Perspective Vol. 1, No.

1 / 25 October 2023 / Optica Quantum 35

Light in quantum computing and simulation:

perspective
Ian Walmsley
Blackett Laboratory, Department of Physics, Imperial College London, London, SW7 2AZ, UK

Received 25 September 2023; revised 18 October 2023; accepted 18 October 2023; published 26 October 2023

A summary is given of recent progress in photonic quantum simulation and computation. Non-error-corrected
machines performing specialised tasks have already demonstrated a quantum advantage over the best algorithms
running on conventional computers, and practical applications for such machines are being explored. Mean-
while, designs for error-corrected fault-tolerant quantum computers based on light are reducing the performance
requirements for individual components and systems, although the engineering challenges are severe. Light also
plays a central role in other platforms for quantum computing and simulation, from control of individual atomic
qubits to remote entanglement of separate processing nodes, along with an important role in communications and
other long-distance networks.
Published by Optica Publishing Group under the terms of the Creative Commons Attribution 4.0 License. Further distribution of
this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI.
https://doi.org/10.1364/OPTICAQ.507527

Light is a unique medium for quantum technologies: it exhibits algorithm that seeks to sample the measurement statistics of
quantum features, such as entanglement and noise below the a multi-particle, multi-mode Bosonic quantum state. If such
familiar shot noise, even in ambient conditions, and its large a state has been prepared by colliding a set of independent
bandwidth provides high capacity for encoding and manip- bosons with one another, then it turns out that even approxi-
ulating information. For these reasons optical systems are a mating the distribution of bosons at the output modes is likely
critical element of future quantum technologies, from obvi- a computationally hard task; that is, one that cannot be effi-
ous applications such as imaging and remote sensing, to ciently undertaken by a classical computer [1]. Therefore, direct
optical communications and even to computing and quantum measurements are the most efficient means to draw samples from
simulation. the distribution.
It is in these latter areas that recent progress has been sig- A Boson Sampling device using light, as shown in Fig. 1(a),
nificant, achieving clear quantum advantage in some limited can be built by taking a large collection of single photons, each
processing tasks as well as opening the door to new algo- in a separate mode so forming a set of independent particles,
rithms relevant to real-world applications. Further, design of mixing the modes on a set of beamsplitters—a linear optical
increasingly loss-tolerant schemes for producing large-scale network—and then counting the number of photons that come
entangled states necessary for quantum computing, along with out of each port of this network. Under specific achievable cir-
rapid progress in the performance of components needed to cumstances this mimics the Arkhipov–Aaronson protocol and
realise them, has spurred large investment in a business sec- allows testing whether sampling using such a quantum device is
tor dedicated to constructing a future fault-tolerant, scalable more rapid that of calculation on a super-computer to estimate
photonic quantum computer. the output distribution from the network design.
Quantum advantage in computing—the ability of a quantum The physical origin of the hardness of this problem arises
machine to perform a task more rapidly than a conventional from quantum interference. Consider a simple 2 × 2 network—a
computer (optimally, and in principle, exponentially faster)—is single 50:50 beamsplitter—into which one photon enters at each
a benchmark that machines achieving this can add real value to port. A well-known result is that if the photons are in a sepa-
information processing tasks. It is not a hard limit, but contingent rable state in identical modes then they will both exit the same
on the task being performed. Recently such advantage has been port. The probability that they exit through different ports is
achieved in quantum simulators. These are devices that seek to zero [2]. This probability is proportional to the Permanent [3] of
encode a particular problem directly in the architecture of the the transfer function of this network, which in this case is zero.
computing machine. The problem may be a representation of a Calculating this function is possible for a smallish network.
physical system the dynamics or structure of which is sought, or However, the known algorithms to calculate the Permanent [4]
it may be a more abstract calculation. do not scale favorably with the size of the network and number
Two recent demonstrations use optical implementations of of photons. (That is, the time they take to run increases more
a version of Boson Sampling (BS). Boson Sampling is an

2837-6714/23/010035-06 Journal © 2023 Optica Publishing Group

 36 Vol. 1, No. 1 / 25 October 2023 / Optica Quantum Perspective

a
1

0
Photon detection
Quantum light
2 c

1
Elastic scattering,
no interaction Photon detection
probability
b

Possible Output Combinations

Squeezed states (S )
T
Courtesy C-Y Lu

Coherent states (D )

