The Physical Implementation of Quantum Computation

David P. DiVincenzo
IBM T.J. Watson Research Center, Yorktown Heights, NY 10598 USA
(February 1, 2008)

arXiv:quant-ph/0002077v3 13 Apr 2000

After a brief introduction to the principles and promise of quantum information processing, the
requirements for the physical implementation of quantum computation are discussed. These five
requirements, plus two relating to the communication of quantum information, are extensively explored and related to the many schemes in atomic physics, quantum optics, nuclear and electron
magnetic resonance spectroscopy, superconducting electronics, and quantum-dot physics, for achieving quantum computing.

routers, scanners, etc.), in which the embodiment and

transformations of this information can be described using the language of classical mechanics. But the final
physical theory of the world is not Newtonian mechanics,
and there is no reason to suppose that machines following the laws of quantum mechanics should have the same
computational power as classical machines; indeed, since
Newtonian mechanics emerges as a special limit of quantum mechanics, quantum machines can only have greater
computational power than classical ones. The great pioneers and visionaries who pointed the way towards quantum computers, Deutsch [8], Feynman [9], and others,
were stimulated by such thoughts. Of course, by a similar
line of reasoning, it may well be asked whether machines
embodying the principles of other refined descriptions of
nature (perhaps general relativity or string theory) may
have even more information processing capabilities; speculations exist about these more exotic possibilities, but
they are beyond the scope of the present discussion.
But computing with quantum mechanics really deserves a lot more attention than wormhole computing or
quantum-gravity computing; quantum computing, while
far in the future from the perspective of CMOS roadmaps
and projections of chip fab advances, can certainly be
seen as a real prospect from the perspective of research
studies in quantum physics. It does not require science
fiction to envision a quantum computer; the proposals
discussed later in this issue paint a rather definite picture of what a real quantum computer will look like.
So, how much is gained by computing with quantum
physics over computing with classical physics? We do
not seem to be near to a final answer to this question,
which is natural since even the ultimate computing power
of classical machines remains unknown. But the answer
as we know it today has an unexpected structure; it is
not that quantum tools simply speed up all information
processing tasks by a uniform amount. By a standard
complexity measure (i.e., the way in which the number
of computational steps required to complete a task grows
with the size n of the task), some tasks are not sped
up at all [10] by using quantum tools (e.g., obtaining the
nth iterate of a function f (f (...f (x)...)) [11]), some are
sped up moderately (locating an entry in a database of n
entries [12]), and some are apparently sped up exponen-

I. INTRODUCTION

The advent of quantum information processing, as an

abstract concept, has given birth to a great deal of new
thinking, of a very concrete form, about how to create
physical computing devices that operate in the hitherto
unexplored quantum mechanical regime. The efforts now
underway to produce working laboratory devices that
perform this profoundly new form of information processing are the subject of this book.
In this chapter I provide an overview of the common
objectives of the investigations reported in the remainder of this special issue. The scope of the approaches,
proposed and underway, to the implementation of quantum hardware is remarkable, emerging from specialties in
atomic physics [1], in quantum optics [2], in nuclear [3]
and electron [4] magnetic resonance spectroscopy, in superconducting device physics [5], in electron physics [6],
and in mesoscopic and quantum dot research [7]. This
amazing variety of approaches has arisen because, as we
will see, the principles of quantum computing are posed
using the most fundamental ideas of quantum mechanics,
ones whose embodiment can be contemplated in virtually
every branch of quantum physics.
The interdisciplinary spirit which has been fostered as
a result is one of the most pleasant and remarkable features of this field. The excitement and freshness that has
been produced bodes well for the prospect for discovery,
invention, and innovation in this endeavor.
II. WHY QUANTUM INFORMATION
PROCESSING?

The shortest of answers to this question would be, why

not? The manipulation and transmission of information
is today carried out by physical machines (computers,

Prepared for Fortschritte der Physik special issue, Experimental Proposals for Quantum Computation, eds. H.-K. Lo
and S. Braunstein.

state as |0i and the other as |1i. The essential feature

that distinguishes a qubit from a bit is that, according
to the laws of quantum mechanics, the permitted states
of a single qubit fills up a two-dimensional complex vector space; the general state is written a|0i + b|1i, where
a and b are complex numbers, and a normalization convention |a|2 + |b|2 = 1 is normally adopted. The general
state of two qubits, a|00i + b|01i + c|10i + d|11i, is a fourdimensional vector, one dimension for each distinguishable state of the two systems. These states are generically entangled, meaning that they cannot be written as
a product of the states of two individual qubits. The
general state of n qubits is specified by a 2n -dimensional
complex vector.
A qubit being well characterized means several different things. Its physical parameters should be accurately known, including the internal Hamiltonian of the
qubit (which determines the energy eigenstates of the
qubit, which are often, although not always, taken as the
|0i and |1i states), the presence of and couplings to other
states of the qubit, the interactions with other qubits,
and the couplings to external fields that might be used
to manipulate the state of the qubit. If the qubit has
third, fourth, etc., levels, the computers control apparatus should be designed so that the probability of the
system ever going into these states is small. The smallness of this and other parameters will be determined by
the capabilities of quantum error correction, which will
be discussed under requirement 3.
Recognizing a qubit can be trickier than one might
think. For example, we might consider a pair of oneelectron quantum dots that share a single electron between them as a two-qubit system. It is certainly true
that we can denote the presence or absence of an electron on each dot by |0i and |1i, and it is well known
experimentallyhow to put this system into the entangled state 1/ 2(|01i + |10i) in which the electron is in
a superposition of being on the left dot and the right
dot. But it is fallacious to consider this as a two-qubit
system; while the states |00i and |11i are other allowed
physical states of the dots, superselection principles forbid the creation of entangled
states involving different
particle numbers such as 1/ 2(|00i + |11i).
It is therefore false to consider this as a two-qubit system, and, since there are not two qubits, it is nonsense
to say that there is entanglement in this system. It is
correct to say that the electron is in a superposition of
different quantum states living on the two different dots.
It is also perfectly correct to consider this system to be
the embodiment of a single qubit, spanned by the states
(in the misleading notation above) |01i (electron on the
right dot) and |10i (electron on the left dot). Indeed, several of the viable proposals, including the ones
by Sch
on, Averin, and Tanamoto in this special issue,
use exactly this system as a qubit. However, false lines
of reasoning like the one outlined here have sunk various
proposals before they were properly launched (no such
abortive proposals are represented in this book, but they

