Loss–DiVincenzo Proposal

The Loss–DiVincenzo proposal is the blueprint that turned a chip of semiconductor into a quantum computer. Published by Daniel Loss and David DiVincenzo in 1998, it makes one economical bet: store each qubit in the spin of a single electron trapped in a gate-defined quantum dot — an artificial atom held by metal-gate voltages on a two-dimensional electron gas — and let qubits interact through a coupling the material supplies for free. The basis is the two Zeeman-split spin states, ⟨0⟩ = ⟨↑⟩ and ⟨1⟩ = ⟨↓⟩. Spin is the deliberate choice: the orbital (charge) degree of freedom couples violently to electric fields and phonons and decoheres in picoseconds, whereas an isolated spin holds phase for milliseconds. The intellectual core is a piece of quantum-mechanical jiu-jitsu. Bring two dots close and lower the tunnel barrier with a gate voltage; the wavefunctions overlap, and the Pauli exclusion principle chains the spatial symmetry of the two-electron state to its spin symmetry. The low-energy Hamiltonian is then purely a spin coupling, the Heisenberg exchange $H = J(t)\,\mathbf{S}_1\cdot\mathbf{S}_2$ , whose strength $J$ is the singlet–triplet splitting, switched on and off in real time by the barrier gate. A charge interaction has become a spin interaction, controllable with nothing but voltage pulses. Loss and DiVincenzo then proved this one knob is enough: a calibrated exchange pulse generates SWAP, a half-pulse the entangling $\sqrt{\mathrm{SWAP}}$ , and two of those with single-spin rotations build a CNOT. With a static field for single-qubit gates, spin relaxation for initialization, and spin-to-charge conversion for readout, the scheme satisfies all five DiVincenzo criteria on one platform — and launched the field of semiconductor spin qubits, which today reaches fidelities above 99% in isotopically purified silicon.

Key takeaways

The Loss–DiVincenzo proposal (1998) encodes each qubit in the two spin states ⟨↑⟩, ⟨↓⟩ of one electron confined in a gate-defined semiconductor quantum dot, split by a static Zeeman field.
Two-qubit gates come free from the Heisenberg exchange $J(t)\,\mathbf{S}_1\cdot\mathbf{S}_2$ that appears when adjacent dots' orbitals overlap; the barrier gate switches $J$ electrically.
Exchange is the singlet–triplet splitting $J=E_T-E_S$ , estimated by the two-site Hubbard model as $J\approx 4t^2/U$ — a virtual, Pauli-selected double occupancy.
A pulse of accumulated phase $\phi=\int J\,\mathrm dt/\hbar=\pi/2$ realizes $\sqrt{\mathrm{SWAP}}$ ; with single-spin rotations this is universal (two $\sqrt{\mathrm{SWAP}}$ make a CNOT).
Spin is the quiet store, charge the loud messenger: information idles in spin coherence but borrows charge only during the exchange pulse — long memory, fast gates.
Its weaknesses — leakage to doubly-occupied states, charge noise on $J$ , hyperfine dephasing — are fought with symmetric (barrier) control and $^{28}\mathrm{Si}$ purification.
Key idea: the Pauli principle turns the overlap of two electron orbitals (a charge effect) into a Heisenberg spin coupling $J(t)\,\mathbf{S}_1\cdot\mathbf{S}_2$ , whose time integral drives a universal two-qubit gate — all by purely electrical gate-voltage pulses.

The core insight

The Pauli principle converts a charge interaction — the overlap of two electron orbitals — into a spin interaction: lowering the inter-dot tunnel barrier turns on a Heisenberg exchange $J(t)\,\mathbf{S}_1\cdot\mathbf{S}_2$ , whose time integral controls a universal two-qubit gate, all by purely electrical gate-voltage pulses.

Where this lives today

The proposal is no longer a proposal — it is a working platform. Its most direct embodiment is the silicon spin qubit, in Si/SiGe and Si-MOS devices, where electrons sit in gate-defined dots inside isotopically purified $^{28}\mathrm{Si}$ . Removing the $^{29}\mathrm{Si}$ nuclear spins strips away the dominant hyperfine dephasing, pushing $T_2^*$ into the tens of microseconds and echoed coherence toward milliseconds; single-qubit fidelities exceed 99.9% and exchange-based two-qubit gates exceed 99%, above the surface-code threshold. The symmetric (barrier) control point derived above is now standard, and CMOS foundry compatibility makes silicon the leading bet for dense, manufacturable arrays. The same $J(t)\,\mathbf{S}_1\cdot\mathbf{S}_2$ interaction underlies the singlet–triplet qubit (two spins, magnetic gradient plus exchange) and the all-electrical exchange-only qubit (three spins, exchange pulses alone), both trading away fast local magnetic fields. Readout is spin-to-charge conversion — Elzerman tunnelling or Pauli-spin-blockade discrimination — sensed by a quantum point contact, single-electron transistor, or dispersive gate reflectometry. The open frontiers are the ones the proposal always implied: scaling the wiring, extending coherence against residual charge noise, and coupling distant spins via microwave resonators or shuttling.

