\pi\approx5.0 GHz and α/2π≈+250\alpha/2\pi\approx+250 MHz
  2. ω01/2π≈2.5\omega_{01}/2\pi\approx2.5 GHz and α/2π≈−50\alpha/2\pi\approx-50 MHz
  3. ω01/2π≈4.75\omega_{01}/2\pi\approx4.75 GHz and α/2π≈−250\alpha/2\pi\approx-250 MHz
  4. ω01/2π≈12.5\omega_{01}/2\pi\approx12.5 GHz and α/2π≈−12.5\alpha/2\pi\approx-12.5 GHz
Show the answer

C. ω01/2π≈4.75\omega_{01}/2\pi\approx4.75 GHz and α/2π≈−250\alpha/2\pi\approx-250 MHz

8EJEC=8×12.5×0.25\sqrt{8E_JE_C} = \sqrt{8\times12.5\times0.25} GHz =5= 5 GHz, so ω01/2π≈5−0.25=4.75\omega_{01}/2\pi\approx5-0.25 = 4.75 GHz and α≈−EC\alpha\approx-E_C. Exact diagonalization gives 4.74 GHz and −0.29-0.29 GHz.

Question 3

In the two-level approximation, a qubit detuned by Δ/2π=1\Delta/2\pi = 1 GHz from its readout resonator, with coupling g/2π=50g/2\pi = 50 MHz, has dispersive shift

  1. χ/2π=50\chi/2\pi = 50 MHz
  2. χ/2π=2.5\chi/2\pi = 2.5 MHz
  3. χ/2π=0.05\chi/2\pi = 0.05 MHz
  4. χ/2π=20\chi/2\pi = 20 MHz
Show the answer

B. χ/2π=2.5\chi/2\pi = 2.5 MHz

χ=g2/Δ\chi = g^2/\Delta: (50 MHz)2/(1000 MHz)=2.5(50\ \text{MHz})^2/(1000\ \text{MHz}) = 2.5 MHz. The resonator moves by ±2.5\pm2.5 MHz, comparable to typical resonator linewidths, which is what makes the two states distinguishable.

Question 4

In a flux-activated CZ gate, which pair of two-transmon states is brought into resonance?

  1. ∣01⟩\ket{01} and ∣10⟩\ket{10}
  2. ∣00⟩\ket{00} and ∣11⟩\ket{11}
  3. ∣10⟩\ket{10} and ∣20⟩\ket{20}
  4. ∣11⟩\ket{11} and ∣20⟩\ket{20} (or ∣02⟩\ket{02})
Show the answer

D. ∣11⟩\ket{11} and ∣20⟩\ket{20} (or ∣02⟩\ket{02})

Tuning so that ω1+ω2=2ω1+α1\omega_1+\omega_2 = 2\omega_1+\alpha_1 makes ∣11⟩\ket{11} degenerate with ∣20⟩\ket{20}. A full oscillation ∣11⟩→∣20⟩→∣11⟩\ket{11}\to\ket{20}\to\ket{11} multiplies ∣11⟩\ket{11} by −1-1, while no other computational state has a partner at that frequency: a conditional phase.

Go deeper

Further reading

  • J. Koch et al., "Charge-insensitive qubit design derived from the Cooper pair box", Physical Review A 76, 042319 (2007).
  • P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gustavsson & W. D. Oliver, "A quantum engineer's guide to superconducting qubits", Applied Physics Reviews 6, 021318 (2019); arXiv:1904.06560.
  • A. Blais, A. L. Grimsmo, S. M. Girvin & A. Wallraff, "Circuit quantum electrodynamics", Reviews of Modern Physics 93, 025005 (2021).