A trapped ion or a lone atom is a qubit nature hands you — every one identical, forever. A superconducting qubit is different: it is a qubit we build, a tiny electrical circuit etched onto a chip out of aluminium and photographable under a microscope. It holds billions of atoms, yet cooled to a whisker above absolute zero it stops behaving like a classical wire and starts behaving like a single quantum object with discrete energy levels. This is the leading solid-state platform — the one behind Google's Sycamore, IBM's processors and Rigetti's chips — precisely because we can engineer it and print it by the hundred.

Start with a circuit that oscillates

The simplest quantum circuit is an LC resonator: an inductor L and a capacitor C wired in a loop. Charge sloshes back and forth at a definite frequency, exactly like a mass on a spring — a quantum harmonic oscillator. Its energy comes in discrete rungs, and that is the good news. The bad news is that those rungs are equally spaced:

Equal spacing is fatal. If a microwave pulse tuned to lift the circuit from |0\rangle to |1\rangle also perfectly fits the |1\rangle\!\to\!|2\rangle step, the circuit climbs straight past the two levels we wanted to use as our qubit. A plain LC circuit makes a lovely oscillator but a useless qubit — you cannot single out two levels.

The Josephson junction: the essential nonlinearity

The fix is one component: a Josephson junction — two superconductors separated by an ultra-thin insulating barrier that superconducting current tunnels across. Electrically it acts as a nonlinear inductor, and replacing the ordinary inductor with it warps the energy ladder so the rungs are no longer equally spaced. The gap shrinks as you climb:

Now the two lowest levels |0\rangle and |1\rangle are separated by a frequency \omega_{01} that differs from the next step \omega_{12}. A pulse tuned to \omega_{01} drives only the transition we want and leaves |2\rangle alone. That difference is called the anharmonicity, and it is the whole reason a superconducting circuit can be a qubit at all. The dominant design, the transmon, is a junction shunted by a large capacitor, shaped deliberately to keep this anharmonicity while making the qubit almost immune to stray charge noise — the flaw that plagued earlier designs.

Picturing the anharmonic ladder

On the left, the harmonic LC circuit's rungs are a uniform staircase. On the right, the Josephson junction squeezes the upper rungs together — so the |0\rangle\!\to\!|1\rangle and |1\rangle\!\to\!|2\rangle transitions ring at different frequencies, and we can pick out just two levels to be our qubit.

Worked example: why the nonlinearity is non-negotiable

Suppose the anharmonicity is \alpha = \omega_{12} - \omega_{01} = -2\pi \times 300\ \text{MHz} (a typical transmon), while a fast microwave gate is only about 20\ \text{ns} long. A short pulse is spectrally broad, so its frequency spread \sim 1/(20\ \text{ns}) = 50\ \text{MHz} must be narrower than the 300\ \text{MHz} gap — otherwise the pulse's skirts also excite |1\rangle\!\to\!|2\rangle and population leaks out of the qubit. It fits comfortably here, so the pulse is a clean single-qubit gate. In a harmonic oscillator the gap would be 0\ \text{MHz}: any pulse leaks. That is the one-sentence case for the Josephson junction — no anharmonicity, no addressable qubit.

Operating and controlling the chip

These circuits only turn quantum when they are extremely cold. They live at the bottom of a dilution refrigerator at about 10–20 millikelvin — far colder than deep space — so that thermal energy cannot randomly kick the qubit up its ladder. From there, everything is done with microwaves:

• Single-qubit gates: shaped microwave pulses tuned to \omega_{01}, sent down a control line, rotate the qubit; a pulse's length and phase set the rotation angle and axis.
• Two-qubit gates: neighbouring qubits are joined through a shared resonator or tunable coupler that switches their interaction on to entangle them.
• Readout: each qubit is coupled to a dispersive readout resonator whose resonant frequency shifts by a tiny amount that depends on whether the qubit is |0\rangle or |1\rangle; bouncing a microwave tone off it and measuring the phase reads the qubit out without destroying its neighbours.

Worked example: how many gates before it forgets?

A qubit's quantum information survives only for its coherence time — call it T_2, typically tens to a few hundred microseconds for a good transmon. A single-qubit gate takes tens of nanoseconds. The rough budget of operations before decoherence scrambles the state is just the ratio:

A few thousand gates sounds like a lot, but useful algorithms with error correction want millions — which is exactly why longer coherence and error correction dominate the field. Fast gates (\sim 10\text{–}100\ \text{ns}) are the platform's great strength; short coherence is its matching weakness, and the two race each other.

The trade: fast and printable, but short-lived and finicky

Because the qubits are lithographically fabricated on a chip — the same toolkit that prints computer processors — they scale readily to hundreds of qubits, and their gates are among the fastest of any platform. The price is paid in coherence and overhead: coherence times sit at tens to hundreds of microseconds, the whole chip needs a big, power-hungry dilution fridge, connectivity is usually only nearest-neighbour, and because each qubit is a hand-built circuit rather than an identical atom, no two are quite the same — every qubit must be individually calibrated.

It is one concrete answer to the DiVincenzo criteria; rivals such as trapped-ion qubits and photonic qubits answer them with a completely different set of trade-offs.