COVER FEATURE OUTLOOK 2023

Quantum Computing:
Progress and Innovation
Gregory T. Byrd , North Carolina State University
Yongshan Ding , Yale University

The field of quantum computers has reached the point

of having real systems available for experimentation
to a broad community. In this article, we review recent
developments in hardware technology, outline challenges
and innovations in quantum systems, and discuss efforts
to cultivate and educate a quantum-ready community.

Q
uantum computing has made significant and several open source software environments for
strides over the last decade, transitioning writing quantum programs. In this article, we summa-
from a niche area of interest to quantum rize the current state of quantum computing hardware
physicists and theoretical computer scien- technology, discuss a number of innovations in systems
tists into a commercial enterprise attracting large invest- and software that enhance the usability of the hardware,
ments and spawning numerous startup companies. The and describe ongoing efforts to create a quantum-ready
surge in interest is due in no small part to the public workforce through education and community building.
availability of quantum computers via cloud services,
BACKGROUND
Digital Object Identifier 10.1109/MC.2022.3217021
Although a complete introduction to quantum comput-
Date of current version: 9 January 2023 ing is not appropriate for this article, we will spend some

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
20 CO M PUTE R P U B LISHED BY THE IEEE COMP UTER SOCIE T Y 0018-9162/ 23©2023IEEE
 time introducing some basic concepts where Σ | αi |2 = 1. When we measure data; this is usually a superposition. 2)
and terminology that will be helpful in an n-qubit state, we observe state | i 〉 Operate on the quantum state in a way
understanding the rest of the content. with probability | αi |2. that amplifies the logical state(s) that
For a complete introduction, there are Multiqubit states may exhibit a property represent an answer to the problem
several well-known textbooks1,2 and known as entanglement, in which the states of and minimizes the states that are not
online resources3 available. individual qubits are correlated with one interesting. 3) Measure the quantum
another. For example, consider the state system, observing a state with high
Qubits | θ 〉 = (1 / 2) |00 〉 + (1 / 2) |11 〉. When probability that solves the problem or
There are several models of using we measure, we will get either | 00 〉 or gives some useful information. This
quantum properties to represent infor- | 11 〉, each with probability 1/2; even is a major simplification, of course,
mation and perform computation, but though we don’t know what the out- but step 2) illustrates one component
the most common one involves the come will be, we will always see that of the benefit of quantum compu-
use of quantum bits, known as qubits, the two qubits are equal. Entangle- tation, known as quantum parallel-
and operations performed on qubits, ment is a key component of quantum ism: Because we operate on all com-
known as gates. Like a classical bit, a computation and communication, ponents of the superposition, in some
qubit has two fundamental states, rep- and there is no classical counterpart to sense we are performing many opera-
resenting logical states of 0 and 1. In this behavior. tions in parallel. The caveat is that we
the quantum realm, we represent these
as | 0 〉 and | 1 〉, and a general quantum
state is normally represented using
a Greek letter, such as | ψ 〉. Unlike a
classical bit, a quantum state can rep-
ENTANGLEMENT IS A KEY COMPONENT
resent a superposition of logical states: OF QUANTUM COMPUTATION AND
| ψ 〉 = α| 0 〉 + β| 1 〉, where α and β are COMMUNICATION, AND THERE IS
complex numbers, and | α |2 + | β |2 = 1. NO CLASSICAL COUNTERPART
Although a qubit state is more gen-
TO THIS BEHAVIOR.
eral than a one-bit classical bit, the
properties of quantum mechanics do
not allow us to directly observe this
state. Instead, when we observe or mea-
sure the state, we will only see either | 0 〉 Gates and algorithms don’t get to use all of those results. We
(with probability | α |2) or | 1 〉 (with prob- Quantum gates are operations that only see one final state when the mea-
ability | β |2). Moreover, the act of mea- manipulate the coefficients of each surement occurs.
suring the qubit is destructive: When basis state. For example, an X oper-
we observe | 0 〉, the qubit’s state is ates on one qubit and swaps the coeffi- Quantum annealing
changed to | 0 〉 and the superposition is cients of the | 0 〉 and | 1 〉 states, so that While this article focuses on gate-based
destroyed. Likewise, when we observe α | 0 〉 + β | 1 〉 becomes α | 1 〉 + β | 0 〉. This quantum computing, there are other
| 1 〉, the qubit’s state is changed to | 1 〉. is analogous to a Boolean NOT opera- computational models. D-Wave, for
When generalizing to n qubits, tion. A relatively small number of 1- example, provides a quantum annealer
there are 2n logical (basis) states, cor- and 2-qubit gates have been proven to that solves certain kinds of optimization
responding to the n-bit binary values be universal, in the sense that they can problems.4 While quantum annealing
i = 0,  1,  . . . ,  2n−1, and the quantum combine to approximate any trans- is not universal, in the sense that it can-
state is a superposition as follows: formation of a quantum state within not solve every kind of computational
some error bounds. problem, it does address a very interest-
2 n −1 Many quantum algorithms follow a ing class of business-relevant problems.
|ψ〉 = ∑α | i〉 
i=0
i
pattern: 1) Create a quantum state that
encodes an initial condition or set of
It is a close relative to adiabatic quantum
computing,5 which is universal.

