Simple Algorithm
Quantum Computing · Rydberg State · 32 min

Transmon Physics

Quantizing a circuit: the Cooper-pair box and the transmon regime.

Why does a millimeter-scale circuit obey the Schrödinger equation, and why did adding a big capacitor to a fragile qubit make it one of the most coherent? This lesson quantizes a circuit from its Lagrangian, maps the Cooper-pair box onto a particle in a periodic potential, and shows that the transmon's protection is a tunneling effect: its sensitivity to charge is the width of a Bloch band.

What you’ll learn

  • Quantize a lumped circuit from its Lagrangian and derive the Cooper-pair-box Hamiltonian 4EC(n−ng)2−EJcos⁡φ4E_C(n-n_g)^2-E_J\cos\varphi.
  • Map the spectrum onto Bloch bands and derive the e−8EJ/ECe^{-\sqrt{8E_J/E_C}} scaling of the charge dispersion from the instanton action.
  • State Koch et al.'s asymptotic formulas for the levels, the anharmonicity and the charge dispersion, and use them for design.
  • Analyze flux-tunable transmons, sweet spots and the main T1T_1 and T2T_2 mechanisms, and place the transmon among alternative superconducting qubits.

Quantizing a circuit

Lumped superconducting circuits are quantized like mechanical systems (Devoret 1995; Vool & Devoret 2017). Take node fluxes Φa(t)=∫tVa dt′\Phi_a(t) = \int^tV_a\,dt' as coordinates: capacitors contribute kinetic energy, inductors and junctions potential energy.

DerivationFrom a Lagrangian to the Cooper-pair box
  1. L=CΣ2Φ˙2−CgVgΦ˙+EJcos⁡(2πΦΦ0)\mathcal L = \frac{C_\Sigma}{2}\dot\Phi^2-C_gV_g\dot\Phi+E_J\cos\Big(\frac{2\pi\Phi}{\Phi_0}\Big)

    A junction (critical current IcI_c, capacitance CJC_J) shunted by a capacitor CBC_B, with a gate capacitor CgC_g to a voltage source VgV_g, and CΣ=CJ+CB+CgC_\Sigma = C_J+C_B+C_g. Constant terms are dropped.

  2. Q=∂L∂Φ˙=CΣΦ˙−CgVg,H=QΦ˙−L=(Q+CgVg)22CΣ−EJcos⁡(2πΦΦ0)Q = \frac{\partial\mathcal L}{\partial\dot\Phi} = C_\Sigma\dot\Phi-C_gV_g,\qquad H = Q\dot\Phi-\mathcal L = \frac{(Q+C_gV_g)^2}{2C_\Sigma}-E_J\cos\Big(\frac{2\pi\Phi}{\Phi_0}\Big)

    The Legendre transform; QQ is the charge conjugate to the node flux Φ\Phi.

  3. n=Q2e,φ=2πΦΦ0,ng=−CgVg2e  ⟹  H=4EC(n−ng)2−EJcos⁡φn = \frac{Q}{2e},\quad\varphi = \frac{2\pi\Phi}{\Phi_0},\quad n_g = -\frac{C_gV_g}{2e}\;\Longrightarrow\;H = 4E_C(n-n_g)^2-E_J\cos\varphi

    Count charge in Cooper pairs and flux in units of Φ0/2π\Phi_0/2\pi. With EC=e2/2CΣE_C = e^2/2C_\Sigma, (Q+CgVg)2/2CΣ=4EC(n−ng)2(Q+C_gV_g)^2/2C_\Sigma = 4E_C(n-n_g)^2. (The sign of ngn_g is a convention; only its value modulo 1 matters.)

  4. [Φ,Q]=iℏ  ⟹  [φ,n]=2πΦ0 12e iℏ=i[\Phi,Q] = i\hbar\;\Longrightarrow\;[\varphi,n] = \frac{2\pi}{\Phi_0}\,\frac{1}{2e}\,i\hbar = i

    Because 2e Φ0=h2e\,\Phi_0 = h. In the phase representation, n=−i ∂/∂φn = -i\,\partial/\partial\varphi.

