Quantum Computing

Circuits and Devices

Travis S. Humble Fahd A. Mohiyaddin
Oak Ridge National Laboratory Oak Ridge National Laboratory,
IMEC
Himanshu Thapliyal and
Edgard Muñoz-Coreas Ryan S. Bennink
University of Kentucky Oak Ridge National Laboratory

such as IBM, Google,

Editor’s note: D-Wave, Rigetti, and IonQ
As an emerging technology, quantum computing brings unique promises are among a growing
in creating a fundamentally different paradigm of information processing. list of devices that have
This article reviews introductory concepts and building blocks—quantum demonstrated the funda-
logic gates and memory, i.e., qubits, for quantum computing and depicts mental elements required
the challenges as well as prospects associated with design, test, and for quantum computing
fabrication of quantum devices. [5]. This progress in pro-
—Swarup Bhunia, Case Western Reserve University
totype QPUs has opened
up new discussions about
 Quantum computing promises new capabili- how to best utilize these nascent devices [6]–[8].
ties for processing information and performing compu- Quantum computing poses several new challenges
tationally hard tasks. This includes significant algorith- to the concepts of design and testing that are unfamil-
mic advances for solving hard problems in computing iar to conventional CMOS-based computing devices.
[1], sensing [2], and communication [3]. The break- For example, a striking fundamental challenge is the
through examples of Shor’s algorithm for factoring inability to interrogate the instantaneous quantum
numbers and Grover’s algorithms for unstructured state of these new devices. Such interrogations may
search have fueled a series of more recent advances be impractically complex within the context of con-
in computational chemistry, nuclear physics, and ventional computing, but they are physically impos-
optimization research among many others. However, sible within the context of quantum computing due
realizing the algorithmic advantages of quantum com- to the no-cloning principles. This physical distinc-
puting requires hardware devices capable of encod- tion fundamentally changes how QPUs are designed
ing quantum information, performing quantum logic, and their operation tested relative to past practice.
and carrying out sequences of complex calculations This tutorial provides an overview of the principles
based on quantum mechanics [4]. For more than 35 of operation behind quantum computing devices
years, there has been a broad array of experimen- as well as a summary of the state of the art in QPU
tal efforts to build quantum computing devices to development. The continuing development of quan-
demonstrate these new ideas. Multiple state-of-the-art tum computing will require expertise from the con-
engineering efforts have now fabricated functioning ventional design and testing community to ensure the
quantum processing units (QPUs) capable of carrying integration of these nontraditional devices into exist-
out small-scale demonstrations of quantum comput- ing design ­workflows and testing infrastructure. There
ing. The QPUs developed by commercial vendors is a wide variety of technologies under consideration
Digital Object Identifier 10.1109/MDAT.2019.2907130
for device development, and this article focuses on
Date of publication: 03 April 2019; date of current version: the current workflows surrounding quantum devices
29 May 2019. fabricated in semiconducting, superconducting, and

May/June 2019 Copublished by the IEEE CEDA, IEEE CASS, IEEE SSCS, and TTTC 2168-2356/19©2019 IEEE
69
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

trapped-ion technologies. We also discuss the design Logically, the qubit is defined over a basis of binary
of logical circuits that quantum devices must execute states labeled as “0” and “1,” respectively, such that
to perform computational work. an arbitrary state of a qubit may be expressed as the
While the tutorial captures many of the introduc- linear combination
tory topics needed to understand the design and test-
|ψ 〉 = ​c​ 0​​  |0〉 + ​c​ 1​​|1〉.​
​ (1)
ing of quantum devices, several more advanced topics
have been omitted due to space constraints. Foremost The superposition of these orthogonal basis states
is the broader theory of quantum computation, which is fundamental to quantum mechanics. The expan-
has developed rapidly from early models of quan- sion coefficients are complex-valued numbers nor-
tum Turing machines to a number of different but malized as |c0|2 + |c1|2 = 1 and a convenient graphi-
equally powerful computational m ­ odels. In addition, cal representation of the qubit is given in spherical
we have largely omitted the sophisticated techniques coordinates. As shown in Figure 1, the surface of a
employed to mitigate the occurrence of errors in quan- unit sphere represents all possible qubit values,
tum devices. Quantum error correction is an important where the points of |0​〉​ and |1​〉​ are located at the
aspect of long-term and large-scale quantum comput- north and south poles, respectively. While the abso-
ing, which uses redundancy to overcome the loss in lute phase of a quantum state is arbitrary [9], |​Ψ​〉​
information from noisy environments. Finally, our is normalized to unity and must lie on the surface
review of quantum computing technologies is inten- of the sphere. In Figure 1, the amplitudes c0 and c1
tionally narrowed to three of the leading candidates represent the projection of the quantum state onto
capable of near-term experimental demonstrations. the corresponding basis states and the example
However, there is a great diversity of experimental qubit |​Ψ​〉​ has expansion coefficients c0 = cosθ and
quantum physical systems that can be used for encod- c1 = ei​φ​sinθ. This ­representation of the qubit state on
ing and processing quantum information. a unit sphere is commonly called the Bloch sphere in
quantum mechanics.
Principles of quantum computing
The principles of quantum computing derive
from quantum mechanics, a theoretical frame-
work that has accurately modeled the microscopic
world for more than 100 years. Quantum computing
draws its breakthroughs in computational capabili-
ties from the many unconventional features inher-
ent to quantum mechanics, and we provide a brief
overview of these features while others offer more
exhaustive explanations [4].
In quantum mechanics, all knowable informa-
tion about a physical system is represented by a
quantum state. A prominent example of a quantum
state within the context of quantum computing is
the case of a qubit. A qubit, or quantum bit, refers
to the quantum state of an isolated two-level quan-
tum mechanical system. Informally, the qubit is the Figure 1. The Bloch sphere with a
quantum analog of bit that serves as the fundamen- unit radius provides a geometrical
tal unit of information within quantum computing. representation of a qubit. The north
Methods for storing a qubit of information require and south poles of the sphere define the
the control of a physical two-level system, and we orthonormal basis states |0〉 and |1〉,
denote those physical systems as quantum register respectively, while the surface defines
elements that have the ability to store a single qubit the set of all possible qubit values. In
of information. We will discuss some of the differ- spherical coordinates, the example qubit
ent physical systems as quantum register elements |Ψ〉 has expansion coefficients c0 = cosθ
in the “Devices for quantum computing” section. and c1 = eiφ sinθ.

