The Transmon Qubit

The transmon is not a candidate technology; it is the incumbent. IBM, Google, Rigetti, and many academic groups run processors with tens to hundreds of fixed-frequency or flux-tunable transmons, wired through circuit-QED readout and coupled by resonators, tunable couplers, or cross-resonance drives. The 2019 random-circuit-sampling experiment used 53 transmons, and the frontier problem now is quantum error correction: transmons are the physical substrate for the surface code, where a two-dimensional lattice of data and ancilla qubits undergoes repeated stabilizer measurements, and recent experiments have shown the logical error rate falling as the code distance grows — the long-sought "below threshold" regime. A parallel line uses a single transmon as a nonlinear ancilla to control a high-Q microwave cavity, encoding a logical qubit in bosonic cat or GKP states. The open engineering problems are exactly the ones the physics predicts: frequency collisions and two-level-system defects (managed by flux-tuning qubits to safe frequencies), leakage into $|2\rangle$ (managed by DRAG and by careful anharmonicity budgeting), residual $ZZ$ crosstalk between always-coupled neighbours (managed by tunable couplers), and the relentless push on coherence times, now routinely $100\ \mu\text{s}$ to milliseconds, toward the fault-tolerance threshold. Every one of these knobs is a line in the transmon Hamiltonian above.

Imagine a particle whose coordinate is the superconducting phase difference $\varphi$ across the junction, sliding in the washboard potential $-E_J\cos\varphi$. When $E_J\gg E_C$ the particle is heavy and the well is deep, so it sits far down near the bottom, executing small, almost-harmonic oscillations at the plasma frequency $\omega_p\approx\sqrt{8E_JE_C}/\hbar$. But a cosine is not a parabola: its curvature softens as you climb, so successive energy levels crowd closer together. That gentle crowding is the anharmonicity, and it is precisely what lets a microwave drive tuned to $\omega_{01}$ address the $|0\rangle\!\to\!|1\rangle$ transition without also driving $|1\rangle\!\to\!|2\rangle$. A pure harmonic $LC$ oscillator has a perfectly even ladder and can never be a qubit; the Josephson junction's nonlinear inductance is what breaks that degeneracy.

The conjugate side of the same story is charge. Phase and Cooper-pair number are conjugate variables, $[\hat\varphi,\hat n]=i$, so localizing $\varphi$ deep in the well delocalizes $\hat n$: the ground-state wavefunction spreads over many charge states. A wavefunction smeared across many values of $\hat n$ automatically averages over the offset charge $n_g$, so a slow drift in $n_g$ barely moves the energy levels — charge noise is motionally narrowed away, exactly as a fast-tumbling molecule narrows its NMR line. And because tunnelling between neighbouring cosine wells (which is what makes the levels wobble with $n_g$ at all) is a barrier-penetration process, that wobble shrinks exponentially with the barrier height $\sim\sqrt{E_J/E_C}$. The anharmonicity, by contrast, is a low-order correction to a parabola and fades only algebraically. The second figure is the entire design philosophy on one pair of axes.

Compared with a trapped ion — whose electronic levels differ by hundreds of terahertz — the transmon's anharmonicity is minuscule, a few hundred megahertz sitting on a multi-gigahertz transition. That is the price of charge immunity, and it is why transmon gates use carefully shaped pulses (DRAG) to keep population from leaking into $|2\rangle$, a worry an ion never has. The rest of this article turns these two pictures into equations, derives the trade quantitatively, and follows it out to the processors it now powers.

The starting point is the circuit Hamiltonian. Quantize a Josephson junction shunted by a capacitor and you get a "particle" of charge moving in a cosine potential — a kinetic (charging) term plus a Josephson potential term.

where: $E_C=e^2/2C_\Sigma$ — charging energy, with $C_\Sigma$ the total island capacitance and $e$ the electron charge; $E_J=I_c\Phi_0/2\pi$ — Josephson energy, with $I_c$ the junction critical current and $\Phi_0=h/2e$ the flux quantum; $\hat n$ — number operator for Cooper pairs transferred across the junction; $n_g$ — offset (gate) charge in units of $2e$, the knob that charge noise jitters; $\hat\varphi$ — superconducting phase difference, conjugate to $\hat n$ with $[\hat\varphi,\hat n]=i$. Reading it: the first term wants a definite charge, the second a definite phase, and the ratio $E_J/E_C$ decides which one wins. Deep in the well the phase barely strays from zero, so expanding the cosine and keeping the leading quartic term turns the circuit into a weakly anharmonic oscillator whose spectrum you can read off. The lowest transition sits just below the bare plasma frequency.

where: $\omega_{01}$ — angular frequency of the $|0\rangle\!\to\!|1\rangle$ transition; $\sqrt{8E_JE_C}=\hbar\omega_p$ — the harmonic plasma spacing at the bottom of the well; $-E_C$ — the leading Duffing correction that pulls the transition down. Reading it: pick $E_C$ and $E_J/E_C$ and you have set the qubit frequency, typically $4$ to $6$ GHz. The same expansion fixes how unequal the ladder is. The anharmonicity — the gap between the first and second transitions — is set entirely by the charging energy and is negative, because the levels compress as they rise.

