The Superconducting Transmon

An IBM Quantum Heron chip at room temperature fits in your palm. It is a thin square of sapphire, perhaps the size of a one-rupee coin, patterned with curling niobium wires, a few dozen tiny cross-shaped structures, and a forest of microwave transmission lines spreading out from the centre. At 20 millikelvin — colder than outer space — each of those cross-shaped structures is a qubit. There are 133 of them on the chip.

What is a superconducting qubit, physically? It is not an atom. It is not a photon. It is not a trapped ion. It is a specially engineered electrical circuit — an anharmonic oscillator built from a superconducting wire, a tiny patch of insulating aluminium oxide, and a shunt capacitor — that behaves quantum-mechanically when cooled deep below the temperature of intergalactic space.

The specific oscillator IBM uses is called a transmon, and it is the dominant superconducting qubit in the industry. Google's Willow chip uses transmons. Rigetti's Ankaa-3 uses transmons. IBM's Condor, Heron, and every announcement of theirs for the next decade will use transmons. Nearly every university group that fabricates its own chips fabricates transmons.

This chapter builds the transmon from scratch. Start with a regular LC oscillator; then add the Josephson junction that makes it quantum; then add the shunt capacitor that makes it robust. By the end you will know what physical object a transmon is, what numbers describe it, and why it keeps winning.

A regular LC oscillator first

Before the transmon, the LC oscillator. If you put a capacitor C and an inductor L in a loop, you get a circuit that rings: charge sloshes back and forth between the capacitor plates at a natural frequency \omega = 1/\sqrt{LC}. The energy alternates between electric energy stored in C and magnetic energy stored in L.

The Hamiltonian is

where Q is the charge on the capacitor and \Phi is the magnetic flux through the inductor. Two squared terms added together — this is exactly the form of a harmonic oscillator (think of Q as momentum and \Phi as position), and at low enough temperatures the whole thing is quantum.

Why LC oscillators need to be quantised: above the superconducting transition temperature, an LC circuit is just a bit of resistive ringing — the energy leaks out before quantum coherence can matter. Below T_c (a few kelvin for aluminium, about 9 K for niobium), the resistance vanishes and the circuit becomes lossless. Then the Hamiltonian above is the full story, and its spectrum — the allowed energies — has the quantum-mechanical evenly-spaced ladder.

The energy levels of a quantum LC oscillator are

— the familiar harmonic-oscillator ladder, with spacing \hbar\omega between adjacent levels.

Why this won't work as a qubit. A qubit needs exactly two addressable states, |0\rangle and |1\rangle. If you send a microwave pulse at frequency \omega_{01} = E_1 - E_0 to drive |0\rangle \to |1\rangle, the same pulse also drives |1\rangle \to |2\rangle (same spacing), |2\rangle \to |3\rangle, and so on. You cannot isolate the lowest two levels of a harmonic oscillator with a microwave pulse — they are inextricably coupled to an infinite ladder of higher levels. You need anharmonicity: unequal energy spacings.

The Josephson junction — the magic ingredient

The problem: the LC oscillator is too regular. The fix: replace the inductor with a Josephson junction.

A Josephson junction is two pieces of superconductor separated by an insulating barrier a few atoms thick (typically aluminium oxide, ~1 nm). At first glance this is just a capacitor — insulator between two conductors. But because the conductors are superconductors, something remarkable happens: Cooper pairs can tunnel through the barrier as a coherent, macroscopic quantum-mechanical current.

The result is two Josephson relations, derived by Brian Josephson in 1962:

Here \delta is the phase difference across the junction (a quantum-mechanical variable — the difference in the phase of the superconducting wavefunction on the two sides), I_c is the critical current (a constant of the junction), and V is the voltage across it.

Why these relations make the junction a nonlinear inductor: a regular inductor has V = L\,dI/dt, which gives I \propto \int V\,dt \propto \delta — the current is linear in the phase. The Josephson relation has I \propto \sin\delta — nonlinear in the phase. Expand in powers: \sin\delta = \delta - \delta^3/6 + \ldots The linear term looks like a regular inductor; the cubic (and higher) terms are the nonlinearity that makes the oscillator anharmonic.