70 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 More formally, a quantum state is defined as A multiqubit register is an addressable array of n
a vector within a Hilbert space, which is a com- two-level physical systems. The principle of superpo-
plex-valued vector space supporting an inner prod- sition may be extended to the register as the quan-
uct. By convention, the quantum state with label ​Ψ​ tum state for the composite physical system is also
is expressed using the “ket” notation as |​Ψ​〉​, whereas given by (2). For an n-qubit register, the computa-
the dual vector is expressed as the “bra” 〈​Ψ|​. The tional basis is expressed in binary notation as
inner product between these two vectors is ​〈Ψ​|Ψ ​ ​〉​and ​ |j 〉 = | ​j1​  ​​ ​j​ 2​​  …​  j​ n​​〉 = | ​j1​  ​​〉 ⊗ |​j 2​  ​​  〉 … ⊗ |​jn​  ​​  〉 ,​ (4)
normalized to one. An orthonormal basis for an
N-dimensional Hilbert space satisfies 〈i| j​  〉​ = δi,j, and where the binary values jk correspond to the binary
an arbitrary quantum state may be represented expansion of j. The dimensionality of the underly-
within a complete basis as ing Hilbert space is N = 2n and any normalized vec-
N−1 tor represents a valid quantum state. In particular,
​ |Ψ〉 = ​ ∑ ​​cj​  ​​  | j 〉​,​  (2) there are composite quantum states that cannot be
j=0
where cj = 〈 j |Ψ​〉​ is the corresponding coefficient. expressed as separable products of n single-qubit
Within a chosen basis, the coefficients of the quan- states. Such states are known as entangled states
tum state are interpreted as probability amplitudes and they are a hallmark of quantum mechanics and,
such that the squared magnitude of this amplitude therefore, quantum computing. For example, con-
yields the probability to lie along the chosen basis, sider the quantum state of a two-qubit register as
that is, pj = |cj |2. The mathematical theory of quan- ​ |Ψ〉 = ​  __ ( 1 2
​√ 2 ​ 1 2 ) ​
1__  ​ ​ | ​0​  ​​ ​0​  ​​ 〉 + | ​1​  ​​ ​1​  ​​ 〉 ​. (5)
tum mechanics is exceedingly rich and draws from
Measuring the individual elements of the register will
aspects of linear algebra, probability, and complex
generate binary outcomes 0 or 1 with equal proba-
analysis. Additional details on these aspects points
bility. Accordingly, the classical expectation for joint
are found, for example, in [9].
measurement of the register is a uniform distribution
The fundamental equation of motion for the
of four possible outcomes. However, measurements of
quantum state is the Schrodinger equation, a partial
this quantum state are always correlated such that both
differential equation defined as
results are either (0,0) or (1,1), where the probability
​ˆ(t ) | Ψ(t ) 〉,​(3)
∂ |Ψ(t ) 〉
​ih̄ ______
​
∂t
​ = H​  for each of these outcomes is 1/2. Notably, there is no
where the time-dependent operator H​  ˆ
​ (t )​defines the
possibility for observing anticorrelated ­outcomes for
this quantum state, for example, (0, 1). The presence
energetic interactions governing the physical system
of these correlations in the measurement statistics is
and is referred to as the Hamiltonian. Consequently,
known as entanglement and the underlying quantum
the Hamiltonian is important for manipulating the
state is said to be entangled. Fundamentally, entan-
quantum state and its control plays a prominent
glement is a limitation on the ability to describe
role in the design and testing of quantum comput-
states of a register solely by specifying the value of
ing technologies. It is important to note that a quan-
each register element, and entangled states are nota-
tum state cannot be directly observed by physical
ble for the ability to violate the local, causal relations
measurement. Rather, the measurements of a quan-
predicted by classical mechanics [11].
tum state must be performed relative to a basis set,
The no-cloning principle represents a fundamental
for example, {| j​  〉​}. The probability to observe the i th
constraint placed on quantum information processing.
outcome corresponds to the probability pi defined
The no-cloning principle is a consequence of the line-
above, such that a series of repeated measurements
arity of quantum mechanics [12], in which the ability
over an ensemble of identically prepared quantum
to perfectly clone, aka copy, an arbitrary quantum state
states will generate a distribution of outcomes that
is not permitted. In particular, given a quantum register
approximates the set of probabilities {pj}. Thus, the
storing an arbitrary state |Ψ​ 
​​ 1​​​​〉​, this information cannot
accurate characterization of this distribution can be
be copied into a second register without loss of infor-
exceedingly difficult due to a large number of basis
mation. Efforts to optimally approximate the value of
states and the infrequent occurrence of measure-
the first register, known as quantum cloning [13], can
ment outcomes corresponding to low probabilities.
be evaluated by measuring the fidelity defined as
A survey of methods for measuring quantum state is
provided in [10]. ​ f = |​〈Ψ​  2​​ ​|Ψ​  1​​  〉​|​​  2​​​​  ,(6)

May/June 2019

Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
71
 Tutorial

where |Ψ2​〉​ is the value of the second register and the remainder of this article will focus on the gate
f ∈ [0,1]. model for quantum computing, we refer the reader
The principles of operation for a quantum com- interested in adiabatic quantum computing to the
puter are based on Schrodinger’s equation in (3), recent review by Albash and Lidar [16].
in which the time-dependent Hamiltonian H​ ˆ
​ (t )​ can We now summarize the basic criteria that define
be directly controlled through the use of externally the expected functionality of quantum comput-
applied fields. Depending on the specific technology ing devices. As first presented by DiVincenzo
in place, these controls will consist of electrical, mag- et al. [17], these criteria represent the minimal behav-
netic, or optical fields designed to drive the dynam- iors needed to perform general-purpose quantum
ics toward a specific response. In the “Devices for computing in the presence of likely architectural
quantum computing” section, we present examples constraints. First is the ability to address the elements
for devices based on semiconductors, superconduc- in a scalable register of quantum systems. Scalability
tors, and trapped-ion technologies. In some compu- implies a manufacturing capability to fabricate and
tational models, the time-dependent controls are layout as many register elements as needed for a spe-
realized as pulsed fields that act discretely on the cific computation. Second, these register elements
quantum register elements.   These discrete periods must be capable of being initialized with high fidelity,
of field interaction are known as gates and the effect as the starting quantum state of the computation must
of the gate on the quantum register is described by a be well known to ensure accurate results. Third is the
unitary operator that transforms the stored quantum ability to measure register elements in a well-specified
state.   This is known as the gate or circuit model since basis. As discussed above, measurement samples the
a diagrammatic sequence of gates acting on registers statistical distribution encoded by the quantum state
provides a design for instruction execution. according the probabilities pi over a given basis set.
An alternative computational model applies the A measurement sample represents readout from the
time-dependent field as continuous interaction sub- register of the quantum computer and this value may
ject to constraints on the rate of change for the overall be subsequently processed.
Hamiltonian. This constraint imposes an adiabatic Fourth, the control over the register must include
condition on the dynamics of the quantum system [14], the ability to apply sequences of gates drawn from
such that the Hamiltonian slowly modifies the interac- a universal set. Universality of the gate set charac-
tions between quantum physical subsystems, that is, terizes the potential to perform an arbitrary unitary
register elements, relative to the internal energy scales operation on the quantum state using a sufficiently
describing those subsystems. As a result, the register long series of gates from that set. In particular, it is
state can be driven toward the desired outcome. This known that a finite set of gates is sufficient to approx-
is known as the adiabatic model given the constraints imate universality and, moreover, that a finite set of
on the controls. A device design based on the adiaba- addressable one- and two-qubit gates are sufficient
tic model has been implemented in superconducting for universality [18]. The latter result, known as the
technology by the commercial vendor D-Wave Sys- Solovay–Kitaev theorem, provides a constructive
tems, Inc. In the realization of that design, the Ham- method for composing arbitrary gates from a finite,
iltonian control is restricted to a specific functional universal gate set. Selection of a universal gate set
form, namely, the transverse Ising model, which limits raises the question of the optimal instruction set
the device operation to computing discrete optimiza- architecture for an intended application within a
tion problems. In addition, the physics of the device specific device technology [19]. The fifth criterion
is not well modeled by the Schrodinger equation (3) is that the gate operation times must be much shorter
but rather require a more sophisticated model that than the characteristic interaction times on which
includes nontrivial interactions with the surround- the register couples to other unintended quantum
ing quantum physical systems as well as finite- physical systems. These interactions induce deco-
temperature effects [15]. Nonetheless, the device has herence of the stored quantum superposition states,
been observed to correctly compute the solution to which leads to the loss of information [20], [21]. In
a wide variety of discrete optimization problems and order to maintain the stored quantum state with suf-
has been characterized as having some advantages ficient accuracy, the duration of the gate sequence
relative to conventional computing devices. While must be shorter than the characteristic decoherence