where: $\alpha$ — anharmonicity, negative because the transmon spectrum compresses upward; $\omega_{12}$ — frequency of the $|1\rangle\!\to\!|2\rangle$ transition; $E_C$ — charging energy, which sets the absolute anharmonicity scale. Reading it: $|\alpha|\approx E_C/h$ is what a drive must stay spectrally narrower than, so it directly bounds gate speed. The reason the transmon regime is safe from charge noise is the charge dispersion — the amount each level swings as $n_g$ drifts. From the exact Mathieu-equation band structure this swing is exponentially small in $\sqrt{E_J/E_C}$.

where: $\epsilon_m(n_g)$ — deviation of level $m$ from its $n_g$-averaged value, i.e. the charge-dispersion band; $\epsilon_m$ — peak-to-peak amplitude of that band, which sets charge-noise dephasing; $\sqrt{8E_J/E_C}$ — the effective WKB tunnelling action through the cosine barrier. Reading it: at $E_J/E_C=50$ the exponent is $20$, so $\epsilon\sim e^{-20}\approx2\times10^{-9}$ — the qubit frequency stops depending on $n_g$ altogether. Set the exponential of (4) against the algebraic decay of the anharmonicity and the whole trade-off becomes one ratio. Relative anharmonicity — what gate speed really cares about — falls only as an inverse square root of the design ratio.

where: $\alpha/\omega_{01}$ — anharmonicity as a fraction of the transition frequency; $E_J/E_C$ — the design ratio. Reading it: doubling $E_J/E_C$ costs only a factor $\sqrt2$ in relative anharmonicity but buys an exponential drop in charge dispersion — the bargain in the second figure. To read a transmon out you couple it to a microwave resonator far off resonance. Eliminating the resonator gives a qubit-state-dependent pull on the cavity frequency; keeping the transmon's third level puts an anharmonicity factor into that pull.

where: $\omega_r$ — bare resonator frequency; $g$ — transmon–resonator (vacuum Rabi) coupling; $\Delta=\omega_{01}-\omega_r$ — qubit–cavity detuning, large in the dispersive limit $|\Delta|\gg g$; $\chi$ — dispersive shift, so the cavity sits at $\omega_r\pm\chi$ depending on the qubit state; $\alpha$ — anharmonicity, whose appearance in $\chi$ is the fingerprint of a multilevel transmon rather than a true two-level atom. Reading it: measure the cavity's frequency and you have measured the qubit, without destroying it — a quantum-non-demolition readout. Finally, replacing the single junction with a two-junction SQUID loop makes $E_J$ — and hence the qubit frequency — tunable with an applied magnetic flux, the standard way to bring transmons in and out of resonance for gates.

where: $\Phi$ — external flux threading the SQUID loop; $\Phi_0=h/2e\approx2.07\times10^{-15}$ Wb — the superconducting flux quantum; $E_{J,\max}$ — the Josephson energy at $\Phi=0$. Reading it: the frequency is flat (first-order flux-insensitive) at $\Phi=0$, the "sweet spot", and steepens away from it, trading tunability for flux-noise dephasing.

The transmon is defined by where it sits on the $E_J/E_C$ axis, so its limiting cases are literally its family tree — the charge qubit it escaped, the harmonic oscillator it must not become, and the flux-tunable cousin it spawns.

1 · Transition frequency and anharmonicity from the design ratio

A transmon is designed with $E_C/h=250$ MHz and $E_J/E_C=50$. Find its qubit frequency $\omega_{01}/2\pi$, its anharmonicity $\alpha/2\pi$, and the naive minimum single-qubit gate time set by leakage.

Step 1. First get the Josephson energy: $E_J/h=50\times250\ \text{MHz}=12.5$ GHz. This is the deep-well regime, so equations (2) and (3) apply.

Step 2. The plasma spacing is $\sqrt{8E_JE_C}/h=\sqrt{8\times12.5\times0.25}\ \text{GHz}=\sqrt{25}=5.0$ GHz. Subtract the Duffing correction $E_C/h=0.25$ GHz to get $\omega_{01}/2\pi\approx5.0-0.25=4.75$ GHz — a textbook transmon frequency.

Step 3. The anharmonicity is simply $\alpha/2\pi\approx-E_C/h=\mathbf{-250}$ MHz, so the $|1\rangle\!\to\!|2\rangle$ transition sits at $\omega_{12}/2\pi\approx4.50$ GHz.

Step 4. The leakage speed limit is $t_{\text{gate}}\gtrsim2\pi/|\alpha|=1/|\alpha/2\pi|=1/(250\times10^6\ \text{Hz})=\mathbf{4}$ ns. Real DRAG-shaped gates run at a comfortable $20$–$40$ ns, a few times this bound. Both the $4.75$ GHz frequency and the $-250$ MHz anharmonicity are exactly the values a lab would quote.

