Optical Quantum Computing

Jeremy L. O’Brien∗
Centre for Quantum Photonics, H. H. Wills Physics Laboratory & Department of Electrical and Electronic Engineering,
University of Bristol, Merchant Venturers Building, Woodland Road, Bristol, BS8 1UB, UK

In 2001 all-optical quantum computing became feasible with the discovery that scalable quantum
computing is possible using only single photon sources, linear optical elements, and single photon
detectors. Although it was in principle scalable, the massive resource overhead made the scheme
practically daunting. However, several simplifications were followed by proof-of-principle demonstra-
tions, and recent approaches based on cluster states or error encoding have dramatically reduced this
worrying resource overhead, making an all-optical architecture a serious contender for the ultimate
goal of a large-scale quantum computer. Key challenges will be the realization of high-efficiency
sources of indistinguishable single photons, low-loss, scalable optical circuits, high efficiency single
photon detectors, and low-loss interfacing of these components.
arXiv:0803.1554v1 [quant-ph] 11 Mar 2008

Over the last few decades quantum information science

has emerged to consider what additional power and func-
tionality can be realised in the encoding, transmission
and processing of information by specifically harnessing
quantum mechanical effects [1]. Anticipated technolo-
gies include: quantum key distribution [2], which offers
perfectly secure communication; quantum metrology [3],
which allows more precise measurements than could ever
be achieved without quantum mechanics; and quantum
litography [4], which could enable fabrication of devices
with features much smaller than the wavelength of light.
Perhaps the most startling and powerful future quantum
technology is a quantum computer, which promises ex-
ponentially faster computation for particular tasks [1, 5]. FIG. 1: Fig. 1. Single photon qubits. (A) A horizon-
The quest to develop a quantum computer will require tal (H) photon represents a logical “0” and a vertical (V)
formidable technical mastery of the fabrication of devices photon represents a logical “1”: |0i ≡ |Hi; |1i ≡ |V i. (B)
An arbitrary state can be plotted on the Bloch (or Poincaré)
at the nano and possibly atomic scale, and precision con- sphere. Examples of diagonal (|Di ≡ |0i + |1i), anti-diagonal
trol of their quantum mechanical states. The task is (|Ai ≡ |0i − |1i), right circular (|Ri ≡ |0i + i|1i), and left
also daunting owing to the inherent fragility of quantum circular (|Li ≡ |0i − i|1i) are shown. (C) Single qubit gates
states and the fact that quantum entanglement, and its are easily realized using birefringent waveplates that retard
role in a quantum computer, is not yet fully understood. one polarization by a fraction of a wavelength λ relative to an
As we engineer devices that exploit quantum mechani- orthogonal polarization, causing a rotation of the state on the
cal effects, we will gain an unprecedented control over Bloch sphere, with the axis of rotation determined by the ori-
entation of the waveplate. For example, a Hadamard (H) gate
the fundamental workings of nature as well as a deeper (defined by its operation on the logical states: |0i → |0i + |1i;
understanding of them. |1i → |0i − |1i or |Hi → |Di; |V i → |Ai) causes a π rota-
The requirements for realizing a quantum computer tion about an axis running through the midpoint between the
are confounding: scalable physical qubits—two state |Hi and |Di states, and can be realized by a λ/2 waveplate
quantum systems—that can be well isolated from the oriented at 22.5◦ . An arbitrary rotation requires a λ/4–λ/2–
environment, but also initialised, measured, and control- λ/4 sequence. (D) Converting between polarization and path
encoding requires a polarizing beam splitter (PBS), which
lably interacted to implement a universal set of quantum
transmits H and reflects V , and a λ/2 waveplate oriented at
logic gates [6]. However, a number of physical implemen- 45◦ , which transforms |V i ↔ |Hi.
tations are being pursued, including nuclear magnetic
resonance, ion, atom, cavity quantum electrodynamics,
solid state, and superconducting systems [7]. Over the
last few years single particles of light—photons—have polarization, time bin, path, etc. Figure 1A shows how
emerged as one of several leading approaches [7]. a qubit can be encoded in the polarization of a single
Single Photons as Qubits photon. An arbitrary state of a single qubit α|Hi + β|V i
Single photons are largely free of the noise—or (|α|2 + |β|2 = 1) can be represented on the Poincaré
decoherence—that plagues other systems; can be eas- (or Bloch) sphere (Fig. 1B). One-qubit logic gates are
ily manipulated to realize one-qubit logic gates; and en- straightforward using birefringent waveplates (Fig. 1C)
able encoding in any of several degrees of freedom— and converting between polarisation and path encoding
 2

