Single-qubit gates and measurements in the surface acoustic wave quantum computer

S. Furuta,∗ C. H. W. Barnes, and C. J. L. Doran

Cavendish Laboratory, Department of Physics, University of Cambridge, Madingley Road, Cambridge, CB3 0HE, UK.
(Dated: August 22, 2010)
In the surface acoustic wave quantum computer, the spin state of an electron trapped in a moving
arXiv:cond-mat/0406167v1 [cond-mat.mes-hall] 7 Jun 2004

quantum dot comprises the physical qubit of the scheme. Via detailed analytic and numerical
modeling of the qubit dynamics, we discuss the effect of excitations into higher-energy orbital states
of the quantum dot that occur when the qubits pass through magnetic fields. We describe how
single-qubit quantum operations, such as single-qubit rotations and single-qubit measurements, can
be performed using only localized static magnetic fields. The models provide useful parameter
regimes to be explored experimentally when the requirements on semiconductor gate fabrication
and the nanomagnetics technology are met in the future.

PACS numbers: 03.67.Lx, 73.21.La, 72.50.+b, 72.25.Dc

I. INTRODUCTION

Quantum computation promises enormous technolog-

ical advances in the field of information processing1,2,3,4
and the quest for its realization has attracted many
strong contenders in the field of physics and engineer-
ing. This paper is concerned with a scheme for quan-
tum computation put forward by Barnes, Shilton and
Robinson,5 which falls into the semiconductor quan-
tum dot category.6,7,8,9,10,11,12 The proposal for quantum
computation is based on the results of ongoing exper-
iments that have demonstrated the capture and trans-
port of single electrons in moving quantum dots.13,14,15 FIG. 1: (Color online) Schematic diagram of an experimen-
The dots are formed when a surface acoustic wave (SAW) tal device for producing quantized acoustoelectric currents
through a narrow Q1DC constriction.
travels along the surface of a piezoelectric semiconductor
containing a two-dimensional electron gas (2DEG). See
Fig. 1 for a schematic diagram of the device. When the
SAW is made to pass through a constriction in the form performs time-ensemble computation, in much the same
of a quasi-one-dimensional channel (Q1DC), the induced way NMR quantum computation performs molecular-
piezoelectric potential drags electrons into and along the ensemble computations.24,25,26 Time-ensemble computa-
Q1DC. In certain parameter regimes the device trans- tion alleviates the demand for single-shot spin measure-
ports one electron per potential minimum of the SAW.14 ments and has the advantage of being robust against
The spin on the trapped electron represents the physical small random errors.
qubit. Quantum computation involves performing qubit
operations on the trapped electrons as they move with This paper considers proposals5 for implementing
the speed of the SAW. quantum gates on single SAW qubits using only static
Many schemes for quantum computation, such as con- magnetic fields generated by surface magnetic gates. De-
ventional quantum dots,6,16 doped silicon,17 supercon- tailed modeling of the gate operation has been accom-
ducting boxes,18 and ion traps,19 involve static qubits. plished by means of both analytic solutions and numeri-
The surface acoustic wave quantum computer on the cal simulations of the Pauli equation. We show how elec-
other hand is of the “flying qubit” type, which include trostatically confined moving electrons behave under the
linear optics schemes,20 some ion trap schemes with ion influence of various magnetic fields and discuss the im-
shuttling21 and schemes based on coherent electron trans- plications for quantum computing with surface acoustic
port in quantum wires.5,22,23 All these have in common wave electrons.
that the carriers of quantum information physically move The physics of the SAW-guided qubit is explained in
through space during the computation. Flying qubits more detail in section II. In sections III and IV, we
have the advantage of being able to distribute infor- present results on single-qubit unitary gates and single-
mation quickly over large distances across the quan- qubit readout gates based on the Stern-Gerlach effect.
tum circuit when decoherence times are short and to Section V discusses some of the decoherence processes
interface with quantum memory registers at fixed loca- involved in quantum-dot based schemes. Section VI is a
tions. Another advantage of the SAW quantum com- summary of the results with parameter regimes of inter-
putation scheme is its ensemble nature. It intrinsically est for future experiments.
 2

II. SCHEME FOR QUANTUM COMPUTATION

We begin by summarizing the quantum computation

scheme proposed by Barnes, Shilton and Robinson.5 Fig-
ure 1 shows a schematic diagram of the experimental
setup originally designed to demonstrate quantized cur-
rents in semiconductors.13,14,15 A NiCr/Al interdigitated
transducer is patterned on a GaAs/AlGaAs heterostruc-
ture. A narrow depleted Q1DC splits the 2DEG into two
regions, the source and drain. When a high frequency
AC signal is applied to the transducer, a SAW propagates
through the 2DEG, producing a periodic piezoelectric po-
tential across the 2DEG. The potential drags electrons in
the source region through the narrow Q1DC constriction
into the drain. It has been shown experimentally that
over a range of SAW power and gate voltages, the cur-
rent passing through the Q1DC is quantized in units of
ef , where e is the electronic charge and f is the frequency
of the SAW.13,14,15 The smallest quantized current ob-
served corresponds to the transport of a single electron
in each SAW minimum. Typically, the SAW in GaAs
moves at 2700 ms−1 at a frequency of around 2.7 GHz, FIG. 2: (Color online) Schematic diagram of a quantum gate
with an applied power of 3 − 7 dBm.14 These parameters network in a SAW quantum computer: (a) Two-qubit tun-
produce currents in the range of nanoamps. neling gates; (b) One-qubit magnetic gates (in various ori-
Given the ability to trap single electrons in the SAW entations); (c) Gate network for quantum computation with
minima, the scheme for quantum computation is as fol- SAW electron spins. Gray lines running horizontally repre-
lows. It is possible for an array of N Q1DCs in paral- sent Q1DCs; blackened regions indicate the SAW minima
lel to capture N qubits in every M th minimum, with a where the qubits reside; arrows represent spin polarization;
rings represent magnetic surface gates; white squares repre-
single electron in each Q1DC, producing a qubit regis-
sent readout gates.
ter along the SAW wavefront. M can be chosen suffi-
ciently large to ensure that the Coulomb interaction be-
tween successive qubit registers do not interfere with each
separate computation. The qubits move with the min- SAW amplitude is typically 40 meV,31 which is suffi-
ima of the SAW, passing through a sequence of static ciently large to prevent qubits being lost via tunneling
one- and two-qubit gates before arriving at an array of into neighboring SAW minima.
spin readout devices. One-qubit gates may be operated There are two important aspects of the SAW quan-
by nanoscale electromagnetic fields. Where two-qubit tum computation scheme that distinguishes it from other
gates are needed, neighboring Q1DCs are allowed into similar quantum dot schemes. First, the scheme pro-
a tunnel contact controlled by a potential on a surface vides repetitions of the same quantum computation with
gate.5 The use of the Coulomb coupling between neigh- each passing of a single wavefront of the SAW. There-
boring Q1DCs is a common tool in spintronics which can fore, a statistical time-ensemble of identical computa-
be used to generate entangled states in dual-rail qubit tions can be read out at the end of the Q1DCs as a mea-
representations.27,28 Figure 2 illustrates the network of surable current, alleviating the need for single-electron
Q1DCs and qubit gates envisaged for performing a par- measurements. Two sources of noise in the measured
ticular quantum computation. current may be estimated as follows: The shot noise is
The SAW-trapped electron is well confined in all three largely determined by how well the current is quantized
spatial directions. In this paper we define Cartesian co- to I = efSAW and precisions of < 0.1% can be exper-
ordinates such that the SAW propagates along the x axis imentally achieved.32 Johnson noise arises from the re-
with the z axis normal to the 2DEG. The 2DEG is pro- sistance of the ohmic contacts and the 2DEG which are
duced by a band-energy mismatch at the GaAs/AlGaAs on the order of ∼ 100 − 1000 Ω. At temperatures of
interface which gives rise to a confining potential in the z 1 K they produce a rms√voltage noise spectral density on
direction. The energy level spacings in the 2DEG well are the order of 10−10 V/ Hz at most. This cannot drive
on the order of 50 − 100 meV.29 Further confinement in a current noise through the SAW device since its effec-
the y direction is provided by an extended Q1DC, etched tive intrinsic impedance of 10 MΩ is comparatively very
into the surface so as to avoid screening the SAW-induced large. See experimental papers13,14,15,32,33 for more de-
potential with metallic surface gates.30 Finally, the con- tailed discussions. The second key aspect of the scheme
finement in the x direction is due to the SAW poten- is the static nature of the gate components of the quan-
tial minimum which is approximately sinusoidal. The tum circuit. This alleviates the need for strong, targeted
 3