tially (Shors algorithm for factoring an n-digit number

[13]).
In other types of information processing tasks, particularly those involving communication [14], both quantitative and qualitative improvements are seen [15]: for
certain tasks (choosing a free day for an appointment between two parties from out of n days) there is a quadratic
reduction of the amount of communicated data required,
if quantum states rather than classical states are transmitted [16]. For some tasks (the set disjointness problem, related to allocating non-overlapping segments of
a shared memory in a distributed computation) the reduction of required communication is exponential [17].
Finally, there are tasks that are doable in the quantum world that have no counterpart classically: quantum
cryptography provides an absolute secrecy of communication between parties that is impossible classically [18].
And for some games, winning strategies become possible
with the use of quantum resources that are not available
otherwise [19,20].
This issue, and this chapter, are primarily concerned
with the hows of quantum computing rather than the
whys, so we will leave behind the computer science
after this extremely brief mention. There is no shortage
of other places to obtain more information about these
things; I recommend the recent articles by Aharonov [21]
and by Cleve [22]; other general introductions [23] will
give the reader pointers to the already vast specialized
literature on this subject.
III. REALIZING QUANTUM COMPUTATION

Let me proceed with the main topic: the physical realization of quantum information processing. As a guide
to the remainder of the special issue, and as a means
of reviewing the basic steps required to make quantum
computation work, I can think of no better plan than
to review a set of basic criteria that my coworkers and
I have been discussing over the last few years [24] for
the realization of quantum computation (and communication), and to discuss the application of these criteria to
the multitude of physical implementations that are found
below.
So, without further ado, here are the
Five (plus two) requirements for the implementation of quantum computation
1. A scalable physical system with well characterized
qubits
For a start, a physical system containing a collection
of qubits is needed. A qubit (or, more precisely, the embodiment of a qubit) is [25] simply a quantum two-level
system like the two spin states of a spin 1/2 particle,
like the ground and excited states of an atom, or like
the vertical and horizontal polarization of a single photon. The generic notation for a qubit state denotes one
2

a standard state: the system can either be naturally

cooled when the ground state of its Hamiltonian is the
state of interest, or the standard state can be achieved
by a measurement which projects the system either into
the state desired or another state which can be rotated
into it. These approaches are not fundamentally different from one another, since the projection procedure is a
form of cooling; for instance, the laser cooling techniques
used routinely now for the cooling of ion states to near
their ground state in a trap [1] are closely connected to
the fluorescence techniques used to measure the state of
these ions. A more natural kind of cooling is advocated
in many of the electron spin resonance based techniques
(using quantum dots or impurities) [7,4] in which the
spins are placed in a strong magnetic field and allowed
to align with it via interaction with their heat bath. In
this kind of approach the time scale will be a problem.
Since the natural thermalization times are never shorter
than the decoherence time of the system, this procedure
will be too slow for the needs of error correction and a
conveyor belt scheme would be required. Cooling by
projection, in which the Hamiltonian of the system and
its environment are necessarily perturbed strongly, will
have a time scale dependent on the details of the setup,
but potentially much shorter than the natural relaxation
times. One cannot say too much more at this point,
as the schemes for measurement have in many cases not
been fully implemented (see requirement 5). In the NMR
quantum computer implementations to date (see Cory
below), cooling of the initial state has been foregone altogether; it is acknowledged [3] that until some of the
proposed cooling schemes are implemented (a nontrivial
thing to do), NMR can never be a scalable scheme for
quantum computing.

can be found occasionally in the literature).