The physical picture

Picture two artificial hydrogen atoms side by side, each holding one electron. Far apart, the spins are strangers — the pair's energy is indifferent to whether they point the same way or opposite. Now lower the wall between them: the electrons share orbitals, and one hopping onto its neighbour's dot suddenly matters. But Pauli forbids two electrons in one orbital unless their spins are antiparallel. So parallel spins (the triplet) cannot make the virtual hop and gain nothing; antiparallel spins (the singlet) briefly double up, borrowing energy of order $t^2/U$ and dropping down. That singlet–triplet difference is the exchange $J$ , and the barrier gate that sets the overlap is its on/off switch.

Figure

Two gate-defined dots hold one electron each in the (1,1) charge state. Lowering the middle barrier gate raises the tunnel coupling t, letting the antiparallel singlet virtually visit the (2,0) doubly-occupied state and drop in energy while the parallel triplet is Pauli-blocked. The residual spin coupling is the Heisenberg exchange J ≈ 4t²/U, switched purely by voltage.

The second lesson is that this interaction generates SWAP, which is why it entangles: a full SWAP merely trades the spins' states and is not entangling, but the halfway $\sqrt{\mathrm{SWAP}}$ sends ⟨↑↓⟩ into a superposition of ⟨↑↓⟩ and ⟨↓↑⟩ — a Bell state, the resource a CNOT needs. How strong $J$ is near the charge transition, and how quiet against voltage noise, is set by the singlet–triplet spectrum in the second figure, which fixes the numbers used below.

Figure

Two-electron spectrum versus inter-dot detuning ε. The triplet stays flat (Pauli-blocked); the singlet bends down as it hybridizes with the (2,0) singlet across an anticrossing of width 2√2 t ≈ 57 μeV for t = 20 μeV. The exchange is the gap J = E_T − E_S. Detuning control (large ∂J/∂ε) is charge-noise sensitive; barrier control at the symmetric point flattens that slope.

The core equations

The single-qubit story is Zeeman splitting. A static field $B$ sets the quantization axis and splits ⟨↑⟩ from ⟨↓⟩ by $g\mu_B B$ ; rotations are driven by an oscillating field on resonance (electron spin resonance) or electrically through a micromagnet gradient. This term defines what " $z$ " means for every gate that follows.

H_{\mathrm{qubit}} = \tfrac{1}{2}\, g\mu_B B\, \sigma_z \qquad (1)

where: $g$ — electron g-factor ( $\approx -0.44$ in GaAs, $\approx 2$ in Si); $\mu_B$ — Bohr magneton, $9.274\times10^{-24}\,\mathrm{J/T}$ ; $B$ — static field defining the quantization axis; $\sigma_z$ — Pauli- $z$ ; eigenstates ⟨↑⟩, ⟨↓⟩ are the computational basis. Reading: the splitting $g\mu_B B$ is the spin's clock frequency, tens of GHz at sub-tesla fields. The entangling interaction is the Heisenberg exchange: the isotropic, rotation-invariant coupling of two spin- $\tfrac12$ operators, and the only two-qubit term the scheme needs. Everything nonlocal in the architecture is built from it.

H_{\mathrm{ex}}(t) = J(t)\,\mathbf{S}_1\cdot\mathbf{S}_2 = \tfrac{1}{4}J(t)\,\boldsymbol{\sigma}_1\cdot\boldsymbol{\sigma}_2 \qquad (2)

where: $J(t)$ — time-dependent exchange coupling, set by gate voltages; $\mathbf{S}_i=\tfrac12\boldsymbol{\sigma}_i$ — spin operator of electron $i$ ; $\boldsymbol{\sigma}_i$ — Pauli vector. Reading: being a scalar, it commutes with total spin — it reshuffles ⟨↑↓⟩ and ⟨↓↑⟩ but never touches ⟨↑↑⟩ or ⟨↓↓⟩. What sets the size of $J$ is a quantum-statistical accident: it is the energy gap between the spin singlet and triplet of the two-electron system. Antisymmetry of the total wavefunction ties spin symmetry to orbital symmetry, and the two orbital arrangements cost different energies.

