ECE 571: Seminar on Nano-devices and Technology

Quantum Computing with Quantum Dots

Shane Vitarana
Illinois Institute of Technology
Introduction
The trend of miniaturization in the semiconductor industry, in a few years, will
cause devices to start operating in the quantum regime, where bulk properties of materials
are no longer reliable. Current semiconductor technology will not scale to the levels
required for business productivity as predicted by Moores Law. While the
semiconductor industry strives to hinder quantum mechanical effects, researchers and
scientists are trying to exploit quantum mechanical effects to create new types of devices
and an entirely new type of computing. Quantum Computing is based on the principle
that, in the quantum regime, information can be represented in a superposition of states.
Therefore, the quantum analogue of a bit, or qubit, can represent both 1 and 0 at the
same time. So a quantum computation can be viewed as a superposition of digital
computations. Physically, a qubit can be represented by any two-state system such as a
spin-half particle (i.e.: an electron with up and down spins) or a two-level atom [4]. In a
two-level atom, an electron can be in a superposition of two different orbitals. This kind
of quantum parallelism paves the way for new types of algorithms which are not possible
to implement with classical computers.
There are currently many different physical realizations of quantum computers
being researched, including cold ion-traps, nuclear magnetic resonance (NMR), all-
optical gates, and Josephson junctions [5]. However, the implementation that will be
most feasible will probably be a solid-state solution, since most of the semiconductor
industry is based around this approach. One such solid-state implementation involves the
use of quantum dots. A two-state system utilizing the free valence electron in a quantum
dot can be used to represent a qubit. It has also been shown that a system based on
quantum dots is much more scalable than other implementations (Table 1) [1].

Technology Switching time (s) Decoherence time (s) Figure of merit Scalability
Ion traps 10-7 10-1 106 50 qubits
Cavity QED 10-14 10-5 109 2-5 qubits

!2
 NMR 10-3 104 107 10-50 qubits
Quantum Dots 10-9 10-6 103 1,000 qubits
Table 1: Comparisons of Quantum Computing Methodologies

Quantum Dots
Quantum Dots have gained a lot of attention in research circles recently due to
their strong potential for use in future technological applications. Including quantum
computing, many other application proposals have been made; such as laser emitters,
charge-storage devices, and fluorescent biological markers [2]. A quantum dot is small
structure, typically 30 nm in length, which can confine a single electron or more in
discrete energy levels. What makes quantum dots interesting is that they have atom-like
properties, thus sometimes being referred to as artificial atoms. Like atoms, quantum
dots have an energy band spectra that can be modeled in a similar fashion to the Bohr-
model of the atom. Quantum dots are generally fabricated as one or more electrons
between source and drain contacts, with a gate electrode. This structure allows for
current and voltage (I-V) measurements and enables the addition and removal of
electrons by varying the gate voltage. Thus one can scan through the entire periodic table
of artificial atoms simply by changing the gate voltage [12].
The energy needed to add an electron to a dot is called the addition energy, and is
analogous to the difference between the ionization potential and the electron affinity of a
real atom [17]. This addition energy is governed by the Coulomb repulsive forces within
the dot. Therefore, the addition energy for a dot with N+1 electrons is greater than the
addition energy for a dot with N electrons, a phenomenon known as the Coulomb
blockade. The gate voltage of the dot has to be increased by a value proportional to the
number of electrons contained in the dot and the number of new electrons being added.
The property of quantum dots which makes them ideal for semiconductor
quantum computing applications is their weak interaction with the environmental
obstacles, such as photons, phonons, and plasmons. Quantum dot spins also form a

!3
natural two-level system for performing quantum operations. Also, unlike in
conventional semiconductors where the charge carries are free to move, the charge
carriers in quantum dots are confined in all three spatial dimensions, making them easier
to manipulate. For example, laser pulses can be used to flip the electron between discrete
energy levels, or into a superposition of states between two energy levels. Also, magnetic
fields can be applied to alter the spin direction of the electron. Note that this cannot be
done with regular atoms because their lengths are too small to be effected by a typical
magnetic field created using current technology. Typical magnetic fields around 1 T
correspond to magnetic lengths of close to 10 nm, which is well into the scale of quantum
dots [9].