An amazing variety of realizations of the qubit are represented in this volume. There is a very well developed
line of work that began with the proposal of Cirac and
Zoller [1] for an ion-trap quantum computer, in which,
in its quiescent state, the computer holds the qubits in
pairs of energy levels of ions held in a linear electromagnetic trap. Various pairs of energy levels (e.g., Zeemandegenerate ground states, as are also used in the NMR
approach [3] discussed by Cory) have been proposed
and investigated experimentally. The many neutral-atom
proposals (see chapters by Kimble [2], Deutsch [26], and
Briegel [27]) use similar atomic energy levels of neutral
species. These atomic-physics based proposals use other
auxiliary qubits such as the position of atoms in a trap
or lattice, the presence or absence of a photon in an optical cavity, or the vibrational quanta of trapped electrons,
ions or atoms (in the Platzman proposal below [6] this
is the primary qubit). Many of the solid-state proposals exploit the fact that impurities or quantum dots have
well characterized discrete energy level spectra; these include the spin states of quantum dots (see chapters by
Loss [7] and Imamoglu [2]), the spin states of donor impurities (see Kane [4]), and the orbital or charge states
of quantum dots (see Tanamoto [7]). Finally, there are a
variety of interesting proposals which use the quantized
states of superconducting devices, either ones involving
the (Cooper-pair) charge (see Sch
on, Averin), or the flux
(see Mooij) [5].
2. The ability to initialize the state of the qubits to a
simple fiducial state, such as |000...i
This arises first from the straightforward computing requirement that registers should be initialized to a known
value before the start of computation. There is a second
reason for this initialization requirement: quantum error
correction (see requirement 3 below) requires a continuous, fresh supply of qubits in a low-entropy state (like
the |0i state). The need for a continuous supply of 0s,
rather than just an initial supply, is a real headache for
many proposed implementations. But since it is likely
that a demonstration of a substantial degree of quantum
error correction is still quite some time off, the problem
of continuous initialization does not have to be solved
very soon; still, experimentalists should be aware that
the speed with which a qubit can be zeroed will eventually be a very important issue. If the time it takes to do
this initialization is relatively long compared with gateoperation times (see requirement 4), then the quantum
computer will have to equipped with some kind of qubit
conveyor belt, on which qubits in need of initialization
are carried away from the region in which active computation is taking place, initialized while on the belt, then
brought back to the active place after the initialization is
finished. (A similar parade of qubits will be envisioned
in requirement 5 for the case of low quantum-efficiency
measurements [28].)
There are two main approaches to setting qubits to

3. Long relevant decoherence times, much longer than

the gate operation time
Decoherence times characterize the dynamics of a
qubit (or any quantum system) in contact with its environment. The (somewhat overly) simplified definition of
this time is that it is the characteristic time for a generic
qubit state |i = a|0i + b|1i to be transformed into the
mixture = |a|2 |0ih0|+ |b|2 |1ih1|. A more proper characterization of decoherence, in which the decay can depend
on the form of the initial state, in which the state amplitudes may change as well, and in which other quantum
states of the qubit can play a role (in a special form
of state decay called leakage in quantum computing
[29,30]), is rather more technical than I want to get here;
but see Refs. [31] and [32] for a good general discussion
of all these. Even the simplest discussion of decoherence
that I have given here should also be extended to include the possibility that the decoherence of neighboring
qubits is correlated. It seems safest to assume that they
will be neither completely correlated nor completely uncorrelated, and the thinking about error correction has
taken this into account.
Decoherence is very important for the fundamentals

tolerant quantum computation. The result is that, if the

decoherence time is 104 105 times the clock time of
the quantum computer, that is, the time for the execution of an individual quantum gate (see requirement
4), then error correction can be successful. This is, to
tell the truth, a rather stringent condition, quantum systems frequently do not have such long decoherence times.
But sometimes they do, and our search for a successful
physical implementation must turn towards these. At
least this result says that the required decoherence rate
does not become ever smaller as the size and duration
of the quantum computation grows. So, once the desired threshold is attainable, decoherence will not be an
obstacle to scalable quantum computation.
Having said this, it must be admitted that it will be
some time before it is even possible to subject quantum
error correction to a reasonable test. Nearly all parts of
requirements 1-5 must be in place before such a test is
possible. And even the most limited application of quantum error correction has quite a large overhead: roughly
10 ancillary qubits must be added for each individual
qubit of the computation. Fortunately, this overhead
ratio grows only logarithmically as the the size of the
quantum computation is increased.
In the short run, it is at least possible to design
and perform experiments which measure the decoherence times and other relevant properties (such as the
correlation of decoherence of neighboring qubits) of candidate implementations of qubits. With such initial test
experiments, caution must be exercised in interpreting
the results, because decoherence is a very system-specific
phenomenon, depending on the details of all the qubits
couplings to various environmental degrees of freedom.
For example, the decoherence time of the spin of an impurity in the bulk of a perfect semiconductor may not
be the same as its decoherence time when it is near the
surface of the solid, in the immediate neighborhood of device structures designed to manipulate its quantum state.
Test experiments should probe decoherence in as realistic
a structure as is possible.

of quantum physics, as it is identified as the principal

mechanism for the emergence of classical behavior. For
the same reason, decoherence is very dangerous for quantum computing, since if it acts for very long, the capability of the quantum computer will not be so different
from that of a classical machine. The decoherence time
must be long enough that the uniquely quantum features
of this style of computation have a chance to come into
play. How long is long enough is also indicated by the
results of quantum error correction, which I will summarize shortly.
I have indicated that the relevant decoherence times
should be long enough. This emphasizes that a quantum
particle can have many decoherence times pertaining to
different degrees of freedom of that particle. But many
of these can be irrelevant to the functioning of this particle as a qubit. For example, the rapid decoherence of
an electrons position state in a solid state environment
does not preclude its having a very long spin coherence
time, and it can be arranged that this is the only time
relevant for quantum computation. Which time is relevant is determined by the choice of the qubit basis states
|0i and |1i; for example, if these two states correspond
to different spin states but identical orbital states, then
orbital decoherence will be irrelevant.
One might worry that the decoherence time necessary
to do a successful quantum computation will scale with
the duration of the computation. This would place incredibly stringent requirements on the physical system
implementing the computation. Fortunately, in one of
the great discoveries of quantum information theory (in
1995-6), it was found that error correction of quantum
states is possible [33] and that this correction procedure
can be successfully applied in quantum computation [34],
putting much more reasonable (although still daunting)
requirements on the needed decoherence times.
In brief, quantum error correction starts with coding;
as in binary error correction codes, in which only a subset of all boolean strings are legal states, quantum error correction codes consist of legal states confined to
a subspace of the vector space of a collection of qubits.
Departure from this subspace is caused by decoherence.
Codes can be chosen such that, with a suitable sequence
of quantum computations and measurements of some ancillary qubits, the error caused by decoherence can be
detected and corrected. As noted above, these ancillary
qubits have to be continuously refreshed for use. I will
not go much farther into the subject here, see [31] for
more. It is known that quantum error correction can be
made fully fault tolerant, meaning that error correction
operations can be successfully intermingled with quantum computation operations, that errors occurring during the act of error correction, if they occur at a sufficiently small rate, do no harm, and that the act of quantum computation does not itself cause an unreasonable
proliferation of errors.
These detailed analyses have indicated the magnitude
of decoherence time scales that are acceptable for fault-