and carefully-timed electromagnetic pulses that can be

difficult and expensive to implement. The requirement
of such expensive control resources often limit the scala-
bility of most quantum computing implementations.
It would certainly be convenient, though not essential,
to have a means of preparing a pure fiducial qubit state.
In NMR quantum computing, operations are carried out
on ensembles of replica qubits which remain close to a
highly mixed state of thermal equilibrium. Nevertheless,
a successful readout of the computation is obtained be-
cause the sum over many identical computations provides
a measurable signal. Similarly in the SAW quantum com-
puting scheme, states close to the maximally-mixed state
are still useful because of the time-ensemble nature of the
scheme. However, in contrast to NMR schemes, it is in
principle possible to read out single electrons in the SAW FIG. 3: (Color online) Energy diagram showing polarized
scheme. We will therefore begin with nonensemble quan- electron capture by means of Zeeman band splitting in the
tum computation in mind and only later exploit the ad- presence of a uniform magnetic field: Arrows on electrons in-
dicate spin polarizations; Ef is the Fermi energy. The SAW
vantages of ensemble computation to deal with noise. Of
is strongly screened in the bulk 2DEG below the Fermi en-
course, the more pure the qubit states remain, the faster ergy. As the SAW enters the Q1DC constriction, the confining
the computation will converge to the result. For these potential begins to form. The probability of capturing spin-
reasons, we describe in this paper two simple methods down is small at the point the minority spin-type dot forms
for preparing pure fiducial qubit states. (upper curve). In contrast, the probability of capturing spin-
In the original proposal for the SAW quantum up electrons is high when the majority spin-type dot forms
computer,5 it was noted that the application of an exter- (lower curve).
nal magnetic field of about 1 T will influence which spin
states of electrons are favored in the capture process.31
This polarized capture process can be summarized as fol- will be demonstrated that spin-polarized electrons can be
lows. The SAW is strongly screened in the bulk 2DEG prepared and measured using a gate driven by the Stern-
until it is raised above the Fermi energy and the quan- Gerlach effect. First we turn to the implementation of
tum dot begins to form. When the higher-energy minor- single qubit rotations using local static magnetic gates.
ity quantum dot forms for the higher-energy sub-band of
polarized electrons, the probability of capturing minor-
ity electrons is small (see Fig. 3), whilst the probability III. SINGLE-QUBIT UNITARY GATES
of capturing electrons from the lower-energy sub-band is
large. Once a cloud of approximately polarized electrons Single-qubit unitary operations may be carried out,
is captured, the exchange interaction will generally en- in principle, by allowing the trapped single electrons
tangle electrons in the same dot together, so that the to pass through regions of uniform magnetic field. If
subsequent loss of electrons into the Fermi sea will lead there is no spin-orbit coupling effect, the spin state of
to a decoherent process that could relax the remaining the qubit will evolve according to the Zeeman term in
electrons into the low-energy polarized state. A more de- the Hamiltonian: 21 gµB B · σ ψ, where B is the mag-
tailed multiparticle analysis will be needed to determine netic field, σ is the vector form of the 3 Pauli oper-
the final state of the remaining single electron, but the ators and ψ = (α, β) is a spinor. The Bloch vector
combination of the above two processes is likely to lead n = −2 Im[αβ ∗ ], 2 Re[αβ ∗ ], |α|2 − |β|2 precesses about


to a high level of polarization. the direction of the magnetic field with angular frequency
The above method is conceptually simple and would gµB |B|/~. A local static magnet may be produced by
be easy to implement in the laboratory. However, the a magnetic force microscope or by evaporative deposi-
macroscopic magnetic field required in the capture re- tion of a ferromagnetic material such as Cobalt, or a
gion will need to be shielded from the rest of the device permalloy such as NiFe. Inevitably, different samples
where the quantum computation is to be carried out, and will produce different strengths of magnetic field. But
this may present a nontrivial problem. If we chose to there are methods to vary the strength and pattern of
drive spin flips using microwave pulses, then the macro- the field once the sample has been fabricated. It has been
scopic field is actually required across the whole device. demonstrated that ferromagnetic properties of thin-film
However, the problem with using microwaves is their rel- 3d transition metals can be modified via ion irradiation.34
atively long wavelength which tends to affect every part Another method would be to use oxidation techniques
the computation. Alternatively, we may use local, static with the atomic force microscope.35,36,37 Only two inde-
magnetic fields to initialize, rotate and read out single pendent directions of the B-field are necessary to pro-
qubits without a global magnetic field. In Section IV, it duce an arbitrary single-qubit manipulation and we will
 4

choose these to be perpendicular to the direction of the acting on the particle, the Schrödinger equation simpli-
SAW, one aligned with and the other perpendicular to fies significantly and there would be no need to concern
the 2DEG. ourselves with the spatial behavior of the qubit.
For idealized qubits with no spatial degree of freedom, Before proceeding further, it is convenient to transform
the above model for single-qubit rotations is complete. into the rest frame of the electron moving with the SAW
However, the trapped electron is a charged particle with speed v. We may use simple Galilean transformations
a spatial distribution within the dot. The fields couple x′ = x − vt, y ′ = y, t′ = t, since the SAW velocity is
to both the spatial and spin degrees of freedom, causing non-relativistic. The potentials in the electron rest frame
the electron to experience the Lorentz force as well as become, in the harmonic oscillator approximation,
spin-precession. It is therefore clear that one cannot in-
crease magnetic fields arbitrarily, since the Lorentz force A(x′ , t′ ) = −β(x′ + vt′ ) (3)
2
will upset the confinement properties of the electron. Nor y′ 2πx′
  
′ ′
can the direction of the field be chosen arbitrarily without V (x , y ) = V0 2 + A 1 − cos (4)
2w λ
consequences for the robustness of the gate. To address
V0 ′ 2 A 2 ′ 2
these concerns, we analyze the behavior of the full wave- ≃ y + k x , (5)
function: a two-component spinor field defined over space 2w2 2
and time, ψσ (x, t), under the action of gates operated by with k = 2π/λ. A further convenience is to use a system
static magnets. of natural units such that ~, m∗ , v, e are unity. The units
of length, time and energy become ~/m∗ v = 0.640 µm,
~/m∗ v 2 = 0.237 ns and m∗ v 2 = 2.78 µeV, respectively.
A. Pauli Hamiltonian with uniform magnetic fields The natural unit of magnetic field is 1.61 mT. Parame-
ters can now be assigned with dimensionless values with
Assuming a uniform magnetic field in the lab frame, respect to the above units. In the electron rest frame,
we solve the Pauli equation for the spin field from which 1

