Simple Algorithm
Quantum Computing · Rydberg State · 28 min

Spin Qubits in Semiconductors

Electrons in quantum dots, and riding the chip industry.

An electron trapped in a puddle of charge a few tens of nanometers across, inside a transistor-like device, has a spin that can hold a qubit for milliseconds. Spin qubits promise to ride on the semiconductor industry's ability to make billions of nearly identical devices, if the remaining differences between them can be tamed.

What you’ll learn

  • State the Loss–DiVincenzo proposal and prove that exchange pulses plus single-spin rotations are universal.
  • Derive the exchange coupling from a Hubbard model, its detuning dependence, and the symmetric operating point.
  • Derive the CZ condition in the regime of large Zeeman-energy difference.
  • Analyze dephasing from quasi-static noise, including the effect of 28^{28}Si enrichment, and explain valley splitting.
  • Derive the synchronization condition that cancels crosstalk between frequency-crowded qubits.

The Loss–DiVincenzo blueprint

Loss and DiVincenzo (1998) proposed one electron per quantum dot, with its spin as the qubit, controlled by the Hamiltonian

H(t)=∑igμB Bi(t)⋅Si+∑⟨ij⟩Jij(t) Si⋅Sj.H(t) = \sum_i g\mu_B\,\mathbf B_i(t)\cdot\mathbf S_i + \sum_{\langle ij\rangle}J_{ij}(t)\,\mathbf S_i\cdot\mathbf S_j.

Local fields (or their electrical equivalents) rotate single spins, and gate voltages pulse the exchange JijJ_{ij} between neighbors. Kane (1998) proposed a donor-based variant using the nuclear spins of phosphorus atoms in silicon.

DerivationProof
  1. S1⋅S2=14 σ1⋅σ2=12 SWAP−14 I\mathbf S_1\cdot\mathbf S_2 = \tfrac14\,\boldsymbol\sigma_1\cdot\boldsymbol\sigma_2 = \tfrac12\,\mathrm{SWAP} - \tfrac14\,\id

    From σ1⋅σ2=2 SWAP−I\boldsymbol\sigma_1\cdot\boldsymbol\sigma_2 = 2\,\mathrm{SWAP} - \id: triplets have SWAP eigenvalue +1+1 and S1⋅S2=14\mathbf S_1\cdot\mathbf S_2 = \tfrac14; the singlet has −1-1 and −34-\tfrac34.

  2. UJ(θ)=eiθ/4exp⁡ ⁣(−iθ2 SWAP)=eiθ/4(cos⁡θ2 I−isin⁡θ2 SWAP)U_J(\theta) = e^{i\theta/4}\exp\!\big(-i\tfrac\theta2\,\mathrm{SWAP}\big) = e^{i\theta/4}\big(\cos\tfrac\theta2\,\id - i\sin\tfrac\theta2\,\mathrm{SWAP}\big)

    Because SWAP2=I\mathrm{SWAP}^2 = \id. At θ=π\theta = \pi this is −i SWAP-i\,\mathrm{SWAP} up to a phase; at θ=π/2\theta = \pi/2 it squares to SWAP, so it is SWAP\sqrt{\mathrm{SWAP}}.

  3. Multiplying out the five factors in the basis above gives diag⁡(1,−1,−1,−1)\diag(1,-1,-1,-1) up to a global phase. Its phases satisfy φ11−φ01−φ10≡π\varphi_{11}-\varphi_{01}-\varphi_{10} \equiv \pi, the criterion for being a CZ dressed with ZZ rotations, and CZ plus single-qubit gates is universal. ■\blacksquare

Exchange from tunneling, and the CZ

DerivationJ(ε) in a double dot
  1. HS=(0−2 t−2 t−2 tU−ε0−2 t0U+ε)on{∣S(1,1)⟩, ∣S(0,2)⟩, ∣S(2,0)⟩}H_S = \begin{pmatrix}0 & -\sqrt2\,t & -\sqrt2\,t\\ -\sqrt2\,t & U-\varepsilon & 0\\ -\sqrt2\,t & 0 & U+\varepsilon\end{pmatrix}\quad\text{on}\quad\big\{\ket{S(1,1)},\ \ket{S(0,2)},\ \ket{S(2,0)}\big\}

    The singlet sector of a two-site Hubbard model with tunneling tt, charging energy UU and interdot detuning ε\varepsilon, which tilts the double dot toward (0,2)(0,2). The (1,1)(1,1) triplets do not couple: two parallel spins cannot share an orbital.