Quantum Computing
One motivation for the study of quantum computing is because it has been shown
that there is a class of problems which cannot be solved using a Turing machine, the
mathematical model of a classical computer. In terms of complexity theory, a classical
computer can only solve problems in the P domain efficiently, and needs certain
heuristics to solve problems that are NP-complete. However, a quantum computer has
the capability to solve problems that are in the Exponential domain, which includes the
class of NP-complete problems. In 1982, the famous American physicist, Richard
Feynman, proved that no classical Turing machine can simulate quantum phenomena
without an exponential slow down. A few years later, in 1985, David Deutsch of Oxford
University, proposed a new type of computing paradigm called a quantum Turing
machine (QTM) [14].
Armed with a new model of computing, in 1994, Peter Shor came up with an
algorithm to find the prime factors of an integer, a problem known as the discrete
logarithm problem which has no known efficient solution on a classical computer. The
entire field of public-key cryptography is based on the difficulty of classical computers to
solve this problem. If a quantum computer were to be constructed today, algorithms such
as RSA, which form the backbone of internet security, will be of no use. Another

!4
algorithm known as Grovers algorithm can potentially speed up the problem of
searching large databases. This algorithm will definitely be useful in an age where the
amount of information is growing at an exponential rate. Both of these algorithms show
that quantum computers are more powerful than Turing machines [13]. Theoretical
research has also been done on the outcomes of distributed quantum computation,
essentially trying to achieve the quantum analogue of how modern classical computers
work over networks. Studies show that quantum computers can require exponentially
less communication between computers to solve certain problems in a distributed
environment [13].
The theory of quantum computation is a mathematical abstraction for all quantum
computing implementations. Therefore, if a quantum computer is realized using
electron-spin or orbital-state configurations, the logical operations will be exactly the
same as long as qubits are achievable. It is worthy to note that all quantum operations are
probabilistic, meaning that the same algorithm can have two different outcomes if run
twice. However, this can be circumvented by running the same algorithm in parallel and
voting on the outcome.
Quantum operations are generally represented in Dirac notation, which is the
standard notation for representing states in quantum mechanics. Using this notation, a
qubit in a superposition of states, such as having a spin up component as well as a spin
down component, can be represented as:
= 0 + 1
! ,

where ! and ! are complex weighted amplitudes, allowing an infinite number of

superpositions. Therefore, this equation is essentially a vector in a complex two-

dimensional space. According to the laws of quantum mechanics, the values of ! and !
cannot directly be measured in a reliable way. A measurement of the above question
2 2

would result in a value of 0 with probability ! or a value of 1 with probability !
[13]. However, indirect measurements of qubit states are possible by manipulating and
transforming the qubit, as will be shown later. Since the sum of probabilities must sum to

!5
 2 2
+ =1
1, a normalizing condition! must be applied to each measured state. Also,
the act of measuring a qubit causes its wavefunction to collapse from a superposition of
states into its specific state corresponding to the outcome of the calculation. Thus, after
measuring a qubit, its state will always be the outcome of the first measurement.
A 3-qubit quantum register can store individual numbers such as 3 or 7,

011 = 3 111 = 7
represented as ! and ! where each bit in the sequence is associated
with one qubit in the register. However, a quantum register can also store both 3 and 7 at
the same time by creating a superposition of the two numbers. Since the only difference
between 011 and 111 is the left most significant bit, we can represent it as a superposition

1/ 2 ( 0 + 1 ).
of both 1 and 0 at the same time, or ! Therefore, the final content of the
register will be [15]:
1
( 0 + 1 ) 1 1 1 ( 011 + 111 ),
2 2
1
(3 + 7 )
! 2

Note that the ! 1/ 2 term arises from the normalizing condition mentioned above. Since
2 2
0 1 1 +1 = 2
both coefficients of ! and ! are 1, ! . The ability to put qubits into
superpositions of 0 and 1 is one of the most useful operations for a quantum computer. In
3
the above example, if all three qubits are in a superposition of 0 and 1, then ! 2 , or eight
possible combinations of numbers can be represented in one quantum register. Following
30
this logic, a 100 qubit register would be able to hold approximately ! 10 different
numbers simultaneously, which translates into 100 million trillion gigabytes of
information in classical terms [1].
The millions of transistors on chips today are components of the physical
representation of Boolean logic gates, such as AND, NOT, and OR. The NOT gate is the
simplest one of these and only operates on one bit at a time. In the classical world, the
NOT gate is the only gate which takes in one bit as an input, however, in the quantum