4. A universal set of quantum gates

This requirement is of course at the heart of quantum
computing. A quantum algorithm is typically specified
[8] as a sequence of unitary transformations U1 , U2 , U3 ,...,
each acting on a small number of qubits, typically no
more than three. The most straightforward transcription
of this into a physical specification is to identify Hamiltonians which generate these unitary transformations, viz.,
U1 = eiH1 t/h , U2 = eiH2 t/h , U3 = eiH3 t/h , etc.; then, the
physical apparatus should be designed so that H1 can be
turned on from time 0 to time t, then turned off and H2
turned on from time t to time 2t, etc.
Would that life were so simple! In reality what can be
done is much less, but much less can be sufficient. Understanding exactly how much less is still enough, is the
main complication of this requirement. In all the physical
implementations discussed in this volume, only particular

the quantum computer will be determined by the time

interval needed such that two consecutive pulses have
negligible overlap.
Another consideration, which does not seem to present
a problem with any current implementation schemes, but
which may be an issue in the future, is the classicality
of the control apparatus. We say that the interaction
Hamiltonian H(t) has a time profile which is controlled
externally by some classical means, that is, by the intensity of a laser beam, the value of a gate voltage, or
the current level in a wire. But each of these control
devices is made up themselves of quantum mechanical
parts. When we require that these behave classically, it
means that their action should proceed without any entanglement developing between these control devices and
the quantum computer. Estimates indicate that this entanglement can indeed be negligible, but this effect needs
to be assessed for each individual case.
In many cases it is impossible to turn on the desired
interaction between a pair of qubits; for instance, in the
ion-trap scheme, no direct interaction is available between the ion-level qubits [1]. In this and in other cases,
a special quantum subsystem (sometimes referred to as a
bus qubit) is used which can interact with each of the
qubits in turn and mediate the desired interaction: for
the ion trap, this is envisioned to be the vibrational state
of the ion chain in the trap; in other cases it is a cavity photon whose wavefunction overlaps all the qubits.
Unfortunately, this auxiliary quantum system introduces
new channels for the environment to couple to the system and cause decoherence, and indeed the decoherence
occurring during gate operation is of concern in the iontrap and cavity-quantum electrodynamics schemes.
Some points about requirement 4 are important to
note in relation to the implementation of error correction. Successful error correction requires fully parallel
operation, meaning that gate operations involving a finite
fraction of all the qubits must be doable simultaneously.
This can present a problem with some of the proposals in
which the single bus qubit is needed to mediate each
interaction. On the other hand, the constraint that interactions are only among nearest neighbors in a lattice,
as in many of the solid-state proposals, does allow for
sufficient parallelism [41].
Quantum gates cannot be implemented perfectly; we
must expect both systematic and random errors in the
implementation of the necessary Hamiltonians. Both
types of errors can be viewed as another source of decoherence and thus error correction techniques are effective for producing reliable computations from unreliable
gates, if the unreliability is small enough. The tolerable
unreliability due to random errors is in the same vicinity
as the decoherence threshold, that is, the magnitude of
random errors should be 104 105 per gate operation
or so. It might be hoped that systematic errors could be
virtually eliminated by careful calibration; but this will
surely not always be the case. It seems harder to give
a good rule for how much systematic error is tolerable,

sorts of Hamiltonians can be turned on and off; in most

cases, for example, only two-body (two-qubit) interactions are considered. This immediately poses a problem
for a quantum computation specified with three-qubit
unitary transformations; fortunately, of course, these can
always be re-expressed in terms of sequences of one- and
two-body interactions [35], and the two-body interactions can be of just one type [36], the quantum XOR
or cNOT. There are some implementations in which
multi-qubit gates can be implemented directly [37].
However, this still leaves a lot of work to do. In some
systems, notably in NMR (see Cory), there are two-body
interactions present which cannot be turned off, as well
as others which are switchable. This would in general
be fatal for quantum computation, but the particular
form of the fixed interactions permit their effects to be
annulled by particular refocusing sequences of the controllable interactions, and it has recently been discovered
[38] that these refocusing sequences can be designed and
implemented efficiently.
For many other systems, the two-body Hamiltonian
needed to generate directly the cNOT unitary transformation is not available. For example, in the quantumdot proposal described by Loss below [7], the only twobody interaction which should be easily achievable is the
exchange interaction between neighboring spins, H
~i S
~i+1 ; in the Imamoglu chapter [2], the attainable
S
interaction is of the XY type, i.e., H Six Sjx + Siy Sjy .
An important observation is that with the appropriate
sequence of exchange or XY interactions, in conjunction with particular one-body interactions (which are assumed to be more easily doable), the cNOT transformation can be synthesized [39]. It is incumbent on each
implementation proposal to exhibit such a sequence for
producing the cNOT using the interactions that are naturally realizable.
Often there is also some sophisticated thinking required about the time profile of the two-qubit interaction. The naive description above uses a square pulse
time profile, but often this is completely inappropriate;
for instance, if the Hamiltonian can also couple the qubit
to other, higher-lying levels of the quantum system, often
the only way to get the desired transformation is to turn
on and off the interaction smoothly and slowly enough
that an adiabatic approximation is accurate [29,30] (in a
solid-state context, see also [40]). The actual duration of
the pulse will have to be sufficiently long that any such
adiabatic requirement
is satisfied; then typically only the
R
time integral dtH(t) is relevant for the quantum gate
action. The overall time scale of the interaction pulse
is also controlled by the attainable maximum size of the
matrix elements of H(t), which will be determined by
various fundamental considerations, like the requirement
that the system remains in the regime of validity of a
linear approximation, and practical considerations, like
the laser power that can be concentrated on a particular
ion. Given these various constraints, the clock time of