72 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
time. Fault-tolerant protocols for gate operations are (normally aluminum or polysilicon) that can control
designed to counter the losses from decoherence the energy landscape in the silicon substrate. These
and other errors by redundantly encoding informa- electrodes are appropriately designed and biased
tion with quantum error correction codes [22]. such that a single electron is confined in a quantum
Two additional functional criteria are necessary dot at the interface. Examples of a silicon quantum
for a quantum computer with geometrical con- dot include the MOS device as shown in Figure 2a
straints on the layout of the quantum register. In par- or the Si/SiGe device as shown in Figure 2b. Similar
ticular, layout constraints may impose restrictions on electrostatic control is used for silicon donor devices
which register elements can be addressed by mul- like the example shown in Figure 2c of a phosphorus
tiqubit gates, for example, nearest neighbors within donor implanted inside a silicon substrate. In all of
a 2D rectangular lattice design. Physical layout these examples, the electrons are strongly confined
restrictions may be overcome by moving the stored such that the lowest electronic orbital energy in the
quantum states between register elements. This is quantum dot or the donor is well isolated from other
accomplished using the SWAP gate, a unitary opera- excited electronic states. The confinement length
tion that exchanges the quantum state between two for the donor electron is ∼1.5 nm in all three dimen-
register elements. In addition, a MOVE operation can sions, whereas for the dot electron, these dimensions
support long-distance transport of a stored value, in are ∼10 and ∼2 nm in the lateral and vertical direc-
which the register element itself is displaced. The tions, respectively. These characteristic dimensions
latter proves useful for distributed quantum registers make silicon qubits the most compact technology
that may require interconnects, aka communication as compared to the qubit technologies discussed in
buses, to SWAP register values. The necessity of these later sections.
functions depends on the purpose of the quantum Addressing silicon spin qubits uses an applied
computer and especially the limitations of the tech- static magnetic field B0 to split the orbital degeneracy
nology. Presently, all technologies for quantum com- of the dot electron at the interface. Due to the Zeeman
puting face some constraints on register layout. effect, the orbital for the confined electron is split into
the distinct spin states |↑​〉​ and |↓​〉​. These spin states
Devices for quantum computing encode the computational states |0​〉​ and |1​〉​, where
There are many different possible technologies the energy splitting is given by the Zeeman energy
available for building quantum computers, and γe B0 with γe, ∼28 GHz/T, being the gyromagnetic
these are typically classified by how qubits of infor- ratio of the electron. For 31P donors, the electron and
mation are stored [23]. As discussed in the “Princi- nuclear spins are coupled by the hyperfine interac-
ples of quantum computing” section, these devices tion, A ∼117 MHz [28]. The donor qubits are gener-
must meet several functional criteria to carry out ally operated under large magnetic fields B0 > 1 T,
reliable quantum computation. In this section, we such that ​​(​γe​  ​​  + ​γ n​  ​​)​​B​  0​​    A​, where γn ∼ 17 MHz/T is the
provide an overview of three technologies that are gyromagnetic ratio of the nucleus. In this limit, the
currently used for developing quantum computing eigen spin states are the tensor products of the elec-
devices and we discuss the progress toward meeting tronic spin (|↑​〉​, |↓​〉​) and the nuclear spin (|⇑​〉,​ |⇓​〉​)
the functional criteria. states. The resulting energies are shown in Figure 2d,
where the electron spin qubit splitting depends
Silicon spin qubits on the nuclear spin states, and vice versa. Typical
Silicon spin qubits denote a technology imple- energy splittings are of the order of tens of gigahertz
mentation by which quantum information is and megahertz for the electron and nuclear spins,
encoded either in the spin states of an electron in respectively [29], [30]. The hyperfine interaction A
a silicon quantum dot, or in the spin state of the and the electron gyromagnetic ratio γe depend on
electron or nucleus of a single-dopant atom (typi- the orbital wavefunction of the electron, which can
cally group V donors) in a silicon substrate. In par- be tuned with electric fields [31], [32]. As a result,
ticular, the orientation of the spin in these systems is the qubit splittings are electrically tunable after the
used to encode the |0​〉​ and |1​〉​ states. Notably, these silicon qubit devices are fabricated.
silicon devices are fabricated with conventional Electron spin qubits are commonly initial-
CMOS techniques and consist of gate electrodes ized and measured using spin-charge conversion

May/June 2019

Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
73
 Tutorial

Figure 2. (a) Bottom: Scanning electron microscope (SEM) image of a MOS quantum dot
device similar to the one where single and two-qubit gates were demonstrated.
Top: Cross-sectional schematic of the device illustrating the location of qubits at the
Si/SiO2 interface. (Reprinted from [24] with permission from Nature.) (b) Bottom: SEM
image of a Si/SiGe double quantum dot device, where two-qubit operations were
implemented. Middle: Variation of the static magnetic field along the axis of quantum
dots. Top: Cross sectional device schematic highlighting the position of the quantum
dots. (Reprinted from [25] with permission from the American Association for the
Advancement of Science (AAAS).) (c) SEM image of an ion-implanted 31P device similar
to the one used for demonstrating record spin-coherence times [26], [27]. (d) 31P donor
electron (|↑〉, |↓〉) and nuclear (|⇑〉, |⇓〉) spins states [34].

techniques [33]. Charge sensors such as quantum on the principles of magnetic r­esonance, t­ransitions
point contacts and single-electron transistors (SETs) between the spin states are then achieved at a rate
are located adjacent to the quantum dot (or donor) proportional to the amplitude of the driving field [35].
and are then capacitively coupled to them, cf., The driving field is pulsed appropriately to obtain a
Figure 2. The charge sensors are biased appropri- specific rotation of the spin state, for implementing
ately with gate voltages, such that the current passing a single qubit gate. A microwave transmission line
through them is strongly sensitive to the electrostatic antenna (see Figure 2a and 2c) is normally used to
environment in their vicinity. The orbital energy of generate the driving field [36], yielding magnetic field
the electron is then electrically tuned such that the amplitudes of ∼0.1 mT, and single qubit gate times of
electron can preferentially tunnel to the same or few microseconds [29] (or milliseconds [30]) for the
another nearby charge reservoir, depending on its electron (or nucleus). Alternatively, a micromagnet
spin. The presence or absence of the electron on the producing a dc magnetic field gradient (Figure 2b)
donor or dot can then be detected via a change in can also be embedded on chip [37]. In the presence
current passing through the charge sensors, which of an additional oscillating electric field (from gate
aids to readout the electron spin state. The protocol voltages), the electron feels an effective oscillating
will also initialize the electron spin state in the dot or magnetic field, resulting in spin resonance with faster
the donor to |↓​〉​[33]. gate times. Note that the frequency of the control field
For spin control, an oscillating (driving) magnetic is different for both the electron [electronic spin reso-
field is applied to the qubits. The frequency of the oscil- nance (ESR) frequencies ∼ tens of gigahertz] and the
lating field is chosen to be equivalent to the energy nucleus (NMR frequencies ∼ tens of megahertz). The
difference between the two spin qubit l­evels. Based ability to c
­ ontrol and readout the electron spin state