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
JANUARY 2023 21
 OUTLOOK 2023

Physical implementations cannot control. When the changes research labs around the world. While
To build a quantum computer, we are enough to be observed, they cause superconducting qubits and trapped
must have some physical system that errors in the processing and mea- ions are perhaps the best known,
behaves like a qubit, and we must be surements of the state. Examples there are several other approaches
able to precisely control the initializa- of errors include decoherence, a | 1 〉 that are expected to compete in the
tion, manipulation, and measurement changing to | 0 〉, and dephasing, where near term, as illustrated in Table 1. In
of these qubits. We’ll see in the next α | 0 〉 + β | 1〉 changes to α | 0 〉 − β | 1〉. this section, we provide a brief over-
section that there are several technol- These error rates are improving as the view of different types of qubit tech-
ogies that can achieve these expecta- technology matures, but they still nologies used in contemporary quan-
tions, but we first discuss some com- limit the number of operations that tum computers.
mon issues with building a quantum can practically be performed.
computing system. We know how to encode infor- Superconducting qubits
First, quantum states are fragile. mation using many physical qubits A superconducting qubit is an electronic
The theory tells exactly what will hap- to implement fault-tolerant logical circuit in which energy levels, such
pen for an isolated quantum system, qubits, through techniques known as as charge or flux, take on quantized
but no system is ever completely iso- quantum error correction (QEC), but values. Standard components such as
lated: it interacts with the environ- that brings us to our second limita- capacitors and inductors, together with
ment. Such interactions mean that tion: We don’t yet have a lot of qubits nonlinear Josephson junctions, create
we should expect the quantum state in one system. The challenges of man- an anharmonic oscillator, in which the
to change over time in ways that we ufacturing, isolating, controlling, and difference among energy levels is non-
measuring qubits have limited current uniform. Typically, the lowest two lev-
quantum computers to tens or hun- els are used to represent | 0 〉 and | 1 〉,
TABLE 1. Quantum computing dreds of qubits. Scaling to the thou- although higher energy levels can be
implementation technologies. sands or millions of qubits needed for used to implement multilevel quantum
game-changing applications is still a digits (qutrits, qudits).
Representative
work in progress. Superconducting qubits are manu-
Technology* companies
For these reasons, today’s quantum factured using 2D films of a supercon-
Superconducting Google Quantum AI computers are known as noisy, inter- ducting material, such as aluminum
qubits IBM mediate-scale quantum (NISQ), a term or niobium, on silicon or sapphire sub-
Rigetti Computing first coined by John Preskill.6 Research strates. The materials and techniques
Atlantic Quantum and development in quantum comput- are compatible with silicon CMOS man-
ing is following two paths. The first is ufacturing. The quantum chip must be
Trapped ions IonQ
Quantinuum to continue on the path to error-cor- cooled to cryogenic temperatures, so
Oxford Ionics rected, fault-tolerant systems, which that thermal energy does not disturb the
will require many more high-quality state of the qubit. The qubit is controlled
Neutral atoms Atom Computing qubits. The second is to understand through electromagnetic coupling
ColdQuanta what kinds of computational advan- with microwave pulses. The amplitude,
Pasqal tage can be gained in the near-term phase, and duration of the pulse deter-
Photonics PsiQuantum using NISQ hardware. mines the evolution of the quantum
Xanadu state. Each qubit is also capacitively cou-
HARDWARE TECHNOLOGY pled to a fixed number of neighboring
Silicon qubits Intel One of the interesting aspects of quan- qubits, and two-qubit interactions are
Silicon Quantum tum computing is that the choice of performed using through control pulses.
Computing
implementation technology is still The most common type of supercon-
*T his is a sampling, and is therefore not an exhaus- ver y much undecided. A number ducting qubit is known as the transmon,
tive list of the technologies being explored, nor
does it reflect the entire list of companies that are
of different approaches are under used by IBM, Google, and Rigetti. IBM’s
developing them. active development in companies and Eagle chip7 has 127 qubits connected