rerunning the calculation three times. Much better, actually, is to copy the single output qubit to three, by
applying two cNOT gates involving the output qubit and
two other qubits set to |0i, and measuring those three.
(Of course, qubits cannot be copied, but their value in
a particular basis can.) In general, if quantum efficiency
q is available, then copying to somewhat more than 1/q
qubits and measuring all of these will result in a reliable
outcome. So, a quantum efficiency of 1% would be usable
for quantum computation, at the expense of hundreds of
copies/remeasures of the same output qubit. (This assumes that the measurement does not otherwise disturb
the quantum computer. If it does, the possibilities are
much more limited.)
Even quantum efficiencies much, much lower than 1%
can be and are used for successful quantum computation: this is the bulk model of NMR (see Cory and
[3]), where macroscopic numbers of copies of the same
quantum computer (different molecules in solution) run
simultaneously, with the final measurement done as an
ensemble average over the whole sample. These kinds
of weak measurements, in which each individual qubit is
hardly disturbed, are quite common and well understood
in condensed-matter physics.
If a measurement can be completed quickly, on the
timescale of 104 of the decoherence time, say, then its
repeated application during the course of quantum computation is valuable for simplifying the process of quantum error correction. On the other hand, if this fast
measurement capability is not available, quantum error
correction is still possible, but it then requires a greater
number of quantum gates to implement.
Other tradeoffs between the complexity and reliability
of quantum measurement vs. those of quantum computation have recently been explored. It has been shown
that if qubits can be initialized into pairs of maximally
entangled states, and two-qubit measurements in the socalled Bell basis ( = |01i |10i, = |00i |11i) are
possible, then no two-qubit quantum gates are needed,
one-bit gates alone suffice [44]. Now, often this tradeoff
will not be useful, as in many schemes a Bell measurement would require two-bit quantum gates.
But the overall message, seen in many of our requirements, is that more and more, the theoretical study
of quantum computation has offered a great variety of
tradeoffs for the potential implementations: if X is very
hard, it can be substituted with more of Y. Of course,
in many cases both X and Y are beyond the present experimental state of the art; but a thorough knowledge
of these tradeoffs should be very useful for devising a
rational plan for the pursuit of future experiments.

the conservative estimates give a very, very small number

(the square of the above) [31], but on the other hand there
seems to be some evidence that certain important quantum computations (e.g., the quantum Fourier transform)
can tolerate a very high level of systematic error (over- or
under-rotation). Some types of very large errors may be
tolerable if their presence can be detected and accounted
for on the fly (we are thinking, for example, about charge
switching in semiconductors or superconductors).
Error correction requires that gate operations be done
on coded qubits, and one might worry that such operations would require a new repertoire of elementary
gate operations for the base-level qubits which make up
the code. For the most important error correction techniques, using the so called stabilizer codes, this is not
the case. The base-level toolkit is exactly the same as
for the unencoded case: one-bit gates and cNOTs, or
any gate repertoire that can produce these, are adequate.
Sometimes the use of coding can actually reduce the gate
repertoire required: in the work on decoherence free subspaces and subsystems, codes are introduced using blocks
of three and four qubits for which two-qubit exchange interactions alone are enough to implement general quantum computation [42,43]. This simplification could be
very useful in the quantum-dot [7] or semiconductorimpurity [4] implementations.
5. A qubit-specific measurement capability
Finally, the result of a computation must be read out,
and this requires the ability to measure specific qubits.
In an ideal measurement, if a qubits density matrix is
= p|0ih0| + (1 p)|1ih1| + |0ih1| + |1ih0|, the measurement should give outcome 0 with probability p and
1 with probability 1 p independent of and of any
other parameters of the system, including the state of
nearby qubits, and without changing the state of the rest
of the quantum computer. If the measurement is nondemolition, that is, if in addition to reporting outcome
0 the measurement leaves the qubit in state |0ih0|, then
it can also be used for the state preparation of requirement 2; but requirement 2 can be fulfilled in other ways.
Such an ideal measurement as I have described is said
to have 100% quantum efficiency; real measurements always have less. While the fidelity of a quantum measurement is not captured by a single number, the single
quantum-efficiency parameter is often a very useful way
to summarize it, just as the decoherence time is a useful if
incomplete summary of the damage caused to a quantum
state by the environment.
While quantum efficiency of 100% is desirable, much
less is needed for quantum computation; there is, in fact,
a tradeoff possible between quantum efficiency and other
resources which results in reliable computation. As a
simple example, if the quantum efficiency is 90%, then,
in the absence of any other imperfections, a computation
with a single-bit output (a so-called decision problem,
common in computer science) will have 90% reliability.
If 97% reliability is needed, this can just be achieved by