Exact spectrum: Bloch bands and charge dispersion

In the charge basis the Hamiltonian is tridiagonal, H=∑n4EC(n−ng)2∣n⟩ ⁣⟨n∣−EJ2∑n(∣n⟩ ⁣⟨n+1∣+∣n+1⟩ ⁣⟨n∣)H = \sum_n4E_C(n-n_g)^2\dyad{n}-\tfrac{E_J}{2}\sum_n\big(\ketbra{n}{n+1}+\ketbra{n+1}{n}\big). In the phase basis it is Mathieu's equation, solved exactly by Mathieu characteristic values (Koch et al. 2007). The physics is clearest after a gauge transformation.

DerivationOffset charge as quasimomentum
  1. H=4EC(−i∂∂φ−ng)2−EJcos⁡φ,ψ(φ+2π)=ψ(φ)H = 4E_C\Big(-i\frac{\partial}{\partial\varphi}-n_g\Big)^2-E_J\cos\varphi,\qquad\psi(\varphi+2\pi) = \psi(\varphi)

    The phase representation.

  2. ψ(φ)=eingφχ(φ):(−i∂∂φ−ng) eingφχ=eingφ(−i∂χ∂φ)\psi(\varphi) = e^{in_g\varphi}\chi(\varphi):\qquad\Big(-i\frac{\partial}{\partial\varphi}-n_g\Big)\,e^{in_g\varphi}\chi = e^{in_g\varphi}\Big(-i\frac{\partial\chi}{\partial\varphi}\Big)

    This substitution removes ngn_g from the Hamiltonian ...

  3. [−4EC∂2∂φ2−EJcos⁡φ]χ=Eχ,χ(φ+2π)=e−2πingχ(φ)\Big[-4E_C\frac{\partial^2}{\partial\varphi^2}-E_J\cos\varphi\Big]\chi = E\chi,\qquad\chi(\varphi+2\pi) = e^{-2\pi in_g}\chi(\varphi)

    ... at the price of a twisted boundary condition. Read φ∈R\varphi\in\R as the position of a particle of "mass" 1/8EC1/8E_C in the periodic potential −EJcos⁡φ-E_J\cos\varphi: this is Bloch's theorem, with ngn_g playing the role of quasimomentum.

  4. Em(ng)≈Eˉm−ϵm2cos⁡(2πng)E_m(n_g)\approx\bar E_m-\frac{\epsilon_m}{2}\cos(2\pi n_g)

    Each transmon level mm is a Bloch band. In the transmon regime the wells are deep and the tight-binding approximation applies: the bandwidth, which is the charge dispersion ϵm\epsilon_m, is set by the amplitude for tunneling from one well to the next, φ→φ±2π\varphi\to\varphi\pm2\pi.

DerivationThe tunneling exponent
  1. 4ECn2=n22M,M=18EC4E_Cn^2 = \frac{n^2}{2M},\qquad M = \frac{1}{8E_C}

    Identify the effective mass, with ℏ=1\hbar = 1 and dimensionless variables.

  2. S0=∫02π2M[U(φ)−U(0)] dφ=EJ4EC∫02π1−cos⁡φ  dφS_0 = \int_0^{2\pi}\sqrt{2M\big[U(\varphi)-U(0)\big]}\,d\varphi = \sqrt{\frac{E_J}{4E_C}}\int_0^{2\pi}\sqrt{1-\cos\varphi}\;d\varphi

    The WKB (instanton) action for tunneling under the barrier between neighboring minima of U=−EJcos⁡φU = -E_J\cos\varphi.

  3. ∫02π1−cos⁡φ  dφ=2∫02πsin⁡φ2 dφ=42  ⟹  S0=8EJEC\int_0^{2\pi}\sqrt{1-\cos\varphi}\;d\varphi = \sqrt2\int_0^{2\pi}\sin\frac\varphi2\,d\varphi = 4\sqrt2\;\Longrightarrow\;S_0 = \sqrt{\frac{8E_J}{E_C}}

    Using 1−cos⁡φ=2sin⁡2(φ/2)1-\cos\varphi = 2\sin^2(\varphi/2).