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
22 COMPUTER  W W W.CO M P U T E R .O R G /CO M P U T E R
 in a “heavy-hexagonal” pattern, with in a vacuum to minimize interactions the presence of a photon in the other
an average coherence time of around with stray atoms in the environment. represents | 1 〉. Single-qubit operations
90 μs. The latency of a single-qubit gate Doppler cooling and other laser-based are performed using beam-splitters and
is around 500 μs, and two-qubit gates techniques are used to cool the atoms, phase shifters, and measurements are
are significantly longer, which means which means that the overall system performed by single-photon detectors.
100–150 operations could potentially can operate at room temperature. PsiQuantum uses a computational
be performed during the lifetime of Because of the free-space nature of mode called fusion-based quantum com-
a quantum state. However, there are optical control, two-qubit operations puting,11 in which small groups of entan-
other sources of error which practically can be performed on any qubits within gled qubits (known as resource qubits)
limit the number of operations. the same trap, essentially creating a are subject to two-qubit fusion mea-
IBM has published a roadmap8 that completely connected interconnection surements with their neighbors. The
promises a chip with 1,121 qubits by graph. This can be a significant advan- resulting measurements are then used
the end of 2023, and the other vendors tage over locally connected topologies, during the error correction process as
are similarly on a path to scalability. as described previously for supercon- well as the measurements specified by
Notably, Rigetti demonstrated a cross- ducting qubits because extra gates are the quantum computation. PsiQuan-
chip interconnect that will support the not needed to move quantum state to tum does not have a publicly available
building of multichip systems.9 While nearest-neighbor qubits. IonQ recently system at this point; their strategy is to
improvements continue to be made introduced its Forte system,10 with 31 produce a large-scale error-corrected
in the transmon, other companies are addressable qubits. system from the beginning, rather than
pursuing alternative superconducting Other companies are using neutral release incremental NISQ computers.
qubit design. Atlantic Computing, for atoms, rather than ions, where the qu- Another photonics-based approach
example, is developing a system based bit state is represented by the nuclear is to not use qubits at all, but to use
on the fluxonium, which allows for spin. Because the atoms are not electri- the continuous-variable model, dealing
longer coherence time at the expense cally charged, the trapping is done opti- with properties of light as opposed to
of additional Josephson junctions. cally. These systems share many of the individual photons. A squeezed state
same benefits as ions: long coherence of light is one in which the uncertainty
Trapped ions and atoms times, flexible interconnectivity, and of the electric field is reduced for some
Another qubit implementation uses the room-temperature systems. phases. Xanadu’s Borealis system 12
quantum properties of individual atoms, uses pulses of squeezed light to repre-
rather than the collective properties of Photonics sent quantum states, and time division
a group of electrons. In a trapped ion Using photons for quantum computing multiplexing to allow states to interact
system, a dynamic electric field is used is promising because they have some with each other. It is not a general-pur-
to isolate and hold in place a single elec- natural advantages. They are easy pose quantum computer, but it can
trically charged atom (ion). The ion can to measure, and they don’t naturally perform certain sampling tasks much
transition between a ground state | 0 〉 interact with one another, reducing faster than classical computations. The
and an excited state | 1 〉 by interaction concerns about crosstalk. They also system is available for public use.
with a laser pulse. Multiqubit operations operate at room temperature (though
also take advantage of coupled vibra- detectors may be based on supercon- Silicon qubits
tional modes across the ions. ductors which require cryogenic tem- Progress continues in the space of sili-
Ion traps generally have the low- peratures) and are not geometrically con-based qubits. Intel’s quantum chip
est error rates and longest coherence constrained. Silicon photonics is quite is based on quantum dots, nanofeatures
time of the commercially available mature and compatible with CMOS in which an electron can be trapped and
systems. Depending on the type of manufacturing processes. controlled. Intel and QuTech recently
system, coherence time can be on the One approach for representing a demonstrated a two-qubit gate operat-
order of seconds or longer. Gate times, qubit is to associate two waveguides for ing at 1.1 K, much higher than the mil-
on the other hand, are longer, taking each qubit. The presence of a photon likelvin temperatures previous required
tens of microseconds. The ions are kept in one waveguide represents | 0 〉 and for such control.13 Silicon Quantum

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
JANUARY 2023 23
 OUTLOOK 2023