!6
world, there is a whole class of gates known as single-qubit gates which operate on only
one qubit.
0 + 1
One such gate is the quantum NOT gate, which translates ! into
0 + 1
! . Since a single qubit can be represented by a unit vector in the complex
plane, it can also be visualized as a 3D sphere with Cartesian coordinates, known as a
!
Bloch sphere [13]. Therefore, a quantum NOT operation would cause a 180! rotation of
the unit vector on the Bloch sphere. This is equivalent to the spin of an electron in a
quantum dot reversing direction. However, now imagine if the electron were
manipulated, such as by applying a large magnetic field, so it will be in a superposition of
!
spin up and spin down. Then this operation translates into a 90! rotation in the Bloch
sphere, which is exactly half of the NOT operation. This fundamentally important gate is
known as the square-root of NOT gate because two of these gates in succession will

produce a NOT operation, namely ! NOT NOT = NOT .

The quantum NOT operation can also be described using a matrix. For example,
if X is taken to represent the quantum NOT gate, and:
0 1
X
! 1 0 ,

then the output of the NOT gate is:

X =
! .
Similarly, all single qubit gates can be described by a two-dimensional unitary matrix
[13]. Phase components can also be included in the matrix to produce any kind of
rotation along the Bloch sphere. This representational form for single-qubits gates is very
important as it provides a mapping between the physical implementation of quantum dots
and logical quantum gates.
However, quantum algorithms cannot be designed with single-qubit gates alone
since they do not allow operations between different qubits. The most important multi-

!7
qubit gate is the controlled-NOT or c-NOT gate. The c-NOT gate simply forwards the
input to the output when the control bit is 0, and flips the output bit, also known as the
target bit, when the control bit is 1. Thus, the truth table for this gate resembles that of
the classical XOR operation. The quantum schematic symbol for this gate is shown in
Figure 1. The transfer of the input into the output in this gate may seem redundant, but it
is an essential part of quantum gates. All quantum gates must be reversible in order to be
physically implemented. Any loss of information in a quantum operation is analogous to
particle being lost and causing decoherence. Therefore, the operation of every quantum
gate can be reversed by applying another quantum gate.
In classical computers, the combination of the gates NOT and AND make up the

NAND gate, which is known as a universal gate because all logical operations can be
Figure 1: Controlled-NOT gate
created solely with various combinations of this gate. Similarly, all quantum logical
operations can be performed by concatenating c-NOT and single-qubit gates.
A quantum circuit which swaps the states of two qubits can be created with three
successive calls to a c-NOT gate. Figure 2 shows the circuit symbol for this gate, known
as the swap gate.

Figure 2: Swap gate

The following example illustrates the behavior of this gate [13]:
a, b a, a b
a (a b ), a b = b, a b
b, (a b ) b = b, a
! .

!8
As can be seen, a quantum gate is executed left to right. The lines in the circuit do not
represent wires, but are more like a representation of the path along time. It is also
worthy to note that the concept of fan-out is impossible in quantum logic due to the No
cloning rule of quantum mechanics.

Quantum Computer Implementation using Quantum Dots

The research community has come up with a set of five requirements which must
encompass a quantum computer: (i) the machine should have a collection of bits, (ii) it
should be possible to initialize all bits to zero before the start of each computation, (iii)
the error rate must be sufficiently low, (iv) it must be possible to perform elementary
logic operations between pairs of bits, and (v) reliable read-out of the final result should
be possible [8].
The most heavily researched type of quantum dot coupling is spin-based, mainly
because the decoherence times are much less for electron spin compared to electron
orbital states. As shown in Table 1, the decoherence times for quantum dot spins is in
the order of microseconds, while for electron orbital states it is in the order of
nanoseconds. Another reason that spin-based quantum computation is popular is because
a spin is a natural two-state system, compared to the orbital degree of freedom of a
confined electron, which can have more than two states since there can be more than two
orbitals. In quantum mechanical terms, this means that a spin qubit can only exist in the
two-dimensional Hilbert space, and there are no extra
dimensions into which the state can leak [9].
Figure 3 illustrates a spin-coupled quantum dot model,
first proposed by Loss and DiVincenzo [11] from the University
of California, Santa Barbara, and IBM T.J. Watson Research
Center. The two coupled quantum dots are labeled 1 and 2, each
Figure 3: Coupled
containing an electron (e) oriented according to its spin. The quantum dots

solid borders of the dots represent the state when the tunneling
barrier between the dots is lowered and no interaction takes place between the qubits.