  2. J(ε)=ET−ES≈2t2U−ε+2t2U+ε=4t2UU2−ε2J(\varepsilon) = E_T - E_S \approx \frac{2t^2}{U-\varepsilon} + \frac{2t^2}{U+\varepsilon} = \frac{4t^2U}{U^2-\varepsilon^2}

    Second-order perturbation theory for t≪U±εt \ll U\pm\varepsilon. At ε=0\varepsilon = 0 this is J=4t2/UJ = 4t^2/U.

  3. ∂J∂ε=8t2Uε(U2−ε2)2 → ε=0  0\frac{\partial J}{\partial\varepsilon} = \frac{8t^2U\varepsilon}{(U^2-\varepsilon^2)^2}\ \xrightarrow{\ \varepsilon = 0\ }\ 0

    At the symmetric operating point ε=0\varepsilon = 0, detuning noise affects JJ only at second order. Experiments therefore switch JJ with the barrier gate, which tunes tt, rather than with detuning (Martins et al., 2016; Reed et al., 2016).

DerivationFrom exchange to CZ
  1. H=ΔEZ2(S1z−S2z)+J(S1⋅S2−14),ΔEZ≫JH = \frac{\Delta E_Z}{2}\big(S^z_1 - S^z_2\big) + J\Big(\mathbf S_1\cdot\mathbf S_2 - \tfrac14\Big),\qquad \Delta E_Z \gg J

    Two spins with different Zeeman energies (from a micromagnet gradient or different gg-factors), in the frame rotating at their mean frequency.

  2. E↑↑=E↓↓=0,E↑↓, ↓↑=−J2±12ΔEZ2+J2≈±ΔEZ2−J2E_{\uparrow\uparrow} = E_{\downarrow\downarrow} = 0,\qquad E_{\uparrow\downarrow,\,\downarrow\uparrow} = -\frac J2 \pm \frac12\sqrt{\Delta E_Z^2 + J^2} \approx \pm\frac{\Delta E_Z}{2} - \frac J2

    The flip-flop term mixes ∣↑↓⟩\ket{\uparrow\downarrow} and ∣↓↑⟩\ket{\downarrow\uparrow} only weakly, at order J/ΔEZJ/\Delta E_Z; to first order, exchange simply lowers both antiparallel states by J/2J/2.

  3. φ00+φ11−φ01−φ10=−1ℏ∫J dt⟹CZ when ∫J dt=πℏ\varphi_{00} + \varphi_{11} - \varphi_{01} - \varphi_{10} = -\frac{1}{\hbar}\int J\,dt\quad\Longrightarrow\quad\text{CZ when}\ \int J\,dt = \pi\hbar

    The conditional phase is the local-equivalence invariant (with ∣0⟩=∣↑⟩\ket0 = \ket\uparrow), and single-qubit ZZ corrections absorb the rest. This adiabatic CZ is the standard two-qubit gate in micromagnet-equipped silicon devices.

Predict before you look

In the large-ΔEZ\Delta E_Z regime the exchange is pulsed to J/h=5J/h = 5 MHz. How long must it stay on for a CZ?

Show the answer

100 ns

∫J dt=πℏ\int J\,dt = \pi\hbar gives t=πℏ/J=h/2J=1/(2×5 MHz)=100t = \pi\hbar/J = h/2J = 1/(2\times5\ \text{MHz}) = 100 ns. Faster gates need larger JJ, but JJ must stay well below ΔEZ\Delta E_Z or the flip-flop term causes errors; pulse shaping relaxes this.

Decoherence: nuclei, charges and valleys

DerivationQuasi-static noise gives Gaussian decay
  1. δω∼N(0,σ2),constant during each shot\delta\omega\sim\mathcal N(0,\sigma^2),\quad\text{constant during each shot}

    Slow noise (nuclear-spin fluctuations, low-frequency charge noise) shifts the qubit frequency by a random but nearly constant amount in each repetition of a Ramsey experiment.

  2. ⟨ei δω t⟩=e−σ2t2/2≡e−(t/T2∗)2,T2∗=2σ\big\langle e^{i\,\delta\omega\,t}\big\rangle = e^{-\sigma^2t^2/2} \equiv e^{-(t/T_2^*)^2},\qquad T_2^* = \frac{\sqrt2}{\sigma}

    Averaging the accumulated phase over shots gives Gaussian decay of the coherence.