1

1
the probability density field and the Bloch vector field − ∇2 ψ − i A · ∇ + (∇ · A) ψ + A2 ψ
2 2 2
can be obtained. (6)
The qubit is trapped in a net electrostatic potential 1 ∂
+ gµB σ · Bψ + V ψ = i ψ,
with contributions from the Q1DC split gates, the 2DEG 2 ∂t
confining potential and the SAW piezoelectric potential. where the primes will subsequently be dropped from the
The parameter regimes we consider allow us to neglect coordinates. We will further simplify the Hamiltonian by
motion out of the 2DEG plane.67 The net confining po- adopting the Coulomb gauge ∇ · A = 0.
tential in the xy plane is modeled by31

V = VQ1DC + VSAW B. Qubit rotation: Uniform transverse magnetic

y2  field
= V0 2 + A 1 − cos [2π(x/λ − f t)] . (1)
2w
For a magnetic field in the y direction in the plane of
The Q1DC split gate voltages are such that typically the 2DEG and transverse to the Q1DC, we model the
V0 ∼ 2800 meV. The width of the Q1DC w is typically magnetic field in the region of the single-qubit gate with
between 1 and 2 µm. The amplitude of the SAW is A ∼ a vector potential of the form
40 meV with wavelength λ ∼ 1 µm and frequency f ∼
2.7 GHz. Az (x) = −β x, (7)
The appropriate nonrelativistic equation for the two-
with Ax = Ay = 0. Clearly this does not vanish at in-
component spinor field is the Schrödinger equation with
finity, but we only consider interactions over regions of
a Pauli Hamiltonian. For an electron moving in an ar-
finite extent. This potential generates a uniform mag-
bitrary vector potential A and potential energy V , the
netic field of strength β in the y direction. With this
Pauli Hamiltonian H is
potential Eq. (6) becomes
1
1
p̂2 + e (p̂ · A) + 2eA · p̂ + e2 A2 + gµB σ · B + V, 1 1 1 ∂
2m ∗ 2 − ∇2 ψ+ β 2 (x+t)2 ψ+ gµB β σy ψ+V ψ = i ψ, (8)
(2) 2 2 2 ∂t
where e > 0 is the electronic charge, g ≃ 0.44 is the in which the potential V is given by (5). An effective
Landé g-factor and m∗ ≃ 0.067me is the effective mass potential V + A2 /2 can be identified, which resembles a
in GaAs. By simulating in the plane of the 2DEG we sim- harmonic oscillator but is time-dependent. Anticipating
plify to a 2+1 dimensional model, with variables (x, y, t). an analytic solution by separation of variables we apply
If the particle were chargeless but with an anomalous the ansatz
magnetic moment, terms in H that explicitly involve e
can be dropped. In such a case, there is no Lorentz force ψ± (x, y, t) = χ(x, t) φn (y) e−iEn t |sy i e−isy ∆E t/2 , (9)
 5

where φn are the harmonic oscillator eigenstates with dragged a distance λ away from the center of the dot in
energies En = ωy (n + 21 ), and |sy i are the eigenstates the x direction:
of σy with eigenvalues sy = ±1. √ We have introduced c0 + c1
the oscillator frequency ωy = V0 /w and the Zeeman Tmax ≃ λ. (15)
energy gap ∆E = gµB β. The energy eigenstates in the y 2c1
direction are exactly the harmonic oscillator modes. The When this is compared with the time required for the
wavefunction in the x direction is time-dependent and Bloch vector to rotate by some appreciable angle such as
it is the solution of χ(x, t) to which we now turn. On a π rotation, we obtain the ratio
substituting (9) into (8), we obtain a PDE for χ(x, t):
A k2 + β 2
 
Tmax /Tπ = gµB , (16)
 
1 1
− ∂x2 χ + (c0 + c1 ) x2 + c1 xt χ = i ∂t χ, (10) k
2 2
which is about 2000 for a 1 T field. This means that the
with c0 = A k 2 and c1 = β 2 . Terms which depend only Bloch vector is allowed to precess approximately 1000
on t have been dropped, as they merely contribute global, times before the drag effect destabilizes the qubit confine-
time-dependent phases that do not affect the dynamics. ment. In the limit of weak fields, this ratio is no smaller
A Gaussian solution of (10) can be found with a further than gµB A k ∼ 800. Therefore, whilst it is possible that
ansatz higher orbital states are excited as a consequence of the
interaction, there is little danger of the electron escap-
χ(x, t) = exp f1 (t) x2 + f2 (t) x + f3 (t) .
 
(11) ing the SAW quantum dot during the operation of the
gate. We now turn to the behavior of δ(t), the width of
This system of time-dependent functions can be deter- the Gaussian wavefunction, which in the limit c1 ≪ c0
mined self-consistently, assuming a Gaussian groundstate oscillates according to the simple expression
of the SAW dot for the initial condition at t = 0. The s √
resulting expressions for f1 , f2 , f3 are complicated, but c0

c1 √
1/2
by noting a few general properties of the solution we can δ(t) = 1+ cos2 ( c0 + c1 t) , (17)
understand all the important features of the dynamics. 2(c0 + c1 ) c0
f3 (t) takes account of the normalization but is otherwise
of no more interest. f1 (t) and f2 (t) together describe a with c0 = A k 2 and c1 = β 2 . The frequency of the oscilla-
groundstate Gaussian wavefunction evolving in the time- tion increases with the energy of the dot and is typically
dependent vector potential. The solution is particularly much faster than the Zeeman spin precession frequency,
simple in that it remains Gaussian throughout, so we will i.e., rate of precession of the qubit Bloch vector. These
only need to keep track of the position of the central peak results are plotted for a specific case in Fig. 4. The top
µ and the width (standard deviation) δ of the probability two plots show the evolution of µ and δ. As the Bloch
distribution |χ(x, t)|2 : vector rotates about the field along the y-axis, the Gaus-
sian translates in the x direction with a rapidly oscillating
1 Re[f2 (t)] width. The bottom plot of Fig. 4 shows probability am-
µ(t) = − (12) plitudes of excitation into higher-energy simple harmonic
2 Re[f1 (t)]
oscillator (SHO) modes in the x direction of the quantum
1p
δ(t) = −Re[f1 (t)]. (13) dot.
2 Our conclusions are as follows. In addition to rotat-
The typical energy scales for the SAW electron encoun- ing the Bloch vector as required, the gate has the ef-
tered in experiment puts the system in the regime where fect of displacing the center of the Gaussian wavefunction
s ≪ 1 and c1 ≪ c0 . Expanding f1 and f2 to first order in which will increase the energy of the bound state by an
s and c1 /c0 , we obtain the asymptotic behavior for the amount depending on both the gate time and the field
position of the peak: strength. Higher-energy orbital states are likely to be
excited, which could lead to decoherence via spin-orbit
−c1 couplings or dipole coupling to phonons and photons.
µ(t) → t. (14) According to Fig. 4, the electron is almost completely
c0 + c1
out of the groundstate after a π/2 rotation of the Bloch
From this result we see that the magnetic field introduces vector. In the extreme case after many spin precessions
a constant drift velocity of the peak in the −x direction. (typically thousands), the qubit will leave the SAW quan-
On exiting the interaction region, the peak will be off- tum dot. However, for quantum computations the gate
center with respect to the SAW dot and the probability time need only be long enough to conduct a single or-
distribution will subsequently oscillate in the x direction, bit of the Bloch sphere, in which case the excitation into
perhaps exciting higher-energy orbital states of the dot. higher orbital states is negligible. These observations re-
If the charge distribution is pulled too far off-center, it veal some important features of the gate operation that
is likely to escape the SAW quantum dot. We could ask are not revealed by an idealized spin-only model of the
how long the electron can remain in the field before it is qubit.
 6