!9
The dotted outline of the dots represent the state when the tunneling barrier is raised,
allowing a quantum mechanical phenomenon known as an exchange interaction to occur.
Exchange interactions occur due to a combination of Coulomb interactions and the Pauli
exclusion principle. The Pauli Exclusion principle states that no two electrons in an atom
can be in the same state and in the same configuration. This principle invokes restrictions
on the electron wavefunctions and, along with Coulomb interactions, forms a sensitive
window of time where there states of the electrons can be controlled. One unique
property of this model is that the tunneling barrier between the dots can be controlled by
varying gate voltages, rather than by spectroscopic manipulation as in other models [11].
The spins of each of the two qubits can be temporarily coupled via exchange
interactions to produce two-qubit quantum logic, such as the swap operation. Quantum
gates in this model are created by identifying Hamiltonians which generate the unitary

matrices mentioned earlier. For example, the exchange coupling ! J (t ) between the two
! !! !!! ! !!! !!
S S
spins, ! and ! , represented by the Heisenberg Hamiltonian!
1 2 Hs (t ) = J (t ) S1 S
! 2 , can be
iHt / !
used to create the unitary transformation ! U = e [16]. Essentially, the operation of a
quantum gate is reduced to the act of controlling the exchange interaction between the
quantum dots. For example, to produce the swap operation mentioned earlier, the
s
dtJ (t ) / != J 0 s / =! (mod 2 )
exchange coupling must be pulsed such that ! 0 . This is

associated with the unitary transformation !

( t
U (t ) = T exp i H s ( )d / !
0 )which
corresponds to the swap operation [9]. This representation of the exchange coupling is
a mathematical abstraction for any type of coupling. Each type of exchange coupling

implementation will have specific parameters which ! J (t ) will depend on.

This model by Loss and DiVincenzo can be scaled up to a full solid-state quantum
computer by having an array of quantum dots with each dot connected to its nearest
neighbor by a gated tunneling barrier, as shown in Figure 4 [10]. The dotted circles
represent the quantum dots with the direction of their free electron shown as an arrow.
The right two quantum dots are closer together to represent coupling between them.

!10
 Although much of the technology of this implementation is in the solid-state,
initialization of the qubits must still be done with a large magnetic field, as shown by the
arrow on the left. Once a field has been applied, the spins converge to the ground state,

0
or ! . Without having to incorporate bulky electro-magnetics into this implementation,
various proposals have been suggested to create this magnetic field in the solid-state.
One technique is to force the electrons into a magnetized layer as shown by the middle
gray layer. This technique is also shown in Figure 3, where the electron in dot 1 can
tunnel into an auxiliary insulating ferromagnetically ordered dot (FM), where it can be
initialized. Another approach is to force the electrons into a layer with polarized nuclear
spins. Due to the hyperfine coupling between nuclear spins and electron spins, a
phenomenon known as the Overhauser effect can force the electron spins into the ground
state [10].

Figure 4: Solid-state quantum dot array

The task of applying single-qubit rotations also requires a magnetic field, that has
to be pulsed onto the quantum dot. This is represented by the BAC wave at the left in
Figure 4. The length of this pulse determines what unitary transformation will be
applied, and consequently, which gate operation will be performed. As in the case of
initializing the dots, a magnetized layer may eliminate the need for an external magnetic
pulse. However, currently this can be achieved by using a scanning-probe tip [11]. The
layout of this array may make it seem like operations can only be performed between
adjacent dots. However, using the tunneling capability of electrons, the qubit can be

