Skip to content

Latest commit

 

History

History
206 lines (139 loc) · 20.2 KB

File metadata and controls

206 lines (139 loc) · 20.2 KB

03 · The Josephson Junction & Anharmonicity

In the previous chapter we built an LC oscillator out of superconducting circuit elements. It has a beautiful, clean problem: it is harmonic. Its energy levels are perfectly evenly spaced, like rungs on a ladder where every step is the same height. That sounds nice, but for a qubit it is fatal. If you send in a microwave pulse tuned to drive the $0 \to 1$ transition, the exact same photon also drives $1 \to 2$, $2 \to 3$, and so on. You cannot isolate a clean two-level system. You need one circuit element that breaks this even spacing, and that element is the Josephson junction.

A junction is almost embarrassingly simple to picture: two superconductors separated by a thin (~1 nm) insulating barrier. Classically nothing should flow. But the superconducting condensate on each side is described by a single macroscopic wavefunction $\psi = \sqrt{n_s},e^{i\theta}$ with a well-defined phase $\theta$. Only the gauge-invariant phase difference $$ \varphi = \theta_L-\theta_R-\frac{2\pi}{\Phi_0}\int_R^L \mathbf A\cdot d\mathbf l $$ across the barrier matters. This convention takes the positive branch direction to be $R\to L$; positive current and voltage below use the same branch orientation. Reversing the branch sends $\varphi\to-\varphi$ and $I\to-I$ but leaves $U=-E_J\cos\varphi$ unchanged. In a gauge with negligible vector potential across the barrier this reduces to $\theta_L-\theta_R$, and through it Cooper pairs tunnel coherently across the gap. The junction is the one circuit element that is simultaneously nonlinear and non-dissipative, and you cannot build a good qubit without both.

Here is the causal story the rest of the chapter tells, end to end:

flowchart TD
    A["Cooper pairs tunnel<br/>(condensate phase phi)"] --> B["First relation<br/>I = I_c sin phi"]
    B --> C["Nonlinear<br/>inductance L_J(phi)"]
    C --> D["Cosine potential<br/>-E_J cos phi"]
    D --> E["Quartic term<br/>flattens the well"]
    E --> F["Unequal level spacing<br/>E12-E01 approx -E_C"]
    F --> G["Address one<br/>0->1 transition<br/>= a qubit!"]
    H["Run at large<br/>E_J/E_C"] --> I["Charge dispersion<br/>~ exp(-sqrt(8 E_J/E_C))"]
    I --> J["Dephasing<br/>suppressed<br/>(long T2)"]
    G --> K["Transmon<br/>sweet spot"]
    J --> K
Loading

The two Josephson relations

Let $\varphi$ be the gauge-invariant phase difference across the junction, and define the branch flux $\Phi=(\Phi_0/2\pi)\varphi$ so that $V=\dot\Phi$. The entire device is governed by two relations:

$$ I = I_c \sin\varphi \qquad\text{(DC / first relation)} $$ $$ \frac{d\varphi}{dt} = \frac{2e}{\hbar},V \qquad\text{(AC / second relation)} $$

Where the first relation comes from. Each superconductor carries a condensate with phase $\theta_{L,R}$. Coupling the two through the barrier (Feynman's two-mode tunnelling model) makes the supercurrent depend only on the phase difference, and the simplest such periodic, odd function is $\sin\varphi$. The amplitude is the critical current $I_c$, set by the barrier transparency and the gap. Plain meaning: push a phase across the junction and a dissipationless current flows with no voltage, until you try to exceed $I_c$.

Where the second relation comes from. Each condensate phase evolves as $\dot\theta = -E/\hbar$, where $E$ is the energy of a charge carrier on that side. A voltage $V$ biases the two sides by an energy $2eV$, note the $2e$, because the carriers are Cooper pairs, not single electrons. So the difference winds at $\dot\varphi = 2eV/\hbar$. Integrated, a DC voltage produces an AC supercurrent at the Josephson frequency $f = 2eV/h$, with the universal slope $2e/h \approx 483.6\ \text{MHz}/\mu\text{V}$. That factor of $2e$ is the experimental fingerprint of Cooper pairing.