C. Qubit rotation: Uniform perpendicular

magnetic field

To move the qubit state to an arbitrary point on the

Bloch sphere a second axis of rotation on the Bloch
sphere is needed, and to this end we consider a gate im-
plemented by a uniform magnetic field in the z direction.
This will not be just a trivial extension of the preceding
analysis, since the 3D rotation symmetry is broken by
the motion of the SAW in the x direction. In practice,
the y magnetic field may be easier to fabricate than the
z magnetic field, since the latter passes perpendicularly
through the 2DEG structure. It could feasibly be pro-
duced by layering oppositely aligned thin-film magnets
just beneath and just above the 2DEG, or alternatively
by applying a global z magnetic field which is shielded in
regions where it is not needed.
A vector potential generating the uniform Bz magnetic
field is

Ax (y) = −βy, (18)

with Ay = Az = 0. It should be noted that although

the field is uniform and static in the laboratory frame,
the electron sees a moving uniform field. In the electron
rest frame, Ax is time-independent and z-independent,
allowing a Hamiltonian in x and y only. The chosen gauge
facilitates numerical simulations which follow shortly. An
effective scalar potential that is quadratic in y arises from
the A2 term in the Hamiltonian (6). It is interesting to
compare the strength of this confining potential with the
Q1DC potential, which is also approximately quadratic
in y:

Energy of A2 term β2
= . (19)
Energy of Q1DC term V0 /w2

With magnetic fields of order 1 Tesla and typical Q1DC

energies, this ratio is of the order unity. This means that
for the parameter values being considered the effective
scalar potential arising form the A2 term is compara-
FIG. 4: Gaussian evolution of the probability distribution ble to the Q1DC confinement potential. A stronger field
|χ(x, t)|2 in a constant magnetic field By . The top plot shows would start to significantly deform the shape of the dot.
peak position µ and the middle plot shows the width δ. The
This will not be a problem as long as the evolution has
bottom plot shows the probability amplitudes |Cn |2 of the nth
SHO mode in the x direction: |C1 |2 (long dash), |C2 |2 (short occurred adiabatically during the deformation. However,
dash). The plot also shows 1−|C0 |2 (solid). Typical values for as we shall discuss in a moment, the probability of exci-
the SAW dot are taken: A = 40 meV; λ = 1.0 µm; magnetic tation into higher orbital states due to the Bz field is not
field of 1 T. T = π/gµB β ≈ 0.08 ns is the time taken for a negligible.
π rotation of the Bloch vector about the y-axis. The initial The other interesting term in the Hamiltonian is the
Gaussian wavefunction was taken to be the SAW quantum asymmetric coupling −iβy ∂x between the x and y vari-
dot groundstate (C0 = 1). In the limit c1 /c0 ≪ 1 (weak ables. This is expected to introduce rotational behavior,
field), µ(t) moves approximately linearly
√ and δ(t) undergoes as we would intuitively anticipate some form of Landau
bounded oscillations about δ(0) = s/ 2. orbital motion due to the Lorentz force.
A numerical simulation implementing a Crank-
Nicholson,38 finite difference algorithm (alternating di-
rection method) was used to simulate the operation of the
gate. We started with an initial Gaussian groundstate of
the dot with spin state | ↑x i and subjected it to 1 T of
magnetic field for a duration of Tπ/2 = π/2gµB β, which
 7

is the gate time required for a π/2 rotation of the Bloch

vector about the z axis. The evolution of the Bloch vec-
tor is simple due to the uniformity of the magnetic field:
nx (t) = cos(gµB βt), ny (t) = sin(gµB βt), and nz (t) = 0.
All other parameters were assigned those values given
just after Eq. (1). The result of the simulation is shown
in Fig. 5, which shows time-shots of the probability dis-
tribution in the 2DEG (xy) plane. Losing its initial el-
liptic contours, the distribution develops two lobes which
rotate about its midpoint. The density at the center in-
creases due to the spatial squeezing from the A2 term
in the Hamiltonian. This clearly shows excitation into
higher-energy orbital states.
Using perturbation theory to second order in the
strength of the field β, with harmonic oscillator modes
as the basis, we found the following facts: (i) To sec-
ond order of perturbation theory, only the second excited
state with SHO quantum numbers nx = ny = 1 becomes
populated; (ii) The probability ratio with respect to the
groundstate is
 C11 2 2
 
  = β 2 ωx sin [t (ωx + ωy )/2] , (20)
 C00  ωy (ωx + ωy )2
where ωx and ωy are frequencies arising from the har-
monic oscillator approximation; (iii) The amplitude of
the ratio approaches unity when the field approaches 1 T;
(iv) The ratio of the spin precession frequency to the fre-
quency of the |C11 | oscillation is about 1.6 × 10−5 for
typical SAW parameters (see just after Eq. (1)). FIG. 5: Contour plots of electron probability density during
The above model is useful in assessing the robustness the operation of a qubit rotation gate. This is a numerical
of the qubit and its susceptibility to decoherence due to simulation in the xy (2DEG) plane of a SAW electron under-
orbital motion. Population into this excited state is a going a π/2 rotation gate about the z axis, under a uniform
problem for decoherence, since the oscillating charge in magnetic field Bz of strength 1 T. Snap-shots are shown at
the dot couples via dipole interactions to phonons and the following times (ps): (a) 0, (b) 2.89, (c) 11.6, (d) 20.3,
other charges outside the dot. However, provided we (e) 28.9, (f) 37.6. The initial probability distribution is a
Gaussian groundstate with standard widths σx = 26.0 and
have sufficient control of the gate time, we can use (20)
σy = 20.6. The probability density rotates about the z axis
to make the electron exit from the gate in its ground- under the influence of the Lorentz force acting in the xy plane.
state. Otherwise, a gate driven by radiofrequency pulses Moreover, an effective scalar potential contributed by the A2
in the presence of a global magnetic field could provide term in the Hamiltonian is quadratic in y, which spatially
an alternative means to implement spin rotations about squeezes the initial Gaussian distribution into the center of
the z axis. the dot.

IV. SINGLE-QUBIT INITIALIZATION AND Rabi oscillations.42,43 However, recent theoretical results
MEASUREMENTS by Stace and Barrett44 argue the absence of coherent os-
cillations in a continuously measured current noise, con-
In addition to single-qubit rotation gates, we require trary to previous results and assumptions,45,46 and there-
the ability to initialize and measure qubits at the be- fore raise concerns about the measurability of charge os-
ginning and end of the computation. Spin-polarized cillations in similar scenarios. In any case, it is difficult to
electrons can be obtained from injection through a apply these methods to qubits in moving quantum dots.
ferromagnetic contact.39,40 There is also a method to We therefore turn towards a quite different approach,
polarize spin using nondispersive phases (Aharonov- one which enables the initialization and readout of SAW
Bohm and Rashba) without the need for ferromag- electron spin qubits solely with the aid of nanomagnets
netic contacts.41 In the field of quantum computing, a and ohmic contacts.5 The readout gate we consider is
well-known method for achieving readout of solid state based on the Stern-Gerlach effect.47,48,49 In the 1920s
spin qubits is to convert spin information into charge Bohr and Pauli asserted that a Stern-Gerlach measure-
information6 and subsequently use single-electron tran- ment on free electrons was impossible,50,51 using argu-
sistors or point contacts to detect charge displacements or ments which combined the concept of classical trajecto-
 8