Linear Optical
Circuit

Fig. 1. (a) A linear quantum optical network operating as a Boson Sampling machine. Individual photons are inserted at the input nodes.
The photons scatter from effective internal beamsplitters, and the number of photons at output nodes of the network are counted. Quantum
interference of the Bosonic fields occupied by the photons means that it is computationally hard to calculate the output photon distribution
if the network is randomly chosen. (b) A Gaussian Boson Sampling device consisting of a set of squeezed (S) or classical (laser) light
(D) sources, a linear optical network, and a set of photon-number-resolving detectors. The resulting joint photocount distribution across all
channel combinations is also hard to estimate using known algorithms running on conventional computers. (c) Laboratory implementation of
a 100-mode GBS machine using build optics to minimise mode-coupling losses. (Photograph courtesy of Chaoyang Lu, USTC, China)

rapidly than a polynomial function of these parameters.) There- optical circuit transfer function parameters)—a so-called Haf-
fore, when there are enough photons and a big enough network, nian. There is also no known efficient classical algorithm for
it takes longer to run the calculation of the output photon dis- finding the values of this function, and similar arguments to the
tribution on a supercomputer than to record a sufficient number case of BS show that it is also therefore hard to sample from the
of samples in a physical device to approximate the distribution. output photon number distribution efficiently.
There are challenges to building such a system, of course, Interestingly, the use of photon counting as a measurement
and there is also the difficulty of knowing what is the best algo- scheme is critical for GBS, since it turns out that direct measure-
rithm that can run on a conventional computer against which the ments of the field quadrature amplitudes at the output of the
quantum device may be tested. First, it is important that each machine would render the problem amenable to efficient emu-
input photon is the same as all the others in its spatio-temporal lation on a classical computer. The quality of the input light for
shape (and polarization), and it is important that as few as possi- GBS is specified by how close it is to a single-mode squeezed
ble are lost in propagating through the network. Then they must state—this is often called the state purity—as well as how sim-
be detected with high efficiency. Such imperfections facilitate ilar the light pulses are in spatial, temporal, and spectral shape,
calculating the output photon statistics, since they mean that the and whether this changes from run to run of the machine—this
system operates partly in a classical regime. In parallel, algo- is often called the indistinguishability.
rithms that make use of these imperfections to shorten the run An important question is exactly how big a network must be
time for estimating the output photon distribution on a classical constructed to demonstrate operation beyond that of a classi-
computer are being developed. cal computer and, therefore, how many quantum light sources
A more recent version of a Boson Sampler uses not individ- and how many detectors are needed. This is a bit of a moving
ual photons, but another form of non-classical light, primarily target, since there is no formal definition of “advantage.” Fur-
squeezed states [5]. These are states for which the noise in the ther, supercomputers, and better algorithms for them, continue
two quadratures of the optical field is not equal, as it is in, say, to improve at a rapid pace. Nevertheless, there are some reason-
laser light, but is lower than shot noise in one quadrature. Cor- able expectations that at a scale of many tens of photons and
respondingly it must be larger than shot noise in the other. A similarly sized networks, it is possible to do better than the best
squeezed state may be characterized by the joint distribution current and foreseeable next generation of supercomputers. Two
of its quadrature amplitudes, which turns out to be Gaussian. experiments in the past two years have implemented GBS in dif-
Since a linear optical network preserved this Gaussian char- ferent ways and have shown this advantage. They illustrate well
acter, the protocol is called Gaussian Boson Sampling (GBS), two very different physical arrangements, both of which suggest
as illustrated in Fig. 1(b). It nevertheless still relies on counting directions of improvement for the future.
photons at the output, where the joint photon number distribution First, a group at USTC in China built a machine consisting of
is specified by evaluating a different function of the machine’s 100 free-space optical beams connected by a bulk-glass mono-
properties (the strength of the squeezed light sources and the lithic beamsplitter array, together with squeezed light sources
based on parametric downconverters [6]. The output beams
 Perspective Vol. 1, No. 1 / 25 October 2023 / Optica Quantum 37

