Simple Algorithm
Quantum Computing · Excited State · 24 min

Photons, Spins & Topological Qubits

Light-based computing, silicon spins and topological protection.

Three very different bets on quantum hardware share this lesson. Photons barely interact with anything, which makes them perfect messengers and frustrating logic gates. Electron spins in silicon ride on the chip industry's manufacturing machine. Topological qubits would hide information from noise altogether, if the exotic quasiparticles they need can be shown to exist.

What you’ll learn

  • Describe polarization, dual-rail and time-bin photonic qubits, and compute the action of beam splitters and phase shifters.
  • Explain why photonic two-qubit gates are probabilistic, using Hong–Ou–Mandel interference and the KLM scheme.
  • Outline measurement- and fusion-based photonic computing with cluster states.
  • Derive the exchange interaction between quantum-dot spins, and name the main decoherence sources in silicon.
  • Explain how Majorana zero modes would encode a qubit non-locally, why braiding gives only Clifford gates, and what the experimental status is.

Qubits made of light

A photon can carry a qubit in several degrees of freedom:

  • Polarization: ∣H⟩\ket H and ∣V⟩\ket V. Easy to control with wave plates; natural for free-space links.
  • Path (dual-rail): one photon in two waveguides, ∣1,0⟩\ket{1,0} (upper) or ∣0,1⟩\ket{0,1} (lower). The workhorse of integrated photonic chips.
  • Time bin: early or late arrival, ∣e⟩\ket{e} or ∣ℓ⟩\ket{\ell}. Robust in optical fiber, where polarization drifts.

All three are dual-rail codes: one photon shared between two optical modes. Losing the photon leaves zero photons, a state outside the code space, so loss is detectable by counting photons.

Single-qubit gates are linear optics. A phase shifter on one mode applies diag⁡(1,eiφ)\diag(1, e^{i\varphi}), a ZZ-type rotation. A balanced (50:50) beam splitter mixes the two modes:

B=12(111−1)=HB = \frac{1}{\sqrt2}\begin{pmatrix}1 & 1\\ 1 & -1\end{pmatrix} = H
With a suitable phase convention a balanced beam splitter acts on a dual-rail qubit as a Hadamard. Beam splitters and phase shifters together implement any single-qubit unitary, for example as a Mach–Zehnder interferometer.
Check your understanding

A dual-rail photonic qubit loses its photon in a lossy waveguide. What do the detectors see, and why is that useful?

  1. A random qubit value: loss acts like a bit flip you cannot detect
  2. Two photons, because loss creates a photon in the other rail
  3. No photon at all, a state outside the code space, so the loss is heralded and can be treated as an erasure
  4. The same statistics as before: loss does not affect photonic qubits
Show the answer

C. No photon at all, a state outside the code space, so the loss is heralded and can be treated as an erasure

Loss maps ∣1,0⟩\ket{1,0} and ∣0,1⟩\ket{0,1} to ∣0,0⟩\ket{0,0}, which no valid codeword produces. An error at a known location (an erasure) is far easier to correct than an unknown flip.

Making photons interact without interaction

Photons pass through each other, and optical nonlinearities at the single-photon level are extremely weak in ordinary materials. A deterministic two-photon gate would need such a nonlinearity, so photonic computing uses a different resource: measurement. The key ingredient is two-photon interference.

DerivationHong–Ou–Mandel interference
  1. a†→12(c†+d†),b†→12(c†−d†)a^\dagger \to \tfrac{1}{\sqrt2}\big(c^\dagger + d^\dagger\big),\qquad b^\dagger \to \tfrac{1}{\sqrt2}\big(c^\dagger - d^\dagger\big)

    A balanced beam splitter maps the creation operators of the input modes a,ba, b to those of the output modes c,dc, d.

  2. a†b†∣0⟩→12(c†2−c†d†+d†c†−d†2)∣0⟩=12(c†2−d†2)∣0⟩a^\dagger b^\dagger\ket{0} \to \tfrac12\big(c^{\dagger2} - c^\dagger d^\dagger + d^\dagger c^\dagger - d^{\dagger2}\big)\ket0 = \tfrac12\big(c^{\dagger2} - d^{\dagger2}\big)\ket0

    Two identical photons, one in each input. Bosonic creation operators commute, so the cross terms cancel.