2 · How dead is the charge noise

For the same device ($E_J/E_C=50$), estimate the charge-dispersion suppression, and compare with a shallow device at $E_J/E_C=5$.

Step 1. The exponent is $\sqrt{8E_J/E_C}$. At $E_J/E_C=50$ this is $\sqrt{400}=20$, so the dispersion amplitude is $\propto e^{-20}\approx\mathbf{2\times10^{-9}}$ of its band scale — the qubit frequency is utterly deaf to $n_g$.

Step 2. At $E_J/E_C=5$ the exponent is $\sqrt{40}\approx6.3$, giving $e^{-6.3}\approx1.8\times10^{-3}$. That is about six orders of magnitude larger dispersion than at $50$ — enough charge sensitivity to reintroduce serious dephasing.

Step 3. Meanwhile relative anharmonicity only changed from $1/\sqrt{40}\approx0.16$ to $1/\sqrt{400}=0.05$ — a factor of about three. Six orders of magnitude of noise immunity bought for a threefold anharmonicity cut is the transmon bargain in one line.

3 · The dispersive readout shift

A transmon with $\omega_{01}/2\pi=4.75$ GHz and $\alpha/2\pi=-250$ MHz is coupled with $g/2\pi=100$ MHz to a resonator at $\omega_r/2\pi=5.75$ GHz. Find the dispersive shift $\chi$ and the cavity pull between qubit states.

Step 1. The detuning is $\Delta/2\pi=\omega_{01}/2\pi-\omega_r/2\pi=4.75-5.75=-1.0$ GHz, and $|\Delta|=1$ GHz $\gg g=100$ MHz, so the dispersive limit (6) holds.

Step 2. The two-level piece is $g^2/\Delta$: with $(g/2\pi)^2=(0.1\ \text{GHz})^2=0.01\ \text{GHz}^2$ and $\Delta/2\pi=-1.0$ GHz, this gives $-0.01$ GHz $=-10$ MHz.

Step 3. The multilevel factor is $\alpha/(\alpha+\Delta)=(-0.25)/(-0.25-1.0)=(-0.25)/(-1.25)=0.2$. Multiply: $\chi/2\pi=(-10\ \text{MHz})\times0.2=-2$ MHz.

Step 4. The cavity therefore sits at $\omega_r-\chi$ or $\omega_r+\chi$ depending on the qubit state, a total pull of $|2\chi|/2\pi=\mathbf{4}$ MHz — comfortably resolvable against a few-hundred-kHz cavity linewidth, which is why this readout is fast and high-fidelity.

The transmon is a Yale invention, and no single name owns it, but one figure sits at the centre of the circuit-QED program that made it matter: Robert Schoelkopf, whose lab turned superconducting qubits from fragile curiosities into the readable, couplable, scalable devices that now fill dilution refrigerators worldwide.

Texts and papers. Koch et al., "Charge-insensitive qubit design derived from the Cooper pair box", Phys. Rev. A 76, 042319 (2007) — the founding paper · Krantz et al., "A Quantum Engineer's Guide to Superconducting Qubits", Appl. Phys. Rev. 6, 021318 (2019) — the practical bible · Blais, Grimsmo, Girvin, Wallraff, "Circuit Quantum Electrodynamics", Rev. Mod. Phys. 93, 025005 (2021) — the cQED reference · Devoret & Schoelkopf, "Superconducting Circuits for Quantum Information: An Outlook", Science 339, 1169 (2013) · Nielsen & Chuang, Quantum Computation and Quantum Information, Ch. 7 (physical realizations).

Provenance. The Hamiltonian, transition frequency, anharmonicity, and charge-dispersion exponent follow Koch et al. (2007); the multilevel dispersive shift and cQED treatment follow Blais et al. (2021) and Krantz et al. (2019). A verifier should: (1) recover $\hbar\omega_{01}=\sqrt{8E_JE_C}-E_C$ and $\alpha=-E_C/\hbar$ by perturbative expansion of the cosine, cross-checked against exact charge-basis diagonalization; (2) confirm the worked numbers ($\omega_{01}/2\pi=4.75$ GHz, $\alpha/2\pi=-250$ MHz, $e^{-20}\approx2\times10^{-9}$, $t_{\text{gate}}\gtrsim4$ ns, $\chi/2\pi=-2$ MHz); (3) check the charge-dispersion exponent $\sqrt{8E_J/E_C}$ against the Mathieu band structure; (4) verify $\Phi_0=h/2e\approx2.07\times10^{-15}$ Wb. No device-specific numbers are asserted as universal; the quoted $E_C$, $E_J/E_C$, coupling, and frequency values are representative textbook ranges.

Prerequisites. The Josephson Junction · Superconducting Qubits · The DiVincenzo Criteria. Leads to. Circuit Quantum Electrodynamics · Dispersive Readout.

Ask the tutor: Ask me to diagonalize the transmon Hamiltonian numerically, tune $E_J/E_C$ to trade anharmonicity against charge dispersion, or work through DRAG pulse shaping and dispersive readout for a device you specify.