74 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
also allows measurement of the nuclear spin state. As (such as Si-29) in the lattice, and their fluctuations,
the ­electron spin resonance frequency is determined can result in decoherence of the electron spins [41].
by the nuclear spin state (see Figure 2d), probing Hence, isotopic purification of silicon from spin-
frequencies at which the electron can be controlled containing nuclei allows for long-coherence times
allows readout of the nuclear spin [30]. (T2) of milliseconds and seconds for the electron
Since the splittings are dependent on A and γe, and nuclear spins, respectively [27]. Additional
they can be tuned electrically and it is possible to sources of decoherence include charge or electric
independently control each donor located within field noise arising from nearby defects or traps, con-
a precisely positioned array [38]. In their idle state, trol signals, gate electrodes, and thermal radiation
the qubits are electrically detuned from the control from the microwave antenna [27].
field by appropriately tuning A and γe. When opera- Although the methods used to address and couple
tions need to be performed on the qubits, they are silicon qubits can be integrated with the microelec-
brought in resonance with the control field, that is, tronics industry, the qubits are very sensitive to atomic
the energy splitting is tuned to the frequency of the details that have not yet been addressed in the indus-
control field. try. These details strongly affect the qubit operation,
The coupling between two electron spin qubits and hence, it is essential to design devices that mini-
occurs via the intrinsic exchange interaction mize their influence on the qubits. First, the exchange
between them [38]. The exchange coupling Je is pri- coupling between donor electrons is extremely sen-
marily determined by the overlap between the two- sitive to the position of donors, necessitating precise
electron wave functions. Je can hence be tuned by donor placement accuracies and/or large exchange
either modifying the tunnel barrier between the two coupling tunability [42], [43]. Efforts are underway
electrons or by shifting the relative orbital energies to demonstrate qubits with single-donor atoms in
of the two electrons [39]. Both these methods can silicon that are placed precisely with scanning tun-
be realized by appropriately tuning the gate voltages neling microscopy [44], as well as to explore alter-
that control the potential landscape in the device. nate means of coupling between the qubits (such
To perform a CNOT gate, the electron spin qubits as dipolar interactions [45], [46]) that are less sen-
are operated in a regime where Je is smaller than sitive to donor placement inaccuracies. In addition,
the energy difference between the qubit splittings atomic roughness and step edges at the interface can
of the two electrons (often termed as the detuning). result in the excited orbital states coming close to the
In such a regime, each electron spin qubit will have ground orbital state in silicon quantum dots, acceler-
two resonance frequencies, which are determined ating relaxation and even resulting in a nonspin-1/2
by the state of the other qubit. Hence, an oscillating ground state [40]. The energy separation between the
control field at one resonant frequency will con- ground and excited orbital states (also referred to as
ditionally rotate the qubit dependent on the state valley splitting) can be tuned with electric field to an
of the other qubit, resulting in a CNOT gate [24], extent [47], yet it is always desirable to obtain larger
[25]. To perform SWAP, the qubits are initialized and uniform valley splittings with a smooth interface.
in a regime, where the exchange coupling is much Finally, uncontrolled strain in the lattice arises from
smaller than their detuning. The exchange coupling the thermal mismatch between the gate and substrate
is then increased to a value much larger than their materials when the device is cooled from room tem-
detuning, such that the two qubits exchange infor- perature to milli-Kelvin temperatures [48]. This modi-
mation with each other. After an appropriate time fies the potential landscape in the device, altering the
that determines the angle of SWAP, the exchange position and confinement of the quantum dots, along
coupling is brought back to a low value. with introducing accidental dots. Ref. [48] highlights
The spin-orbit coupling is weak for electrons in sil- that using gate materials (such as polysilicon rather
icon, resulting in long spin-relaxation times T1. The than aluminum), which have similar thermal expan-
relaxation time has been shown to be dependent on sion coefficients to that of silicon, can aid to reduce
the temperature and magnetic field [40]. Operating the lattice strain.
the qubits at low temperatures (< 1 K) and magnetic The exchange interaction between the qubits is
fields (< 5 T) yield T1 exceeding several seconds and short-range (within a few tens of nanometers) can
even hours. The presence of spin-containing nuclei only result in nearest neighbor couplings. To scale

May/June 2019
75
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

up silicon qubit devices to a large-scale architecture, Trapped ion qubits

it is beneficial to have connectivity between qubits Trapped ion qubits represent an implementation
that are separated by much larger distances. Methods where quantum information is encoded in the elec-
to couple silicon qubits to a photonic mode spanning tronic energy levels of ions suspended in vacuum.
∼ centimeter in a microwave resonator have been To obtain trapped ions, metals such as Calcium
proposed [45], [49], and recently demonstrated in (Ca) or Ytterbium (Yb) are first resistively heated
Si/SiGe quantum dots [50], [51]. Through the pho- and vaporized with a current passing through them,
tonic mode, two qubits, separated by as far as a and then directed to the trap. While loading these
centimeter, can be virtually coupled to each other, ions into the trap, these vaporized neutral atoms are
enhancing the qubit-connectivity significantly. Cou- simultaneously photo-ionized, where their outer-
pling the spins to the resonator also provides a path- most electron is removed, resulting in ions that have
way to readout the spin states [45]. The transmission a single valence electron. As the ions are charged
frequency of the resonator then depends on the spin particles, appropriate voltages applied to gate elec-
state of the qubit. Hence, applying a microwave sig- trodes in their vicinity and resulting electric fields
nal to the resonator, and measuring its transmission can then confine the ions in the trap. The most com-
aids to detect the spin state. mon gate electrode configuration for ion trapping
Designing silicon spin qubit devices requires is the (rf   ) Paul trap (Figure 3a), which consists of
modeling several classical and quantum mechani- four electrodes (two with oscillating voltages and
cal parameters with a range of techniques that are two grounded) that induce an effective harmonic
adapted from the semiconductor industry [34]. potential in the xy plane, and additional two dc gate
Classical variables that are relevant and need to be electrodes to induce harmonic confinement in the
z plane [55]. In the harmonic oscillator potential,
solved for include the electrostatic potential land-
there are several eigenstates corresponding to the
scape, electric fields, electron densities, capaci-
vibrational modes of the trapped ions. To ensure
tances, magnetic fields, and strain. The electrostatic
that thermal effects and fluctuating electromagnetic
parameters in silicon devices can be obtained by
fields do not cause random excitation of these states
solving Poisson’s equation with the finite-element
and thereby motion of the ions, the ions are laser-
method with traditional technology computer-aided
cooled to their vibrational ground state [56]. For a
design (TCAD) packages such as Sentaurus TCAD, or
small number of ions (∼50), the ions will then be
a general multiphysics package like COMSOL. Solv-
arranged in a linear chain along the z-direction such
ing Maxwell’s equations with high-frequency electro-
that overall forces from the external fields cancel out
magnetic solvers (such as CST-Microwave Studio or
the forces from their Coulomb interaction. Typical
ANSYS-HFSS) aids estimating the driving magnetic
ion separation in the trap is ∼10 µm.
fields generated by the microwave antenna in such
As mentioned above, a qubit is defined using the
devices. Thermal strain while cooling such devices energy levels of individual ions in the trap to encode
can also be simulated by solving the stress–strain the basis states |0​〉 a​ nd |1​〉​. Depending on the orbital
equations with COMSOL [48]. In addition to the clas- energy levels used for encoding, there are two popu-
sical parameters, it is also essential to solve the elec- lar implementations of trapped-ion qubits: hyperfine
tronic structure in silicon qubit devices and estimate and optical. For hyperfine qubits, the states corre-
the electron orbital energies and wave functions. spond to the hyperfine levels in the atomic s-orbital.
Effective mass theory and tight-binding techniques For example, as shown in Figure 3b, the ion 171Yb+
have been extensively used for such calculations [40]. has a nuclear spin of 1/2 and the qubit is encoded
The orbital energies and wave functions act as a using the singlet |S ​〉​ and |T0 ​〉​ configurations of the
handle to the hyperfine, exchange, and tunnel cou- electron and nuclear spins [57]. A small dc mag-
plings, along with the electron gyromagnetic ratio netic field is applied to separate the |T0​〉​ state from
and electron spin relaxation times. These parameters other triplet states |T−​〉​ and |T+​〉​. The qubit splitting
are ultimately fed into a simplified spin Hamiltonian, of 12.6 GHz for 171Yb+ is determined by the hyperfine
which is solved with mathematical packages (such interaction between the electron and the nucleus,
as MATLAB, Mathematica, or QuTiP), to simulate the and insensitive to magnetic field fluctuations up to
instantaneous spin states and quantum gate fidelities. first order [58]. Alternatively, for the optical qubit