The nonlinear inductance. Differentiate the first relation and substitute the second:

$$ \frac{dI}{dt} = I_c\cos\varphi,\dot\varphi = I_c\cos\varphi\cdot\frac{2e}{\hbar}V. $$

An inductor obeys $V = L,dI/dt$, so $L = V/(dI/dt)$, giving

$$ L_J(\varphi) = \frac{\hbar}{2e,I_c\cos\varphi} = \frac{\Phi_0}{2\pi I_c\cos\varphi}, \qquad \Phi_0 = \frac{h}{2e}. $$

A linear inductor has constant $L$; here $L_J$ stiffens and softens with $\varphi$, diverges at $\varphi=\pi/2+k\pi$, and has negative incremental sign where $\cos\varphi&lt;0$. That phase-dependence is exactly the ingredient the bare LC circuit was missing.

Pitfall. $L_J$ is only a literal lumped inductor in the small-oscillation linearisation around an operating point. Near $\varphi=\pi/2+k\pi$ it blows up; where $\cos\varphi&lt;0$ it describes negative curvature of the cosine potential, not a stable standalone inductor.

The cosine potential

Integrate the energy delivered to the junction, $U = \int I,V,dt$. Using $I = I_c\sin\varphi$ and $V = (\hbar/2e)\dot\varphi$ gives $I,V,dt = I_c\sin\varphi,(\hbar/2e),d\varphi$, and integrating over $\varphi$:

$$ U(\varphi) = -E_J \cos\varphi, \qquad E_J = \frac{\hbar I_c}{2e} = \frac{I_c\Phi_0}{2\pi}. $$

$E_J$ is the Josephson energy, the amplitude of the cosine potential. With the zero chosen as written, the minimum is at $-E_J$ and the zero-bias barrier to the neighboring maximum is $2E_J$. Taylor-expanding,

$$ -E_J\cos\varphi \approx -E_J + \tfrac{1}{2}E_J\varphi^2 - \tfrac{1}{24}E_J\varphi^4 + \dots $$

Using $\Phi=(\Phi_0/2\pi)\varphi$, the $\varphi^2$ term reproduces a harmonic oscillator: $\tfrac12E_J\varphi^2=\Phi^2/(2L_{J0})$, so $L_{J0}=(\Phi_0/2\pi)^2/E_J=\Phi_0/(2\pi I_c)$. The $\varphi^4$ term is the crucial correction: it is negative, so the cosine well is flatter than a parabola away from the bottom. A flatter well means the energy rungs get closer together as you climb, the ladder is no longer evenly spaced.

 U(φ)
   │      .                              .
   │       \           parabola         /     ← harmonic approx (steeper)
   │        \  (½E_Jφ²)               /
   │   ──────\───────────────────── E₂   ⎫ ħω₁₂  (smaller)
   │     ─────\─────────────────── E₁    ⎬ ─────
   │       ────\─────────────── E₀       ⎭ ħω₀₁  (larger)
   │            \....         ..../        gaps shrink as you climb
   │             `-._ -E_Jcosφ _.-'        = anharmonicity
   │                  `-._____.-'  ← min at φ=0, U=-E_J; barrier height 2E_J
   └──────────────┴───────┴──────────── φ
                 -π/2     0    +π/2

The cosine (solid) hugs the parabola (the steeper arms) only at the bottom; the levels $E_0, E_1, E_2$ live in the flatter cosine, so $\hbar\omega_{12} &lt; \hbar\omega_{01}$.