FIG. 2: Fig. 2. An optical controlled-NOT gate. (A)

Schematic of a possible realization of an optical CNOT gate.
See text for details. (B) In the notation of quantum circuits
the BSs implement a Hadamard (H) gate.

can be easily achieved using a polarizing beam splitter

(Fig. 1 D), where |0i or |1i now represents a photon in
the upper or lower path, respectively.
FIG. 3: Fig. 3. An optical CNOT gate via teleportation (A)
A major difficulty for optical quantum computing is in A cartoon of a measurement assisted nondeterministic CNOT
realizing the entangling logic gates required for univer- gate. (B) Quantum interference of two photons at a BS. (C)
sal quantum computation. The canonical example is the Teleportation of a CNOT: Ignoring the “Probabilistic Gate”,
controlled-NOT gate (CNOT), which flips the state of a a qubit in an unknown state |Ci and one of two photons pre-
target (T) qubit conditional on a control (C) qubit being pared in a maximally entangled state |φi are subjected to
in the logical state “1”. Figure 2A shows why this opera- a Bell measurement (B). This measurement leaves the third
tion is difficult. The two paths used to encode the target qubit in the state |Ci or the bit (X) and/or phase (Z) flipped
version of |Ci, depending on which of the four maximally en-
qubit are mixed at a 50% reflecting beamsplitter (BS) tangled states is measured. An unwanted X and/or Z flip
(or half-silvered mirror), which performs the Hadamard can be trivially corrected by applying a second X and/or Z
operation (see Fig. 1C caption). If the phase shift is as required. Still ignoring the “Probabilistic Gate”, the un-
not applied, the second Hadamard (BS) undoes the first, known input state of the control and target qubits can both
returning the target qubit to exactly the same state it be teleported and the CNOT performed on the output qubits.
started in (this is an example of (classical) wave inter- This seems like a lot of extra work for no gain, however, per-
ference). If, however, a (π) phase shift is applied—i.e. forming the CNOT before the (possible) X and Z flip has the
tremendous advantage that we could repeatedly attempt the
|0i+|1i ↔ |0i−|1i—the target qubit undergoes a bit-flip CNOT on the two halves of two entangled states, and only
or NOT operation. A CNOT must implement this phase when the gate works would we proceed with teleportation. In
shift only if the control photon is in the “1” path. No this way the control and target qubits are preserved until the
known or foreseen material has an optical non-linearity gate works (on average 32 entangled states will be consumed)
strong enough to implement this conditional phase shift and we can implement the CNOT deterministically. In quan-
(although tremendous progress has been made with sin- tum mechanics the order in which operations are performed
gle atoms in high-finesse optical cavities [8, 9, 10]). is important; performing the CNOT earlier means we must
add the X and Z flips indicated in red in Fig. 3C [13].
In 2001 a major breakthrough showed that scalable
quantum computing is possible using only single photon
sources and detectors, and simple (linear) optical circuits
consisting of BSs [11]. This is a truly remarkable dis- tons emerge, having had the CNOT logic operation ap-
covery since the argument above suggests that a strong plied to their state, conditional on a single photon being
optical non-linearity is required to realize the most basic detected at both detectors. This detection event occurs
logic element. with probability P < 1 (1/16 in the original scheme); the
Linear Optical Quantum Computing rest of the time (P = 15/16) another detection pattern
A cartoon of a non-deterministic (probabilistic with suc- is recorded (none, only one, two photons at one detector,
cess signal) CNOT is shown in Fig. 3A. The control etc) and the CNOT logic is not applied—in fact a sin-
and target qubits (encoded in polarization, say), together gle photon may not even emerge from the control and/or
with two auxiliary photons, enter an optical network of target outputs in these cases.
BSs, where the four photons’ paths are combined. At A nondeterministic CNOT is of little use for quan-
the output of this network, the control and target pho- tum computing as the probability that a computation
 3