The energy stored in the junction is

where E_J = \hbar I_c / (2e) is the Josephson energy. Again, compare to a regular inductor: U_L \propto \Phi^2, quadratic in the flux. The Josephson junction has U_J \propto -\cos\delta, which is quadratic for small \delta but becomes non-quadratic for larger amplitude oscillations.

Building the transmon Hamiltonian

Put the Josephson junction in parallel with a capacitor C. Also let n be the number of Cooper pairs that have tunneled across the junction (so Q = 2e\,n is the charge, and n is the conjugate variable to the phase \delta). Let n_g be an externally controlled offset (from a gate voltage — the "g" stands for gate). The Hamiltonian of the whole circuit is

where E_C = e^2 / (2C) is the charging energy (the energy cost of putting one extra electron on the capacitor) and \phi is the superconducting phase difference across the junction (same as \delta — notation drift in the literature). Two energy scales, E_C and E_J, and their ratio is what decides what kind of qubit you have.

• E_J \ll E_C (charge qubit regime). The states of definite n are nearly energy eigenstates. The qubit is very sensitive to the gate voltage n_g — tiny charge fluctuations in the environment cause large frequency shifts. This was the early Cooper-pair-box design — it worked but suffered from severe charge noise.
• E_J \approx E_C (intermediate). Mixed regime. Various early superconducting designs sat here.
• E_J \gg E_C (transmon regime). The Josephson term dominates. The charge sensitivity is exponentially suppressed — shift n_g and the qubit frequency barely moves. This is the transmon regime, introduced by Koch, Yu, Gambetta, Houck, Schuster, Majer, Blais, Devoret, Girvin, and Schoelkopf at Yale in 2007.

The transmon paper's key insight was this: in the limit E_J / E_C \gg 1, the wavefunction of each energy eigenstate is delocalised over many values of n, and its sensitivity to n_g decays as \exp(-\sqrt{8 E_J/E_C}) — exponentially small. Meanwhile the anharmonicity, which also decreases with E_J/E_C, only decays as a power law: \alpha \approx -E_C. So by picking E_J / E_C \approx 50–100, you get nearly zero charge noise sensitivity and keep hundreds of MHz of anharmonicity. Best of both worlds.

Transmon energy levels

In the transmon regime, expand \cos\phi around its minimum at \phi = 0:

The quadratic term is the harmonic-oscillator part; the quartic is the leading nonlinearity. The energy levels come from diagonalising the full Hamiltonian and are well-approximated (for E_J \gg E_C) by

The first term is a constant offset. The second term is the harmonic ladder with "plasma frequency" \hbar\omega_p = \sqrt{8 E_J E_C}. The third term is the anharmonic correction.

From this, the relevant frequencies are

so the anharmonicity is

The picture. The transmon has an evenly spaced ladder at zeroth order (like an LC oscillator) but each successive gap is smaller by E_C. The |0\rangle \to |1\rangle transition is at \omega_{01}; the |1\rangle \to |2\rangle transition is at \omega_{01} - E_C/\hbar, detuned by the anharmonicity. With E_C/h \approx 250 MHz and \omega_{01}/2\pi \approx 5 GHz, a microwave pulse at 5 GHz drives |0\rangle \to |1\rangle on resonance but is 250 MHz off resonance from |1\rangle \to |2\rangle. Fast pulses (bandwidth \sim 1/T_g) are slow enough to resolve this gap if T_g \gg 1/(250 \text{ MHz}) \approx 4 ns — which is why transmon gates are typically 20–100 ns long, not faster.

Typical parameters

These are representative numbers for a modern IBM Heron-class transmon (2026):

The cryogenic stack

A transmon lives at 15 millikelvin, inside a dilution refrigerator — a multi-stage cooling apparatus that first uses liquid nitrogen (77 K), then helium-4 (4 K), then a helium-4 / helium-3 mixture whose phase separation provides cooling power down to 10 mK. The chip sits at the coldest stage, wrapped in many layers of radiation shielding to block stray infrared photons. Microwave lines deliver control pulses to the chip; each line is heavily attenuated at each stage to thermalise it (so that noise from room temperature does not propagate down and excite the qubits).

Why so cold? At 5 GHz the qubit's energy gap is \hbar\omega_{01} \approx 3\times 10^{-24} J. The thermal-excitation condition k_B T \ll \hbar\omega_{01} requires T \ll 250 mK. Going well below this — to 20 mK — makes thermal occupation of |1\rangle less than 0.1\%, which is necessary for reliable initialisation in |0\rangle.