  3. ∣1,1⟩→12(∣2,0⟩−∣0,2⟩)\ket{1,1} \to \tfrac{1}{\sqrt2}\big(\ket{2,0} - \ket{0,2}\big)

    Using c†2∣0⟩=2∣2⟩c^{\dagger2}\ket0 = \sqrt2\ket{2}. The photons always leave together: the coincidence rate vanishes.

Think first What happens if the two photons are perfectly distinguishable, say orthogonally polarized, with polarization-blind detectors?

No interference: each photon transmits or reflects independently, and they leave through different ports half the time (14+14\tfrac14 + \tfrac14). The depth of the "HOM dip" in the coincidence rate therefore measures how indistinguishable two photons are, a key benchmark for photon sources.

HOM interference is a conditional effect: what one photon does depends on the other, through their indistinguishability. Knill, Laflamme and Milburn (2001) showed that auxiliary photons, photon-number-resolving detectors and feed-forward (adapting later optics to earlier detection results) suffice for universal, scalable quantum computing. The price is that gates succeed only with some probability, heralded by the detectors: KLM's basic controlled-ZZ succeeds with probability 1/161/16, and teleporting gates into the circuit boosts this toward 1 at the cost of more photons.

Computing by measuring

Probabilistic gates are awkward inside a circuit: one failure can ruin the computation. Measurement-based quantum computing (Raussendorf & Briegel, 2001) separates entangling from computing. First build a large entangled cluster state: put every qubit in ∣+⟩\ket + and apply CZ between neighbors on a graph. Then compute using only single-qubit measurements, choosing each basis according to earlier outcomes. Entanglement can be made offline, by trial and error, and only finished resources are consumed.

Photonic architectures grow cluster states from small entangled states by fusion, a two-photon Bell-type measurement that merges two clusters into one (Browne & Rudolph, 2005). Basic fusions succeed with probability 12\tfrac12, but failures are heralded, so the architecture can route around them. Fusion-based quantum computing (Bartolucci et al., 2023) makes this the organizing principle of a fault-tolerant design.

Spin qubits in silicon

A spin qubit stores information in the spin of a single electron (or hole) confined in a semiconductor quantum dot, a small puddle of charge defined by voltages on nanoscale gate electrodes, or bound to a donor atom such as phosphorus in silicon. In a magnetic field BB the spin states split by the Zeeman energy EZ=gμBBE_Z = g\mu_BB, about 28 GHz per tesla for g≈2g \approx 2, so fields of a fraction of a tesla give qubit frequencies of several to tens of GHz.

Single-qubit gates use electron spin resonance (an oscillating magnetic field) or, more often, electric-dipole spin resonance: an oscillating electric field shakes the electron in the field gradient of a micromagnet, or acts through spin–orbit coupling. Readout uses spin-to-charge conversion: a spin-dependent tunneling event moves a charge that a nearby sensor detects.

DerivationExchange from tunneling
  1. H=−t∑σ(c1σ†c2σ+h.c.)+U∑i=1,2ni↑ni↓H = -t\sum_\sigma\big(c_{1\sigma}^\dagger c_{2\sigma} + \text{h.c.}\big) + U\sum_{i=1,2}n_{i\uparrow}n_{i\downarrow}

    Two neighboring dots with one electron each, tunnel coupling tt and on-site Coulomb repulsion U≫tU \gg t: a two-site Hubbard model.

  2. ∣S(1,1)⟩↔∣S(0,2)⟩, ∣S(2,0)⟩  (matrix elements of size 2 t),∣T(1,1)⟩ blocked\ket{S(1,1)} \leftrightarrow \ket{S(0,2)},\ \ket{S(2,0)}\ \ \text{(matrix elements of size } \sqrt2\,t\text{)},\qquad \ket{T(1,1)}\ \text{blocked}

    Hopping into a doubly occupied dot requires opposite spins in the same orbital, a singlet. The Pauli principle blocks the triplets.

  3. ES≈−2t2U−2t2U=−4t2U,ET=0E_S \approx -\frac{2t^2}{U} - \frac{2t^2}{U} = -\frac{4t^2}{U},\qquad E_T = 0

    Second-order perturbation theory: virtual hops into each doubly occupied state lower the singlet.

  4. Heff=J(S1⋅S2−14),J=4t2UH_{\rm eff} = J\Big(\mathbf S_1\cdot\mathbf S_2 - \tfrac14\Big),\qquad J = \frac{4t^2}{U}

    The singlet–triplet splitting is an effective exchange interaction. Gate voltages tune tt by orders of magnitude within nanoseconds, switching JJ on and off.