  3. σ2=∑kAk2 ⟨Ik,z2⟩∝f Ndot(ANdot)2=fA2Ndot\sigma^2 = \sum_k A_k^2\,\langle I_{k,z}^2\rangle \propto f\,N_{\rm dot}\Big(\frac{A}{N_{\rm dot}}\Big)^2 = \frac{fA^2}{N_{\rm dot}}

    For hyperfine noise the Overhauser field sums over the fNdotfN_{\rm dot} spinful nuclei the electron overlaps (spinful fraction ff), each coupled with strength Ak∼A/NdotA_k \sim A/N_{\rm dot}. At fixed dot size, the hyperfine-limited T2∗T_2^* therefore scales as f−1/2f^{-1/2}.

  4. A spin echo refocuses the quasi-static part, so T2echo≫T2∗T_2^{\rm echo} \gg T_2^*. What remains is set by the noise spectrum at higher frequencies, which dynamical-decoupling sequences both probe and filter.

Check your understanding

Holding the dot fixed, you reduce the 29^{29}Si fraction from natural abundance (4.7%) to 800 ppm. If hyperfine noise limits T2∗T_2^*, by about what factor does it grow?

  1. 1.7×
  2. 7.7×
  3. 59×
  4. 3400×
Show the answer

B. 7.7×

T2∗∝f−1/2T_2^*\propto f^{-1/2}, and ff falls by 4.7%/0.08%≈594.7\%/0.08\%\approx59, so T2∗T_2^* grows by 59≈7.7\sqrt{59}\approx7.7. Comparisons between real devices also reflect dot size and other differences, and once hyperfine noise is suppressed, charge noise typically takes over.

Charge noise is the other universal enemy. Two-level fluctuators in oxides and interfaces produce a roughly 1/f1/f spectrum of potential fluctuations, which reaches the spin through exchange (∂J/∂ε\partial J/\partial\varepsilon), through the micromagnet gradient (an electron pushed by an electric field sees a different magnetic field) and through spin–orbit coupling. Designs balance speed against sensitivity: a steeper gradient gives faster electric-dipole spin resonance but also couples more charge noise into dephasing.

Noise sourceTypical effectMitigation
Nuclear spins (hyperfine)T2∗∼10T_2^*\sim10 ns in GaAs, ∼1\sim1 μs in natural Si, tens of μs or more in enriched 28^{28}SiIsotopic enrichment; echoes; nuclear-spin-free host materials
Charge noise (roughly 1/f1/f)Dephasing through JJ, magnetic gradients and spin–orbit coupling; slow driftsSymmetric operation; dynamical decoupling; cleaner oxides and interfaces
Small valley splittingLeakage, relaxation hot spots, readout errorsHeterostructure and interface engineering; device selection
Device variabilityDifferent gate voltages and frequencies for each qubitAutomated tuning; more uniform fabrication

Control, crosstalk and scale

Qubits driven through shared lines are told apart by frequency: each spin has its own resonance, set by the micromagnet gradient or by gg-factor differences. A pulse resonant with qubit A (Rabi frequency Ω\Omega) also drives a neighbor B detuned by Δ\Delta, which reaches an excitation of at most Ω2/(Ω2+Δ2)\Omega^2/(\Omega^2+\Delta^2). Packing more qubits into a limited frequency band makes this worse, unless the pulse is synchronized.

At scale, several engineering threads converge. Silicon spin qubits have been operated above 1 K (Petit et al., 2020; Yang et al., 2020), where cryogenic control electronics have far more cooling power available; devices have been fabricated on industrial 300-mm lines (Zwerver et al., 2022); two-qubit gate fidelities above 99% were reported by several groups in 2022; and electrons can be shuttled between dots or coupled at a distance through superconducting microwave resonators.