A typical dilution fridge for a transmon chip is two metres tall and weighs a few hundred kilograms, and uses \sim 15 kW of electrical power and \sim 1 litre per hour of liquid helium to maintain base temperature. The cost is \sim \$500,000 per fridge, plus an integrated microwave-electronics rack (another \$500,000+). This is why you rent transmon time through IBM's cloud rather than buy the chip.

Advantages and disadvantages

Why transmons win.

• Lithography. The qubits are patterned on a chip using the same photolithography and electron-beam lithography that the semiconductor industry has used for decades. IBM's 1121-qubit Condor chip was fabricated in a standard cleanroom using conventional tools. Scaling a transmon chip to a thousand qubits is not a physics problem; it is a fab problem, and we know how to solve fab problems.
• Fast gates. Transmon gates run in tens of nanoseconds. Compared to ion-trap gates (microseconds to milliseconds), this gives roughly 10^4\times more operations per unit of wall-clock time — so circuits that would take an ion trap a minute complete on a transmon in microseconds.
• Strong coupling. Transmons couple easily to microwave resonators (for readout) and to other transmons (for two-qubit gates), because the circuit elements that implement the coupling are just wires and capacitors — the same lithography that makes the qubits.
• A mature software stack. Qiskit, Cirq, Cuda-Q, and the academic package Qutip all have deep support for transmon hardware. IBM's and Google's cloud access has made transmons the default target for a decade of algorithm research.

Why transmons still struggle.

• Coherence-limited. T_1 and T_2 are ~100 μs on the best chips — short compared to trapped ions. Every transmon algorithm runs against the wall of "finish before the qubit relaxes." Materials research is slowly improving this; more aggressive designs (Google's "coaxmon," IBM's low-loss tantalum chips) push T_1 toward the millisecond regime.
• Cryogenic cost. A transmon chip is useless without its $1M dilution fridge. Trapped ions need a vacuum chamber and lasers but no dilution refrigerator; photonic qubits run at room temperature. Cryogenics is the transmon's structural tax.
• Frequency crowding. Each transmon has its own fixed \omega_{01} (set by fabrication variation in E_J, E_C), and on a chip with hundreds of qubits, the frequencies crowd together — two neighbouring qubits with accidentally-close frequencies have crosstalk (driving one slightly drives the other) and are harder to isolate. Tunable transmons (next section) partly solve this but add their own complications.
• Two-level-system defects. Tiny defects in the junction oxide and substrate form unwanted two-level systems that couple to the transmon and limit T_1. These are a materials problem, and solving them fully is an open research area.

Tunable transmons and couplers

A pure transmon has a single Josephson junction and a fixed frequency. Replace the junction with a SQUID (Superconducting Quantum Interference Device) — two junctions in a loop — and you can tune the effective Josephson energy E_J(\Phi) by threading an external magnetic flux \Phi through the SQUID loop:

where \Phi_0 = h/(2e) is the superconducting flux quantum. Adjusting \Phi adjusts E_J, which adjusts \omega_{01}. This gives the experimenter a knob to move the qubit frequency in and out of resonance with neighbours or with readout resonators.

The downside: SQUID-based tunable transmons are sensitive to flux noise, which shortens T_2. Modern architectures balance this: the qubits themselves are fixed-frequency (for long T_2), and a separate tunable circuit between qubits — a tunable coupler, often a third tunable transmon — is used to turn two-qubit interactions on and off. IBM's Heron, Rigetti's Ankaa-3, and Google's Willow all use tunable-coupler designs.

Indian context

India's National Quantum Mission (NQM), launched in 2023 with a ₹6,003 crore budget, funds domestic superconducting-qubit development at several institutions. At IISc Bangalore, the Centre for Nano Science and Engineering hosts a transmon fabrication and characterisation programme; at TIFR Mumbai, the historic NMR-quantum-computing group has been joined by a superconducting-qubit group that has fabricated and measured prototype transmons. IIT Delhi and IIT Madras host related efforts. The explicit goal: indigenous Indian superconducting quantum processors of 50–100 qubits by 2030, competitive with current IBM Heron-class chips.