ries and the uncertainty principle. This subsequently led

physicists to analyze single-electron Stern-Gerlach mea-
surements within increasingly more rigorous quantum
settings, ultimately ending the debate by showing that
the measurement can indeed be done, albeit with cer-
tain caveats.52,53,54 Thus Stern-Gerlach measurements on
free electrons have been extensively investigated, but lit-
tle attention has been given to such measurements on
confined electrons.55 An interesting semiclassical analy-
sis of a Stern-Gerlach type experiment with conduction
electrons has been reported,56 in which the authors jus-
tifiably neglect the Lorentz force effects. In contrast, we
analyze a single-electron Stern-Gerlach device, providing
a full quantum mechanical treatment and including all
Lorentz force effects.
In the SAW electron system, single electrons are trans-
ported in moving quantum dots. Can this system un-
dergo a Stern-Gerlach type measurement, and would it
be of any use in a scheme for quantum computation using FIG. 6: Schematic diagram of the spin readout/polarizing de-
qubits trapped in surface acoustic waves? vice based on the Stern-Gerlach effect: SAW propagates from
left to right transporting a single electron in a moving quan-
The electron confinement to low dimensions allows us tum dot. Q1DC relaxes with a gradient tan(θ) to partially
to guide the electron through the magnetic field in ways delocalize the particle during the gate operation; C↑ (C↓ ) la-
that enhance the spin measurement and suppress the bels the Q1DC receiving electron flux in the spin up(down)
deleterious effects of delocalization and the Lorentz force. state of the σy operator; Magnets (of any geometry and polar-
Surface magnets can be arranged in such a way as to pro- ity) need to produce a localized, inhomogeneous distribution
duce a local magnetic field inhomogeneity. For example, of magnetic field.
two north poles placed on either side of the Q1DC will
produce a region of intense magnetic field gradient in-
between (see Fig. 6). Via the Zeeman interaction term A. Stern-Gerlach gate using a linearly
∝ σ · B in the Pauli Hamiltonian, the field inhomogene- inhomogeneous field
ity has the effect of correlating the spatial location of a
wavepacket to its spin state.57 In most situations it is In the first model we will analyze a simple, unidirec-
necessary to continue confining the qubit during the op- tional and linearly inhomogeneous field pointing in the y
eration of the gate, because the spreading time68 for a direction:
free Gaussian wavefunction is comparatively short – on
the order of 1 ns. But the gate must cause a wavepacket By = −β y, (22)
splitting in order for the spin states to be resolved, hence
the Q1DC must relax to allow for motion in the y di- with Bx = Bz = 0. A wedge-shaped single domain sur-
rection. This is achieved by patterning the Q1DC in a face magnet of appropriate dimensions can produce an in-
funnel shape, with an opening angle θ (see Fig. 6), such homogeneous magnetic field of this form near its center of
that in the electron rest frame the potential looks like symmetry. A vector potential for this field is Ax = −β yz,
with Ay = Az = 0. This field exerts a spin-dependent
force in the y direction and it is the simplest field that in-
VQ1DC = V0 y 2 /2 [w + tan(θ)u(t)t]2 , (21) duces the Stern-Gerlach effect. Although a z dependence
enters into the vector potential, by considering motion
only in the z = 0 plane we may avoid contributions from
where u(t) is the step function: u(t < 0) = 0 and terms involving A in the Hamiltonian, as well as the z
u(t ≥ 0) = 1. A negatively biased surface gate placed on component of magnetic field. The absence of x in the
the x axis can be used to guide the electron into the chan- potential immediately allows us to write down harmonic
nels. If necessary, both the position of the electrode and oscillator modes for the x dependence. The remaining
the opening angle θ could be optimized for a particular (y, t) dependent part obeys
sample device under low-temperature and high-vacuum  
conditions by erasable electrostatic lithography.58 We 1 2 1
− ∇ − gµB βsy y + VQ1DC (y, t) − i ∂t ψ(y, t) = 0
will study the quality of the readout obtained from the 2 2
gate for two different magnetic field configurations, both (23)
sufficiently simple so as to be realizable in the near fu- where sy is the eigenvalue ±1 of the spinor |sy i, and
ture: the linearly inhomogeneous field and the 2D dipole VQ1DC is given by (21). The initial state is again the
field. Gaussian ground state of the Q1DC. A solution to (23)
 9

is obtained by a time-dependent Gaussian ansatz (11) to the Q1DC broadening, allowing a small current to be
detected in the wrong channel C↓ . Thus this gate is a
ψ(y, t) = exp f1 (t) y 2 + f2 (t) y + f3 (t) .
 
(24) spin polarizing filter with some intrinsic error rate, inde-
pendent of decoherence effects, shot noise and Johnson
In a similar way as before, we solve the system of cou- noise. It remains for us to judge how well it performs,
pled ordinary differential equations and derive the time and to this end we introduce a fidelity measure of the
dependence of the standard deviation δ(t) and the posi- gate.
tion µ(t) of the probability distribution.
Let c0 = V0 /w2 and c1 = gµB βsy . The width of the
initial Gaussian groundstate wavefunction is determined B. Fidelity of the Stern-Gerlach gate with linearly
−1/4
by s = c0 . The behavior of the Gaussian solution is inhomogeneous field
then characterized by the following time-dependent pa-
rameters As an input to the Stern-Gerlach gate, we consider a
τ √ density matrix of the form
δ(τ )2 = − 4 c0 + α2 cosh( γ ln(τ )/α)

√
2 γ c0
√ √  ρ(y, y ′ ) = g(y) g(y ′ )∗ ⊗ σ, (31)
−α γ sinh( γ ln(τ )/α) (25)
√ where g(y) is the Gaussian groundstate of the dot and σ
τ c1  3/2 √
µ(τ ) = τ − cosh( γ ln(τ )/α) is the initial spin density matrix. In the inhomogeneous
2 (2 α2 + c0 )
field the spatial degrees of freedom couple to the spin
3α √ 
degrees of freedom, leading to
− √ sinh( γ ln(τ )/α) , (26)
γ X
ρ(y, y ′ , t) = gs (y, t) gs′ (y ′ , t)∗ σs s′ |sihs′ |. (32)
where we have introduced ancillary variables α =
s,s′ =↑,↓
tan(θ)/w, γ = α2 − 4 c0 and τ = 1 + α t. The parameter
γ is useful in determining when the trigonometric func- After a gate time t, we post-select the state in, say, the
tions become oscillatory.
√ If the angle θ is critical such C↑ channel and renormalize:
that tan(θ) = 2 V0 , then γ = 0 and the width has an
especially simple behavior: 1 X
ρ̃(y, y ′ , t) = σs s′ |sihs′ | ×
N (t) ′s,s
 