Computing, based in Australia, uses error mitigation schemes—that permit synthesis and architectural design from
donor-based qubits, where both the large-scale applications to be realized. the early stage of classical computing
electron and nuclear spins of phospho- That is the topic of the next section. around the 1980s,15 when hardware
rus atoms implanted in a silicon sub- resources were scarce, prone to errors,
strate can be controlled.14 Silicon-based SYSTEMS AND SOFTWARE and expensive. Unlike the old times,
approaches have demonstrated long Quantum computers hold enormous however, QC development can benefit
coherence time, and offer higher den- potential in solving problems beyond from the lessons learned in classical
sity of qubits than other approaches. the reach of their classical (digital) computing, and, more importantly, can
counterparts. Emerging quantum sys- be integrated into the existing classical
Outlook tems, however, have not yet reached a computing infrastructure.
It is clearly very early in the implemen- scale compatible with practical appli- Layers of abstraction similar to that
tation phase of quantum computing sys- cations. In fact, a significant gap exists of digital computers have begun to
tems, and the best technology is yet to between the reality of these noisy form: from domain-specific program-
be determined. While superconducting devices and the requirements of a prac- ming languages, to quantum compiler
qubits and ion traps have an early lead in tical application. On the one hand, software, to instruction set architec-
terms of deployment and public aware- today’s quantum systems are typically ture, then to quantum hardware. A
ness, the first technology to demonstrate small in size, sensitive to environmental classical software toolchain across the
a large number of error-corrected logi- disturbance, and difficult to control.6 layers is developed to help translate
cal qubits will have a significant advan- On the other hand, known applications the high-level quantum programs to
tage in widespread adoption. Just as require devices with capabilities such instructions on quantum hardware.
important will be systems-level consid- as 1) stable and abundant qubits, 2) effi- The abstraction layer is useful because
erations—compilers, control systems, cient and universal programmability, it sufficiently hides the working
and 3) fault-tolerant logical operations. details of the underlying systems.
Fortunately, state-of-the-art systems From a usability perspective, we
and software techniques have shown also see compelling reasons for creat-
Application Layer
great potential in addressing these ing abstraction layers to sufficiently
Quantum Algorithm
requirements. The landscape of quan- hide details of hardware limitations
Quantum Classical
tum systems engineering research has from programmers. For instance, pro-
begun to take shape. grammers want qubits to be abundant,
Systems Software Layer Our goal in this section is to first noiseless, universal, and cheap, but
Quantum DSL, clarify how a quantum stack can emerging quantum systems do not yet
Compilation, bridge the gap between applications satisfy these requirements. From past
Unitary Synthesis, and hardware, then identify the top work, we observe that applications that
Pulse Control,
obstacles and opportunities in the next are adapted to the underlying hardware
Noise Mitigation,
Error Correction five to ten years, and finally highlight via cross-layer compiler optimizations
some promising architectural meth- can be executed with orders of magni-
odologies for building utility-scale tude higher success rate.16 However,
Hardware Layer quantum computers. implementing a quantum algorithm in
Quantum Hardware a noise-resilient way would entail atten-
System Controller
Building a quantum stack tion to hardware constraints in painstak-
Qubits Quantum computers need a new full- ing detail. A quantum stack can auto-
stack system software because they mate the cross-layer optimization needed
operate in a fundamentally different for noise mitigation and error correction,
FIGURE 1. A quantum stack consisting
way than classical computers. Fortu- thus allowing programmers to use qubits
of an application layer, systems soft-
nately, their design principles are not fault-tolerantly and economically. Fig-
ware layer, and hardware layer. DSL:
completely distinct. The current state ure 1 illustrates the components and
domain-specific language.
of QC systems design resembles circuit functionalities of a quantum stack.