J = E_T - E_S \qquad (3)

where: $E_S$ — singlet energy, $\lvert S\rangle=(\lvert\uparrow\downarrow\rangle-\lvert\downarrow\uparrow\rangle)/\sqrt2$ (symmetric orbital, lower); $E_T$ — triplet energy (antisymmetric orbital). Reading: a positive $J$ (singlet below triplet) is the antiferromagnetic sign a simple two-electron double dot always produces. A concrete estimate comes from the two-site Hubbard model at half filling. Treating tunnelling $t$ as a perturbation on the singly-occupied configuration, only the singlet makes the virtual round trip to double occupancy, and second-order perturbation theory gives a compact result.

J \simeq \frac{4t^2}{U} \qquad (4)

where: $t$ — inter-dot tunnel coupling (hopping matrix element); $U$ — on-site Coulomb charging energy of double occupancy. Reading: $J$ grows quadratically with overlap and is throttled by $U$ , so a small barrier-voltage change (hence $t$ ) swings $J$ over orders of magnitude. Gates are the time-ordered exponential of the exchange. Because $\mathbf{S}_1\cdot\mathbf{S}_2$ is constant in operator form, only the accumulated phase $\phi$ survives the integral — the pulse shape is irrelevant, only its area matters.

U_{\mathrm{ex}}(\tau) = \exp\!\Big(-\tfrac{i}{\hbar}\!\int_0^\tau\! J(t)\,\mathbf{S}_1\cdot\mathbf{S}_2\,\mathrm{d}t\Big),\qquad \phi \equiv \tfrac{1}{\hbar}\!\int_0^\tau\! J\,\mathrm{d}t \qquad (5)

where: $\tau$ — pulse duration; $\phi$ — accumulated exchange phase, the one quantity that matters. Reading: calibrating a gate means calibrating a pulse area — why exchange gates forgive pulse-shape distortion. Two special areas name the primitive gates: a full $\pi$ of exchange phase gives SWAP; a half, $\pi/2$ , gives the entangling square-root-SWAP that Loss and DiVincenzo identified as the workhorse.

U_{\mathrm{SWAP}} = U_{\mathrm{ex}}(\phi=\pi),\qquad U_{\sqrt{\mathrm{SWAP}}} = U_{\mathrm{ex}}(\phi=\pi/2) \qquad (6)

where: SWAP — exchanges the qubit states, $\mathrm{SWAP}\lvert ab\rangle=\lvert ba\rangle$ ; $\sqrt{\mathrm{SWAP}}$ — the entangling half-swap, the platform's native two-qubit gate. Reading: SWAP alone is not entangling, but its square root is — the derivation below makes this precise. To reach a canonical CNOT you sandwich two $\sqrt{\mathrm{SWAP}}$ pulses between single-spin $z$ -rotations — the explicit universal-gate recipe from the original paper.

\mathrm{CNOT} = e^{i\pi/4}\, e^{-i\frac{\pi}{4}\sigma_z^{(1)}}\, e^{+i\frac{\pi}{4}\sigma_z^{(2)}}\, U_{\sqrt{\mathrm{SWAP}}}\, e^{i\frac{\pi}{2}\sigma_z^{(1)}}\, U_{\sqrt{\mathrm{SWAP}}} \qquad (7)

where: $\sigma_z^{(i)}$ — Pauli- $z$ on qubit $i$ , applied as single-spin rotations. Reading: single-qubit rotations plus one entangler is universal, so this line certifies the whole scheme. Finally, whether any of this is usable depends on a hierarchy of coherence times: the gate must be fast compared with the times over which the spin loses phase or relaxes.

\Gamma_1^{-1}=T_1,\qquad T_2^*,\qquad T_{2,\mathrm{echo}} \qquad (8)

where: $T_1$ — spin relaxation time (phonon-mediated; ms–s in Si); $T_2^*$ — inhomogeneous dephasing time (nuclear/charge-noise limited); $T_{2,\mathrm{echo}}$ — refocused coherence, recovering much of $T_1$ in purified $^{28}\mathrm{Si}$ . Reading: the figure of merit is coherence time over gate time — thousands to millions of operations per coherence window in modern silicon.

Derivation ASWAP and square-root-SWAP from the exchange pulse

The goal is to prove that the Heisenberg exchange is literally a generator of the SWAP operation, so that pulsing it for the right area produces SWAP and half that area produces the entangling square-root-SWAP the whole scheme rests on. We work with $\hbar=1$ and constant $J$ , rewrite $\mathbf{S}_1\cdot\mathbf{S}_2$ in terms of SWAP, exponentiate exactly using $\mathrm{SWAP}^2=\mathbb 1$ , and read off the two gates and their phases.