!11
transported to a place where it can be coupled with a second qubit. Using this
mechanism, each cell in the array need not contain the technology to apply magnetic
fields. Qubits can be transported to a special location where single-qubit rotations are

applied. The exchange coupling! J (t ) in this scheme is a function of the magnetic field,
the associated electric field, and the inter-dot distance.
The final hurdle to complete this implementation is to arrive at a method of
reading-out the value of a qubit or set of qubits. A straightforward method of
accomplishing this with great accuracy is to use a spin filter. A spin filter is a tunnel
barrier device which only allows either spin-up electrons or spin-down electrons to pass
through at one time [10]. For example, if the filter was configured to only allow spin-up
electrons to pass through, then a detector near the dot would register a charge for spin-up
electrons, and no charge for spin-down electrons. The spin filter is represented by dot 3
in Figure 3 which shows a tunneling barrier with a spin valve (SV). Recent
experiments have shown that a single-electron transistor (SET) can be used as a sensitive
electrometer to determine the charging energy of a quantum dot, shown by E in the same

figure [18].
Some of the parameters of this system, such as the pulse duration to achieve a
rotation, and the decoherence times, can be determined by numerical analysis. The

authors of this system state that, if an exchange constant of ! J 0 = 80 eV were achievable,

then a pulse duration of s = 25 ps with a magnetic field of 0.6 T, and a decoherence time
of 1.4 ns would be needed. These parameters have been shown to be achievable in solid-
state spin systems [11].
An issue of great concern to quantum computer designers is if all of these
operations can be carried out in room temperature. Experiments by Michler and
Imamoglu provide encouraging evidence. The team was able to observe photon
antibunching from a single cadmium selenide quantum dot at room temperature. For the
first time, it was shown that a single quantum dot can be manipulated, and single photon
emission events can be observed with great precision at room temperature [14].

!12
Challenges
The length of time a quantum dot spin can be coherent is one of the biggest
challenges facing the quantum computing community. There are many efforts underway
to increase the coherence/dephasing time of an individual quantum dot. A quantum dot
needs to be able to maintain its state long enough for computation to occur. Detailed
analysis has shown that the minimum acceptable decoherence time is 104 times the time it
takes for the execution of an individual quantum gate [16].
One of the reasons for decoherence is the hyperfine coupling between electron
spin and nuclear spins [10]. Theoretical research by Burkard states that this decoherence
can be controlled by creating an Overhauser field, which is a magnetic field that can
influence the polarization of nuclear spins [10]. Experimental results using transient
four-wave mixing (TFWM) has shown that decoherence times in excess of 1 ns are
attainable in quantum dots [7]. More recent magneto-optical experiments show that a
lower bound on the decoherence time can be as high as 100ns [10]. This should
definitely suffice for the demonstration of quantum logic.
Recently, methods have been proposed to eliminate the decoherence problem by
creating a Decoherence Free Subspace (DFS) [3]. A DFS is created by carefully placing
other quantum dots around the quantum dot which is performing the operation. These
outer quantum dots can shield the computing dot from environmental effects. This
technique can be plugged into Loss and DiVincenzo solid-state model by creating extra
layers of shielding quantum dots.
Another equally important challenge is the proper control and positioning of the
individual dots [7]. New discoveries in the manipulation of nano-particles are being
made at a rapid rate, making the construction of a solid-state quantum dot array a
possibility. According to the Internal Technology Roadmap for Semiconductors, a
quantum dot array is manufacturable using the specifications for the proposed 70 nm
process [19]. The layout and addressing of a quantum dot array is also similar to current
methods used in the production of DRAMS [19].

!13
 There is also room for improvement in the development of quantum dots in
magnetic semiconductors. Advances in this regime can greatly reduce the complexity of
the quantum dot array by removing the apparatus needed to create large magnetic fields.
This can make the magnetized layer of Figure 4 a reality.
Furthermore, it would be beneficial to see an increase in the number of useful
quantum algorithms that are available. Currently, if a quantum computer were built, it
would only serve specific tasks such as encryption for the military and industrial database
searches in fields such as bioinformatics. It can play a big role in companies that are
working on computationally intensive tasks such as gene expression and protein folding.
More algorithms are needed to show that quantum computers can indeed perform many
tasks much better than classical computers, if they are to be used commercially. It is also
important that the costs of creating a quantum computer does not outweigh its benefits.