2 τ 1 2
δ(t) = 1 − ln τ + (ln τ ) . (27) Z
α 2 dz dz ′ E↑ (y, z) gs (z, t) gs′ (z ′ , t)∗ E↑ (z ′ , y ′ ), (33)
This model predicts the width and position of the qubit
wavefunction after a time t in the linearly inhomogeneous The probability yield in the channel is the normalization
field with field gradient β. In practice we cannot produce N (t) of the post-selected state. The simplest projection
arbitrary magnetic fields in arbitrary configurations. For kernel projecting into the support of C↑ is E↑ (y, y ′ ) =
a weak magnetic field such that c0 ≫ c1 ≫ α, δ(y − y ′ ) u(y), where u(y) is the step-function defined in
r (21). We finally define our fidelity as
τ t
r
Z
δ(τ ) = √ = δ(0) 1 + tan(θ) (28)
2 c0 w F↑ = dy h↑ |ρ̃(y, y, t)| ↑i (34)
√
√ c0
 
c1 σ↑↑
Z ∞
µ(τ ) = τ 2 − τ cos ln τ . (29) = |g↑ (y, t)|2 dy. (35)
2 c0 α N (t) 0
√
The width increases as ∼ t and the position moves The yield N (t) can be calculated without explicit knowl-
as ∼ t2 as expected. We may compare δ(t) with the edge of g↓ , since the Gaussian solutions are related by
dispersion of a Gaussian in free space reflection symmetry: g↑ (y, t) = g↓ (−y, t). We choose var-
s ious magnetic field gradients ranging from 0.1 T µm−1 to
t2 100 T µm−1 and plot the output fidelity of a maximally-
δ(t) = δ(0)2 + , (30)
4 δ(0)2 mixed input state as a function of the Q1DC angle θ and
the gate time t (see Fig. 7). Perfect spin filtering corre-
which is linear in t at large times. The broadening of the sponds to a fidelity F = 1, and the worst-case fidelity is
width is suppressed due to the confinement potentials. F = 1/2. The density plots show how the optimal param-
The effect of the Stern-Gerlach can be pictured as fol- eter regions increase with larger magnetic field gradients.
lows. Consider a qubit in the state | ↑y i entering the gate. This model is useful in determining the appropriate gate
The Q1DC broadens as the qubit enters the field, allow- geometry and field gradients required to achieve good fi-
ing the Stern-Gerlach effect to produce a spin-dependent delities.
shift in the center of mass towards the channel C↑ . How- The technology of nanoscale single domain magnets
ever, the initial localized distribution will deloclize due is not yet capable of producing large fields in arbitrary
 10

FIG. 8: Vector field plot in the xy (2DEG) plane showing

the magnetic field due to an infinite string of dipole moments
running parallel to the z-axis at a distance d from the Q1DC.

so that the qubit moves in the region −T /2 < x < T /2

in the laboratory frame. The question of how we are to
FIG. 7: Fidelity F↑ in the C↑ channel for a Stern-Gerlach shield the extraneous field from the qubit does not have
gate operated by a linearly inhomogeneous and unidirectional
a simple solution. It has been suggested that a supercon-
magnetic field in the y direction. White regions ⇒ F↑ = 1. θ
is the angle between the Q1DC and SAW direction. t is the ducting material such as Niobium could be used to shield
gate time. Input state is a maximally-mixed spin state in the the magnetic field where it is not needed.5
groundstate of the SAW quantum dot. Various magnetic field The results of a 1D simulation of the wavefunction
gradients have been chosen to show the increase in the optimal along the x = z = 0 direction are shown in Fig. 9. The
parameter region (white) in the (t, θ) plane with increasing numerical method used was a Crank-Nicholson algorithm
field gradient. tmax ≈ 100 ns is a crude order of magnitude adapted to the Pauli Equation. The plot shows three
estimate for the GaAs spin relaxation lifetime. time-shots of the probability density |ψ(y)2 |, normalized
to peak. The initial state is a Gaussian groundstate with
spin state | ↑y i. The simulation shows no significant spa-
configurations. The above simple linearly inhomogeneous tial displacement of the electron density in the y direc-
magnetic field is therefore a simple approximation to the tions, in contrast to the behavior exhibited under the lin-
real magnetic field configuration on any particular device. early inhomogeneous field (c.f. Eq. (26)). This is partially
In the next subsection we consider the effect of a different due to the effect of the Lorentz force generated by the A2
(perhaps more realistic) field configuration, namely, the term in the Hamiltonian, which contributes an effective
field near one of the poles of a dipole magnet. potential that is highly confining in the y direction. The
A2 confinement is so strong that the Q1DC opening an-
gle θ becomes irrelevant. Bohr and Pauli’s claim that
C. Stern-Gerlach gate with dipole field the Lorentz force washes out the Stern-Gerlach effect are
upheld in this particular scenario.
In this subsection we study the quality of the readout Another source of problems for this field configuration
obtained from a Stern-Gerlach type gate driven by a sim- is that the | ↑y i, | ↓y i states are not eigenstates of the field
ple 2D dipole field lying in the 2DEG plane, as shown in as they were with the linearly inhomogeneous field. Fig-
Fig. 8. In Appendix A, we derive the magnetic vector ure 10 shows the evolution of the Bloch vector in a spin-
potential only model of the electron spin qubit. It demonstrates
how the Bloch vector begins precessing about the x com-
βx ponent of the magnetic field, causing the Stern-Gerlach
Az = , (36)
x2 + (y − d)2 force to act in different directions. Although careful tim-
ing of the gate and possible compensation mechanisms
where β is a strength parameter and d is the distance of could be put in place, this gate is essentially not robust.
the 2D dipole from the Q1DC. We draw the conclusion that in parameter regimes rele-
Figure 8 shows that the gradient of the field in the y di- vant to single-electron transport by SAW in GaAs, a field
rection reverses direction at two points, x = ±d. In order of this type will not be suitable for a Stern-Gerlach mea-
to prevent the moving qubit from experiencing opposite surement gate, and that field unidirectionality is gener-
field gradients, we will restrict the gate time to T = 2 d ally an important requirement for quantum gates driven
 11

impurities;31,59 the coupling of qubits to phonons,60,61

nuclear spins7 and radio-frequency photons62 .
The models we have presented here give rise to behav-
ior which was previously neglected in the original pro-
posal for the quantum computation scheme.5 They allow
us to predict the contribution to decoherence from ex-
citation into higher orbital states and give their proba-
bilities of occupation. Any potential decoherence due to
tunneling between neighboring dots is negligible, and if
needed the spatial separation of qubits can be increased
by either introducing higher harmonics of the SAW fun-
damental frequency or by active gating at the entrance to
each Q1DC. A major source of error is the current fluctu-
ations in the quantized single-electron SAW current. In
recent experiments,32,33 error in the current quantization
FIG. 9: Three time-shots (see legend) showing the spatial
electron probability density (normalized to peak). y is the di-
(including shot and Johnson noise) of less than 0.1% were
rection transverse to the Q1DC. This is a Stern-Gerlach type observed.
readout gate using the 2D dipole field (1 Tµm−1 at x = y = 0) The relevant decoherence timescales that underpin our
shown in Fig. 8. Initial condition is a Gaussian wavepacket proposals is the spin lifetime (T1 and T2 ) of single elec-
with spin | ↑y i. Sideways translation of the probability density trons in quantum dots. For n-type bulk semiconduc-
is suppressed due to both the strongly confining effective po- tors, T1 spin lifetimes in GaAs of 100 ns have been
tential arising from the A2 term and the precessional motion reported, which gives a very crude ball-park figure.63
of the Bloch vector in the non-unidirectional field. An estimate that is more relevant to the SAW electron
system is the spin relaxation lifetime of a few nanosec-
onds, which was obtained from spin-resolved micropho-
toluminescence spectroscopy measurements on photoex-
cited electrons.64 However, this is the worst-case scenario
because the method of electron capture we propose5 pro-
duces a conduction-band hole that is more extended,
short-lived and mixed than the photoexcited hole. In-
deed, recent experiments on static GaAs quantum dots
report a lower bound on T1 as long as 50 µs,65 which
makes an estimate of 100 ns for the T1 lifetime of SAW
electrons quite reasonable in the light of current technol-
ogy.