76 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
Figure 3. (a) Schematic of a Paul trap used to confine ions in vacuum. Inset:
Visualization of ions in the trap with fluorescent techniques. (Reprinted from [52] with
permission from Nature.) (b) Electronic energy levels of a 171Yb+ ion illustrating qubit
encoding (|0〉 and |1〉) with hyperfine energy levels [53]. Transition between qubit states
is achieved by a Raman process via excitation to a virtual state |e〉. (c) Electronic
energy levels of a 40Ca+ ion illustrating qubit encoding with the s- and d-orbital energy
levels. (Reprinted from [54] with permission from Springer.)

encoding with trapped ions, the basis corresponds other basis state |0​〉​, base​​d on spin-selection rules
to s-orbital and d-orbital electronic energy levels. As [60]. Hence, if the initial qubit state is |1​〉​, the result-
shown for 40Ca+ in Figure 3c [54], the energy splitting ing p-state after excitation may spontaneously decay
is then ≈ 411 THz and equivalent to 729 nm. Trapped to states apart from |0​〉​, which are also continuously
ion qubits are highly reproducible [59] provided excited. Photons from the spontaneous emission are
there are no magnetic and electric field inhomoge- then detected with a CCD camera. If the initial qubit
neities in the trap, which may modify the energy lev- state is |0​〉​, the qubit cannot be excited to the p-states
els through Stark and Zeeman effects, respectively. by the laser, as its frequency is far away from reso-
Fluorescent techniques are used to visualize the nance and there is no output at the CCD camera.
ions, where the qubit states are continuously excited For optical qubits, a stable laser (having ∼400-THz
to the p-states with the help of a laser, to induce an frequencies) with a narrow line-width can drive
electric dipole transition [56]. On such a transition, the the transitions between the |0​〉​ and |1​〉​ states via a
ions scatter the photons that are detected by photo-­ quadrupole transition, enabling qubit control [62].
multiplers or a charge-coupled device (CCD) camera The hyperfine qubits can be controlled with two
(see Figure 3a). The required laser frequency is equiv- methods. First, microwave radiation with frequen-
alent to the separation between the energy states used cies (e.g., 12.6 GHz for 171Yb+) matching the qubit
for the transition and depends on the choice of the ion. splitting can drive transitions between |0​〉​ and
The hyperfine and optical qubits are initialized |1​〉​ states [63]. Microwaves can be generated with
with optical pumping. Here, a laser is incident on a microwave horn that is located several centime-
the ions with an appropriate frequency that can con- ters from the trap. However, as microwaves corre-
tinuously drive the |1​〉​ state to the excited p-states. spond to centimeters in wavelength, and the ions
Any spontaneous decay from the excited p-state to are separated by micrometers, it is not possible to
ground states apart from |0​〉​, are also further driven focus microwaves and address individual qubits in
by the laser [60]. Over a period of time (∼microsec- a chain of several ions. Second, stimulated Raman
onds), all the spontaneous emissions result in the transitions with two laser fields (from a pulsed laser)
qubit state being initialized to |0​〉 ​[61]. can be used to control the qubit state [64]. Each
For readout of trapped-ion qubits, the laser is laser field excites the qubit states to a virtual level
tuned to a frequency that continuously drives one |e​〉​ that is well detuned (by δ ) from the excited
of the basis states (e.g., |1​〉​) to an excited p-state. The p-states (see Figure 3b). The frequency difference
polarization of the laser and excited state is chosen between the two laser fields is chosen to match the
such that spontaneous emission cannot occur to the qubit splitting. Based on a Raman process, the qubit

May/June 2019
77
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

is rotated at a frequency proportional to the p ­ roduct interact with them, for connecting distant qubits.
of the individual Rabi frequencies (from |0​〉​ to This would require exquisite control of the shuttling
|e​〉​and from |1​〉​to |e​〉​determined by the laser power) of the atomic ions, as well as the periodically cooling
and inversely proportional to the detuning δ from down the excess motion arising from shuttling ions.
the p-states. This method has the advantage of selec- Although this method could potentially work for a
tively addressing the qubits, where the laser can be larger number of qubits (∼1000), it becomes imprac-
focused individually on each qubit. Typical times- tical for scale-up due to the complexity of intercon-
cales for single qubit operations are of the order of nects, diffraction of optical beams, and extensive
several microseconds. hardware requirements. Photonic interfaces have
The Coulomb interaction between the ions been proposed to connect even larger systems [61].
serves to mediate the coupling between the qubits Here, qubits at the edges of the chain are driven to
[52]. Based on this interaction, the qubit states are an excited state with very fast laser pulses so that
coupled to the vibrational modes of the ion chain. at most one photon emerges from each qubit after
Hence, appropriate laser frequencies can help trans- radiative decay. Following selection rules, the radi-
ferring the qubit states to the vibrational modes. ative decay can lead to entanglement between the
Depending on the vibrational modes of the ion-trap, photonic and trapped ion qubit. Photons from two
a subsequent ion in the chain can be rotated with a separate qubits are mode-matched and interfered on
laser, to demonstrate a CNOT gate. The vibrational a beamsplitter, which is then detected. A successful
modes can also be swapped with the subsequent detection then yields an entangled state between the
qubit, resulting in a SWAP gate. two distant ion trap qubits.
Like silicon spin qubits, trapped ion qubits have The design packages available in the conven-
extremely long relaxation and coherence times. The tional microelectronics industry cannot be directly
relaxation mechanism is via spontaneous decay that extended to design trapped ion qubits, as their
approaches several seconds for optical qubits, and implementation has very little overlap with that
several days for hyperfine qubits. The coherence of of silicon. Nevertheless, the electric fields avail-
the qubits is primarily affected by ambient magnetic able from classical electrostatic solvers (such as
field fluctuations that modify the qubit energy lev- ­COMSOL) can be used to optimize and design the
els through the Zeeman effect, laser intensity, and gate electrode configuration and voltages for the
frequency fluctuations over time, and coupling of trap. As illustrated previously in this section, the
the qubit states to the vibrational degree of freedom electronic orbital levels of single ions (or even a
during two-qubit operations [65]. The sources of cluster of ions) in the trap, determine the laser fre-
decoherence for the vibrational degree of freedom quencies needed for initialization, readout, control,
include unstable trap parameters, the coupling of and coupling of the trapped ion qubits. The orbital
the electric dipole associated with the motion energies and hyperfine interactions for a variety
of ions to thermal radiation in the environment, and of trapped ion candidate materials can be deter-
ion collisions with the residual background gas. Typ- mined from ab initio electronic structure calcula-
ical coherence times of the trapped ion qubits due to tion techniques such as DFT. A significant aspect
these effects is of the order of seconds. of the design also includes the optical setup for the
The coupling rate between the qubit state and lasers, including its power and focus. These param-
vibrational mode (for two-qubit operations) has been eters can be obtained with commercial ray-tracing
shown to be inversely proportional to the square software packages such as Zemax, Code V, or Oslo.
root of the number of ions in the chain [61]. Hence, The dynamics of the trapped ion qubits upon inter-
increasing the ion number in the chain beyond action with a laser can be mapped onto a simpli-
∼50 slows down the two-qubit operations, where fied Hamiltonian, which can then be solved with
decoherence (heating) of the motional modes and commercial mathematical packages, such as MAT-
fluctuating electric fields become significant. Archi- LAB. While there are several analytical expressions
tectures for scale-up with a larger number of ions and mathematical models for light-matter interac-
include quantum charge coupled device (QCCD) tions, a device simulator capable of capturing the
architectures [66] where individual ions at the edges nonidealities in realistic trapped ion devices is cur-
of a trap are shuttled to nearby traps and made to rently nonexistent.