succeeds decreases exponentially with the number of

CNOTs. Fortunately, the success probability of the non-
deterministic CNOT can be boosted by harnessing quan-
tum teleportation [12]—a process whereby the unknown
state of a qubit can be transfered to another qubit. The
idea is to teleport a non-deterministic gate that has al-
ready worked onto the control and target qubits [13] (see
Fig. 3 caption for details). Quantum teleportation has
been realized with single photons [14].
An important omission from the above discussion is
that since the Bell measurements required for telepor-
tation (Fig. 3C) measure maximally entangled states
they requires a similar optical nonlinearity to a CNOT FIG. 4: Fig. 4. Cluster state quantum computing. For pho-
tons it is practical to start measuring the qubits, while the
(although the photons can be destroyed in the measure- cluster is still being grown: The blue qubits are in a cluster
ment) and therefore fail some of the time. When they fail, state, where the bonds between them represent entanglement.
they measure the state of the control and target photons The green qubits are being added to the cluster, while the grey
in the {|0i, |1i} basis. The final component is an en- qubits have been measured and are no longer entangled. The
coding against this “measurement error”: a single logical measurement outcome determines the basis for the measure-
qubit is encoded in several physical qubits such that if ment on the next qubit.
one of the physical qubits is measured, the original log-
ical qubit can still be recovered. These encoded states
are entangled and therefore require entangling gates to
realize them. However, by using more and more photons deterministic CNOTs. It turns out that a similar advan-
a CNOT with success probability approaching one can tage can also be gained in the circuit forumulation of op-
be realized [11]. tical quantum computing by using more sophisticated er-
Reducing the Resource Overhead ror encoding techniques [29]. These, and other techniques
These developments were expanded upon [15, 16, 17, 18, that dispense with CNOT gates entirely [30], reduce the
19] and soon followed by several proof-of-principal ex- resources required by 3-4 orders of magnitude, making
perimental demonstrations of CNOTs [20, 21, 22, 23] an all-optical approach far more attractive. There have
and encoding against measurement error [24, 25]. De- already been experimental proof-of-principle demonstra-
spite this great progress, optical quantum computing was tions of these new schemes (eg. [31, 32, 33, 34]).
still widely regarded as impractical owing to the large re- Fault Tolerance
source overhead required to realize a near-deterministic The final, and arguably most important consideration
CNOT: >10,000 pairs of entangled photons to achieve a (for all physical realizations), is fault tolerance [1]. In
success probability of >95%. The reason that all-optical contrast to conventional computers, quantum computers
quantum computing is today a promising route to practi- will be very susceptible to noise, which must be encoded
cal quantum computing is due to new schemes that dra- against (in addition to the encoding described above).
matically reduce this worrying resource overhead. The threshold theorem says that if the noise is below
Quantum computations (regardless of physical realiza- some threshold an arbitrarily long quantum computa-
tion) are typically formulated using the quantum circuit tion can be realized. One of the most encouraging results
model (eg. Fig. 2B), a generalization of the circuit model for all approaches to quantum computing was the high
for Boolean logic: qubits are represented by wires propa- threshold of 1% recently reported by Knill [35]. Because
gating in time from left to right, subjected to a sequence cluster state approaches do not conform to the standard
of quantum logic gates, and finally measured [1]. In model the threshold theorem does not apply; fortunately
2001 a remarkable alternative was proposed in which the analogous thresholds have been shown to exist [36]. Re-
computation starts with a particular massively entangled cent results give cause for optimism: They show that if
state of many qubits—a cluster state—and the computa- the product of source and detector efficiency is > 2/3
tion proceeds via a sequence of single qubit measurements then optical quantum computing is possible, provided all
from left to right that ultimately leave the rightmost col- other components operate perfectly [37] (photon loss can
umn of qubits in the answer state [26] (Fig. 4). in some cases be incorporated into source or detector ef-
In 2004 it was recognized that the cluster approach of- ficiency). More complete treatments that consider more
fered tremendous advantages for optical realizations [27] sources of noise give thresholds of 10−3 − 10−4 [38]. The
(see also [28]). Because preparation the cluster state can true number will likely lie somewhere in between.
be probabilistic, non-deterministic CNOTs are suitable Sources, Detectors, and Circuits
for making it, removing much of the massive overhead There are very stringent requirements for single photon
that arises from the error encoding used to make near- sources for optical quantum computing. In a general lin-
 4

