In short: A qubit is any controllable two-level quantum system whose state is a unit-length vector α|0⟩ + β|1⟩ in a 2-dimensional complex space, so unlike a transistor-based bit it holds continuous amplitudes (with |α|² + |β|² = 1) that collapse to a single 0 or 1 only when measured.
A two-level system, not a switch
A classical bit is an abstraction we impose on a transistor. A CMOS inverter in a 0.8 V process is "0" when its node sits near 0 V and "1" when it sits near VDD; anything in between is a failure that the next gate restores. The bit is a labelled voltage range carried by billions of electrons.
A qubit starts from the opposite end. You find a physical system with exactly two usable energy levels, call the lower one |0⟩ (ground) and the upper one |1⟩ (excited), and you control it so precisely that it can sit in a superposition of both. Candidates include the spin of an electron, the polarisation of a photon, two hyperfine levels of a trapped ion, and the two lowest levels of a superconducting transmon circuit that resonates near 5 GHz and must be held at roughly 15 millikelvin.
Key fact: "Two-level" means the Hilbert space is two-dimensional. Every other level of the device (a transmon has many) is deliberately kept out of play by design and by pulse shaping.
The state vector and the normalisation rule
We write the basis states as column vectors: |0⟩ = (1, 0)ᵀ and |1⟩ = (0, 1)ᵀ. A general pure state is
|ψ⟩ = α|0⟩ + β|1⟩ = (α, β)ᵀ, α, β ∈ ℂ
|α|² + |β|² = 1 (normalisation)
P(0) = |α|², P(1) = |β|² (Born rule)Amplitudes are complex, so each carries a magnitude and a phase. Probabilities come from squared magnitudes, which is why the vector must have unit length: the two outcomes are exhaustive, so their probabilities must sum to one.
Interviewers love a quick normalisation exercise. Take the unnormalised vector (1, 2i). Its squared norm is |1|² + |2i|² = 1 + 4 = 5, so the physical state is (1/√5, 2i/√5), giving P(0) = 20% and P(1) = 80%. Another favourite: is (3/5, 4i/5) valid? Yes, since 9/25 + 16/25 = 1. The equal superposition |+⟩ = (|0⟩ + |1⟩)/√2 gives 50/50, because (1/√2)² = ½. Note that |+⟩ is not "a bit that is secretly 0 or 1"; it is a definite state that is neither, and a rotation can map it back to |0⟩ deterministically.
Global phase versus relative phase
Multiply the whole state by e^(iγ) and every measurement probability is unchanged, so global phase is unobservable. That removes one real parameter. Counting the rest: two complex numbers is four reals, normalisation removes one, global phase removes another, leaving two: θ and φ, exactly the Bloch sphere angles above.
The relative phase is physical. (|0⟩ + |1⟩)/√2 and (|0⟩ − |1⟩)/√2 give identical Z-basis statistics, yet a Hadamard gate sends them to |0⟩ and |1⟩ respectively. That is interference, and it is where quantum speed-ups come from. A common trap answer is "a qubit stores 2 numbers so it stores more data." It does not: measurement returns one bit, and Holevo's theorem caps the retrievable classical information at one bit per qubit.
Key fact: n qubits need 2ⁿ complex amplitudes to describe. 50 qubits is 2⁵⁰ ≈ 10¹⁵ amplitudes, about 16 petabytes at 16 bytes each, which is why classical simulation hits a wall around 45 to 50 qubits.
Qubit versus transistor bit
| Property | CMOS bit | Qubit |
|---|---|---|
| Physical carrier | Thousands to millions of electrons on a node | One quantum object (spin, photon, ion, circuit mode) |
| State space | Two discrete values | Unit vector in ℂ², a sphere of states |
| Restoring logic | Gain restores 0.1 V noise to rail | None; errors accumulate continuously |
| Read-out | Non-destructive, repeatable | Destructive, probabilistic |
| Copying | Fanout is trivial | Forbidden by the no-cloning theorem |
| Lifetime | Static until power is cut | Coherence time T1/T2, often 10 to 500 µs for transmons |
For a deeper side-by-side, see our companion piece qubit vs bit.
Watching amplitudes move
A resonant microwave pulse on a transmon drives a Rabi oscillation: the state rotates about the Bloch sphere, and P(1) = sin²(Ωt/2) swings smoothly between 0 and 1. A "π pulse" (a few tens of nanoseconds on current hardware) is the quantum NOT, an X gate. The animation below shows P(0) and P(1) trading probability while their sum stays fixed at 1.
Why a chip designer should care
Qubits are not built from standard logic, but everything around them is. Control and read-out electronics are increasingly cryo-CMOS, designed and simulated with Cadence Virtuoso and Spectre or Synopsys tools, with foundry models extended to 4 K where mobility rises and threshold voltage shifts. Superconducting resonator and coupler layouts go through electromagnetic solvers, and timing between microwave pulses is verified like any mixed-signal interface. If you want to see how design services meet this problem, Avecas VLSI design services work on exactly this mixed-signal boundary. Next in this series: superposition and measurement.
Key fact: A qubit is defined by its vector, not its hardware. Two qubits on different platforms with the same state vector are, to any algorithm, the same qubit.
Ready to go beyond definitions? Work through gates, circuits and interview problems in the CourseTron quantum track: browse all courses. For classroom-style VLSI training, see ChipXpert training programmes.
FAQ
Can a qubit be 0 and 1 at the same time?
Loosely, but precisely: it is in a state that is a weighted sum of the two basis states. Any measurement still returns a single 0 or 1, with probabilities |α|² and |β|².
Why must |α|² + |β|² equal 1?
Because the two measurement outcomes are the only possibilities, their probabilities must total one. Unitary gates preserve this length, so the constraint holds throughout a circuit.
Why does the Bloch sphere use θ/2?
Orthogonal states |0⟩ and |1⟩ are antipodal on the sphere (θ = π) but orthogonal in Hilbert space, so the amplitudes use the half angle.
Does a qubit hold more information than a bit?
Not retrievable information: Holevo's bound limits it to one classical bit per qubit. The advantage appears in how n entangled qubits evolve in a 2ⁿ-dimensional space.