78 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
Superconducting transmon qubits capacitance between the superconductors___ [68]. The
Transmon qubits encode quantum information qubit splitting is then given by ​E01 ≈ ℏ/ ​√LC ​ ​, where​
in the charge states of superconducting islands con- ℏ​ is the reduced Planck constant. E01 is typically
nected by Josephson junctions. The superconduc- ∼5 GHz in units of frequency.
tors, typically aluminum or niobium, are deposited The total energy of the system is distributed
on a silicon substrate, allowing transmon qubits between the inductor and capacitor, and thereby
to be fabricated on a large scale with techniques consists of two parts: 1) Josephson energy
adapted from the microelectronics industry. A micro- EJ = ​ℏ​2/(4e2L) and 2) charging energy EC = e2/2C
scopic image of a transmon qubit device is shown in of the superconductors [69]. As the charge states
Figure 4a and illustrates that the qubit region spans constitute the qubit, they can heavily be suscep-
length scales of tens of micrometers. tible to electric field noise. The noise sensitivity
The simplified qubit Hamiltonian can be derived can be minimized with appropriate distribution of
as a quantum analog of a classical LC oscillator, energies EJ and EC. Figure 4b plots the energy levels
where L is the Josephson inductance, and C is the for several values of EJ /EC , and indicates that large

Figure 4. (a) The transmon qubit consisting of two superconducting islands that are
coupled through Josephson junctions and a large interdigitated capacitance. Inset: SEM
image of the device in the vicinity of the Josephson junctions. (b) Eigenenergies Em (first
three levels, m = 0, 1, 2) of the superconducting system as a function of the effective
offset charge ng induced by nearby gate electrodes and environment [67]. Energies
are given in units of the transition energy E01 = E1 − E0 evaluated at ng = 1/2, and are
calculated for various values of EJ / EC. The zero point energy is chosen as the bottom of
m = 0 level. For increasing values of EJ /EC, Em becomes more robust against fluctuations
in ng arising from environmental noise, whereas the anharmonicity (Eδ = E01 − E12)
reduces. EJ / EC is chosen between 10 and 50 for transmon qubits in order to obtain
robustness with sufficient anharmonicity. (c) Schematic of a transmon qubit capacitively
coupled to a superconducting resonator for initialization, readout and control [67].
The capacitances between various entities of the transmon-resonator system are also
labeled. (d) Equivalent circuit of a transmon coupled to the resonator [67]. [(b), (c), and
(d) are reprinted from [67] with permission from the American Physical Society (APS).]

May/June 2019
79
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

values of EJ /EC render the qubits robust against Alternatively, direct capacitive coupling between
noise. However, this will also lower the difference two adjacent transmon qubits can also be leveraged
between qubit splitting and other splittings in for demonstrating CNOT gates. However, using only
the system, often called the anharmonicity. A large direct capacitive coupling between the qubits leads
anharmonicity is required to ensure that charge to significant cross talk when they are incorporated
states with higher energy levels are not excited in a large-scale architecture.
while operating the qubit. As a tradeoff, EJ /EC is Compared to silicon and trapped-ion qubits, the
normally chosen between 10 and 50 for sufficient relaxation and coherence times of superconducting
robustness of the qubit, along with anharmonicity qubits are short. The main sources of decoherence
Eδ ≈ EC /2 ∼100 MHz [67]. arise from the coupling of the qubits to additional
To perform quantum operations, the transmon two-level systems present in the bulk or interfaces
qubits are commonly placed adjacent to a super- of the device, nonequilibrium quasi-particles gen-
conducting resonator (Figure 4c) and is capacitively erated from stray infrared light, and radiation to
coupled to it (Figure 4d) [67], [70], [71]. Here, the ­additional modes present in device [73], [74]. The
qubit-resonator system is designed to be in the dis- relaxation rate has also been shown to be expo-
persive regime, where the detuning (∆ ∼100 MHz) nentially dependent on the temperature, due to the
between the qubit and the photonic mode of the res- qubit interaction with thermal photons [67]. As a
onator is much larger than the coupling ( g ∼10 MHz) result, extremely low temperatures, ∼20 mK, are nec-
essary for the high-fidelity operation of qubits. Dif-
between them. In this regime, the shift in the reso-
ferent device designs and operation regimes during
nator transmission frequency from its fundamen-
the last decade have resulted in improvements in the
tal mode frequency is given by ± g2/∆, where the
relaxation and coherence times by several orders of
sign (+ or −) depends on the qubit state [70]. By
magnitude. Dephasing times currently is of the order
applying microwave pulses to the resonator, and
of ∼100 µs.
measuring its transmission, the qubit state can hence
The Josephson energy is strongly determined by
be readout.
the critical current across the junction, which, in
Resonant microwave pulses can be used to con-
turn, is dependent on the superconducting energy
trol the qubits, as the qubit splitting is ∼5 GHz. Qubit
gap and the normal resistance (Rn) of the Josephson
control timescales are a few hundreds of nanosec-
junction when it is operated above the critical tem-
onds depending on the quantum gate operation and
perature [75]. Rn is determined by the thickness (few
are much faster than that of trapped ion and silicon
nm) of the Josephson junction and can be variable
spin qubits. Measurement of the qubit and its subse-
across different devices. This results in nonuniform
quent control also aids in deterministic initialization qubit splittings across devices, with an in-homogeneity
of the qubit state. of ∼300 MHz. Another significant challenge is the
Two qubits, which are significantly detuned from large size (several tens of micrometers) of supercon-
the resonator, can be coupled to each other via the ducting qubits, limiting the number of qubits that
resonator. The coupling rate between the qubits is can be coupled to each other via a single resona-
​g​  ​​g​  ​
given by ___
​​​​ 12 2 ​ ​(​1⁄​∆​  1​+ ​1/∆​  2)​​ ​,​ where g1 and g2 are their tor, which spans about a centimeter. Scaling up the
individual coupling strengths to the resonator, and current demonstrations to a large-scale architecture
∆1 and ∆2 are their detunings with respect to the with millions of well-connected qubits operating at
resonator [72]. However, the effective coupling extremely low temperature will benefit strongly by a
rates (∼megahertz) between the qubits will still be reduction in the size of the qubits [76].
smaller than the detunings (∼300 MHz) between While a standalone tool for designing supercon-
them, caused by differences in the qubit splittings ducting qubits is nonexistent, parameters such as the
during manufacturing. As a result, the resonance fre- capacitance (for determining EC) and inductance
quency of each qubit will be determined by the state (for determining EJ) can be estimated with classical
of the other qubit, similar to the electron/nuclear electrostatic and electromagnetic packages such as
spin qubit splittings shown in Figure 2d. This enables FastCap and FastHenry, respectively. Microwave soft-
conditional rotation of one qubit, dependent on the ware such as TXLINE (in AWR Microwave Office) has
state of the other qubit, and hence a CNOT gate. been used to design and estimate the c ­ haracteristic