ear optical network (eg. Fig. 3A) there are places where gested in Fig. 2, but would nevertheless offer significant
photons arrive at both inputs to a BS where quantum advantages. One is to use a two-photon absorber to im-
interference of two (or more) photons can occur. An ex- plement the quantum Zeno effect, whereby repeated mea-
ample is shown in Fig. 3B where a photon enters each surement inhibits the emission of two photons into one
input of a 50% reflective BS. The probability of detecting of the outputs of a CNOT gate [45]—the failure mode of
a single photon at each output is given by the square of the linear optical CNOT gate proposed in [16]. Another
the sum of the probability amplitude for both photons to is to use a strong optical nonlinearity that is significantly
be transmitted and that for both photons to be reflected: weaker than that required in Fig. 2: single photons are
P = |r.r + t.t|2 . Because a phase shift occurs on reflec- made to interact with one-another via a bright laser pulse
tion r.r = −t.t and so P = 0, in contrast to our (classical) and the nonlinear medium [46]. Finally, recent develop-
expectation: P = 1/2 [39]. In order for quantum interfer- ments suggest that a hybrid approach may have many
ence to occur the two photons must be indistinguishable advantages [47]: because single photon sources are in-
from one another in all degrees of freedom. herently quantum mechanical, it is promising to consider
To date, small scale tests of optical quantum comput- storing quantum information in the sources themselves;
ing have relied on indistinguishable pairs of photons gen- already spins associated with impurities in diamond have
erated by a strong laser pulse in a non-linear crystal. shown great promise in this direction [40, 48]. Such sys-
Unfortunately this process is spontaneous and not read- tems are particularly suited to the small-scale quantum
ily scalable [40]. Solid state sources of single photons hold processors that will be required in the nodes and quan-
the promise of ready integration, and quantum interfer- tum repeaters of quantum communication networks.
ence between subsequent photons emitted from a semi- Future Prospects
conductor quantum dot has been observed [41]. How-
Despite great progress, much work remains to be done
ever, an optical quantum computer will require quantum
if a large-scale optical quantum computer is to be real-
interference between photons emitted from independent
ized. It is not yet known whether the circuit or cluster
sources. This has very recently been achieved for a pair of
model (or some other approach) is most promising; in-
trapped atoms [42] and ions [43], a tremendous advance
deed a combination of these approaches where error en-
that bodes well for optical quantum computing. Impu-
coding is achieved using cluster techniques but the com-
rities in diamond may offer the best of both worlds—a
putation proceeds via conventional CNOT gates has been
solid state host and atom-like energy levels—and have
described [49]. Further, the role of nonlinear optics ap-
emerged as very promising candidates [40].
proaches in any future optical quantum computer will
It is actually the inherent non-linarity of photon mea-
depend on their efficacy and practicality. The majority
surement, combined with quantum interference of pho-
of experimental demonstrations to date have relied on
tons, that makes linear optical quantum computing pos-
a non-scalable single photon sources, large-scale optical
sible. Single photon counting modules are commer-
elements, and modest efficiency single photon detectors;
cially available and have been used almost all demon-
scaling to useful devices will require high efficiency single
strations to date, however, they cannot distinguish be-
photon sources and detectors that are efficiently coupled
tween one or more photons and have a limited efficiency
to low-loss microscopic optical waveguide circuits (opti-
(∼70%). Higher efficiency will be required for scalable
cal memories may not be required [50]).
optical quantum computing, while photon number res-
olution will be desirable. Ongoing work indicates that
such high performance detectors will become available,
with superconductor-based devices holding great promise
[44]. ∗
Electronic address: [email protected]
Finally, almost all demonstrations of linear optical [1] M. A. Nielsen, I. L. Chuang, Quantum Computation
logic circuits have relied on large scale BSs and mirrors, and Quantum Information (Cambridge University Press,
with photons propagating in air; improved performance, 2000).
[2] N. Gisin, G. Ribordy, W. Tittel, H. Zbinden, Rev. Mod.
miniaturisation, and scalability will likely require low-loss
Phys. 74, 145 (2002).
microscopic optical waveguide circuits. A promising ap- [3] V. Giovannetti, S. Lloyd, L. Maccone, Science 306, 1330
proach is integrated optics—an analogue of electrical in- (2004).
tegrated circuits—which has been developed by the pho- [4] A. N. Boto, et al., Phys. Rev. Lett. 85, 2733 (2000).
tonics industry. Outstanding challenges are to realize [5] D. Deutsch, Proc. R. Soc. Lond. A 400, 97 (1985).
quantum interference in these devices and to integrate [6] D. P. DiVincenzo, D. Loss, Superlatt. Micro. 23, 419
them with single photon sources and detectors. (1998).
[7] See the US Advanced Research and Devel-
Nonlinear and Hybrid Approaches opment Activity (ARDA) Quantum Compu-
Recently attention has been given to the idea of combin- tation Roadmap for current state-of-the-art:
ing linear optics with optical nonlinearities that would http : //qist.lanl.gov/qcompm ap.shtml.
not allow a CNOT gate to be realized in the fashion sug- [8] Q. A. Turchette, C. J. Hood, W. Lange, H. Mabuchi,
 5