Step 1. Use the identity $\mathrm{SWAP}=\tfrac12(\mathbb 1+\boldsymbol{\sigma}_1\cdot\boldsymbol{\sigma}_2)$ together with $\boldsymbol{\sigma}_1\cdot\boldsymbol{\sigma}_2=4\,\mathbf{S}_1\cdot\mathbf{S}_2$ to write the interaction as a linear function of SWAP:

\mathbf{S}_1\cdot\mathbf{S}_2 = \tfrac14(2\,\mathrm{SWAP}-\mathbb 1) = \tfrac12\,\mathrm{SWAP}-\tfrac14\,\mathbb 1

Why: the exchange operator and SWAP are the same object up to a shift by the identity, so exponentiating one exponentiates the other.

Step 2. Exponentiate with accumulated phase $\phi=\int J\,\mathrm dt$ and pull the identity term out as a global phase:

U_{\mathrm{ex}}(\phi)=e^{-i\phi(\frac12\mathrm{SWAP}-\frac14\mathbb 1)}=e^{i\phi/4}\,e^{-i\frac{\phi}{2}\mathrm{SWAP}}

Why: the $-\tfrac14\mathbb 1$ piece commutes with everything and only adds an unobservable overall phase.

Step 3. Because $\mathrm{SWAP}^2=\mathbb 1$ , it exponentiates exactly like a Pauli operator (eigenvalues $\pm1$ ):

e^{-i\frac{\phi}{2}\mathrm{SWAP}}=\cos\tfrac{\phi}{2}\,\mathbb 1-i\sin\tfrac{\phi}{2}\,\mathrm{SWAP}

so the full propagator is $U_{\mathrm{ex}}(\phi)=e^{i\phi/4}\big[\cos\tfrac{\phi}{2}\,\mathbb 1-i\sin\tfrac{\phi}{2}\,\mathrm{SWAP}\big]$ . Why: an involution squares to the identity, exactly the algebra that turns an exponential into a cosine-plus-sine of the operator.

Step 4. Set $\phi=\pi$ (full pulse). Then $\cos\tfrac\pi2=0$ , $\sin\tfrac\pi2=1$ :

U_{\mathrm{ex}}(\pi)=e^{i\pi/4}(-i\,\mathrm{SWAP})=e^{-i\pi/4}\,\mathrm{SWAP}

Why: a $\pi$ -pulse is SWAP up to a global phase — and SWAP maps products to products, so alone it creates no entanglement.

Step 5. Set $\phi=\pi/2$ (half pulse). Then $\cos\tfrac\pi4=\sin\tfrac\pi4=1/\sqrt2$ :

U_{\mathrm{ex}}(\pi/2)=e^{i\pi/8}\Big(\tfrac{1}{\sqrt2}\mathbb 1-\tfrac{i}{\sqrt2}\mathrm{SWAP}\Big)=\sqrt{\mathrm{SWAP}}

Applied to ⟨↑↓⟩ this yields a superposition of ⟨↑↓⟩ and an $i$ -weighted ⟨↓↑⟩ — a maximally entangled pair. Why: the half-swap is the entangler, the nonlocal primitive the architecture is built to deliver. *As a consistency check: $(\sqrt{\mathrm{SWAP}})^2=U_{\mathrm{ex}}(\pi/2)^2=e^{i\pi/4}e^{-i\frac{\pi}{2}\mathrm{SWAP}}=U_{\mathrm{ex}}(\pi)=\mathrm{SWAP}$ up to global phase, confirming that two half-pulses compose to a full SWAP.*

Go deeper · Derivation B · 1 minSuperexchange from the two-site Hubbard model

Derivation BSuperexchange from the two-site Hubbard model

The goal is to derive $J\approx 4t^2/U$ from first principles, showing that exchange in coupled dots is a second-order virtual process in which only the singlet borrows energy by momentarily double-occupying a dot, while the triplet is frozen out by Pauli exclusion. We set up the two-site, half-filled Hubbard model, take the (1,1) manifold as the low-energy space, and treat hopping $t$ as a perturbation to the (2,0)/(0,2) states at energy $U$ .

Step 1. Write the two-site Hubbard model with on-site repulsion $U$ and hopping $t$ :

H = U\!\sum_{i} n_{i\uparrow}n_{i\downarrow} - t\!\sum_{\sigma}\big(c^\dagger_{1\sigma}c_{2\sigma}+\mathrm{h.c.}\big)

At half filling the low-energy manifold is the two (1,1) states, one electron per dot. Why: the qubits live in (1,1); double occupancy costs $U$ and is only ever visited virtually.

Step 2. Classify the (1,1) states by total spin. The triplet ⟨↑↑⟩ (and its partners) cannot hop, because moving an ↑ onto a dot already holding an ↑ is Pauli-forbidden:

\text{triplet:}\quad \delta E_T = 0

Why: parallel spins have no accessible (2,0) intermediate, so hopping is blocked and the triplet gains nothing.