80 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Table 1. Summary of qubit implementations.

impedance of the superconducting resonator that but can provide important insights leading to fault
aids to readout, control, and couple the qubits. In mitigation strategies or improved devices. For simply
addition, the electromagnetic fields experienced by assessing the performance of a device, benchmark-
the superconducting qubits can be obtained by solv- ing is more practical.
ing Maxwell’s equations with high-frequency elec-
tromagnetic simulators, such as ANSYS-HFSS. As for Benchmarking: Metrics and techniques
silicon and trapped-ion qubits, the qubit dynamics The most basic performance metric is the prob-
can also be obtained by solving the simplified Ham- ability that the device outputs the correct state. In
iltonian with mathematical packages. the context of quantum mechanics, this corresponds
To conclude this section, we reiterate that dif- to the inner product (or overlap) between the out-
ferent implementations are unique with their qubit put state and the intended state, which is called the
type, and methods for qubit readout, control, and fidelity. The infidelity, defined as 1 minus the fidelity,
interaction. We summarize our description of the quantifies the amount of error in the output state.
different qubit technologies in Table 1. Another common way of quantifying the output
error is in terms of the geometric distance between
Testing and characterization of the output state and the target state in the complex
quantum devices vector space.
In spite of the great progress in fabrication and If a qubit device is used to output a specific
control of qubits, today’s quantum computing quantum state, for example, some reference state or
devices are far noisier and error-prone than conven- resource state, the fidelity of the output with respect
tional digital circuits. Bit error probabilities of 10−3 to this known state can be estimated by measuring
to 10−2 per qubit per operation (or per clock cycle) random subsets of qubits along various directions
are typical. Even with continued progress in qubit of the Bloch sphere [77], [78]. In such cases, the
technologies, it is unlikely that the errors incurred experimental cost scales favorably with the regis-
by physical qubits will ever become negligible. Thus ter size. However, a qubit device would be used to
understanding and mitigating fault processes in perform a wide variety of computations each with
qubit devices is a critical aspect of quantum com- a different output state, and these output states
puter development. Correspondingly, the experi- presumably cannot be computed by any conven-
mental testing of qubit devices primarily concerns tional means. In this case, one desires experimen-
the accuracy and reliability of hardware operation tal metrics that allow one to estimate or bound the
rather than the correctness of the circuit logic. fidelity of the device output for any computation it
Qubit device testing may be divided into two ­performs. The state-of-the-art approach for this pur-
broad categories: characterization, wherein the pose is randomized benchmarking (RB) [79]. RB is a
goal is to obtain a detailed model of a device’s fault technique for assessing how much, on average, each
modes; and benchmarking, wherein the goal is to ­operation decreases the output fidelity. Essentially,
determine a few high-level performance metrics. RB involves measuring the final fidelity of a qubit for
Characterization is the more costly type of testing random operation sequences of varying lengths. For

May/June 2019
81
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

weak uncorrelated errors, the fidelity decays expo- Quantum tomography as just described requires
nentially as a function of sequence length. The RB well-calibrated measurements, whereas qubit meas-
decay constant is broadly interpreted as the aver- urements are among the device operations that need
age error per gate, an obviously useful performance to be characterized. This problem is overcome with
metric. Extensions of RB have been devised to yield Gate Set Tomography [87], [88], the state-of-the-
operation-specific error metrics [80], [81], to incor- art method for detailed characterization of qubit
porate multiqubit operations [82], to include qubit devices. Gate set tomography involves tomographic
loss [83], and to assess cross-talk [84]. While RB measurements of many different sequences of
remains a very popular benchmarking method, its device operations. These sequences, which range in
underlying fault model is not universal; hence, RB in length up to hundreds or thousands of operations,
its current form may not be entirely valid or accurate are carefully chosen to reveal all possible types of
as engineering efforts continue to make the simple qubit errors. The data are then fit to a highly nonlin-
fault modes assumed by RB less and less prominent ear model using a sophisticated procedure, yielding
[85]. Additionally, it has been noted that relating a self-consistent model of all of a device’s operations,
RB decay constants to operation fidelities is subtly including the measurement operations themselves.
problematic [86]. Gate Set Tomography has been used to characterize
and significantly improve the control of t­rapped-ion
Characterization via quantum tomography qubits [89].
An alternative to benchmarking is to thoroughly
characterize the fault modes of the device. Since the Other approaches
output state of a quantum circuit is exponentially In addition to RB and Gate Set Tomography,
large in the number of qubits, characterization of a a number of other testing approaches have been
quantum circuit as a whole is generally infeasible. developed. Some of these remain theoretical pro-
The established strategy is to characterize each oper- posals, while others have had at least limited experi-
ation of a qubit device as completely as possible so mental demonstrations.
that the result of any given sequence of operations One approach is to test a quantum device utiliz-
can (in principle) be predicted accurately. The gen- ing another quantum device, either as a reference or
eral name for this strategy is quantum tomography, a as a resource to perform more powerful quantum-
name derived from the medical imaging technique in based tests [90]. This line of approach stands to
which a 3D image of a subject is reconstructed from greatly reduce the cost of quantum device char-
a set of 2D projections. In a similar manner, quantum acterization, but it requires the availability of well-
tomography reconstructs a quantum state or oper- characterized quantum circuits that are similarly
ation from multiple measurements, each of which difficult to certify.
reveals a particular projection of the state. This recon- Another approach is to exploit prior knowledge
struction is based on the fact that a quantum state is to reduce the cost of conventional benchmarking
uniquely specified by the probability distributions for and tomographic methods. For example, adaptive
certain characteristic quantities of a physical system. testing based on Bayesian principles can signifi-
(For a spin qubit, the characteristic quantities are the cantly accelerate both RB [91] and tomography
projection of the spin along three independent spatial [92], [93]. In the case that the state or operation in
directions.) State tomography is the determination of question has some known characteristics (e.g., it has
the quantum state via statistical estimation of these low rank or belongs to a certain symmetry class),
characteristic distributions. Tomographic methods specialized testing methods that are more efficient
can also be used to characterize qubit operations. A are applicable [94], [95]. Related to this, the tech-
qubit operation can be thought of as a linear transfor- nique of compressive sensing has been adapted to
mation of the characteristic probability distributions. the quantum domain and applied to the characteri-
Quantum process tomography is the determination zation of quantum states [96].
of the transformation matrix by characterizing the Other forms of testing may be categorized as
output state for each possible input state, or more model fitting, for example, determining particular
precisely, for a set of linearly independent states that parameters of qubit dynamics, or assessing particu-
span the state space. lar properties of the device output (e.g., purity or

82 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
entanglement). One recently developed approach
to characterizing the quality of many qubit devices
is to measure the distribution of output states pro-
duced by executing random quantum circuits [97].
This reveals the extent to which the device can cre-
ate and maintain superpositions of computational
states, a key facet of the “quantumness” of quantum
computation. Finally, there is now a rapidly growing
interest in the use of machine learning techniques for
characterizing quantum systems. Instead of attempt-
ing to match the experimental data to an intrinsi-
cally quantum model that is likely to be intractable,
researchers have begun to use neural nets to learn
the behavior of quantum systems from experimental
data [98]–[101]. The learning process implicitly cre-
ates a tractable model of the quantum system.