Key takeaways

  • Loss–DiVincenzo: spins in dots plus pulsed exchange J S1⋅S2J\,\mathbf S_1\cdot\mathbf S_2 give SWAP, SWAP\sqrt{\mathrm{SWAP}} and, with ZZ rotations, CZ, a universal set.
  • J(ε)=4t2U/(U2−ε2)J(\varepsilon) = 4t^2U/(U^2-\varepsilon^2): symmetric operation at ε=0\varepsilon = 0 removes first-order detuning-noise sensitivity, and with ΔEZ≫J\Delta E_Z\gg J a pulse with ∫J dt=πℏ\int J\,dt = \pi\hbar is a CZ.
  • Quasi-static noise gives T2∗=2/σT_2^* = \sqrt2/\sigma; hyperfine-limited T2∗∝f−1/2T_2^*\propto f^{-1/2}, so 28^{28}Si enrichment helps until charge noise dominates.
  • Valley splitting, variability and wiring are the scale-up problems; frequency crowding is managed by synchronization, Δ=Ω4k2−1\Delta = \Omega\sqrt{4k^2-1}.

Check yourself

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

Question 1

Why are exchange gates in double dots often operated at zero detuning, with JJ switched by the barrier gate?

  1. Exchange vanishes away from zero detuning
  2. Zero detuning maximizes the valley splitting
  3. At ε=0\varepsilon = 0, ∂J/∂ε=0\partial J/\partial\varepsilon = 0, so detuning noise affects JJ only at second order
  4. Barrier gates are faster than plunger gates
Show the answer

C. At ε=0\varepsilon = 0, ∂J/∂ε=0\partial J/\partial\varepsilon = 0, so detuning noise affects JJ only at second order

J=4t2U/(U2−ε2)J = 4t^2U/(U^2-\varepsilon^2) is stationary at ε=0\varepsilon = 0, a sweet spot against charge noise in the detuning. Tuning the barrier changes tt instead.

Question 2

In the large-ΔEZ\Delta E_Z regime, the conditional phase produced by an exchange pulse is

  1. ∫J dt/ℏ\int J\,dt/\hbar
  2. ∫J dt/2ℏ\int J\,dt/2\hbar
  3. 2∫J dt/ℏ2\int J\,dt/\hbar
  4. independent of JJ
Show the answer

A. ∫J dt/ℏ\int J\,dt/\hbar

Each antiparallel state is lowered by J/2J/2, so the invariant φ00+φ11−φ01−φ10\varphi_{00}+\varphi_{11}-\varphi_{01}-\varphi_{10} has magnitude ∫J dt/ℏ\int J\,dt/\hbar; a CZ needs ∫J dt=πℏ\int J\,dt = \pi\hbar.

Question 3

A square π-pulse on qubit A (Rabi frequency Ω\Omega) must leave neighbor B untouched. Which detuning of B achieves this exactly?

  1. Δ=Ω\Delta = \Omega
  2. Δ=3 Ω\Delta = \sqrt3\,\Omega
  3. Δ=2 Ω\Delta = 2\,\Omega
  4. Only Δ→∞\Delta\to\infty
Show the answer

B. Δ=3 Ω\Delta = \sqrt3\,\Omega

B's generalized rotation is Ω2+Δ2 π/Ω\sqrt{\Omega^2+\Delta^2}\,\pi/\Omega, which equals 2π2\pi when Δ=3 Ω\Delta = \sqrt3\,\Omega.

Question 4

What goes wrong if the valley splitting of a silicon dot is comparable to the Zeeman energy?

  1. Nothing: valleys do not couple to spin
  2. Exchange becomes impossible
  3. The electron turns into a hole
  4. The excited valley state provides leakage and fast spin-relaxation channels, and can spoil spin-blockade readout
Show the answer

D. The excited valley state provides leakage and fast spin-relaxation channels, and can spoil spin-blockade readout

Spin–orbit and interface effects mix spin and valley states, strongly near EVS≈EZE_{\rm VS}\approx E_Z, and a low-lying valley state breaks the two-level assumptions behind control and readout.

Further reading

  • D. Loss & D. P. DiVincenzo, "Quantum computation with quantum dots", Physical Review A 57, 120 (1998).
  • B. E. Kane, "A silicon-based nuclear spin quantum computer", Nature 393, 133 (1998).
  • R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha & L. M. K. Vandersypen, "Spins in few-electron quantum dots", Reviews of Modern Physics 79, 1217 (2007).
  • F. A. Zwanenburg et al., "Silicon quantum electronics", Reviews of Modern Physics 85, 961 (2013).
  • G. Burkard, T. D. Ladd, A. Pan, J. M. Nichol & J. R. Petta, "Semiconductor spin qubits", Reviews of Modern Physics 95, 025003 (2023).