In short: Superconducting qubits are driven to arbitrary states with shaped microwave pulses at their transition frequency, and read out non-destructively by probing a coupled resonator whose frequency shifts depending on the qubit state — all orchestrated by room-temperature AWGs and increasingly by cryogenic CMOS electronics sitting inside the dilution refrigerator.
If a qubit is the quantum equivalent of a transistor, then control and readout electronics are its clock tree, its I/O pads, and its sense amplifiers rolled into one. Getting them right is the difference between a 99.9% gate and a decohered mess. This article walks through how engineers actually drive and measure a superconducting qubit, with the real numbers and tooling you need to reason about it.
The qubit as a weakly anharmonic oscillator
A transmon qubit is an LC resonator in which a Josephson junction supplies a nonlinear inductance. That nonlinearity is the whole game. A perfect harmonic oscillator has equally spaced energy levels, so any drive that excites |0>→|1> equally excites |1>→|2>, and you can never isolate a two-level system. The junction breaks that symmetry: the 0→1 transition typically sits near 5 GHz while the 1→2 transition is detuned by the anharmonicity α, usually around −200 to −300 MHz. That gap is what lets you address a single transition.
Typical transmon: f₀₁ ≈ 4–6 GHz, anharmonicity α ≈ −250 MHz, T₁/T₂ ≈ 50–200 µs. A single-qubit gate lasts 20–40 ns, so you get thousands of operations inside a coherence window.
Driving the qubit: shaped microwave pulses
To rotate a qubit you send a microwave burst at f₀₁ through a drive line capacitively coupled to it. The pulse envelope — not just the carrier — determines the rotation. A resonant pulse of area π produces an X gate (bit flip); area π/2 produces a superposition. The I/Q (in-phase / quadrature) components of the envelope set the rotation axis in the equatorial plane, which is how you get arbitrary single-qubit gates.
The catch is that short, hard pulses have wide spectral content and will leak population into the |2> state through the nearby 1→2 transition. The standard fix is DRAG (Derivative Removal by Adiabatic Gate): you add a quadrature component proportional to the time-derivative of the main Gaussian envelope, which cancels the off-resonant leakage. DRAG is why a 20 ns gate can still hit 99.9% fidelity instead of bleeding into leakage error.
Dispersive readout: measuring without destroying
You cannot look at a qubit directly without collapsing it uncontrollably. Instead you couple it to a readout resonator and operate in the dispersive regime, where the qubit–resonator detuning Δ is much larger than their coupling g. In that limit the resonator picks up a state-dependent frequency shift of ±χ, where the dispersive shift χ ≈ g²/Δ. Probe the resonator with a tone near its bare frequency and the reflected/transmitted signal acquires a different phase and amplitude for |0> versus |1>.
Concretely: a resonator at 7 GHz with χ/2π ≈ 1 MHz will show two cleanly separated blobs in the I/Q plane. A Josephson Parametric Amplifier (JPA) or Travelling-Wave Parametric Amplifier (TWPA) at the first amplification stage adds near-quantum-limited gain so you can resolve those blobs in a few hundred nanoseconds with high single-shot fidelity (often >99%). This is covered in more depth in our companion piece on quantum measurement.
Dispersive shift rule of thumb: χ = g²/Δ · (α/(α+Δ)). Push χ too high and you speed up readout but open a Purcell decay channel that kills T₁ — a classic engineering trade-off solved with a dedicated Purcell filter.
Watch the readout happen
The electronics stack, from 300 K to 10 mK
Control signals originate in arbitrary waveform generators sampling at 1–2.5 GSa/s, I/Q-mixed up to the carrier. As they descend the dilution refrigerator they pass through ~60 dB of distributed attenuation to strip room-temperature thermal noise, with heavy attenuation at the 4 K and still-plate stages. Readout signals come back up through isolators and the parametric/HEMT amplifier chain.
| Stage | Temp | Role |
|---|---|---|
| Room temp | 300 K | AWG, I/Q upconversion, digitizer, FPGA sequencing |
| 4 K plate | 4 K | HEMT low-noise amplifier, bulk attenuation |
| Mixing chamber | ~10 mK | Qubit chip, resonators, JPA/TWPA, circulators |
The scaling wall is wiring: one qubit needs a drive line, a flux line, and a shared readout line. At a few hundred coaxial cables a fridge is full. This is why cryo-CMOS controllers — control ASICs that run at 4 K inches from the qubits — are a hot research area at Intel, Google and academic groups. They multiplex control, cut cable count, and shorten latency. Designing them leans on the same RF/mixed-signal flows (Cadence Virtuoso/Spectre, Synopsys Custom Compiler, Siemens Calibre for signoff) used for any cryogenic analog IC, with the twist that device models must be re-characterised at 4 K.
Calibration is never done
Qubit frequencies drift, so control stacks run continuous calibration: Rabi sweeps to set pulse amplitude, Ramsey experiments to track frequency, and randomized benchmarking to report an averaged gate error that is immune to state-prep-and-measurement bias. A modern lab treats this as a software pipeline — tools like Qiskit Pulse or Qblox/Quantum Machines sequencers close the loop automatically. If you want the hands-on version, our superconducting qubit fabrication walkthrough shows where these parameters come from physically. Teams building the measurement ASICs behind this often partner with silicon design houses like Avecas for custom mixed-signal design services.
Want to go from "I've heard of a transmon" to actually pulsing one? CourseTron's Quantum Hardware track builds the control-and-readout stack step by step with real lab data. Browse the course catalog →
FAQ
Why microwave frequencies and not optical?
The transmon's energy-level spacing (ħω) corresponds to ~5 GHz, which lands squarely in the microwave band. That also matches commodity RF electronics and keeps thermal photon occupation negligible at 10 mK.
What is single-shot readout?
Resolving the qubit state in one measurement rather than averaging over many repetitions. It requires enough signal-to-noise — typically a parametric amplifier — so the |0> and |1> I/Q distributions do not overlap.
Why not just measure the qubit directly?
A direct projective coupling strong enough to read out quickly would also be a strong decay channel, destroying coherence. Dispersive readout keeps the qubit far-detuned from the resonator so you measure a frequency shift instead of exchanging energy.
What limits how many qubits one fridge can control?
Coaxial cable count and the heat each cable carries down. Cryo-CMOS and frequency-multiplexed readout (many qubits sharing one line via distinct resonator frequencies) are the main paths past this wall.