Future Directions
The same authors of the quantum dot array proposal mentioned above have
recently come up with a scheme that eliminates the magnetic field needed to create single
qubit rotations [21]. However, this scheme calls for an increase in the number of qubits
by a factor of three. It also requires the number of two-qubit operations to grow by a
factor of ten. However, due to the scalability of the quantum dot array proposal, these
additional costs should not be a big issue compared to the benefits of eliminating the
complexity of magnetic field based single qubit rotations.
Another potential improvement may involve changing the material of choice for
fabricating quantum dots. Most of the research involving quantum dots so far have been
done using GaAs since it only has a single valley at the point. The multi-valley nature
of Si causes it to have weaker exchange interactions than GaAs [6], which means that it is
harder to bring two Si quantum dots into a superposition. However, silicons ability to
have weaker exchange interactions also means that it naturally has a lower decoherence
time. Therefore, if a method can be found to couple Si quantum dots efficiently, then Si

!14
would make a better candidate for quantum computing than GaAs. Si is cheaper to
fabricate than GaAs and is much more pervasive in the industry.

Conclusion
The concept of quantum computing is such as paradigm shift that it almost seems
like science fiction compared to modern day computing. Even if quantum computers
dont end up replacing current computers, they definitely can be integrated into current
systems to perform quantum algorithms when needed. The quantum dot spin approach
seems like one of the most feasible methods of quantum computing, because, unlike most
non-solid state solutions, it doesnt necessarily have to depend on large external magnetic
fields or cryogenic temperatures to operate. Quantum dots are incredibly versatile and
may also be used in the fields of quantum cryptography and quantum teleportation [10].
Quantum computers of up to 7-qubits using the non-solid-state NMR method have
already been implemented [20]. However, NMR can only be scaled to work with up to
10 qubits. Therefore, with all the research being done on solid-state methods, a more
practical approach based on quantum dots should not be too far off.

!15
References
[1] Brown, Julian. The Quest for the Quantum Computer, Touchstone Publishing, 2001.
[2] Biolatti, et al., Electro-optical properties of semiconductor quantum dots:
Application to quantum information processing., Physical Review B., Vol 65, 075306,
Jan 2002.
[3] Hellberg, Stephen., Robust Quantum Computation with Quantum Dots, quant-ph/
0304150, April 2003.
[4] Li, Shu-Shen, et al., InAs/GaAs single electron quantum dot qubit, Journal of
Applied Physics Vol. 90, p. 6151-6155, September 2001.
[5] Vrijen, et al., Electron-spin-resonance transistors for quantum computing in silicon-
germanium heterostructures, Physical Review A, Vol. 62, 2000.
[6] Hada, Yoko., Electronic states in silicon quantum dots: Multivalley artificial atoms,
Physical Review B 68, 2003.
[7] Hvam, J.M, et al, Coherence and Dephasing in Self-Assembled Quantum Dots,
IEEE, pp. 122-125, 2003
[9] Burkard, Guido., et al., Coupled quantum dots as quantum gates, Phys. Rev. B 59,
p. 2070., 1999.
[10] Burkard, Guido., et al., Spintronics and Quantum Dots for Quantum Computing
and Quantum Communication, Fortschritte der Physik, April 2000 (cond-mat/0004182).
[11] Loss, Daniel., DiVincenzo, David P., Quantum computation with quantum dots,
Physical Review A, Vol 57, pp. 120-126., January 1998.
[12] Kouwenhoven, Leo., Marcus, Charles., Quantum dots, Physics World, June 1998.
[13] Nielsen & Chuang, Quantum Computation and Quantum Information, Cambridge
University Press, 2000
[14] Meglicki, Zdzislaw., Introduction to Quantum Computing (M743), Indiana
University, http://beige.ucs.indiana.edu/M743/index.html, 2002
[15] Ekert, Basic Concepts in Quantum Computation, www.qubit.org, 2000

!16
[16] DiVincenzo, David., The Physical Implementation of Quantum Computation,
Fortschritte der Physik, February 2003 (quant-ph/0002077).
[17] Kastner, Mark A., Artificial Atoms, Physics Today, January 1993.
[18] Emiroglu, et al., Isolated double quantum dot capacitively coupled to a single
quantum dot single-electron transistor in silicon, Applied Physics Letters Vol 83(19) pp.
3942-3944., November 10, 2003.
[19] Kim, K. W., et al., Quantum Computing: Concept and Realization, North Carolina
State University, www.ece.ncsu.edu/nano/presentations.htm
[20] Vaccaro, John., The Quantum Computer, University of Hertfordshire, http://
strc.herts.ac.uk/tp/info/qucomp/
[21] DiVincenzo, et al., Universal quantum computation with the exchange interaction,
Nature 408, p. 339 342, 2000

!17