  4. ϵm∝e−S0=e−8EJ/EC\epsilon_m\propto e^{-S_0} = e^{-\sqrt{8E_J/E_C}}

    The tunneling amplitude, and hence the charge dispersion, is exponentially small in 8EJ/EC\sqrt{8E_J/E_C}, times an algebraic prefactor that grows with mm because higher levels sit closer to the top of the barrier.

EJ/ECE_J/E_Cω01/2π\omega_{01}/2\piα/2π\alpha/2\piCharge dispersion of ω01\omega_{01}
101.97 GHz−0.64-0.64 GHz63 MHz
202.89 GHz−0.36-0.36 GHz4.4 MHz
504.74 GHz−0.29-0.29 GHz10 kHz
806.06 GHz−0.28-0.28 GHz0.09 kHz
Exact diagonalization in the charge basis with EC/h=250E_C/h = 250 MHz (frequencies at ng=0n_g = 0; the dispersion is the peak-to-peak variation of ω01\omega_{01} with ngn_g). From EJ/EC=20E_J/E_C = 20 to 80 the dispersion falls by about 5×1045\times10^4, while ∣α∣\abs\alpha shrinks by less than 25%.
Check your understanding

A design moves from EJ/EC=32E_J/E_C = 32 to EJ/EC=50E_J/E_C = 50. By roughly what factor does the exponential factor e−8EJ/ECe^{-\sqrt{8E_J/E_C}} in the charge dispersion change?

  1. It halves
  2. It decreases by about e4≈55e^{4}\approx55
  3. It decreases by 50/32≈1.650/32\approx1.6
  4. It increases, because EJE_J is larger
Show the answer

B. It decreases by about e4≈55e^{4}\approx55

8×32=16\sqrt{8\times32} = 16 and 8×50=20\sqrt{8\times50} = 20, so the factor drops by e−4e^{-4}, about 55 times. The algebraic prefactor grows a little, so the net suppression of ϵ1\epsilon_1 is somewhat less, about 30 times.

Anharmonicity, tunability and sweet spots

The leading-order result α≃−EC\alpha\simeq-E_C is slightly too small in magnitude: the exact values in the table exceed it because the next corrections are suppressed only by powers of EC/EJ\sqrt{E_C/E_J}. Design usually starts from target values of ω01\omega_{01} and α\alpha:

Predict before you look

You want ω01/2π=5.0\omega_{01}/2\pi = 5.0 GHz with EC/h=200E_C/h = 200 MHz. Using ℏω01≈8EJEC−EC\hbar\omega_{01}\approx\sqrt{8E_JE_C}-E_C, what EJ/hE_J/h do you need, in GHz?

Show the answer

16.9 GHz

8EJEC=5.2\sqrt{8E_JE_C} = 5.2 GHz, so EJ=5.22/(8×0.2)=16.9E_J = 5.2^2/(8\times0.2) = 16.9 GHz and EJ/EC≈85E_J/E_C\approx85. Exact diagonalization with these values gives ω01/2π≈4.99\omega_{01}/2\pi\approx4.99 GHz and α/2π≈−221\alpha/2\pi\approx-221 MHz. Since EJ=Φ0Ic/2πE_J = \Phi_0I_c/2\pi, this fixes the critical current at about 34 nA, and hence the junction's oxide thickness and area.

To make the frequency tunable, replace the junction by a SQUID: two junctions in a loop threaded by an external flux Φext\Phi_{\text{ext}}. The effective Josephson energy becomes

EJ(Φext)=EJΣ ∣cos⁡πΦextΦ0∣ 1+d2tan⁡2πΦextΦ0,d=EJ2−EJ1EJ1+EJ2E_J(\Phi_{\text{ext}}) = E_{J\Sigma}\,\Big|\cos\frac{\pi\Phi_{\text{ext}}}{\Phi_0}\Big|\,\sqrt{1+d^2\tan^2\frac{\pi\Phi_{\text{ext}}}{\Phi_0}},\qquad d = \frac{E_{J2}-E_{J1}}{E_{J1}+E_{J2}}
EJΣ=EJ1+EJ2E_{J\Sigma} = E_{J1}+E_{J2} and dd is the junction asymmetry (Koch et al. 2007). Then ℏω01(Φext)≈8EJ(Φext)EC−EC\hbar\omega_{01}(\Phi_{\text{ext}})\approx\sqrt{8E_J(\Phi_{\text{ext}})E_C}-E_C.