were measured using photodetectors that told whether there another. Two approaches show promise, but are not yet viable
were either no photons or one photon or more. This machine, for scaling. These are cavity-based single-atom switches and
named Jiŭzhāng, acquired and measured photocount distribu- interactions between photon-excited atomic pairs. In the first of
tions in 200 seconds, and simulations were estimated to take these approaches, the transition of an atom from one state to
more than 500 million years on a current supercomputer. This another by absorption of a single photon can shift the cavity
result stimulated the exploration of new algorithms that could resonance so that a second photon is reflected with a phase
exploit imperfections in the machine (losses, distinguishability) shift compared to a reflection from the “normal” cavity [12]. In
to simplify the calculations. These showed that classical algo- the second approach, the excitation of a pair of atoms by two
rithms could then speed up by many orders of magnitude the photons causes each to experience a phase shift that is different
computational time for a supercomputer to estimate the photon than when a single photon excites the same pair of atoms. This
distributions [7]. Nevertheless, the claim of advantage remains. is because the doubly excited state of the atomic pair has a lower
A second group at Xanadu in Canada, built a time-multiplexed energy, due to the atomic dipole interaction, than the sum of the
network using optical fibers to construct a time-bin coded net- energies of the two singly excited atoms individually [13,14].
work [8]. This machine, called Borealis, has the advantage of The lack of a robust single photon conditional gate means that
reconfigurability, as well as requiring only a single light source. there is no simple gate model for a quantum computer. Further,
It does require photon-number-resolving detectors fast enough photons can easily be lost through scattering or absorption, thus
to distinguish the time bins. erasing the information being processed. Nevertheless, means
While the algorithms to evaluate Hafnians and Permanents to overcome these limitations have been proposed, and the main
are singular in enabling the demonstration of a quantum advan- challenges at present are getting device performance to the point
tage in photonic simulators, they also map to some interesting that they can enable a fully scalable quantum computer.
and potentially useful problems, particularly in the analysis of The lack of a conditional interaction between single photons
graphs [9]. A specific task is the identification of completely is in contrast to the matter-based schemes, where either there is
connected subgraphs of a larger graph [10]. This is related to a direct electromagnetic interaction, or one mediated by radia-
search and optimization problems in logistics and in molecular tion, between stable material qubits which form the register. In
simulation, for instance. It turns out that the number of such sub- optics it turns out that it is possible to replace such interactions
graphs is given by the Hafnian of the graph adjacency matrix. with quantum interference and measurement [15]. The key idea
By constructing a GBS machine with the appropriate squeezing is that if there are multiple routes by which a photon may arrive
and unitary circuits, it is possible to efficiently find the probabil- at a detector then it is unknown which photon and which path
ities for sub-graphs of different sizes. The performance is better was actually registered. Since quantum mechanics demands that
than a random sampling of the graph to locate the sub-graphs all the probability amplitudes for all possible pathways leading
[11]. This has recently been successfully applied to the problem to a particular event must be added, then detection of a photon
of finding the optimal docking structures for molecules in target requires the superposition of all the routes by which all input
receptors, which may benefit research in biomedicine. photons could lead to a single event at the photodetector. This
The BS and GBS protocols have shown that quantum leaves the remaining photons in a superposition of paths that do
machines can perform certain computational tasks better than not lead to a photodetection event. So photons that do not phys-
conventional computers running the best known algorithms. ically encounter one another can become entangled by erasing
The combination of these protocols (e.g., using single pho- the information about where they started from [16]. An exam-
tons as well as squeezers and classical light) enables simulation ple of this is shown in Fig. 2(a), where mixing the paths by
of some physical problems and, therefore, opens the door to which each of two input photons may arrive at detector A or
new applications. How extensive these may be is yet to be A′ , yet registering only one photon, leads to the other being in a
determined, and there is much territory to be explored. Fur- superposition of paths.
ther, the development of more efficient quantum light sources These approaches are necessarily probabilistic, since it is not
(such as deterministic single photon emitters, waveguide cav- always the case that only detector A or A′ records a photon.
ity based squeezers) lower-loss reconfigurable optical networks, Sometimes both do, and sometimes neither do so. These events
and photon-counting detectors (such as integrated, higher- are discarded, and another trial is run. In the example of Fig. 2(a),
temperature superconducting devices) will enable machines of the desired output (a single photon superposed in the two output
even larger scale. modes) occurs only half the time. The very indeterminacy that
In the long term a quantum simulator for larger size problems enables coherent superpositions is just that which renders the
and more general tasks will be needed. This will be a fault- state preparation random.
tolerant, scalable quantum computer, universally programmable This general approach can be extended to many photons,
to execute all quantum algorithms. There are still several plat- enabling large entangled states to be constructed from a col-
forms in contention for the prize of reaching this objective. lection of independent input photons. Two examples are shown
All-optical schemes are among these. Other platforms, such as in Figs. 2(b) and 2(c), where the outputs are a so-called Bell
ions held in an electric trap, often use light for control of the state—an entangled state of two photons in four modes, and a
processing elements. Another very promising approach uses Greenberger–Horne–Zeilinger (GHZ) state—an entangled state
superconducting circuits, in which different matter processing of three photons in six modes. In each case, the probability
units (transmons) are connected by means of a microwave signal. of a successful outcome decreases as the number of photons
Optical systems have the very important feature that they can increases, so that, for example, this circuit in Fig. 2(b) only
operate largely in ambient conditions, can be easily integrated generates a Bell pair in 4/16 of the trials.
to achieve large scale, and can be operated at high speed. They Nevertheless, the concept of trial and error makes it possible
have the considerable drawback that there are few currently to overcome the barrier of indeterminacy. If an event is random,
available means by which single photons can interact with one but you know when it happens, then storing the outcome of
 38 Vol. 1, No. 1 / 25 October 2023 / Optica Quantum Perspective