Step 3. The (1,1) singlet can hop: an electron tunnels to make a (2,0) singlet at energy $U$ , then tunnels back. Second-order perturbation theory over both dots as intermediates gives:

\delta E_S = -\sum_{\text{2 paths}}\frac{t^2}{U}\times(\text{2 spin orderings}) = -\frac{4t^2}{U}

Why: two virtual endpoints (2,0 and 0,2), each reached by matrix element $t$ and lowering the singlet by $t^2/U$ , times a factor of two from the antisymmetrized singlet amplitude.

Step 4. Assemble the exchange as the singlet–triplet gap:

J = E_T - E_S = 0-\Big(-\frac{4t^2}{U}\Big)=\frac{4t^2}{U}

Why: the singlet drops below the triplet by exactly the virtual-hopping energy, so the antiferromagnetic exchange is $4t^2/U$ , tunable through $t$ . *Three quick checks on the result: dimensionally $[t^2/U]$ is energy; the sign is antiferromagnetic ( $J>0$ , singlet ground state), as expected for a simple two-electron double dot; and in the limit $t\ll U$ leakage to (2,0) is $O((t/U)^2)$ , so the pure-spin Heisenberg picture holds throughout the perturbative regime.*

Go deeper · Derivation C · 2 minCharge-noise sensitivity and the symmetric sweet spot

Derivation CCharge-noise sensitivity and the symmetric sweet spot

The goal is to quantify why exchange gates dephase — charge noise jitters the detuning $\varepsilon$ , which modulates $J$ — and to show why operating at the charge-symmetric point, controlling $J$ through the barrier rather than the detuning, gives first-order protection. We reduce the singlet sector near the (1,1)–(2,0) transition to a two-level problem, extract $J(\varepsilon)$ , differentiate to expose the noise coupling, and note what the two-level model misses about the symmetric point.

Step 1. In the singlet sector, keep the (1,1) singlet at energy $0$ and the (2,0) singlet at $+\varepsilon$ , coupled by $\sqrt2\,t$ (the $\sqrt2$ from the two ways to form the doubly-occupied singlet):

H_S=\begin{pmatrix}0 & \sqrt2\,t\\ \sqrt2\,t & \varepsilon\end{pmatrix}

Only singlets appear — the triplet is Pauli-blocked from (2,0). Why: the minimal model of the anticrossing that sets $J$ near the charge transition.

Step 2. Diagonalize; the exchange is minus the lower eigenvalue (triplet sits at $0$ ):

E_-=\tfrac12\big(\varepsilon-\sqrt{\varepsilon^2+8t^2}\big),\qquad J(\varepsilon)=-E_-=\tfrac12\big(\sqrt{\varepsilon^2+8t^2}-\varepsilon\big)

with $\varepsilon>0$ measured from the anticrossing into (1,1). Why: $J$ is how far the hybridized singlet is pushed below the flat triplet.

Step 3. Differentiate to find the coupling of detuning noise into the gate:

\frac{\partial J}{\partial\varepsilon}=\tfrac12\Big(\frac{\varepsilon}{\sqrt{\varepsilon^2+8t^2}}-1\Big),\qquad \delta J\approx\frac{\partial J}{\partial\varepsilon}\,\delta\varepsilon

In the tilted scheme one turns $J$ on by reducing $\varepsilon$ toward the transition, where $|\partial J/\partial\varepsilon|$ is large — so charge fluctuations $\delta\varepsilon$ become gate-error fluctuations $\delta J$ . Why: charge noise dominates exchange-gate dephasing, and it enters entirely through this slope.

Step 4. The remedy is symmetric (barrier) control. Hold the dots charge-symmetric and modulate $J$ through the tunnel coupling $t$ instead of $\varepsilon$ ; at the exact symmetric point the (2,0)↔(0,2) mirror symmetry makes $J$ an even function of $\varepsilon$ , so

\left.\frac{\partial J}{\partial\varepsilon}\right|_{\text{sym}}=0,\qquad \left.\frac{\partial J}{\partial t}\right|_{\varepsilon=0}=\sqrt2\neq0

protecting the gate against detuning noise to first order while leaving a live barrier knob. Why: this "symmetric operation" pushed two-qubit fidelities past 99% in silicon. *Two useful checks on the result: for $\varepsilon\gg t$ the two-level formula reduces to $J\to 2t^2/\varepsilon$ , the expected perturbative superexchange with a single (2,0) intermediate; and the even-in- $\varepsilon$ property at the symmetric point requires the (2,0)/(0,2) mirror pair that lies beyond this minimal model, consistent with the barrier-control experiments of Reed et al. and Martins et al. (2016).*

Limiting cases

The Heisenberg form is an effective Hamiltonian; its validity and the gate's behaviour are governed by ratios of the physical energy scales.