Intuition aside. Think of the cosine as a pendulum. Small swings are nearly harmonic and isochronous; but a real pendulum slows for large swings, its period grows with amplitude. That amplitude-dependent frequency is anharmonicity. Tilt the cosine by adding a bias current and you get the tilted washboard: the bob can roll over barriers (phase slips) once $I \to I_c$. A qubit lives in the lowest swings of this superconducting pendulum.

   pivot ●                tilt with bias current I →  /\  /\  /\
         |\                                          /  \/  \/  \
         | \  angle = phase φ                       running "washboard":
         |  \                                       phase slips when I→I_c
         |   ● bob   gravity well = -E_J cos φ

The Hamiltonian in the charge basis

Pair the cosine potential with the electrostatic charging energy of the island's effective capacitance $C_\Sigma$ (junction plus gate/shunt/network capacitances). The island charge is $Q = 2e(\hat n - n_g)$, so $Q^2/(2C_\Sigma)$ gives the full Hamiltonian:

$$ H = 4E_C,(\hat n - n_g)^2 - E_J\cos\hat\varphi, \qquad E_C = \frac{e^2}{2C_\Sigma}, \qquad [\hat\varphi,\hat n] = i. $$

Here $\hat n$ counts excess Cooper pairs, $n_g$ is the dimensionless offset (gate) charge, and $E_C$ is the charging energy, the single-electron scale $e^2/2C_\Sigma$ that sets the cost of moving charge onto the island (one full Cooper pair costs $4E_C$ at $n_g=0$). The $4E_C(\hat n - n_g)^2$ term is the "kinetic" energy; $-E_J\cos\hat\varphi$ is the "potential." Their ratio $E_J/E_C$ governs everything.

The chapter title promised "in the charge basis," so let's make it concrete. Because $\hat\varphi$ and $\hat n$ are conjugate, $e^{\pm i\hat\varphi}$ shift Cooper-pair number by one. Therefore $\cos\hat\varphi$ is the Hermitian nearest-neighbor hopping operator:

$$ \cos\hat\varphi = \tfrac{1}{2}\sum_n \big(|n\rangle\langle n{+}1| + |n{+}1\rangle\langle n|\big). $$

It hops between adjacent charge states $n \leftrightarrow n\pm 1$, literally one Cooper pair tunnelling. In this basis $H$ is tridiagonal: diagonal entries $4E_C(n-n_g)^2$, off-diagonal entries $-E_J/2$. You could type that matrix into NumPy, truncate at $|n|\le 10$, and use eigh to get the spectrum; that is the standard numerical route for plots of the CPB/transmon spectrum. The sketches below are schematic, not generated data.

$E_J/E_C$: the master control knob

Everything about the device follows from one ratio. It interpolates a whole continuum of qubits:

Regime $E_J/E_C$ Eigenstate character Charge dispersion Anharmonicity Trade-off
Cooper-pair box / charge qubit $\lesssim 1$ charge (number) states large (strong $n_g$ sensitivity) large very charge-noise sensitive
Intermediate $\sim 1$-$10$ mixed moderate moderate partially charge-protected, larger anharmonicity
Transmon $\gg 1$ ($\sim 20$-$100$) phase-localized exponentially suppressed weak (power-law) charge-insensitive, modest anharmonicity

In the transmon limit $E_J/E_C\gg1$, the charge dispersion scales as $\sim e^{-\sqrt{8E_J/E_C}}$ while the relative anharmonicity scales as $\alpha_r \sim -(8E_J/E_C)^{-1/2}$. The transmon is just one limit of this family, not the only option.

Contrast the two oscillators a newcomer might confuse:

Property Linear LC oscillator Josephson (transmon)
Potential shape $\tfrac12 L^{-1}\Phi^2$ parabola $-E_J\cos\varphi$ cosine
Inductance constant $L$ $L_J(\varphi)=\Phi_0/(2\pi I_c\cos\varphi)$
Level spacing equal ($\hbar\omega$) unequal (shrinks with $m$)
Usable as a qubit? No (cannot address one transition) Yes
Key parameter $\omega = 1/\sqrt{LC}$ ratio $E_J/E_C$
Dissipation non-dissipative if built from ideal superconducting $L,C$ non-dissipative in the ideal junction model