Exchange makes two-qubit gates. Since S1⋅S2=12SWAP−14\mathbf S_1\cdot\mathbf S_2 = \tfrac12\mathrm{SWAP} - \tfrac14, an exchange pulse with ∫J dt=πℏ/2\int J\,dt = \pi\hbar/2 gives SWAP\sqrt{\mathrm{SWAP}}, which together with single-spin rotations is universal (Loss & DiVincenzo, 1998). With a large Zeeman-energy difference between the dots (from a micromagnet), exchange instead produces a controlled phase, and ∫J dt=πℏ\int J\,dt = \pi\hbar gives a CZ.

Predict before you look

Two dots have tunnel coupling t=5t = 5 μeV and charging energy U=2U = 2 meV. Estimate the exchange J/hJ/h in MHz, using 1 μeV≈h×2421\ \mu\text{eV} \approx h\times242 MHz.

Show the answer

12 MHz

J=4t2/U=4×25/2000 μeV=0.05 μeVJ = 4t^2/U = 4\times25/2000\ \mu\text{eV} = 0.05\ \mu\text{eV}, i.e. J/h≈12J/h \approx 12 MHz. A CZ in the large-Zeeman-difference regime then takes h/2J≈40h/2J \approx 40 ns. Because J∝t2J \propto t^2 and tt depends exponentially on the barrier-gate voltage, JJ can be switched over orders of magnitude.

Two-qubit fidelities above 99% were reported in silicon by several groups in 2022, and spin-qubit devices have been fabricated on industrial 300-mm lines. The remaining challenges are uniformity, wiring and the charge environment.

Topological qubits: hiding information in Majoranas

Every platform so far fights noise with error correction applied from outside. A topological qubit would build protection into the hardware by storing information non-locally, where no local disturbance can read or change it.

The proposed carriers are Majorana zero modes (MZMs): zero-energy excitations bound to the ends of a one-dimensional topological superconductor. A Majorana operator is its own adjoint, γ=γ†\gamma = \gamma^\dagger with γ2=1\gamma^2 = 1. Two MZMs γ1,γ2\gamma_1, \gamma_2, possibly far apart, together form one ordinary fermion mode f=(γ1+iγ2)/2f = (\gamma_1 + i\gamma_2)/2, which can be empty or occupied at no energy cost. Its occupation, the fermion parity −iγ1γ2=±1-i\gamma_1\gamma_2 = \pm1, is stored in the pair as a whole, not at either end.

Braiding (exchanging) two MZMs applies U=exp⁡(π4γiγj)U = \exp(\tfrac\pi4\gamma_i\gamma_j) to the qubit space. These operations depend only on the topology of the exchange, which is the source of their robustness. But they generate only Clifford gates: universality still needs a non-Clifford resource such as magic states, prepared outside the topological protection.

Check your understanding

Exchanging γ1\gamma_1 and γ2\gamma_2 applies exp⁡(π4γ1γ2)=exp⁡(iπ4ZL)\exp(\tfrac\pi4\gamma_1\gamma_2) = \exp(i\tfrac\pi4 Z_L) to the qubit, where ZL=−iγ1γ2Z_L = -i\gamma_1\gamma_2. Up to a global phase, which gate is that?

  1. TT, a π/8\pi/8 gate
  2. SS or S†S^\dagger, a Clifford phase gate
  3. HH, a Hadamard
  4. The identity
Show the answer

B. SS or S†S^\dagger, a Clifford phase gate

exp⁡(iπ4Z)=diag⁡(eiπ/4,e−iπ/4)=eiπ/4diag⁡(1,−i)=eiπ/4S†\exp(i\tfrac\pi4 Z) = \diag(e^{i\pi/4}, e^{-i\pi/4}) = e^{i\pi/4}\diag(1,-i) = e^{i\pi/4}S^\dagger: a quarter turn about ZZ, a Clifford gate (the orientation of the exchange decides between SS and S†S^\dagger). No braid of Majoranas produces TT.