Case	Result	Meaning
Weak coupling (Heisenberg regime)	$J \ll U,\ \hbar\omega_0$	Tunnelling small versus $U$ and orbital spacing $\hbar\omega_0$ ; the Hubbard model collapses to pure spin exchange $J\approx4t^2/U$ , leakage $O((t/U)^2)$ .
Strong coupling (leakage onset)	$J \sim U$	Double occupancy admixes strongly; the two-level spin picture fails, the singlet gains charge character, and population leaks to (2,0)/(0,2).
Symmetric operation point	$\partial J/\partial\varepsilon = 0$	At the charge-symmetric point $J$ is even in $\varepsilon$ , first-order insensitive to charge noise; the barrier ( $t$ ) drives the gate, sharply raising fidelity.
Hyperfine-dominated dephasing	$J \lesssim g\mu_B\,\Delta B_{\mathrm{nuc}}$	When exchange is comparable to the random nuclear Overhauser gradient, that gradient scrambles the gate; purifying $^{29}\mathrm{Si}$ suppresses $\Delta B_{\mathrm{nuc}}$ .

Common trapExchange is a knob you can never leave on for free

It is tempting to picture $J$ as an independent dial you set and forget. It is not. Turning on $J$ requires real wavefunction overlap, and that same overlap simultaneously exposes the spins to charge noise on the barrier and detuning, and admixes the doubly-occupied leakage states (2,0)/(0,2) into the computational subspace. The clean form $J\,\mathbf{S}_1\cdot\mathbf{S}_2$ holds only while $J$ stays far below the orbital spacing and charging energy; push $J$ too high for speed and the two-level approximation breaks, leakage grows, and the "gate" stops being a gate. Fast and clean are in tension — and the whole engineering art (symmetric operation, pulse shaping, isotopic purification) is about buying speed without paying it all back in noise and leakage.

Worked examples

1 · Exchange strength and gate time from device parameters (fully worked)

A silicon double dot has tunnel coupling $t = 20\ \mu\mathrm{eV}$ and charging energy $U = 1\ \mathrm{meV}$ . Estimate $J$ in frequency and temperature units, and the times to run $\sqrt{\mathrm{SWAP}}$ and SWAP.

Step 1. Superexchange (equation 4): $J\approx 4t^2/U = 4(20)^2/1000\ \mu\mathrm{eV} = \mathbf{1.6\ \mu eV}$ — three orders below the charging energy, from a modest 20 μeV tunnel coupling.

Step 2. In frequency, $f_J = J/h = (1.6\times10^{-6}\,\mathrm{eV})/(4.136\times10^{-15}\,\mathrm{eV\,s}) = \mathbf{387\ MHz}$ ; in temperature, $J/k_B = \mathbf{19\ mK}$ — below dilution-fridge base, so thermal escape from the singlet is negligible.

Step 3. A gate is a pulse area (equation 5): $\sqrt{\mathrm{SWAP}}$ needs $\phi=\pi/2$ , so at constant $J$ , $\tau_{\sqrt{\mathrm{SWAP}}} = (\pi/2)\hbar/J = 1/(4f_J) = \mathbf{0.65\ ns}$ , and SWAP ( $\phi=\pi$ ) takes $1/(2f_J) = \mathbf{1.3\ ns}$ . A sub-nanosecond entangler from a 20 μeV barrier coupling — with echoed coherence reaching milliseconds in purified silicon, that is room for $\sim10^{6}$ operations per coherence time.

2 · Zeeman splitting and the single-qubit drive frequency

Compare the qubit frequency in purified silicon ( $g\approx2$ ) and GaAs ( $g\approx-0.44$ ) at $B = 0.5\ \mathrm{T}$ .

Step 1. The splitting is $E_Z = |g|\mu_B B$ (equation 1). Silicon: $E_Z = 2\times(9.274\times10^{-24})\times0.5 = 9.27\times10^{-24}\ \mathrm{J}$ , so $f = E_Z/h = \mathbf{14.0\ GHz}$ — an X-band electron-spin-resonance drive.

Step 2. GaAs's negative, smaller g-factor gives $f = 0.44\times14.0/2 = \mathbf{3.1\ GHz}$ , and the sign of $g$ flips the Larmor sense. Silicon needs faster (X-band) electronics but rewards it with far weaker nuclear coupling; GaAs drives slower yet suffers strong hyperfine dephasing — why the field migrated to Si.

3 · Why the half-swap entangles but the full swap does not

Confirm that $\sqrt{\mathrm{SWAP}}$ entangles a product input while SWAP does not.

Step 1. $\mathrm{SWAP}\lvert\uparrow\downarrow\rangle=\lvert\downarrow\uparrow\rangle$ is still a product state — SWAP maps every product to a product, no entanglement.