Energy levels and anharmonicity

In the transmon regime $\varphi$ is localized near $0$, so expand the cosine to quartic order and treat the $\varphi^4$ term in first-order perturbation theory using harmonic-oscillator matrix elements. The result (Koch et al. 2007):

$$ E_m \simeq -E_J + \sqrt{8E_J E_C},\big(m+\tfrac12\big) - \frac{E_C}{12}\big(6m^2 + 6m + 3\big). $$

The harmonic part is $4E_C\hat n^2+\tfrac12E_J\hat\varphi^2$. With $[\hat\varphi,\hat n]=i$, choose $\hat\varphi=\varphi_{\rm zpf}(a+a^\dagger)$ and $\hat n=\tfrac{i}{2\varphi_{\rm zpf}}(a^\dagger-a)$; balancing the two quadratures gives $\varphi_{\rm zpf}=(2E_C/E_J)^{1/4}$ and the plasma frequency $\hbar\omega_p=\sqrt{8E_JE_C}$. The $m^2$ term bends the ladder. Subtracting levels:

$$ \hbar\omega_q = E_1 - E_0 \simeq \sqrt{8E_J E_C} - E_C, \qquad \alpha_E \equiv (E_2-E_1)-(E_1-E_0) \simeq -E_C, $$ $$ \alpha_\omega \equiv \omega_{12}-\omega_{01} = \frac{\alpha_E}{\hbar} \simeq -\frac{E_C}{\hbar}, \qquad \alpha_r \equiv \frac{\alpha_\omega}{\omega_{01}} \simeq -\Big(\frac{8E_J}{E_C}\Big)^{-1/2}. $$

So the energy anharmonicity is $\alpha_E\approx -E_C$; quoted as a cyclic frequency, $\alpha_E/h$ is typically a few hundred MHz, and the relative anharmonicity shrinks only as a weak power law of $E_J/E_C$. Why does this finally give a qubit? The $0\to1$ and $1\to2$ transitions now sit at different frequencies, detuned by $\alpha_\omega$. A pulse resonant with $\omega_{01}$ is off-resonant from $\omega_{12}$, so it leaves higher levels alone, provided its bandwidth $\sim 1/\tau$ stays below $|\alpha_\omega|/2\pi$. That is a hard gate-speed limit: too-fast pulses leak population into level $2$. Pulse-shaping (e.g. DRAG) is the practical fix.

Pitfall. A transmon is not a true two-level system, it is a weakly anharmonic multi-level oscillator. Levels $2, 3, \dots$ are always there, which is exactly why $|\alpha_\omega|$ and DRAG matter. And the nonlinearity comes entirely from the cosine; $E_C$ only sets the oscillator scale and the size of the anharmonicity, the charging term itself is not nonlinear.

Charge dispersion: the whole point of the transmon

Each level's energy depends on the gate charge $n_g$. Define the signed charge dispersion $\epsilon_m \equiv E_m(n_g=1/2)-E_m(n_g=0)$; the peak-to-peak width is $|\epsilon_m|$, and it is suppressed exponentially (Koch et al. 2007):

$$ \epsilon_m \simeq (-1)^m,E_C,\frac{2^{4m+5}}{m!}\sqrt{\frac{2}{\pi}}\Big(\frac{E_J}{2E_C}\Big)^{\frac{m}{2}+\frac34} e^{-\sqrt{8E_J/E_C}}. $$

The physics: making the band depend on $n_g$ requires the phase to tunnel between adjacent cosine wells ($\varphi\to\varphi+2\pi$). A WKB/instanton estimate of that tunnelling action gives the $e^{-\sqrt{8E_J/E_C}}$ factor (the exact problem maps onto Mathieu's equation, which supplies the algebraic prefactor). Flat bands mean $df_q/dn_g \approx 0$: stray $1/f$ charge noise barely shifts the qubit frequency, so dephasing is suppressed and $T_2$ is long. This is the transmon's defining payoff, and it costs only a weak power-law penalty in anharmonicity.