Losses and the coherence budget

Relaxation rates add over channels, 1/T1=∑iΓi1/T_1 = \sum_i\Gamma_i. It is convenient to write each channel as a quality factor, Γi=ω01/Qi\Gamma_i = \omega_{01}/Q_i, and the dielectric part as a participation-weighted sum, 1/Qdiel=∑kpktan⁡δk1/Q_{\text{diel}} = \sum_kp_k\tan\delta_k, where pkp_k is the fraction of the qubit's electric energy stored in region kk (the substrate and the metal–air, metal–substrate and substrate–air interfaces) and tan⁡δk\tan\delta_k is its loss tangent.

MechanismSignatureTypical remedy
TLS defectsT1T_1 dips at specific frequencies that wander in time (spectral diffusion)materials and surface treatment; retuning the qubit frequency
Quasiparticlesparity switching; correlated bursts after radiation impactsinfrared shielding; gap engineering; phonon traps; radiation shielding
Purcell decay and radiationT1T_1 limited by coupling to lines, resonators and box modesPurcell filters; careful packaging
Flux noise (1/f1/f)T2T_2 degrades away from the sweet spotsweet spots; echo and dynamical decoupling
Thermal photonsT2T_2 limited by residual photons in the readout resonatorline attenuation and filtering
python
import numpy as np

def transmon_levels(EJ, EC, ng=0.0, ncut=15, k=4):
    """Lowest k energies (relative to the ground state) of H = 4 EC (n - ng)^2 - EJ cos(phi)."""
    n = np.arange(-ncut, ncut + 1)
    H = np.diag(4 * EC * (n - ng) ** 2)
    H -= 0.5 * EJ * (np.eye(len(n), k=1) + np.eye(len(n), k=-1))
    E = np.linalg.eigvalsh(H)
    return E[:k] - E[0]

EC = 0.25  # E_C / h in GHz
for ratio in (10, 20, 50, 80):
    E = transmon_levels(ratio * EC, EC)
    f01, alpha = E[1], E[2] - 2 * E[1]
    print(f"EJ/EC = {ratio:3d}: f01 = {f01:.3f} GHz, alpha = {1e3 * alpha:.0f} MHz")

Exact diagonalization in the charge basis, truncated at ∣n∣≤15\abs n\le15, which is ample for these parameters. Repeating with ng=0.5 and comparing gives the charge dispersion listed in the table.

Key takeaways

  • Circuits are quantized from their Lagrangian; for the Cooper-pair box, H=4EC(n−ng)2−EJcos⁡φH = 4E_C(n-n_g)^2-E_J\cos\varphi with [φ,n]=i[\varphi,n] = i.
  • The offset charge is a Bloch quasimomentum, and the charge dispersion is a tight-binding bandwidth, ∝e−8EJ/EC\propto e^{-\sqrt{8E_J/E_C}}, whose exponent is the instanton action between wells.
  • To leading order, ℏω01≃8EJEC−EC\hbar\omega_{01}\simeq\sqrt{8E_JE_C}-E_C and α≃−EC\alpha\simeq-E_C; the relative anharmonicity falls only as (8EJ/EC)−1/2(8E_J/E_C)^{-1/2}.
  • SQUID-tunable transmons trade first-order flux-noise immunity (kept only at sweet spots) for tunability; DRAG pulses manage leakage from the weak anharmonicity.
  • T1T_1 is dominated by dielectric (TLS) loss at surfaces, plus quasiparticles, Purcell decay and radiation; materials science sets the frontier.

Check yourself

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

Question 1

In circuit quantization of the Cooper-pair box, which pair of variables is canonically conjugate?

  1. Current and voltage
  2. The junction phase φ\varphi and the Cooper-pair number nn, with [φ,n]=i[\varphi,n] = i
  3. Flux and current
  4. Charge and voltage
Show the answer

B. The junction phase φ\varphi and the Cooper-pair number nn, with [φ,n]=i[\varphi,n] = i

Q=∂L/∂Φ˙Q = \partial\mathcal L/\partial\dot\Phi is conjugate to the node flux Φ\Phi; rescaled, φ=2πΦ/Φ0\varphi = 2\pi\Phi/\Phi_0 and n=Q/2en = Q/2e satisfy [φ,n]=i[\varphi,n] = i, because 2eΦ0=h2e\Phi_0 = h.