Single photon
2x2 beamsplitter
Detector

Entangled state

A1 A2

A1 A2 A1 A2 A3 A4

B1 B2 B1 B2
a b c

Fig. 2. Schemes for generating multiphoton entangled states from sets of single independent photons. These superpositions of correlated
multiphoton states are a resource for quantum computing. The scheme works by erasing the information about which photon and which path
leads to a heralding photo-detection event at the banks (An , Bm ) of detectors. Designs are shown for (a) coherent superposition of a single
photon in two modes—a one-photon state; (b) coherent superposition of two photons in four modes—a Bell state; (c) coherent superposition
of three photons in six modes—a 3-GHZ state. The scheme is extensible to produce more complex entangled states (e.g., cluster states) by
combining GHZ states using related path-erasure and detection methods, a process known as “fusion.”

successful trials (or running many trials at once and keeping the interference outlined previously. Rather, current approaches to
successful ones) means that for a machine of sufficient scale it is photonic quantum computers make use of measurement-based
guaranteed that there will be an appropriate state available when architectures, in which a large-scale entangled state is prepared,
needed. measurements are made according to the algorithm being run,
There are proposals for producing some types of cluster states and feedforward from the outputs of these measurements inform
on demand, based on the emission of single photons from atoms the next set of measurements to be made on another part of the
or excitons (in cavities or waveguides) in which an internal state [16].
metastable state of the atom is changed upon emission. By The kind of state that must be prepared is a so-called three-
manipulating this state (in effect creating a superposition of dimensional cluster state, in which photons are entangled with
atomic states analogous to the function of the beamsplitters in several neighbours using interference and detection. An exam-
Fig. 2) and sequentially generating photons from this superpo- ple of such a state is represented in Fig. 3(a), showing a group
sition, a linear cluster state can be prepared. In the long run, of photons each entangled with four of its neighbours (in the
this may prove a more resource efficient way to initiate the X–Y plane) as well as with temporally earlier and later pho-
preparation of cluster states for photonic quantum computers. tons (in the T (time) direction). This approach is well suited to
The loss of photons compounds the problem of indeterminacy, optics, both because of the means by which states are prepared
both because it reduces the rates at which successful operations and because the construction of such states is sequential, with
occur, and, more seriously, means the prepared state is not the different “layers” of the state being prepared at different times.
desired state. However, again with sufficient scale it is possible The cluster state is generally prepared using GHZ states, and
to construct entangled states with the appropriate degree and the number of these required depends on the losses in the optical
connectivity to be useful. system, along with other imperfections, such as distinguishabil-
The original approach to a linear optical quantum computer ity of the photons. The best versions of this approach require
envisioned a gate-based architecture, in which the quantum state end-to-end efficiencies (including detection efficiencies) of well
of one photon could be conditioned on the state of another. above 90%, and require millions of photons per encoded qubit
However, this is not the optimal route since it demands too in the cluster [17]. The demands on individual components are
great an overhead of resources given the improbability of the therefore severe.
success of a gate operation based on the concepts of photon
 Perspective Vol. 1, No. 1 / 25 October 2023 / Optica Quantum 39