Step 2. From Derivation 1, up to global phase $\sqrt{\mathrm{SWAP}}\lvert\uparrow\downarrow\rangle = \tfrac{1}{\sqrt2}(\lvert\uparrow\downarrow\rangle - i\,\lvert\downarrow\uparrow\rangle)$ , whose reduced single-spin state is $\tfrac12\mathbb 1$ — maximally entangled. $\sqrt{\mathrm{SWAP}}$ is the entangler; with single-spin rotations (equation 7) it completes a universal set, while SWAP alone never leaves the product states.

Physics of the portrait

The Loss–DiVincenzo proposal is where mesoscopic physics, quantum information, and materials science met. Its authors invented no new interaction; they recognized that a textbook one — Heisenberg exchange, the coupling behind ferromagnetism and the hydrogen molecule — was already sitting in coupled dots, electrically switchable, waiting to be used as a quantum gate.

Origins · 1 minDaniel Loss and David DiVincenzo and how this was discovered

Daniel Loss and David DiVincenzo

b. 1958 / b. 1959 · quantum computation with quantum dots

Daniel Loss, a Swiss theorist at the University of Basel, and David DiVincenzo, then at IBM's Watson Research Center, published "Quantum computation with quantum dots" in Physical Review A in January 1998 — one of the most-cited proposals in the field. DiVincenzo was already asking what any physical quantum computer must provide, work that crystallized into the five DiVincenzo criteria; Loss brought deep expertise in the spin physics of nanostructures. Their move was conceptual economy: rather than engineer a bespoke two-qubit coupling, use the exchange the Pauli principle guarantees whenever two confined electrons overlap, controlled by the gate voltages that already define the dots. They showed $\sqrt{\mathrm{SWAP}}$ plus single-spin rotations is universal, and sketched initialization by spin relaxation and readout by spin-to-charge conversion — a complete architecture on a chip. The follow-up with Guido Burkard (Phys. Rev. B, 1999) put numbers on $J(t)$ from a microscopic model. The proposal seeded two decades of experiment, from the first coupled dots in GaAs to today's silicon arrays, and remains the reference against which every spin-qubit result is measured.

History

1998

— Loss and DiVincenzo propose spin qubits in coupled quantum dots with exchange gates

1999

— Burkard, Loss and DiVincenzo model J(t) microscopically for a double dot

2000

— DiVincenzo states the five criteria a physical quantum computer must satisfy

2005

— Petta and coworkers demonstrate exchange control and singlet–triplet manipulation in GaAs

2018

— isotopically purified silicon dots reach two-qubit gates via tunable exchange

Going further

Texts and papers. Loss and DiVincenzo, "Quantum computation with quantum dots," Phys. Rev. A 57, 120 (1998) · DiVincenzo, "The physical implementation of quantum computation," Fortschr. Phys. 48, 771 (2000) · Burkard, Loss and DiVincenzo, "Coupled quantum dots as quantum gates," Phys. Rev. B 59, 2070 (1999) · Hanson, Kouwenhoven, Petta, Tarucha and Vandersypen, "Spins in few-electron quantum dots," Rev. Mod. Phys. 79, 1217 (2007) · Nielsen and Chuang, §7.8 and §4. The key relations used throughout are: $\mathbf{S}_1\cdot\mathbf{S}_2=-\tfrac34$ (singlet) and $+\tfrac14$ (triplet), giving $E_T-E_S=J$ ; the propagator $U_{\mathrm{ex}}(\phi)=e^{i\phi/4}(\cos\tfrac\phi2\,\mathbb 1-i\sin\tfrac\phi2\,\mathrm{SWAP})$ with $(\sqrt{\mathrm{SWAP}})^2=\mathrm{SWAP}$ ; the Hubbard superexchange $J=4t^2/U$ ; and the anticrossing formula $J(\varepsilon)=\tfrac12(\sqrt{\varepsilon^2+8t^2}-\varepsilon)$ with $\varepsilon\gg t$ limit $2t^2/\varepsilon$ . Worked numbers use $\mu_B=9.274\times10^{-24}\,\mathrm{J/T}$ and $h=4.136\times10^{-15}\,\mathrm{eV\,s}$ : $t=20\ \mu\mathrm{eV}$ , $U=1\ \mathrm{meV}$ gives $J=1.6\ \mu\mathrm{eV}=387\ \mathrm{MHz}=19\ \mathrm{mK}$ with $\sqrt{\mathrm{SWAP}}$ in 0.65 ns; Zeeman splitting at $B=0.5\ \mathrm{T}$ gives $g\mu_B B=14.0\ \mathrm{GHz}$ in Si and 3.1 GHz in GaAs. Fidelities ( $>99\%$ ) are platform benchmarks, not single-experiment claims.