FIG. 10: (Color online) Evolution of the Bloch vector in a 2D VI. SUMMARY
dipole field (1 Tµm−1 at x = y = 0): hσx i (long-short dash),
hσy i (solid), hσz i (short dash). Time t = 0 corresponds to the We have analyzed in detail a proposal5 for imple-
starting point x = −d (see Fig. 8). We have used a spin-only
menting quantum computation on electron spin qubits
model of the electron spin qubit to illustrate spin precession
in the non-unidirectional field. The Bloch vector changes di- trapped in SAW-Q1DC electrostatically-defined dots, us-
rection rapidly, which in turn causes the Stern-Gerlach force ing only static magnetic gates to perform single-qubit
to fluctuate during the operation of the gate. initializations, rotations and readouts. Applying the full
Pauli Hamiltonian, we described the quantum dynamics
of both the spin and orbital states of the qubit for various
by static magnetic fields. parameter regimes relevant in SAW single-electron trans-
port experiments. In the analysis of single-qubit uni-
tary operations with localized uniform magnetic fields,
we showed that the effect of the Lorentz force puts upper
V. DECOHERENCE EFFECTS bounds on field strengths and gate times. Moreover, sim-
ulations showed that field directions normal to the 2DEG
The major effects of decoherence on this SAW quantum excite rotational states in the dot. Probabilities of exci-
computing system have already been considered qualita- tation into higher-energy orbitals were given. In terms
tively in the original proposal and subsequent papers,5,31 of feasibility, the models indicate that a field strength of
and more specifically by a number of authors. These ef- about 80 mT will be sufficient to conduct a π rotation
fects include: The interaction of the electron qubit with in about 1 ns, without compromising the confinement
other electrons in the 2DEG, surface gates and donor properties of the trapped qubit. Since T1 spin lifetimes
 12

of microseconds have been reported,65 it is feasible for Stace for their ideas and helpful discussions. The authors
hundreds of single-qubits gates to operate before all co- gratefully acknowledge the support of the EPSRC. This
herence is lost in the computation. work was partly funded by the Cambridge-MIT institute.
We also studied a device for single-qubit measurement
and initialization based on the Stern-Gerlach effect. The
problematic Lorentz force can be partly suppressed by
virtue of the geometry of the 2DEG system, and for a
unidirectional, linearly inhomogeneous field, the correla-
tion between the spin states and spatial location of the APPENDIX A: DERIVATION OF THE 2D
qubit leads to a good quantum measurement of its spin. DIPOLE FIELD VECTOR POTENTIAL
For a 2D dipole field, the vector potential has a delete-
rious effect. Namely, it contributes an effective confin-
ing potential via the A2 term in the Pauli Hamiltonian We adopt cylindrical coordinates (r, φ, y) about the y
which suppresses the transverse motion in the y direc- √
axis, with r = x2 + z 2 . We consider the field due to
tion. Furthermore, the component of magnetic field in an infinite string of dipoles passing parallel to the z-axis
the x direction causes the Bloch vector to rotate in un- through y = d as shown in Fig. 8. The resulting mag-
desired directions. The latter problem could be over- netic field has planar symmetry with respect to the xy
come by using unitary gates prior to the Stern-Gerlach (2DEG) plane. The Q1DC runs along the x axis. A
gate to correct any spurious rotations, but it is unlikely vector potential for this field can be constructed as fol-
that such a finely-tuned arrangement can be made ro- lows. A magnetic dipole at y = z = 0 pointing in the +y
bust. Magnetic fields that are good approximations to a direction has a vector potential66
unidirectional and linearly inhomogeneous field can pro-
vide a source of polarized electron qubit states in chan-
nels C↑ and C↓ with high yield. The very same gate can µ0 mB dz r
be used to measure the ratio |α|2 /|β|2 for an input qubit A′φ (r, y) = , (A1)
state |ψi = α| ↑y i + β| ↓y i. The advantages of this read- 4π (r + y 2 )3/2
2

out method are that the averaging time of the measured

current can be made long enough to render most noise
sources (e.g. shot noise, input-referred noise of the cur- where A′r = A′y = 0 and mB is the dipole moment per
rent pre-amp) unimportant, and that missing electrons unit length in the z direction. We now displace this po-
from SAW minima do not contribute any error since we tential by d in the y direction and integrate over all z,
are measuring the ratio of currents out of the two chan- obtaining
nels. Field gradients in the range 0.1 − 100 Tµm−1 lead
to gate times that are easily within the range of T1 spin
lifetimes in GaAs. These parameter regimes will need to βx
Az = , (A2)
be probed by experiment if we are to make progress to- x2 + (y − d)2
wards feasible quantum computation with nanomagnets
in the SAW quantum computer.
with Ax = Ay = 0 and strength parameter β =
µ0 m′B /2π. The magnetic field can be derived by applying
Acknowledgements the curl in cylindrical coordinates: Br = −r−1 ∂y (rAφ )
and By = r−1 ∂r (rAφ ). Figure 8 shows the vector field
The authors would like to thank Sean Barrett, Masaya plot of the resulting magnetic field, which falls off as r−2
Kataoka, Alexander Moroz, Andy Robinson and Tom for large r.

∗
Also affiliated to: Centre for Quantum Computation, De- 2000).
5
partment of Applied Mathematics and Theoretical Physics, C. H. W. Barnes, J. M. Shilton, and A. M. Robinson, Phys.
Wilberforce Road, Cambridge CB3 0WA, UK; email: Rev. B 62, 8410 (2000).
6
[email protected] D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120
1
R. Feynman, Int. J. Theor. Phys. 21, 467 (1982). (1998).
2 7
P. Shor, in Proceedings 35th Annual Symposium on Foun- G. Burkard, D. Loss, and D. P. DiVincenzo, Phys. Rev. B
dations of Computer Science (1994), pp. 124–134, quant- 59, 2070 (1999).
8
ph/9508027. D. Loss, G. Burkard, and E. Sukhorukov, Proceedings of
3
A. Ekert and R. Jozsa, Rev. Mod. Phys. 68, 733 (1996). the XXXIVth Rencontres de Moriond “Quantum Physics
4
M. Nielson and I. Chuang, Quantum Computation at Mesoscopic Scale” (1999), cond-mat/9907133.
9
and Quantum information (Cambridge University Press, F. Troiani, U. Hohenester, and E. Molinari, Phys. Rev. B
 13