 E_m(n_g)   small E_J/E_C                   large E_J/E_C (transmon)
   │   __        __        __          │
   │  /  \      /  \      /  \  m=2     │  ────────────────────  m=2
   │ /    \    /    \    /    \         │
   │ \    /\  /    /\  /                │  ────────────────────  m=1
   │  \__/  \/    /  \/      m=1        │
   │   /\    /\    /\                   │  ────────────────────  m=0
   │  /  \  /  \  /  \       m=0        │   ε_m shrinks ~ e^(-√(8E_J/E_C))
   └──┴────┴────┴──── n_g               └────────────────────── n_g
      0   0.5    1                          0        0.5        1
   strongly curved bands              nearly flat ⇒ df_q/dn_g ≈ 0
   (big ε_m, charge-sensitive)        ⇒ insensitive to charge noise ⇒ long T₂

Pitfall. "Charge-insensitive" does not mean $n_g$ vanishes from $H$, it still appears. The point is the resulting bands are exponentially flat. Likewise "non-dissipative" is an ideal-element statement; real junctions still suffer quasiparticle, dielectric, and radiative losses.

A fully worked transmon (illustrative numbers)

Pick two energy scales (round teaching values, illustrative, not any real device): $E_C/h = 250\ \text{MHz}$ and $E_J/E_C = 50$, so $E_J/h = 12.5\ \text{GHz}$. All energies are quoted as frequencies by dividing by $h$.

Step 1: back out the hardware.

  • Total island capacitance: $C_\Sigma = e^2/(2E_C) = e^2/(2h\cdot 250,\text{MHz}) \approx 78\ \text{fF}$ (a large pad, how you reach big $E_J/E_C$).
  • Critical current: $I_c = 2eE_J/\hbar = 2e,(h\cdot 12.5,\text{GHz})/\hbar \approx 25\ \text{nA}$.
  • Zero-phase inductance: $L_{J0} = \hbar/(2eI_c) \approx 13\ \text{nH}$.

Step 2: qubit frequency. $\omega_q/2\pi \approx \sqrt{8E_JE_C}/h - E_C/h = \sqrt{8\cdot 12.5\cdot 0.25} - 0.25 = \sqrt{25} - 0.25 = 5.000 - 0.250 = 4.75\ \text{GHz}$ (plasma frequency $\omega_p/2\pi = 5.00\ \text{GHz}$).

Step 3: anharmonicity. Angular-frequency anharmonicity: $\alpha_\omega/2\pi \approx -E_C/h = -250\ \text{MHz}$, so $\omega_{12}/2\pi \approx 4.50\ \text{GHz}$, the $1\to2$ transition sits $250\ \text{MHz}$ below the $0\to1$. Relative: $\alpha_r \approx -(8\cdot 50)^{-1/2} = -1/\sqrt{400} = -0.05$, i.e. $-5%$.

Step 4: charge dispersion (the payoff). Exponent $\sqrt{8\cdot 50} = 20$, so $e^{-20}\approx 2.06\times10^{-9}$. The ground level barely moves: $\epsilon_0/h \approx (250,\text{MHz})\cdot 32\cdot 0.798\cdot 25^{0.75}\cdot 2.06\times10^{-9} \approx 150\ \text{Hz}$. But the quantity you actually measure is the $0\to1$ transition, and its dispersion is set by the upper level: $\epsilon_m$ grows fast with $m$ (the $2^{4m+5}$ prefactor), so evaluating the formula at $m=1$ gives $|\epsilon_1|/h \approx 11.8\ \text{kHz}$, about $80\times$ larger than $\epsilon_0$. Because $\epsilon_0$ and $\epsilon_1$ alternate in sign (the $(-1)^m$), the $0\to1$ frequency wiggles peak-to-peak by $|\epsilon_1-\epsilon_0|/h \approx |\epsilon_0|/h+|\epsilon_1|/h \approx 11.9\ \text{kHz}$ across a full $n_g$ period, about one part in $5\times10^5$ of $4.75\ \text{GHz}$. That is why the transmon barely feels charge noise. (Diagonalizing the Mathieu problem directly confirms the $\sim 12\ \text{kHz}$ transition swing.)