The fabrication challenge is non-trivial but well-posed. Every step — niobium film deposition, aluminium-oxide junction growth, photolithography, electron-beam lithography, dilution-refrigerator operation — uses techniques well-known to Indian semiconductor labs. The gap is infrastructure investment (a few dilution fridges, a class-1000 cleanroom dedicated to transmon fab, trained cryogenics engineers), not fundamental science. NQM funding closes exactly that gap.

Examples

Common confusions

• "A transmon is a superconductor." A transmon is made of superconductors (aluminium, niobium, or tantalum films on a sapphire or silicon substrate), but a superconductor by itself is just a lossless wire. A transmon is a specific circuit — capacitor + Josephson junction — that you fabricate out of those superconducting films. It is a macroscopic quantum object.

• "A qubit frequency is fixed." For fixed-frequency transmons, yes — the Josephson energy is set by fabrication of a single junction and cannot be tuned. But tunable transmons (built around a SQUID with two junctions) can be tuned over a range of ~1 GHz by adjusting the flux bias. Modern architectures often mix fixed-frequency qubits with tunable couplers between them.

• "A Cooper pair is two electrons stuck together." Close but important detail: a Cooper pair is two electrons bound in the superconducting ground state by phonon-mediated attraction, with opposite spins and opposite momenta. They are a bosonic quasi-particle (integer spin) and many of them can occupy the same quantum state, which is why supercurrent flows without resistance. A Cooper pair carries charge 2e, which is why the transmon Hamiltonian has factors of 2e everywhere.

• "Transmons are better than trapped ions." Transmons are faster and more manufacturable; trapped ions have longer coherence times and higher gate fidelities. Each platform wins on different metrics. Right now (2026), the transmon lead is in qubit count and gate speed; the ion lead is in fidelity per gate and coherence per qubit. Which matters more depends on the algorithm.

• "The transmon's |1\rangle is always the first excited state." By convention in nearly every paper, yes: |0\rangle is the ground state, |1\rangle is the first excited state. But some experiments encode qubits in other pairs of levels — for example, the qutrit encoding uses |0\rangle, |1\rangle, |2\rangle as three states of a ternary quantum digit. And some error-corrected codes use the higher levels |2\rangle, |3\rangle for leakage detection. The two-level subspace is a choice, not a physical constraint.

• "Anharmonicity is an imperfection." It is the feature that makes the transmon a qubit in the first place. A perfectly harmonic oscillator would be useless as a qubit. The question is not whether to have anharmonicity but how much — enough to address |0\rangle \to |1\rangle without driving |1\rangle \to |2\rangle, but not so much that you lose the charge-noise insensitivity. The transmon regime balances these two.

Going deeper

The charge-noise suppression — where the exponential comes from

The Hamiltonian H = 4 E_C (n - n_g)^2 - E_J\cos\phi has the form of a particle (with position \phi, momentum-like variable n) in a periodic cosine potential. The eigenvalues are the Mathieu characteristic values a_\nu(q) with q = E_J/(2 E_C) and a noninteger characteristic exponent \nu set by the charge offset n_g:

for integer k chosen by level ordering. Tedious but textbook. The charge dispersion — how much the level depends on n_g — is

The key factor is e^{-\sqrt{8 E_J/E_C}} — exponentially small in the parameter ratio. At E_J/E_C = 80, \sqrt{8 \cdot 80} = \sqrt{640} \approx 25.3, so \epsilon_1 \propto e^{-25.3} \approx 10^{-11} — effectively zero compared to the tens-of-MHz scale of typical charge noise. The transmon is charge-noise immune by engineering.

Meanwhile the anharmonicity \alpha \approx -E_C is polynomially small in E_J/E_C (it goes as (E_J/E_C)^0 really — constant), so the anharmonicity survives into the transmon regime. Koch et al.'s 2007 paper spelled this out and renamed the qubit from Cooper-pair-box-with-large-shunt-capacitor to transmon — the "transmission-line shunted plasma oscillation qubit."