IV. DESIDERATA FOR QUANTUM

COMMUNICATION

For computation alone, the five requirements above

suffice. But the advantages of quantum information pro6

loading it by the time-reversed process. Other promising

concepts, like the launching of electrons from quantum
dots into quantum wires such that the spin coherence of
the electrons is preserved, need to be worked out more
fully.

cessing are not manifest solely, or perhaps even principally, for straightforward computation only. There are
many kinds of information-processing tasks, reviewed
briefly at the beginning, that involve more than just computation, and for which quantum tools provide a unique
advantage.
The tasks we have in mind here all involve not only
computation but also communication. The list of these
tasks that have been considered in the light of quantum capabilities, and for which some advantage has been
found in using quantum tools, is fairly long and diverse:
it includes secret key distribution, multiparty function
evaluation as in appointment scheduling, secret sharing,
and game playing [14].
When we say communication we mean quantum communication: the transmission of intact qubits from place
to place. This obviously adds more features that the
physical apparatus must have to carry out this information processing. We formalize these by adding two more
items to the list of requirements:

V. SUMMARY

So, what is the winning technology going to be? I

dont think that any living mortal has an answer to this
question, and at this point it may be counterproductive
even to ask it. Even though we have lived with quantum
mechanics for a century, our study of quantum effects
in complex artificial systems like those we have in mind
for quantum computing is in its infancy. No one can see
how or whether all the requirements above can be fulfilled, or whether there are new tradeoffs, not envisioned
in our present theoretical discussions but suggested by
further experiments, that might take our investigations
in an entirely new path.
Indeed, the above discussion, and the other chapters of
this special issue, really do not cover all the foreseeable
approaches. I will mention two of which I am aware: first,
another computational paradigm, that of the cellular automaton, is potentially available for exploitation. This
is distinguished from the above general purpose approach in that it assumes that every bit pattern throughout the computer will be subjected to the same evolution
rule. It is known that general-purpose computation is
performable, although with considerable overhead, by a
cellular automaton. This is true as well for the quantum
version of the cellular automaton, as Lloyd [47] indicated
in his original work. New theoretical work by Benjamin
[48] shows very explicitly how relatively simple local rules
would permit the implementation of some quantum computations. This could point us perhaps towards some sort
of polymer with a string of qubits on its backbone that
can be addressed globally in a spectroscopic fashion. Experiments are not oriented towards this at the moment,
but the tradeoffs are very different, and I dont believe it
should be excluded in the future.
Second, even more speculative, but very elegant, is the
proposal of Kitaev [49] to use quantum systems with particular kinds of topological excitations, for example nonabelian anyons, for quantum computing. It is hard to see
at the moment how to turn this exciting proposal into an
experimental program, as no known physical system is
agreed to have the appropriate topological excitations.
But further research in, for example, the quantum Hall
effect might reveal such a system; more likely, perhaps,
is that further understanding of this approach, and that
of Freedman and his colleagues [50], will shed more light
on doing quantum computing using the standard approach being considered in this book.
I am convinced of one thing: the ideas of quantum
information theory will continue to exert a decisive in-

6. The ability to interconvert stationary and flying

qubits
7. The ability faithfully to transmit flying qubits between specified locations
These two requirements are obviously closely related,
but it is worthwhile to consider them separately, because
some tasks need one but not the other. For instance,
quantum cryptography [18] involves only requirement 7;
it is sufficient to create and detect flying qubits directly.
I have used the jargon flying qubits [2], which has
become current in the discussions of quantum communication. Using this term emphasizes that the optimal
embodiment of qubits that are readily transmitted from
place to place is likely to be very different from the optimal qubits for reliable local computation. Indeed, almost all proposals assume that photon states, with the
qubit encoded either in the polarization or in the spatial wavefunction of the photon, will be the flying qubit
of choice, and indeed, the well developed technology of
light transmission through optical fibers provides a very
promising system for the transmission of qubits. I would
note, though, that my colleagues and I have raised the
possibility that electrons traveling though solids could
provide another realization of the flying qubit [14,45].
Only a few completely developed proposals exist which
incorporate requirements 6 and 7. Of course, there are a
number of quite detailed studies of 7, in the sense that experiments on quantum cryptography have been very concerned with the preservation of the photon quantum state
during transmission through optical fibers or through the
atmosphere. However, these studies are rather disconnected from the other concerns of quantum computing.
Requirement 6 is the really hard one; to date the only theoretical proposal sufficiently concrete that experiments
addressing it have been planned is the scheme produced
by Kimble and coworkers [46] for unloading a cavity photon into a traveling mode via atomic spectroscopy, and
7

fluence on the further investigation of the fundamental

quantum properties of complex quantum systems, and
will stimulate many creative and exciting developments
for many years to come.

[10]

ACKNOWLEDGMENTS
[11]

I gratefully acknowledge support from the Army Research Office under contract number DAAG55-98-C0041. I thank Alec Maassen van den Brink for a careful
reading of this manuscript.