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
24 COMPUTER  W W W.CO M P U T E R .O R G /CO M P U T E R
 Top five challenges entanglement distribution over measurement (such as a swap
and opportunities networked quantum devices. test) or easy to track via partial
From a systems software point of view, 3. Support for circuit-level fault-tol- simulation or formal methods.
we summarize the top five challenges erance. Recent theoretical and 5. Quantum computing as a service.
and opportunities to utility-scale quan- experimental progress in error Quantum computing hardware
tum computing. The first three serve correction has provided sev- keeps evolving at a fast pace. In
as the building blocks for universal eral promising candidates for order to make the latest quan-
fault-tolerant quantum computation, fault-tolerant qubits with a pro- tum technology accessible to
and the last two affect the usability of longed lifetime but integrating users, quantum computers are
QC in an application development cycle. them into universal computing being deployed on the cloud17,18
systems remains challenging. following the success of clas-
1. Availability of qubits. In an ideal A milestone is to demonstrate sical cloud-based services. As
quantum processor, qubits are end-to-end fault-tolerant quan- such, there is a growing interest
noiseless and abundant. But a tum circuit implementations in developing job schedul-
more realistic picture looks like (with operations from state ing systems for maximizing
this: high-quality error-cor- preparation to logical gates and throughput, runtime systems to
rected qubits are going to be a readout) for a practical applica- support reliable execution and
valuable but scarce resource, tion. A scalable control instru- calibration, and security mech-
whereas many qubits with dif- ment and systems software for anisms to prevent reverse-engi-
ferent levels of noise are avail- coordinating error correction neering or information leakage
able. Under this constraint, operations and real-time decod- (under a multitenant setting).
quantum memory manage- ing protocols will be the key
ment (for example, matching enabler of error-corrected quan- Software and hardware
qubits with best-suited tasks or tum computing at scale. co-design
reusing qubits via uncomputa- 4. Verification and debugging. As the latest hardware technology
tion) becomes crucial. Probing the state of a quantum advances, we are developing new appli-
2. Long-range entanglement under systems mid-flight without cations and software tools to fully har-
limited connectivity. Supporting destroying it is usually prohib- ness its computing power. To do so, we
all-to-all connectivity (that is, itive, due to the fundamental must follow a software and hardware
interacting many-body quan- destructive nature of measure- codesign methodology—applications
tum systems) for large-scale ments. This makes verifying need to be adapted to architectural/
systems is challenging, both and debugging quantum com- hardware limitations, and devices must
fundamentally and technically. putation particularly challeng- be optimized to support the intended
As a result, devices with limited ing. When systems are small applications. In this section, we out-
connectivity are being built. The enough, we can use classical line how to integrate the knowledge of
reduced connectivity inevita- simulation to calculate, track, hardware noise into the compiling and
bly curtails the capabilities of and verify the output of a quan- benchmarking processes.
producing entanglement across tum circuit. For larger systems, Noise-aware quantum compiling: Rely-
systems. Efficient routing proto- however, classical simula- ing on programmers to keep track of all
cols are critical to restoring long- tion becomes intractable, as it usage of qubits is neither scalable nor
range entanglement to meet requires computing transfor- efficient. Design automation will play a
the requirements of the applica- mations in an exponentially crucial role in order to make program-
tions. Examples include lattice large state space. As a result, ming manageable and algorithms prac-
surgery or braiding in a surface emerging verification schemes tical. Figure 1 illustrates some important
code architecture, swap-based or typically focus on properties compiler features including efficient uni-
teleport-based communication of quantum systems that are tary synthesis, control flows, quantum
in a multi-chip processor, and relatively easy to probe via circuit optimizations, benchmarking

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
JANUARY 2023 25
 OUTLOOK 2023

and resource estimation, and hardware/ individual qubits as possible in a hard- increase errors, this is still beneficial
simulator back ends. The job of a quan- ware-efficient way while maintaining because some errors are potentially eas-
tum compiler is to translate a quantum high-precision addressability to individ- ier to detect or correct. As such, these
program written in a high-level pro- ual qubits, nearly perfectly isolating qu- techniques are called error engineering
gramming language or mathematical bits from disturbance of the environment algorithms. The bias-noise qubits are
models such as unitary matrices into while supporting fast external control promising because if we restrict the
native instructions recognizable by the read-out, and allowing qubits to interact operations performed on these noisy
hardware through a series of transforma- strongly with each other for computation qubits, they can still function reliably.
tions and optimizations. The compiled while eliminating all unwanted interac- We are beginning to understand that
instructions must be correct, resource-ef- tions. We are beginning to understand some errors may not be as harmful as
ficient, and robust to hardware noise. how to protect a quantum processor others (for example, no propagation) for
Benchmarking and calibration: Knowl- from faults in its hardware components. some applications.
edge of the structures of noise in a sys- Different classes of techniques have been Error correction: QEC encodes log-
tem can be exploited to improve error developed, namely error mitigation, er- ical information redundantly into a
mitigation and correction protocols. ror engineering, and error correction. set of physical qubits to protect it from
This makes benchmarking algorithms Error mitigation: Error mitigation unwanted interaction with the environ-
(as surveyed in Ding and Chong) for strategies aim to suppress sources of ment. Moreover, to overcome erroneous
physical operations, one needs to apply
only fault-tolerant gates to the encoded
logical qubits. So, how do we encode logi-
cal information in hardware redundantly
FROM A USABILITY PERSPECTIVE, WE ALSO for error correction while, at the same
SEE COMPELLING REASONS FOR CREATING time, fit all logical qubits on an economical
ABSTRACTION LAYERS TO SUFFICIENTLY chip? Experimentally, remarkable prog-
HIDE DETAILS OF HARDWARE LIMITATIONS ress has been made to demonstrate the
basic elements of fault-tolerant quantum
FROM PROGRAMMERS.
error correction.19 Notably, in 2016, the
bosonic qubit, which encodes information
in a superconducting oscillator, was one of
the first hardware that crossed the break-
measuring noise a handy set of tools. noise not only by building better hard- even point for extending qubit lifetime.20
Recent work shows that estimating ware (at fabrication time) but also by
simple noise models for large systems designing better software controls (at Integrated classical-
can be done efficiently, for example, compile or run time). Common soft- quantum computing
estimating Pauli channels for hun- ware strategies include canceling error It has become increasingly clear that
dreds of qubits. Once the noise model coherently, avoiding noisy operations, the power of quantum mechanics lies
has been estimated, the obtained synthesizing robust pulses, engineer- in some specialized tasks, for example,
data can be used for calibration, rang- ing asymmetric errors, and efficient in molecular simulations, combinato-
ing from tuning individual quantum postprocessing. We refer readers to rial optimizations, and so on. So, quan-
gates to updating error models for the Ding and Chong for a survey on these tum computers will most likely act as
decoder of an error correction code. noise mitigation algorithms. an accelerator for parts of a larger com-
Error engineering: Quantum noise putation. As such, to realize efficient
Fault-tolerant quantum is notoriously complex and difficult to classical and quantum information
architecture mitigate, but sometimes we can trade processing, we envision quantum pro-
To harness the full quantum phenom- one type of error for another, bias- cessors being tightly integrated into
enon, a quantum processor requires ing toward desirable noise structures. today’s existing high-performance clas-
capabilities such as packing as many Although these strategies might even sical computing infrastructure. Such