a c

Courtesy S. Yu
Fig. 3. (a) 3D entangled states (a “cluster state”) represented as an FCC lattice. Each vertex of the lattice represents an entangled photonic
resource state as suggested in (b), consisting of multiple photons and created by fusing GHZ states using methods similar to those of Fig. 2.
Measurements are made on the photons at lattice vertices, which enables both error correction and readout of the state of the logical qubit. The
algorithms that run on this type of quantum computer work by specifying the sequence in which measurements are to be made on different
“layers” of the lattice, conditioned on the outcome of measurements on previous layers. (Adapted from Ref. [17].) (c) A 10 × 10 mode SiN
photonic integrated circuit forming a mesh of beamplitters that can implement any operation between input and output mode combinations.
Such chips enable the generation of cluster states by means of the circuits shown in Fig. 2. (Photograph courtesy of S. Yu, Imperial College
London)

The need for a large number of photon sources, interferom- which use transmons as the qubits, connected by microwaves,
eters, and detectors lends itself to integration, since photonic and trapped neutral atoms, which use excited electronic states
integrated circuits can manage and route light using waveg- as qubits, connected by dipolar interactions. These platforms
uides. Each waveguide has a cross-section of about one micron utilise large arrays of cooled ions or atoms in free-space traps,
per mode. This means that a standard SI wafer can accommodate or transmons in superconducting electronic circuits. Scaling
about 300,000 modes, which is not yet sufficient for a scal- these machines to enable the numbers of qubits needed for
able photonic quantum computer. This approach has demanded a fully fault tolerant machine remains a challenge because of
innovations in light sources using novel designs for four-wave the difficult engineering needed to prepare, control, and cor-
mixing, control of photon routing on the chips, since standard rect the qubits especially as a single device. For this reason,
methods tend to introduce unacceptable losses, as well as new there are a number of proposals to connect smaller-scale devices
detector designs, which utilise superconducting materials. This on a network to enable modular architectures for a fully scal-
means that the entire machine should ideally fit on a photonic able quantum computer [18]. For this task, light is an obvious
integrated circuit and all must be kept at low temperatures. choice.
Other approaches leverage telecommunications waveguide Efforts to develop efficient, low-noise interfaces between
devices and materials. These will necessarily be physically different physical types of qubits and photons are therefore
larger than an all-integrated approach and must overcome the underway. These include micro-wave to optical photon inter-
losses of interfacing between components. Nevertheless, the conversion as well as optical wavelength conversion into the
larger scale means that feedforward required for measurement telecommunications bands near 1550 nm. These devices will
configuration is more straightforward, and the majority of the link several quantum computing processors together, including
machines, excepting the detectors, can operate in ambient con- those of different types, and would enable a large-scale computer
ditions. Of course, the larger component scale also necessarily to operate across a multi-node network.
means a larger machine, and the trade-offs inherent in these two Indeed, the network could also link conventional high-
approaches has yet to be fully understood. performance computers to quantum processors, allowing spe-
More generally, it is increasingly likely that future quantum cialist tasks to be devolved to appropriate machines, optimising
computers will be networked. This is driven in part by the need the utility of the network to capitalise on the capabilities of
for significant scale, and the challenges across all potential quan- each class of processor. Such a hybrid network may well be
tum computing platforms of building a single monolithic device among the first to actually achieve quantum advantage in scalable
with sufficient number of logical qubits which each use multiple computing.
physical qubits to protect against errors. Light has a bright future as an enabler of quantum computers,
The leading gate-based architectures for quantum comput- quantum simulators and quantum computing networks, deriving
ing include trapped ions, which use hyperfine states as qubits, from its unique features as a quantum medium.
connected by trap vibrational motion, superconducting circuits,
 40 Vol. 1, No. 1 / 25 October 2023 / Optica Quantum Perspective