Charge qubits vs flux qubits vs transmons

There were earlier superconducting-qubit designs:

• Cooper-pair box (Nakamura et al. 1999): small-island charge qubit, E_C \approx E_J. Directly sensitive to charge offset → charge noise dominates T_2.
• Flux qubit (Mooij et al. 1999): a loop with three Josephson junctions biased near half a flux quantum. Two lowest states are symmetric/antisymmetric combinations of clockwise and counterclockwise supercurrent. Moderate coherence (μs scale).
• Phase qubit (Martinis et al. 2002): a single junction in the zero-voltage regime, used as a qubit by picking out two of the levels of a tilted-washboard potential. Never scaled well.
• Transmon (Koch et al. 2007): large shunt capacitor, E_J/E_C \gg 1. Killed charge sensitivity. Became the standard.

A newer design, fluxonium (Manucharyan et al. 2009), adds a super-inductor in parallel with the junction to push E_L \ll E_J, creating a different level structure with very high anharmonicity (~500 MHz) and long T_1 (sometimes > 1 ms). Fluxonium is gaining traction in some labs and may become a future standard, but transmons still dominate industrial processors.

Readout — the dispersive regime

To measure a transmon, couple it capacitively to a microwave resonator (a \lambda/4 or \lambda/2 transmission line, or a lumped-element LC resonator). In the far-detuned limit |\omega_r - \omega_{01}| \gg g (where g is the qubit-resonator coupling), the resonator's frequency is shifted by a qubit-state-dependent amount:

with +\chi for qubit in |0\rangle and -\chi for qubit in |1\rangle. Typical \chi/(2\pi) \approx 1 MHz. Probing the resonator with a microwave tone (~100 ns long pulse, a few photons) and measuring the reflected/transmitted phase tells you which shift is present, hence which state the qubit is in.

The readout does not directly project the qubit — it measures the resonator. But because the resonator frequency is entangled with the qubit state, measuring the resonator performs an effective projective measurement on the qubit, with readout fidelity \sim 99\% for a well-designed dispersive readout chain.

Where the transmon goes next — logical qubits

A transmon is the physical qubit; the road ahead is the logical qubit. With surface-code error correction at distance d, a logical qubit uses d^2 physical qubits and tolerates physical-gate error rates below a threshold (roughly 10^{-3} per gate). Google's Willow experiment (2024) demonstrated that logical error rates can decrease exponentially with code distance on a 105-transmon chip — the first real demonstration of the "error-correction threshold" in action.

To run Shor's algorithm on 2048-bit RSA, current estimates require roughly 10^6–10^7 physical qubits — a factor of 10^4 beyond today's chips. IBM's 2029 roadmap targets ~200,000 physical qubits in multi-chip modules; getting to 10^7 is the defining engineering project of the 2030s. Whether transmons survive that long or get replaced by fluxonium or something entirely different (a topological qubit, perhaps, or a neutral atom with superior scaling) is an open question. But for the next ten years, the transmon is the platform to know.

Where this leads next

• Superconducting gates and readout — how microwave pulses implement X, Y, Z, H, CNOT, and CZ on transmons, and how dispersive readout distinguishes the qubit states.
• Decoherence — T₁ and T₂ — the two numbers that limit how many operations you can run before a transmon forgets what state it was in.
• Trapped ions — Paul traps — the leading alternative platform, with seconds-long coherence but slower gates.
• Photonic quantum computing — room-temperature, loss-limited, probabilistic-gate competitor.
• IBM Heron — the current flagship transmon processor, and what its datasheet reveals.

References

1. Koch, Yu, Gambetta, Houck, Schuster, Majer, Blais, Devoret, Girvin, Schoelkopf, Charge-insensitive qubit design derived from the Cooper pair box (2007) — arXiv:cond-mat/0703002. The original transmon paper.
2. Wikipedia, Transmon — overview of the circuit and its parameters.
3. Krantz, Kjaergaard, Yan, Orlando, Gustavsson, Oliver, A quantum engineer's guide to superconducting qubits (2019) — arXiv:1904.06560. The definitive modern review.
4. Blais, Grimsmo, Girvin, Wallraff, Circuit quantum electrodynamics — Rev. Mod. Phys. (2021) — arXiv:2005.12667. Dispersive readout, cQED coupling theory, and the full Hamiltonian framework.
5. John Preskill, Lecture Notes on Quantum Computation, Ch. 7 — physical realisations of quantum computing, including superconducting qubits. theory.caltech.edu/~preskill/ph229.
6. IBM Quantum, Hardware documentation — live per-qubit parameters (\omega_{01}, \alpha, T_1, T_2) for Heron and earlier generations.