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
26 COMPUTER  W W W.CO M P U T E R .O R G /CO M P U T E R
 integration has already happened at professionals to be quantum-aware and degree. A recent article22 authored by a
the algorithmic level. quantum-proficient, in order to make group of quantum scientists and engi-
Hybrid classical-quantum approach: sound implementation choices for neers outlines a path for creating quan-
Almost all useful algorithms require future applications. To this end, a new tum pathways at the undergraduate
some amount of classical preprocess- subdiscipline of quantum engineering level. This includes introductory courses
ing or postprocessing. For example, is evolving, defined as the “application for nonmajors, so that the population at
Shor’s algorithm has a series of classi- of engineering methods and principles large becomes more quantum-aware.
cal arithmetic operations before and to quantum information systems and There are also significant efforts
after the quantum order-finding sub- problems.”22 The effort to enlarge the to engage students in quantum infor-
routine. One of the most promising community has both educational and mation at the high school and mid-
examples of hybrid classical-quantum community-building components. dle school levels. A notable example
algorithms is in quantum chemistry.21 On the formal education side, sev- is Qubit by Qubit, an initiative of The
Variational quantum eigensolver (VQE) eral institutions have created gradu- Coding School. 24 Virtual summer
is a heuristic algorithm that iterates ate degrees, minors, and certificates camps for middle school and high
between a quantum circuit and a classi- in quantum-related areas. Examples school students are offered, as well as
cal optimizer. The goal is to find the low- include the Colorado School of Mines a year-long high school course. They
est energy state configuration (that is, (M.S. in quantum engineering), TU Delft also facilitate internships with quan-
ground state) of a molecule. In each iter- (M.Sc. courses at QuTech Academy), Uni- tum computing companies.
ation, we start from an initial configura- versity of Wisconsin (M.S. in physics - Efforts at the undergraduate and
tion and estimate the energy of that con- quantum computing), Stevens Institute high school levels are creating what
figuration using the quantum machine. of Technology (M.S. quantum engineer- Rod Van Meter likes to call “quantum
This estimate is used by the classical ing, in the Physics Department), and Uni- natives,” those who learn about quan-
optimizer to guess a new configuration versité PSL (M.S. quantum engineering). tum as just another alternative com-
with a lower energy. While progress has A Master’s program is usually a good puting paradigm.25 Whether these
been made to improve the application’s place to start, due to flexibility in the students ultimately become quantum
rate of convergence or its resilience curriculum. Some of these students will professionals, or even computing pro-
to noise, much remains to be done on continue to pursue a Ph.D. degree, but an fessionals, they will be comfortable
developing compiler and runtime sup- increasing number of jobs are expected with the potential and practicalities of
port for such variational algorithms. for which M.S. graduates will be sought. quantum technologies.
Part of the challenge is that exist- For practicing engineers who want to
COMMUNITY AND ing quantum materials sometimes are pivot to quantum companies, or who are
EDUCATION not presented in the language familiar just quantum-curious, a number of online
Together with the advances in technol- to engineering students in other disci- resources with excellent introductory,
ogy, the field has reached a turning point plines. The authors have made efforts intermediate, and advanced material.
in the development of a quantum-ready to address this issue with textbooks There are too many to cite in a short arti-
workforce. The field has been dominated geared toward computer engineers16 cle, but IBM’s open source Qiskit textbook
by quantum physicists, mathematicians, and electrical engineers,23 and we is a very good starting point.3 The IEEE
and theoretical computer scientists, expect more such efforts in other areas Quantum Initiative hosts webinars and
who have made enormous strides lay- of engineering. other education material,26 and the IEEE
ing the foundations of this new compu- The undergraduate level moves a bit International Conference on Quantum
tational discipline. As the technology more slowly, but the University of New Computing and Engineering (also known
approaches commercial viability, how- South Wales offers a Bachelor of Engi- as IEEE Quantum Week) includes numer-
ever, there is an increasing need for engi- neering (Honors) (quantum engineer- ous tutorials, workshops, and panel dis-
neers to build scalable, maintainable, ing), and the University of Chicago’s cussions in additional to research presen-
and cost-effective systems for quantum Pritzker School of Molecular Engineer- tations, very suitable for those who want
computing, sensing, and communi- ing offers a quantum engineering track to learn how their skills can be applied to
cation. In addition, we need software in their molecular engineering B.S. the development of quantum systems.27