H. J. Kimble, Phys. Rev. Lett. 75, 4710 (1995). [43] P. Maunz, et al., Nature Phys. 3, 538 (2007).
[9] T. Aoki, et al., Nature 443, 671 (2006). [44] See special issue: Single-photon: detectors, applications,
[10] M. Hijlkema, et al., Nature Phys. 3, 253 (2007). and measurement methods, Eds. A. Migdal and J. Dowl-
[11] E. Knill, R. Laflamme, G. J. Milburn, Nature 409, 46 ing, J. Mod. Opt. 51 (2004).
(2001). [45] J. D. Franson, B. C. Jacobs, T. B. Pittman, Phys. Rev.
[12] C. H. Bennett, et al., Phys. Rev. Lett. 70, 1895 (1993). A 70, 062302 (2004).
[13] D. Gottesman, I. L. Chuang, Nature 402, 390 (1999). [46] S. D. Barrett, et al., Phys. Rev. A 71, 060302 (2005).
[14] D. Bouwmeester, et al., Nature 390, 575 (1997). [47] Y. L. Lim, S. D. Barrett, A. Beige, P. Kok, L. C. Kwek,
[15] M. Koashi, T. Yamamoto, N. Imoto, Phys. Rev. A 63, Phys. Rev. A 73, 012304 (2006).
030301 (2001). [48] See forthcoming special issue: Measurement-Based
[16] T. B. Pittman, B. C. Jacobs, J. D. Franson, Phys. Rev. Quantum Information Processing, Eds. J.-W. Pan and
A 64, 062311 (2001). T. Rudolph, New J. Phys. (2007).
[17] T. C. Ralph, A. G. White, W. J. Munro, G. J. Milburn, [49] A. Gilchrist, A. J. F. Hayes, T. C. Ralph, Phys. Rev. A
Phys. Rev. A 65, 012314 (2001). 75, 052328 (2007).
[18] T. C. Ralph, N. K. Langford, T. B. Bell, A. G. White, [50] K. Kieling, T. Rudolph, J. Eisert, arXiv:quant-
Phys. Rev. A 65, 062324 (2001). ph/0611140 (2006).
[19] H. F. Hofmann, S. Takeuchi, Phys. Rev. A 66, 024308 [51] I am indebted to all my collaborators past and present
(2001). who have helped in my understanding of this subject. Op-
[20] T. B. Pittman, M. J. Fitch, B. C. Jacobs, J. D. Franson, tical quantum computing research at Bristol is supported
Phys. Rev. A 68, 032316 (2003). by the U.K. Engineering and Physical Sciences Research
[21] J. L. O’Brien, G. J. Pryde, A. G. White, T. C. Ralph, Council, the U.K. Quantum Information Processing In-
D. Branning, Nature 426, 264 (2003). terdisciplinary Collaboration, the U.S. Disruptive Tech-
