Quantum Error Correction and Quantum Hardware
Every physical qubit forgets within microseconds or seconds, and the algorithms of the previous unit need billions of operations. This unit shows how a logical qubit spread over many physical ones can be protected by measuring only what the errors did, from the three-qubit code to Shor's code and the surface code, and what the threshold theorem promises. It then compares the superconducting, trapped-ion, neutral-atom, silicon, and photonic machines on which this is being tried, including the 2024 result that ran a surface code below threshold.
Every qubit in every machine built so far forgets. A superconducting circuit loses its excitation to the metal around it within a tenth of a millisecond, an ion in a trap picks up a random phase from a wandering magnetic field, and an atom in an optical tweezer is kicked by a stray photon. The algorithms of Unit 15 need billions of gates, and no physical qubit survives that many. This unit explains how quantum information can nevertheless be protected, by spreading one logical qubit over many physical ones and measuring only what the errors did, never what the state is. It then describes the machines on which this is being tried today, and ends with the question of what a student who has come this far can do next.
1 Decoherence and the errors of a qubit
A qubit is never alone. Its two levels are coupled, however weakly, to the electromagnetic field, to the vibrations of the chip or trap that holds it, and to the control lines that drive it, and these couplings slowly turn a definite quantum state into a random mixture. This process is called decoherence. For a single qubit, two numbers measured in every laboratory summarize it.
The first is the energy relaxation time \(T_1\). A qubit prepared in the excited state \(\ket{1}\) and left alone is found in \(\ket{1}\) at a later time \(t\) with probability \begin{equation} P_1(t) = e^{-t/T_1} , \label{eq:t1} \end{equation} because it hands its energy to the environment at a constant rate \(1/T_1\), in the same way as an excited atom emits a photon. The second is the dephasing time \(T_2\), which measures how long a superposition keeps its phase. Prepare the state \((\ket{0} + \ket{1})/\sqrt2\), which Unit 13 called the state \(\ket{+}\) and which has \(\ev{\mathrm{X}} = 1\). This expectation value is the interference term of the superposition, the part of the state that depends on the relative phase of \(\ket{0}\) and \(\ket{1}\); it equals \(1\) when the phase is the one prepared and \(0\) when the phase is completely random. The environment adds a small random phase to \(\ket{1}\) relative to \(\ket{0}\) in every run, so the runs no longer agree with one another, and the average of the interference term decays, \begin{equation} \ev{\mathrm{X}}(t) = e^{-t/T_2} . \label{eq:t2} \end{equation} Energy relaxation also destroys the phase, because a qubit that has fallen to \(\ket{0}\) has no superposition left, so \(T_2\) can never exceed \(2T_1\) (the factor of two appears because the interference term contains the amplitude of \(\ket{1}\), which decays at half the rate of its probability); when \(T_2\) is much shorter than \(2T_1\), something besides energy loss randomizes the phase, usually a fluctuating field that shifts the qubit’s frequency.
For error correction the continuous decay has to be translated into discrete events. The simplest model of pure dephasing says that in a short time interval the operator \(\mathrm{Z}\) is applied to the qubit with a small probability \(p\) and nothing happens with probability \(1 - p\). A \(\mathrm{Z}\) turns \(\alpha\ket{0} + \beta\ket{1}\) into \(\alpha\ket{0} - \beta\ket{1}\) and reverses the sign of \(\ev{\mathrm{X}}\), so averaged over many runs the interference term shrinks to \((1 - p) - p = 1 - 2p\) of its value. Setting \(1 - 2p\) equal to the decay factor \(e^{-t/T_2}\) of Equation \eqref{eq:t2} and solving for \(p\) gives the probability of a phase flip in the time \(t\), \begin{equation} p_{\mathrm{Z}} = \tfrac12\left(1 - e^{-t/T_2}\right) \approx \frac{t}{2T_2} \quad (t \ll T_2) . \label{eq:pz} \end{equation} In the same rough way, energy relaxation in a time \(t \ll T_1\) acts on a qubit in \(\ket{1}\) like a bit flip \(\mathrm{X}\) with probability \(t/T_1\), although the true process is not symmetric between the two levels, since energy loss takes \(\ket{1}\) to \(\ket{0}\) but never \(\ket{0}\) to \(\ket{1}\). The following example shows the sizes involved for a real qubit.