Step 5: gate-speed sanity check. With $|\alpha_\omega|/2\pi = 250\ \text{MHz}$, a plain resonant pulse must keep its spectral width $\sim 1/\tau$ well under $250\ \text{MHz}$, so $\tau\gtrsim$ a few ns; DRAG relaxes this.

Takeaways (all illustrative): $\omega_q/2\pi\approx 4.75\ \text{GHz}$, $\alpha_\omega/2\pi\approx -250\ \text{MHz}$, $0\to1$ charge dispersion $\approx 12\ \text{kHz}$ (level-0 dispersion $\epsilon_0\approx 150\ \text{Hz}$), $C_\Sigma\approx 78\ \text{fF}$, $I_c\approx 25\ \text{nA}$, $L_{J0}\approx 13\ \text{nH}$.

Common pitfalls

  • It's $2e$, not $e$. Both Josephson relations and $\Phi_0 = h/2e$ carry the Cooper-pair charge. The factor-of-two error is everywhere.
  • Bigger $E_J/E_C$ is not strictly better. It suppresses charge dispersion exponentially but preserves anharmonicity only as a power law; too little $|\alpha_\omega|$ means leakage. The transmon is a deliberate compromise.
  • Two anharmonicities. $\alpha_E$ is an energy ($\approx -E_C$), while $\alpha_\omega=\alpha_E/\hbar$ is angular frequency and $\alpha_E/h$ is the value quoted in Hz; $\alpha_r$ (relative, $\approx -5%$, dimensionless) measures fractional unevenness.
  • The cosine is flatter, not steeper. The negative quartic term lowers upper levels and shrinks spacings, that is what "negative anharmonicity" means.

Key takeaways

  • The Josephson junction supplies the standard nonlinear, ideally non-dissipative circuit element; a bare superconducting LC oscillator can be low-loss, but it remains harmonic and therefore unusable as an addressable qubit.
  • Two relations: $I = I_c\sin\varphi$ and $\dot\varphi = 2eV/\hbar$, giving the phase-dependent inductance $L_J(\varphi) = \Phi_0/(2\pi I_c\cos\varphi)$.
  • The junction energy is the cosine potential $-E_J\cos\varphi$; its quartic term flattens the well and unevenly spaces the ladder.
  • $H = 4E_C(\hat n - n_g)^2 - E_J\cos\hat\varphi$ is tridiagonal in the charge basis and governed by $E_J/E_C$, a continuum from charge qubit to transmon.
  • Anharmonicity $\alpha_\omega\approx -E_C/\hbar$ lets a selective pulse address one $0$-$1$ transition (gate-speed limited by $|\alpha_\omega|$); charge dispersion $\sim e^{-\sqrt{8E_J/E_C}}$ is the exponential charge-noise immunity that defines the transmon.

Go deeper

  • Koch et al., "Charge-insensitive qubit design derived from the Cooper pair box," Phys. Rev. A 76, 042319 (2007), the foundational transmon paper and source of the $E_m$, $\omega_q$, $\alpha_r$, and charge-dispersion formulas above. arXiv:cond-mat/0703002.
  • Krantz et al., "A Quantum Engineer's Guide to Superconducting Qubits," Appl. Phys. Rev. 6, 021318 (2019), the standard modern review. arXiv:1904.06560.
  • Blais et al., "Circuit Quantum Electrodynamics," Rev. Mod. Phys. 93, 025005 (2021), authoritative derivations of the charge-basis Hamiltonian and charge dispersion. arXiv:2005.12667.

← Back to project README · Tutorial index