[12]
[13]

[14]
[1] J.I. Cirac and P. Zoller, Phys. Rev. Lett. 74, 4091 (1995);
T. Pellizzari, S.A. Gardiner, J.I. Cirac, and P. Zoller,
Phys. Rev. Lett. 75, 3788 (1995); C. Monroe, D. M.
Meekhof, B. E. King, W. M. Itano, and D. J. Wineland,
Demonstration of a fundamental quantum logic gate,
Phys. Rev. Lett. 75, 4714 (1995); A. Sorensen and K.
Molmer, Phys. Rev. Lett. 82, 1971 (1999); S. Schneider, D. F. V. James, and G. J. Milburn, Method of
quantum computation with hot trapped ions, quantph/9808012.
[2] Q. A. Turchette, C. J. Hood, W. Lange, H. Mabuchi, and
H. J. Kimble, Measurement of conditional phase shifts
for quantum logic, Phys. Rev. Lett. 75, 4710 (1995); A.
Imamoglu, D. D. Awschalom, G. Burkard, D. P. DiVincenzo, D. Loss, M. Sherwin, and A. Small, Phys. Rev.
Lett. 83, 4204 (1999) (quant-ph/9904096).
[3] N. Gershenfeld and I. Chuang, Science 275, 350 (1997);
D. Cory, A. Fahmy, and T. Havel, Proc. Nat. Acad. Sci.
94 (5), 1634 (1997).
[4] B. Kane, Nature 393, 133 (1998); R. Vrijen et al., Electron spin resonance transistors for quantum computing
in silicon-germanium heterostructures, Phys. Rev. A, in
press (quant-ph/9905096).
[5] D. Averin, Solid State Commun. 105, 659 (1998); A.
Shnirman, G. Sch
on, and Z. Hermin, Phys. Rev. Lett.
79, 2371 (1997); J. E. Mooij, T. P. Orlando, L. Levitov,
L. Tian, C. H. van der Wal, and S. Lloyd, Science 285,
1036 (1999).
[6] P. M. Platzman and M. I. Dykman, Science 284, 1967
(1999).
[7] D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120
(1998) (cond-mat/9701055); M. Sherwin, A. Imamoglu,
and T. Montroy, Quantum computation with quantum dots and terahertz cavity quantum electrodynamics, Phys. Rev. A 60, 3508 (1999) (quant-ph/9903065);
T. Tanamoto, Phys. Rev. A 61, 022305 (2000) (quantph/9902031).
[8] D. Deutsch Proc. R. Soc. London A 400, 97 (1985); 425,
73 (1989).
[9] R. P. Feynman, Int. J. Theor. Phys. 21, 467 (1982);
Found. Phys. 16, 507 (1986). See also Feynman Lectures

[15]
[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

on Computation, eds. A. J. G. Hey and R. Allen (Perseus

Press, 1996).
R. Beals, H. Buhrman, R. Cleve, M. Mosca, and R. de
Wolf, Quantum lower bounds by polynomials, Proc.
of the 39th Annual Symposium on the Foundations of
Computer Science (IEEE Press, Los Alamitos, 1998), p.
352; quant-ph/9802049.
Y. Ozhigov, Quantum computer cannot speed up iterated applications of a black box, quant-ph/9712051; B.
Terhal, PhD Thesis, University of Amsterdam, 1999.
L. K. Grover, Quantum mechanics helps in searching for
a needle in a haystack, Phys. Rev. Lett. 79, 325 (1997).
P. W. Shor, Polynomial time algorithms for prime factorization and discrete logarithms on a quantum computer, SIAM J. Comput. 26, 1484 (1997), and references
therein.
D. P. DiVincenzo and D. Loss, Quantum Computers
and Quantum Coherence, J. Magnetism Magn. Matl.
200, 202-218 (1999).
R. Cleve and H. Buhrman, Phys. Rev. A 56, 1201 (1997).
H. Buhrman, R. Cleve, and A. Wigderson, Quantum vs.
Classical Communication and Computation, in Proc. of
the 30th Ann. ACM Symp. on the Theory of Computing
(ACM Press, 1998), p. 63; eprint quant-ph/9802040.
A. Ambainis, L. Schulman, A. Ta-Shma, U. Vazirani, and A. Wigderson, The quantum communication complexity of sampling, in Proc. of the
39th Annual Symposium on the Foundations of Computer Science (IEEE Press, Los Alamitos, 1998); see
http://www.icsi.berkeley.edu/ amnon/Papers/qcc.ps.
C. H. Bennett and G. Brassard, Quantum Cryptography: Public Key Distribution and Coin Tossing, in Proceedings of the IEEE International Conference on Computers, Systems and Signal Processing, Bangalore, India
(IEEE, New York, 1984), p. 175.
D. A. Meyer, Quantum strategies, Phys. Rev. Lett. 82,
1052 (1999) (quant-ph/9804010); J. Eisert, M. Wilkens,
and M. Lewenstein, Quantum games and quantum
strategies, Phys. Rev. Lett. 83, 3077 (1999) (quantph/9806088); L. Goldenberg, L. Vaidman, and S. Wiesner, Quantum gambling, Phys. Rev. Lett. 82, 3356
(1999) (quant-ph/9808001).
A. M. Steane and W. van Dam, Physicists Triumph
at Guess my Number , Physics Today 53 (2), 35-39
(2000).
D. Aharonov, Quantum Computation, in Annual Reviews of Computational Physics, vol. VI (ed. Dietrich Stauffer, World Scientific, Singapore, 1998) (quantph/9812037).
R. Cleve, An Introduction to Quantum Complexity
Theory, to appear in Collected Papers on Quantum
Computation and Quantum Information Theory (eds. C.
Macchiavello, G. M. Palma, and A. Zeilinger, World Scientific, 2000) (quant-ph/9906111).
C. H. Bennett, Physics Today 48 (10), 24 (1995); D.
P. DiVincenzo, Science 270, 255 (1995); D. P. DiVincenzo, Proc. R. Soc. London A 454, 261 (1998) (and
quant-ph/9705009); A. Barenco, Contemp. Phys. 37, 375
(1996); A. Steane, Rep. Prog. Phys. 61, 117 (1998); C.
H. Bennett and P. W. Shor, IEEE Trans. Info. Theory

44, 2724 (1998).