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
JANUARY 2023 27
 OUTLOOK 2023

6. J. Preskill, “Quantum computing in

the NISQ era and beyond,” Quan-
ABOUT THE AUTHORS tum, vol. 2, p. 79, Aug. 2018, doi:
10.22331/q-2018-08-06-79.
GREGORY T. BYRD is a professor and associate department head of electri- 7. J. Chow, O. Dial, and J. Gambetta,
cal and computer engineering at North Carolina State University, Raleigh, NC “IBM quantum breaks the 100-qubit
27695 USA. His research interests include quantum computing and parallel processor barrier.” IBM. Accessed:
computer architecture. Byrd received a Ph.D. in electrical engineering from Nov. 3, 2022. [Online]. Available:
Stanford University. He is a Senior Member of IEEE and the IEEE Computer https://research.ibm.com/blog/127
Society. Contact him at [email protected]. -qubit-quantum-processor-eagle
8. “Our new 2022 development road-
YONGSHAN DING is an assistant professor of computer science at Yale Univer- map.” IBM. Accessed: Nov. 3, 2022.
sity, New Haven, CT 06520 USA. His research interests include algorithms and [Online]. Available: https://www.
computer architecture for quantum computing. Ding received a Ph.D. from the ibm.com/quantum/roadmap
University of Chicago. Contact him at [email protected]. 9. A. Gold et al., “Entanglement across
separate silicon dies in a modular super-
conducting qubit device,” Npj Quantum