37
62, 2263 (2000). N. J. Curson, R. Nemutudi, N. J. Appleyard, M. Pepper,
10
D. P. DiVincenzo, D. Bacon, J. Kempe, G. Burkard, and D. A. Ritchie, and G. Jones, App. Phys. Lett. 78, 3466
K. B. Whaley, Nature 408, 339 (2000). (2001).
11 38
J. M. Elzerman, R. Hanson, J. S. Greidanus, L. H. Willems J. Crank and P. Nicolson, in Proc. Cambridge Philos. Soc.
van Beveren, S. D. Franceschi, L. M. K. Vandersypen, (1947), vol. 32, p. 50.
39
S. Tarucha, and L. P. Kouwenhoven, Phys. Rev. B 67, S. Datta and B. Das, Appl. Phys. Lett. 56, 665 (1990).
40
161308(R) (2003). Y. Ohno, D. K. Young, B. Beschoten, F. Matsukura,
12
T. Calarco, A. Datta, P. Fedichev, E. Pazy, and P. Zoller, H. Ohno, and D. D. Awschalom, Nature 402, 790 (1999).
41
Phys. Rev. A 68, 012310 (2003). R. Ionicioiu and I. D’Amico, Phys. Rev. B 67, 041307(R)
13
J. M. Shilton, V. I. Talyanskii, M. Pepper, D. A. Ritchie, (2003).
42
J. E. F. Frost, C. J. B. Ford, C. G. Smith, and G. A. C. C. I. Pakes, V. Conrad, J. C. Ang, F. Green, A. S. Dzurak,
Jones, J. Phys. Condens. Matt. 8, L531 (1996). L. C. L. Hollenberg, D. N. Jamieson, and R. G. Clark,
14
V. I. Talyanskii, J. M. Shilton, M. Pepper, C. G. Smith, Nanotechnology 14, 161 (2003).
43
C. J. B. Ford, E. H. Linfield, D. A. Ritchie, and G. A. C. M. Friesen, C. Tahan, R. Joynt, and M. A. Eriksson, Phys.
Jones, Phys. Rev. B 56, 15180 (1997). Rev. Lett. 92, 037901 (2004), cond-mat/0304422.
15 44
V. I. Talyanskii, J. M. Shilton, J. Cunningham, M. Pepper, T. M. Stace and S. D. Barrett, Phys. Rev. Lett. 92, 136802
C. J. B. Ford, C. G. Smith, E. H. Linfield, D. A. Ritchie, (2004), cond-mat/0309610.
45
and G. A. C. Jones, Physica B 251, 140 (1998). H.-S. Goan and G. J. Milburn, Phys. Rev. B 64, 235307
16
R. Vrijen, E. Yablonovitch, K. Wang, H. W. Jiang, A. Ba- (2001).
46
landin, V. Roychowdhury, T. Mor, and D. DiVincenzo, A. N. Korotkov, Phys. Rev. B 63, 085312 (2001).
47
Phys. Rev. A 62, 012306 (2000). W. Gerlach and O. Stern, Z. Phys. 9, 349 (1922).
17 48
B. E. Kane, Nature 393, 133 (1998). A. Challinor, A. Lasenby, S. Gull, and C. Doran, Phys.
18
Y. Makhlin, G. Schon, and A. Shnirman, Rev. Mod. Phys. Lett. A 218, 128 (1996).
49
73, 357 (2001). A. Venugopalan, D. Kumar, and R. Ghosh, Physica A 220,
19
D. J. Wineland, M. Barrett, J. Britton, J. Chiaverini, 576 (1995).
50
B. L. DeMarco, W. M. Itano, B. M. Jelenkovic, C. Langer, Mott and Massley, The Theory of Atomic Collisions (Ox-
D. Leibfried, V. Meyer, et al., Philos. Trans. R. Soc. Lon- ford University Press, 1965), chap. IX.1.
51
don Ser. A 361, 1349 (2003), quant-ph/0212079. W. Pauli, in Proceedings of the Sixth Solvay Conference
20
E. Knill, L. Laflamme, and G. J. Milburn, Nature 409, 46 (Gauthier-Villars, Brussels, 1932), pp. 183–186.
52
(2001). H. Batelaan, T. J. Gay, and J. J. Schwendiman, Phys. Rev.
21
J. I. Cirac and P. Zoller, Nature 404, 579 (2000). Lett. 79, 4517 (1997).
22 53
A. Bertoni, P. Bordone, R. Brunetti, C. Jacoboni, and B. M. Garraway and S. Stenholm, Phys. Rev. A 60, 63
S. Reggiani, Phys. Rev. Lett. 84, 5912 (1999). (1999).
23 54
A. E. Popescu and R. Ionicioiu (2003), cond-mat/0306401. G. A. Gallup, H. Batelaan, and T. J. Gay, Phys. Rev. Lett.
24
I. L. Chuang, N. Gershenfeld, M. G. Kubinec, and D. W. 86, 4508 (2001).
55
Leung, Proc. R. Soc. London A 454, 447 (1998). H. Dehmelt, Science 247, 539 (1990).
25 56
D. G. Cory, A. F. Fahmy, and T. F. Havel, J. Fabian and S. DasSarma, Phys. Rev. B 66, 024436
Proc. Natl. Acad. Sci. 94, 1634 (1997). (2002), cond-mat/0104146.
26 57
J. A. Jones, Phil. Trans. R. Soc. Lond. A 361, 1429 (2003). A. Peres, Quantum theory: concepts and methods (Kluwer,
27
R. Ionicioiu, P. Zanardi, and F. Rossi, Phys. Rev. A 63, 1993).
58
050101(R) (2001). R. Crook, A. C. Graham, C. G. Smith, I. Farrer, H. E.
28
M. Kitagawa and M. Ueda, Phys. Rev. Lett. 67, 1852 Beere, and D. A. Ritchie, Nature 424, 751 (2003).
59
(1991). S. D. Barrett and G. J. Milburn, Phys. Rev. B 68, 155307
29
M. J. Kelly, Low-dimensional semiconductors : materi- (2003).
60
als, physics, technology, devices (Oxford Clarendon Press, V. N. Golovach, A. Khaetskii, and D. Loss (2003), cond-
1995). mat/0310655.
30 61
P. Utko, K. Gloos, J. B. Hansen, and P. E. Lindelof, Acta Y. G. Semenov and K. W. Kim, Phys. Rev. Lett. 92,
Phys. Pol. A 103, 533 (2003). 026601 (2004).
31 62
A. M. Robinson and C. H. W. Barnes, Phys. Rev. B 63, Y. Yamamoto and A. Imamoglu, Mesoscopic Quantum Op-
165418 (2001). tics (John Wiley, 1999), ISBN 0471148741.
32 63
A. M. Robinson, V. I. Talyanskii, M. Pepper, J. E. J. M. Kikkawa and D. D. Awschalom, Phys. Rev. Lett. 80,
Cunningham, E. H. Linfield, and D. A. Ritchie, in 4313 (1998).
64
Inst. Phys. Conf. Ser. (2003), vol. 171. T. Sogawa, P. V. Santos, S. K. Zhang, S. Eshlaghi, A. D.
33
A. M. Robinson, V. I. Talyanskii, M. Pepper, J. E. Cun- Wieck, and K. H. Ploog, Phys. Rev. Lett. 87, 276601
ningham, E. H. Linfield, and D. A. Ritchie, Phys. Rev. B (2001).
65
65, 045313 (2002). R. Hanson, B. Witkamp, L. M. K. Vandersypen, L. H.
34
W. M. Kaminsky, G. A. C. Jones, N. K. Patel, W. E. Booij, Willems van Beveren, J. M. Elzerman, and L. P. Kouwen-
M. G. Blamire, S. M. Gardiner, Y. B. Xu, and J. A. C. hoven, Phys. Rev. Lett. 91, 196802 (2003).
66
Bland, Appl. Phys. Lett. 78, 1589 (2001). J. D. Jackson, Classical Electrodynamics (John Wiley,
35
R. Nemutudi, C. G. Smith, C. J. B. Ford, N. J. Appleyard, 1975), p. 194, 2nd ed.
67
M. Pepper, D. A. Ritchie, and G. A. C. Jones, J. Vac. Sci. In other words, we ignore the z dependence √ of the wave-
Tech. B 20, 2810 (2002). function. This is partly justified if e β ∆x/ ∆E m∗ ≪ 1,
36
R. Nemutudi, N. J. Curson, N. J. Appleyard, D. A. Ritchie, where ∆E is the characteristic energy gap between the
and G. A. C. Jones, Microelectron. Eng. 57-8, 967 (2001). 2DEG energy bands and ∆x is the length of the gate. This
 14

is indeed the case for parameter regimes that we consider. for its standard deviation to double.
68
The spreading time for a Gaussian state is the time taken