[24] D. P. DiVincenzo, in Mesoscopic Electron Transport,
eds. L. Sohn, L. Kouwenhoven, and G. Sch
on (Vol.
345, NATO ASI Series E, Kluwer, 1997), p. 657 (condmat/9612126); D. P. DiVincenzo and D. Loss, Superlattices and Microstructures 23, 419 (1998) (condmat/9710259); D. P. DiVincenzo and D. Loss, J. Magn.
Mag. Matl. 200, 202 (1999) (cond-mat/9901137).
[25] B. Schumacher, Phys. Rev. A 54, 2614 (1996).
[26] G. K. Brennen, C. M. Caves, P. S. Jessen, and I. H.
Deutsch, Phys. Rev. Lett. 82, 1060 (1999).
[27] D. Jaksch, H. J. Briegel, I. J. Cirac, C. Gardiner, and P.
Zoller, Phys. Rev. Lett. 82, 1975 (1999).
[28] I am grateful to R. Schoelkopf and M. Devoret for clarifying discussion on these points.
[29] M. B. Plenio and P. L. Knight, Phys. Rev. A 53, 2986
(1996).
[30] M. B. Plenio and P. L. Knight, Proc. Roy. Soc. Lond.
A453, 2017-2041 (1997).
[31] J. Preskill, Proc. R. Soc. Lond A 454, 385 (1998) (quantph/9705031).
[32] M. A. Nielsen and I. L. Chuang, Quantum Computation
and Quantum Information (Cambridge University Press,
2000); see also M. A. Nielsen, C. M. Caves, B. Schumacher, and H. Barnum, Proc. R. Soc. Lond. A 454,
277-304 (1998) (quant-ph/9706064).
[33] P. Shor, Phys. Rev. A 52, 2493 (1995); A. M. Steane,
Phys. Rev. Lett. 77, 793-797 (1996).
[34] P. Shor, in Proceedings of the 37th Symposium on the
Foundations of Computer Science (IEEE Press, Los
Alamitos, CA, 1996) (quant-ph/9605011); D. Aharonov
and M. Ben-Or, in Proceedings of the 29th Annual
ACM Symposium on the Theory of Computing (ACM
Press, New York, 1997) (quant-ph/9611025); E. Knill,
R. Laflamme, and W. Zurek, Science 279, 342 (1998).
These results are reviewed in [31].
[35] D. P. DiVincenzo, Two-bit gates are universal for quantum computation, Phys. Rev. A 51, 1015 (1995), condmat/9407022.
[36] A. Barenco, C. H. Bennett, R. Cleve, D. P. DiVincenzo,
N. Margolus, P. Shor, T. Sleator, J. A. Smolin, and H.
Weinfurter, Elementary gates for quantum computation, Phys. Rev. A 52, 3457 (1995), quant-ph/9503016.
[37] K. Molmer and A. Sorensen, Phys. Rev. Lett. 82, 1835
(1999).
[38] D. W. Leung, I. L. Chuang, F. Yamaguchi, and Y. Yamamoto, Efficient implementation of selective recoupling in heteronuclear spin systems using Hadamard matrices, quant-ph/9904100.
[39] G. Burkard, D. Loss, D.P. DiVincenzo, and J.A. Smolin,
Phys. Rev. B 60, 11404 (1999); cond-mat/9905230.
[40] G. Burkard, D. Loss, D. P. DiVincenzo, Phys. Rev. B 59,
2070 (1999); cond-mat/9808026.
[41] D. Gottesman, Fault-Tolerant Quantum Computation
with Local Gates, J. Mod. Optics 47, 333 (2000); quantph/9903099.
[42] D. Bacon, J. Kempe, D. A. Lidar, and K. B. Whaley,
Universal fault-tolerant computation on decoherencefree subspaces, quant-ph/9909058.
[43] D. P. DiVincenzo, G. Burkard, D. Loss, and E. V. Sukho-

[44]
[45]
[46]
[47]
[48]
[49]

[50]

rukov, Quantum computation and spin electronics, in

Quantum Mesoscopic Phenomena and Mesoscopic Devices in Microelectronics, eds. I. O. Kulik and R. Ellialtioglu (NATO Advanced Study Institute, Turkey, June
13-25, 1999), to be published; cond-mat/9911245.
D. Gottesman and I. L. Chuang, Nature 402, 390 (1999).
D. Loss and E. V. Sukhorukov, Phys. Rev. Lett. 84, 1035
(2000).
J. I. Cirac, P. Zoller, H. J. Kimble, and H. Mabuchi,
Phys. Rev. Lett. 78, 3221 (1997) (quant-ph/9611017).
S. Lloyd, Science 261, 1569 (1993); 263, 695 (1994).
S. Benjamin, Phys. Rev. A 61, 020301(R) (2000) (quantph/9909007).
A. Yu. Kitaev, Fault-tolerant quantum computation
with anyons, quant-ph/9707021; see also J. Preskill, in
Introduction to Quantum Computation and Information
(eds. H.-K. Lo, S. Popescu, and T. Spiller, World Scientific, Singapore, 1998) pp. 213-269 (quant-ph/9712048).
M. H. Freedman, M. Larsen, and Z. Wang, A modular
functor which is universal for quantum computation,
quant-ph/0001108; M. H. Freedman, Poly-locality in
quantum computing, quant-ph/0001077; M. H. Freedman, A. Yu. Kitaev, and Z. Wang, Simulation of topological field theories by quantum computers, quantph/0001071.