T
Inf., vol. 7, no. 1, Sep. 2021, Art. no. 142,
o summarize, there is a lot of engineers, and entrepreneurs to apply doi: 10.1038/s41534-021-00484-1.
excitement and activity in quan- quantum technologies to a broad set of 10. “Unveiling IonQ forte: The first soft-
tum computing, as well as the application areas and increase quantum ware-configurable quantum com-
related areas of quantum sensing and awareness among computing profession- puter.” IonQ. Accessed: Nov. 3, 2022.
communication. The outlook is very als. The largest breakthroughs are likely to [Online]. Available: https://ionq.
positive, evidenced by the investments occur when these quantum natives seize com/posts/may-17-2022-ionq-forte
by established companies and the cre- on the opportunities seeded by the quan- 11. S. Bartolucci et al., “Fusion-based
ation of new startups. tum pioneers of yesterday and today. quantum computation,” 2021.
Over the coming years, we will see con- arxiv:2101.09310
tinued and accelerated development of REFERENCES 12. L. S. Madsen et al., “Quantum compu-
hardware platforms, focusing on reliabil- 1. M. A. Nielsen and I. L. Chuang, tational advantage with a program-
ity and scalability. Substantial research Quantum Computation and Quantum mable photonic processor,” Nature,
and innovation are needed in the support- Information. Cambridge, UK: vol. 606, no. 7912, pp. 75–81, Jun. 2022,
ing technology (the “uncore”), such as clas- Cambridge Univ. Press, 2010. doi: 10.1038/s41586-022-04725-x.
sical control, decoding for error correction, 2. E. Rieffel and W. Polak, Quantum Com- 13. L. Petit et al., “Universal quantum
and communication of quantum informa- puting: A Gentle Introduction. Cam- logic in hot silicon qubits,” Nature,
tion, as well as system-level approaches to bridge, MA, USA: MIT Press, 2014. vol. 580, no. 7803, pp. 355–359, Apr.
error mitigation. We should also see the 3. A. Abbas et al., “Learn quantum compu- 2020, doi: 10.1038/s41586-020-2170-7.
development and analysis of new quan- tation using Qiskit.” Qiskit. Accessed: 14. J. C. McCallum et al., “Donor-based qubits
tum algorithms and applications, as well Nov. 3, 2022. [Online]. Available: http:// for quantum computing in silicon,”
as new software tools and languages. community.qiskit.org/textbook Appl. Phys. Rev., vol. 8, no. 3, p. 031314,
Eventually, quantum processors will be 4. “How D-wave systems work.” D-Wave Sep. 2021, doi: 10.1063/5.0060957.
integrated into the high-performance Systems. Accessed: Nov. 3, 2022. [Online]. 15. F. T. Chong, D. Franklin, and M. Martonosi,
computing landscape, helping to tackle Available: https://www.dwavesys. “Programming languages and compiler
society’s largest and most challenging com/learn/quantum-computing/ design for realistic quantum hardware,”
problems, and the first steps are being 5. C. McGeoch, Adiabatic Quantum Nature, vol. 549, no. 7671, pp. 180–187,
taken at national labs around the world. Computation and Quantum Annealing: Sep. 2017, doi: 10.1038/nature23459.
Especially encouraging are the efforts Theory and Practice. San Rafael, CA, 16. Y. Ding and F. Chong, Quantum
to establish communities of scholars, USA: Morgan and Claypool, 2014. Computer Systems: Research for Noisy

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
28 COMPUTER  W W W.CO M P U T E R .O R G /CO M P U T E R
 Intermediate-Scale Quantum Comput- in superconducting circuits,” Nature, 24. “Training the future diverse quantum
ers. San Rafael, CA, USA: Morgan and vol. 536, no. 7617, pp. 441–445, Aug. workforce,” Qubit by Qubit, Los Ange-
Claypool, 2020. 2016, doi: 10.1038/nature18949. les, CA, USA, 2022. Accessed: Nov. 3,
17. Azure Quantum. (2022). Microsoft. 21. J. R. McClean, J. Romero, R. Babbush, 2022. [Online]. Available: https://www.
Accessed: Nov. 3, 2022. [Online]. and A. Aspuru-Guzik, “The theory qubitbyqubit.org/
Available: https://azure.microsoft. of variational hybrid quantum-clas- 25. Keio University Quantum Comput-
com/en-us/products/quantum/ sical algorithms,” New J. Phys., vol. ing Center. Accessed: Nov. 3, 2022.
18. “Amazon Braket: Accelerate quan- 18, no. 2, p. 023023, Feb. 2016, doi: [Online]. Available: https://quantum.
tum computing research.” Amazon. 10.1088/1367-2630/18/2/023023. keio.ac.jp/
Accessed: Nov. 3, 2022. [Online]. Avail- 22. A. Asfaw et al., “Building a quan- 26. IEEE Quantum. Accessed: Nov. 3,
able: https://aws.amazon.com/braket/ tum engineering undergraduate 2022. [Online]. Available: https://
19. R. Acharya et al., “Suppressing program,” IEEE Trans. Educ., vol. 65, quantum.ieee.org/
quantum errors by scaling a no. 2, pp. 220–242, May 2022, doi: 27. IEEE International Conference on
surface code logical qubit,” 2022, 10.1109/TE.2022.3144943. Quantum Computing and Engineering,
arXiv:2207.06431. 23. D. D. Stancil and G. T. Byrd, Principles IEEE Quantum Week, Broomfield, CO,
20. N. Ofek et al., “Extending the lifetime of Superconducting Quantum Comput- USA, Sep. 18–23, 2022. [Online]. Avail-
of a quantum bit with error correction ers. Hoboken, NJ, USA: Wiley, 2022. able: https://qce.quantum.ieee.org/

IEEE TRANSACTIONS ON
SUBMIT
TODAY SUSTAINABLE COMPUTING
SUBSCRIBE AND SUBMIT
For more information on paper submission, featured articles, calls for
papers, and subscription links visit: www.computer.org/tsusc

Authorized licensed use limited to: University of Leeds. Downloaded on December 04,2024 at 21:33:40 UTC from IEEE Xplore. Restrictions apply.
Digital Object Identifier 10.1109/MC.2022.3227741