Key takeaways

  • Photonic qubits are dual-rail codes: beam splitters and phase shifters give any single-qubit gate, and loss is the dominant but detectable error.
  • Photons do not interact, so entangling operations are measurement-induced and probabilistic (KLM; Bell measurements at most 50% efficient without ancillas), which favors measurement- and fusion-based architectures.
  • Spin qubits use electrons in quantum dots; exchange J≈4t2/UJ \approx 4t^2/U gives fast two-qubit gates, and coherence is limited by nuclear spins (cured by 28^{28}Si), charge noise and valley physics.
  • Majorana zero modes would store a qubit non-locally in fermion parity; braiding gives only Clifford gates, and an unambiguous demonstration of MZMs and braiding is still outstanding.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

A dual-rail photon passes a balanced beam splitter (acting as HH), a phase shift φ=π\varphi = \pi on one arm, and a second identical beam splitter. Where does it exit?

  1. Always at the first output (D0)
  2. Always at the second output (D1)
  3. 50/50 at random
  4. It depends on the photon's polarization
Show the answer

B. Always at the second output (D1)

HZH=XHZH = X: the phase shift π\pi is a ZZ gate between two Hadamards, which flips the dual-rail qubit. P(D1)=sin⁡2(π/2)=1P(D_1) = \sin^2(\pi/2) = 1.

Question 2

Why can't linear optics alone perform a deterministic Bell-state measurement on two photonic qubits?

  1. Photons are destroyed by detection
  2. Beam splitters are too lossy
  3. Bell states of photons do not exist
  4. With beam splitters, phase shifters and photon counting (no auxiliary photons), at most two of the four Bell states can be distinguished, so success is at most ½
Show the answer

D. With beam splitters, phase shifters and photon counting (no auxiliary photons), at most two of the four Bell states can be distinguished, so success is at most ½

Bosonic interference sorts Ψ−\Psi^- and Ψ+\Psi^+ into distinct click patterns, but Φ+\Phi^+ and Φ−\Phi^- give identical ones. Calsamiglia and Lütkenhaus (2001) proved no linear network does better without ancillas.

Question 3

In measurement-based quantum computing, where does the computation happen?

  1. In adaptive single-qubit measurements on a pre-built entangled cluster state
  2. In the CZ gates that build the cluster state
  3. In one final joint measurement of all qubits
  4. In the classical post-processing alone
Show the answer

A. In adaptive single-qubit measurements on a pre-built entangled cluster state

The cluster state is a fixed, algorithm-independent resource. The choice of single-qubit measurement bases, adapted to earlier outcomes, encodes the algorithm.

Question 4

Why does isotopic enrichment to 28^{28}Si improve spin-qubit coherence?

  1. It increases the electron's g-factor
  2. It enlarges the valley splitting
  3. It removes most spinful 29^{29}Si nuclei, whose fluctuating hyperfine field dephases the electron spin
  4. It eliminates charge noise
Show the answer

C. It removes most spinful 29^{29}Si nuclei, whose fluctuating hyperfine field dephases the electron spin

28^{28}Si has no nuclear spin. Removing 29^{29}Si shrinks the random Overhauser field and lengthens T2∗T_2^*, until charge noise becomes the limit.

Question 5

Why does a Majorana qubit need four zero modes rather than two?

  1. Fermion-parity superselection forbids superposing the even and odd states of one pair, so the qubit uses two pairs at fixed total parity
  2. Two Majoranas cannot be separated
  3. Each Majorana carries half a qubit, so four make two qubits
  4. Braiding needs at least four modes to do anything
Show the answer

A. Fermion-parity superselection forbids superposing the even and odd states of one pair, so the qubit uses two pairs at fixed total parity

One pair's two states differ in fermion parity and cannot be superposed. Two pairs at fixed total parity span a genuine two-dimensional qubit space.

Go deeper

Further reading

  • P. Kok, W. J. Munro, K. Nemoto, T. C. Ralph, J. P. Dowling & G. J. Milburn, "Linear optical quantum computing with photonic qubits", Reviews of Modern Physics 79, 135 (2007).
  • E. Knill, R. Laflamme & G. J. Milburn, "A scheme for efficient quantum computation with linear optics", Nature 409, 46 (2001).
  • G. Burkard, T. D. Ladd, A. Pan, J. M. Nichol & J. R. Petta, "Semiconductor spin qubits", Reviews of Modern Physics 95, 025003 (2023).
  • C. Nayak, S. H. Simon, A. Stern, M. Freedman & S. Das Sarma, "Non-Abelian anyons and topological quantum computation", Reviews of Modern Physics 80, 1083 (2008).