Prerequisites. DiVincenzo Criteria · Pauli exclusion principle and exchange symmetry · Spin-1/2 algebra and Pauli matrices. Leads to. Singlet–triplet and exchange-only qubits · Superconducting qubits · Quantum error correction.

Ask the tutor: Want to see exactly how two √SWAP pulses and a pair of single-spin z-rotations multiply out to a CNOT, matrix by matrix?

Why store the qubit in spin rather than in the electron's charge or orbital state?

Because charge is loud and spin is quiet. Orbital (charge) states couple directly to electric fields and phonons and decohere in picoseconds to nanoseconds — far too fast to compute with. The spin is electrically almost invisible: its relaxation time $T_1$ reaches milliseconds to seconds, and its dephasing time, once nuclear spins are purified away, stretches to milliseconds under echo. The Loss–DiVincenzo bargain keeps information in the quiet spin but summons the loud charge only briefly, during an exchange pulse, when two qubits must interact — long idle memory and fast electrical gates from the same electron.

How can a scalar interaction like S₁·S₂ possibly entangle two qubits?

The key is that $\mathbf{S}_1\cdot\mathbf{S}_2$ is not a sum of single-spin terms — it is a product of operators on both qubits, $\tfrac14\boldsymbol\sigma_1\cdot\boldsymbol\sigma_2$ , with cross terms like $\sigma_x^{(1)}\sigma_x^{(2)}$ . Rewritten it is $\tfrac12\mathrm{SWAP}-\tfrac14\mathbb 1$ , so exponentiating it generates fractional SWAPs. A full SWAP merely relabels the qubits and does not entangle, but the half-pulse $\sqrt{\mathrm{SWAP}}$ takes a product state ⟨↑↓⟩ into a superposition of ⟨↑↓⟩ and ⟨↓↑⟩ that cannot be factored into two single-spin states. That inseparability is entanglement — exactly the nonlocal resource a CNOT needs.

What is the difference between detuning control and barrier (symmetric) control of exchange?

Both raise $J$ by increasing overlap, through different knobs. Detuning control tilts the double well toward the (2,0) transition; $J$ rises steeply, but so does $\partial J/\partial\varepsilon$ , so any charge-noise wobble in $\varepsilon$ feeds straight into gate error. Barrier (symmetric) control keeps the dots charge-balanced and lowers the tunnel barrier to raise $t$ . At that symmetric point $J$ is even in the detuning, so its first-order sensitivity to detuning noise vanishes while the barrier stays a live control. Since charge noise dominates exchange-gate dephasing, this "sweet spot" was decisive in pushing two-qubit fidelities past 99%.

Why does isotopic purification of silicon matter so much?

Natural silicon contains about 4.7% $^{29}\mathrm{Si}$ , the only stable isotope with a nuclear spin. Those spins produce a fluctuating Overhauser field — a random, slowly drifting effective field on the electron — that dephases the qubit and adds an uncontrolled gradient across a double dot, scrambling exchange gates whenever $J\lesssim g\mu_B\,\Delta B_{\mathrm{nuc}}$ . Growing the device in enriched (spinless) $^{28}\mathrm{Si}$ removes nearly all of them. The payoff is dramatic: $T_2^*$ jumps from tens of nanoseconds (GaAs is worse still, every isotope carrying spin) to tens of microseconds, echo reaching milliseconds — the gap between a few and a million operations per coherence time.

Does the shape of the exchange pulse matter, or only its area?

For the ideal Heisenberg model, only the area matters. Because $\mathbf{S}_1\cdot\mathbf{S}_2$ is constant in operator form, the time-ordering is trivial and the gate collapses to the single number $\phi=\int J\,\mathrm dt/\hbar$ . A tall short pulse and a low long pulse of equal area give the same gate — a practical blessing that makes exchange gates forgiving of pulse-shape distortion. The caveats are real-world: if $J$ is pushed high enough to admix leakage states, or if single-qubit terms fail to commute during the ramp, shape re-enters, and pulse engineering (adiabatic ramps, dynamically corrected gates) suppresses the residual error.

How does the Loss–DiVincenzo scheme satisfy all five DiVincenzo criteria?

One platform, five boxes. Scalable qubits: a lithographic array of gate-defined dots, one spin each. Initialization: relax the spin to its ground state in the field, or load a singlet from a reservoir. Long coherence versus gate time: $T_2$ reaches milliseconds in purified silicon while gates run in nanoseconds, so thousands to millions of operations fit one coherence window. Universal gates: single-spin rotations (electron spin resonance or micromagnet-assisted electric drive) plus exchange-generated $\sqrt{\mathrm{SWAP}}$ . Measurement: spin-to-charge conversion read by a nearby electrometer. That completeness on one semiconductor chip is why the proposal became the field's reference architecture.