[22] J. L. O’Brien, et al., Phys. Rev. Lett. 93, 080502 (2004). nologies Office, the E.U. Integrated Project Qubit Ap-
[23] S. Gasparoni, J.-W. Pan, P. Walther, T. Rudolph, plications, the Leverhulme Trust, and the Daiwa Anglo-
A. Zeilinger, Phys. Rev. Lett. 93, 020504 (2004). Japanese Foundation.
[24] J. L. O’Brien, G. J. Pryde, A. G. White, T. C. Ralph,
Phys. Rev. A 71, 060303 (2005).
[25] T. B. Pittman, B. C. Jacobs, J. D. Franson, Phys. Rev.
A 71, 052332 (2005).
[26] R. Raussendorf, H. J. Briegel, Phys. Rev. Lett. 86, 5188
(2001).
[27] M. A. Nielsen, Phys. Rev. Lett. 93, 040503 (2004).
[28] N. Yoran, B. Reznik, Phys. Rev. Lett. 91, 037903 (2003).
[29] T. C. Ralph, A. J. F. Hayes, A. Gilchrist, Phys. Rev.
Lett. 95, 100501 (2005).
[30] D. E. Browne, T. Rudolph, Phys. Rev. Lett. 95, 010501
(2005).
[31] P. Walther, et al., Nature 434, 169 (2005).
[32] N. Kiesel, et al., Phys. Rev. Lett. 95, 210502 (2005).
[33] R. Prevedel, et al., Nature 445, 65 (2007).
[34] C.-Y. Lu, et al., Nature Physics 3, 91 (2007).
[35] E. Knill, Nature 434, 39 (2005).
[36] M. A. Nielsen, C. M. Dawson, Phys. Rev. A 71, 042323
(2005).
[37] M. Varnava, D. Browne, T. Rudolph, arXiv:quant-
ph/0702044 (2007).
[38] C. M. Dawson, H. L. Haselgrove, M. A. Nielsen, Phys.
Rev. A 73, 052306 (2006).
[39] Quantum mechanics tells us that if a particular event can
happen in two or more indistinguishable ways we must
first sum the probability amplitudes before squaring to
obtain a probability. Because these amplitudes can be
negative, it is possible for events that we would intuitively
expect to be possible to have zero probability. If the BS
is more or less reflective than 50% these two amplitudes
no longer have the same magnitude, but are still opposite
in sign, meaning that the probability is nonzero, but less
than one would naively expect.
[40] See special issue: Focus on Single Photons on Demand,
Eds. P. Grangier, B. Sanders, and J. Vuckovic, New J.
Phys. 6 (2004).
[41] C. Santori, D. Fattal, J. Vuc̈kovic̈, G. S. Solomon, Y. Ya-
mamoto, Nature 419, 594 (2002).
[42] J. Beugnon, et al., Nature 440, 779 (2007).