Question 2

In the Bloch-band picture of the transmon, what plays the role of quasimomentum, and what sets the bandwidth?

  1. ngn_g; the amplitude for tunneling between neighboring wells of −EJcos⁡φ-E_J\cos\varphi, ∝e−8EJ/EC\propto e^{-\sqrt{8E_J/E_C}}
  2. φ\varphi; the anharmonicity
  3. EJE_J; the charging energy
  4. nn; thermal fluctuations
Show the answer

A. ngn_g; the amplitude for tunneling between neighboring wells of −EJcos⁡φ-E_J\cos\varphi, ∝e−8EJ/EC\propto e^{-\sqrt{8E_J/E_C}}

A gauge transformation moves ngn_g into the boundary condition χ(φ+2π)=e−2πingχ(φ)\chi(\varphi+2\pi) = e^{-2\pi in_g}\chi(\varphi), exactly as quasimomentum appears in Bloch's theorem. Tight-binding bandwidths are tunneling amplitudes, here exponentially small in the instanton action 8EJ/EC\sqrt{8E_J/E_C}.

Question 3

Why is a flux-tunable transmon usually parked at its maximum frequency when idle?

  1. Its anharmonicity is largest there
  2. Its charge dispersion vanishes there
  3. Its frequency is first-order insensitive to flux noise there, ∂ω01/∂Φext=0\partial\omega_{01}/\partial\Phi_{\text{ext}} = 0
  4. It is farthest from its readout resonator there
Show the answer

C. Its frequency is first-order insensitive to flux noise there, ∂ω01/∂Φext=0\partial\omega_{01}/\partial\Phi_{\text{ext}} = 0

At an extremum of ω01(Φext)\omega_{01}(\Phi_{\text{ext}}) the linear coupling to flux noise vanishes, leaving only a weaker second-order sensitivity, so pure dephasing from 1/f1/f flux noise is minimized.

Question 4

In the transmon regime, how do the charge dispersion ϵ1\epsilon_1 and the anharmonicity α\alpha behave as EJ/ECE_J/E_C grows?

  1. Both decrease exponentially
  2. Both decrease algebraically
  3. ϵ1\epsilon_1 decreases algebraically and α\alpha exponentially
  4. ϵ1\epsilon_1 decreases exponentially, while α\alpha stays close to −EC-E_C (relative anharmonicity ∝(EJ/EC)−1/2\propto(E_J/E_C)^{-1/2})
Show the answer

D. ϵ1\epsilon_1 decreases exponentially, while α\alpha stays close to −EC-E_C (relative anharmonicity ∝(EJ/EC)−1/2\propto(E_J/E_C)^{-1/2})

This asymmetry is the whole point of the transmon. Charge sensitivity is a tunneling effect, exponentially suppressed; anharmonicity comes from the quartic term and decreases only algebraically relative to ω01\omega_{01}.

Further reading

  • J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schuster, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin & R. J. Schoelkopf, "Charge-insensitive qubit design derived from the Cooper pair box", Physical Review A 76, 042319 (2007).
  • V. Bouchiat, D. Vion, P. Joyez, D. Esteve & M. H. Devoret, "Quantum coherence with a single Cooper pair", Physica Scripta T76, 165 (1998); Y. Nakamura, Yu. A. Pashkin & J. S. Tsai, "Coherent control of macroscopic quantum states in a single-Cooper-pair box", Nature 398, 786 (1999).
  • U. Vool & M. Devoret, "Introduction to quantum electromagnetic circuits", International Journal of Circuit Theory and Applications 45, 897 (2017).
  • F. Motzoi, J. M. Gambetta, P. Rebentrost & F. K. Wilhelm, "Simple pulses for elimination of leakage in weakly nonlinear qubits", Physical Review Letters 103, 110501 (2009).
  • P. Krantz et al., "A quantum engineer's guide to superconducting qubits", Applied Physics Reviews 6, 021318 (2019).