Funding. European Union Horizon 2020; UK Engineering and Physical 8. L. Madsen, F. Laudenbach, and M. F. Askarani, et al., “Quantum
Sciences Research Council. computational advantage with a programmable photonic processor,”
Nature 606(7912), 75–81 (2022).
Acknowledgment. IAW received funding for research on the topics 9. A graph is a mathematical structure consisting a set of objects in
of this article from UKRI (EPSRC, through the Quantum Computing and which some pairs of the objects are related. It is specified as a set
Simulation Hub), the European Union Horizon 2020 Marie Slodowska- of “Vertices” representing the objects and “Edges” which connect the
Curie Actions, QuantERA and FET programmes. objects in a uni- or bi-directional sense. The properties of a graph
may be encoded into the parameters of a GBS machine.
Disclosures. IAW is Provost of Imperial College London and Chair and 10. J. M. Arrazola and T. R. Bromley, “Using Gaussian Boson Sampling
co-founder of ORCA Computing (I, C, P). to find dense subgraphs,” Phys. Rev. Lett. 121(3), 030503 (2018).
11. S. Sempere-Llagostera, R. B. Patel, and I. A. Walmsley, et al.,
“Experimentally finding dense subgraphs using a time-bin encoded
REFERENCES Gaussian Boson Sampling device,” Phys. Rev. X 12(3), 031045
(2022).
1. S. Aaronson and A. Arkhipov, “The computational complexity of linear
12. S. Sun, H. Kim, and Z. Luo, et al., “A single-photon switch and transis-
optics,” Proc. STOC, June 2011, pp. 333–342.
tor enabled by a solid-state quantum memory,” Science 361(6397),
2. C. K. Hong, Z. Y. Ou, and L. Mandel, “Measurement of subpicosec-
57–60 (2018).
ond time intervals between two photons by interference,” Phys. Rev.
13. S. Baur, D. Tiarks, and G. Rempe, et al., “Single-photon switch based
Lett. 59(18), 2044–2046 (1987).
on Rydberg blockade,” Phys. Rev. Lett. 112(7), 073901 (2014).
3. The Permanent of a matrix is similar in form to the better-known
14. P. Kok, W. J. Munro, and K. Nemoto, et al., “Linear optical quantum
Determinant, as a sum of the products of matrix entries from sep-
computing with photonic qubits,” Rev. Mod. Phys. 79(1), 135–174
arate rows and columns. The difference is in the weighting of each of
(2007).
these products. This also allows for many variations in between the
15. D. E. Browne and T. Rudolph, “Resource-efficient linear optical
two functions.
quantum computation,” Phys. Rev. Lett. 95(1), 010501 (2005).
4. L. Valiant, “The complexity of computing the permanent,” Theor.
16. S. Bartolucci, P. Birchall, and H. Bombín, et al., “Fusion-based
Comp. Sci. 8(2), 189–201 (1979).
quantum computation,” Nat. Commun. 14(1), 912 (2023).
5. C. S. Hamilton, R. Kruse, and L. Sansoni, et al., “Gaussian boson
17. B. Pankovich, A. Kan, and K. H. Wan, et al., “High photon-loss thresh-
sampling,” Phys. Rev. Lett. 119(17), 170501 (2017).
old quantum computing using GHZ-state measurements,” arXiv,
6. H. S. Zhong, H. Wang, and Y. H. Deng, et al., “Quantum compu-
arXiv:2308.04192 (2023).
tational advantage using photons,” Science 370(6523), 1460–1463
18. C. Monroe and J. Kim, “Scaling the ion trap quantum processor,”
(2020).
Science 339(6124), 1164–1169 (2013).
7. J. F. F. Bulmer, B. A. Bell, and R. S. Chadwic, et al., “The boundary
for quantum advantage in Gaussian boson sampling,” Sci. Adv. 8(4),
eabl9236 (2022).