Quantum circuit design and synthesis Figure 5. The Clifford+T gate set is
Quantum circuits provide representation for how a universal basis for expressing
register elements may be modified by a sequence quantum circuits.
of gates to implement basic computation. As sum-
marized in the “Principles of quantum computing” • CNOT gate: The CNOT gate belongs to the set of
section, gates represent quantum mechanical opera- Clifford+T gates, cf., Figure 5, and the two inputs
tors that address one or more register elements and produce two outputs according to the logical
by design, the gates are reversible and represented mapping A, B → A, A ⊕ B.
by unitary matrices [102]. However, the available • Toffoli gate: Figure 6 presents the circuit dia-
gates are often restricted to well-defined subsets gram and matrix representations of the Toffoli
of available operators from which a quantum cir- gate, while Figure 7 shows an example of how
cuit specification must be constructed. Fixed-point this three-qubit gate may be implemented as a
arithmetic circuits can be used for solving complex series of two-qubit Clifford gates and the T gate.
elementary functions including evaluation of Tay- Notably, the Toffoli gate itself is universal for
lor series [103], [104]. In this section, we review reversible logic and the three inputs produce
the design of quantum circuits with an emphasis three outputs according to the logical mapping
on arithmetic operations, such as addition, sub- A, B, C → A, B, A · B ⊕ C.
traction, and multiplication, which are required in • Fredkin gate: Figure 8 presents the circuit dia-
the implementations of many quantum algorithms gram for the three-qubit Fredkin gate and its ma-
[102], [105]. We also review the steps required for trix representation. The Fredkin gate is universal
the synthesis of quantum circuits into technology-
specific implementations.
The design of quantum arithmetic circuits
based on Clifford+T gates has caught the attention
of researchers [105]–[108]. Figure 5 presents the
quantum gates in the Clifford+T gate set with their
matrix and graphic representations. The Clifford+T
quantum gate set can be used to realize multiqubit
logic gates such as the Toffoli and Fredkin gates
previously presented in [109] and [110]. These
multiqubit gates will prove useful for describing the Figure 6. The circuit diagram for the
implementation of quantum circuits presented in three-qubit Toffoli gate and its matrix
this article. representation.

May/June 2019
83
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

Figure 7. The Toffoli gate may be Figure 9. The Fredkin gate may be
implemented as the series of Clifford implemented as the series of Clifford and
and T gates shown [111]. T gates shown [111].

for reversible logic and, as shown in Figure 9, it computers [117], [118]. We now define the T-count,
can also be realized as a sequence of two-qubit T-depth, and qubit cost resource measures.
Clifford gates and T gates. The Fredkin gate maps
• Qubit cost: The total number of qubits required
_ inputs to three_ outputs as A, B, C → A,
three
to design the quantum circuit.
​​ ​​ · B + A · C, A · B + A
A ​​ ​​ · C.
• T-count: The total number of T gates used in the
Recent proposals for the realizations of reversible quantum circuit.
logic gates and quantum circuits have focused on • T-depth: The number of T gate layers in the cir-
the fault tolerant Clifford+T gate set due to its demon- cuit, where a layer consists of quantum opera-
strated tolerance to noise errors [111], [112]. Potential tions that can be performed simultaneously.
fault-tolerant implementations of these gates could
play an important role in mitigating the noise observed Quantum operators are reversible, and therefore,
in current quantum computing devices [111]–[113]. a quantum circuit must establish a one-to-one
While fault-tolerant implementations can help to mapping between the input and output states.
tolerate limited amounts of noise [114], [115], it is Ensuring a one-to-one mapping between input
important to note that the overhead associated with and output states may require circuit overhead
the implementation of fault-tolerant protocols can that includes the use of ancillae qubits and gar-
be significant [112], [113]. Therefore, an important bage outputs. For example, any constant input
concern for designing quantum circuits is to account required by the quantum circuit may be encoded
for the resource overhead associated with each gate. using ancillae qubits. Garbage output refers to
For example, fault-tolerant T gates are well known output that may not be a useful part of the result
to incur a significant increase in resources, thereby but is necessary for the quantum circuit to pre-
making T-count and T-depth important performance serve a one-to-one mapping. The inputs regen-
measures for fault-tolerant quantum circuit design erated at the circuit output are not considered
[114], [116]. garbage outputs [110]. An ideal quantum circuit
The number of qubits in a quantum circuit is a would be garbageless in nature, and efforts to min-
resource measure of interest because of the limited imize the circuit overhead from ancillae and gar-
number of qubits available on existing quantum bage outputs are active research areas.
When a quantum circuit has garbage outputs,
the garbage outputs can be removed by using Ben-
nett’s garbage removal scheme [119]. Figure 10
illustrates Bennett’s garbage removal scheme. Let
U represent an arbitrary quantum circuit that per-
forms f (x1, x 2,···, xn −1, xn ) and let U −1 represent its
logical inverse.
Bennett’s garbage removal scheme is three-step
process. After U is applied, all desired outputs are
copied to ancillae with CNOT gates. Then, U −1 is
applied to the qubits of the original circuit U. Thus,
Figure 8. The circuit diagram for the at the end of the computation, the garbage outputs
Fredkin gate and its matrix representation. have been restored to their initial values.

84 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Figure 11 illustrates the complete addition circuit for
the case of two 4-bit inputs a and b.
The carry bits ci are produced based on the inputs
ai−1, bi−1 and the carry bit ci−1 from the previous stage.
Each generated carry bit ci is stored at the quantum
register location |Ai 〉​​ that initially stored the value ai
for 0 ≤ i ≤ n − 1. After the generated carry bits are
used in further computation, each quantum register
location |Ai ​〉​ is restored to the value ai while each
quantum register location |Bi​〉​ stores the sum bit si
for 0 ≤ i ≤ n − 1. The restoration of |Ai 〉​ ​ to the value
ai eliminates all garbage outputs and transforming
|Bi 〉​
​ to the sum si cuts the ancillae cost to 1.

Figure 10. An example illustration for

the steps in Bennett’s garbage removal
Quantum circuit for multiplication
We present an example of a quantum integer
scheme described in the text. (a) After
multiplication circuit that is presented in [108].
Step 1. (b) After Step 2. (c) After Step 3.
The quantum circuit is based on a novel design of
a quantum conditional addition (Ctrl-Add) c ­ ircuit
Quantum arithmetic circuits with no input carry and the Toffoli gate array. The
The quantum logic gates presented in the pre- quantum multiplication circuit implements the
vious section can be combined to create quantum shift and add multiplication algorithm. As a result,
circuits that implement quantum algorithms. As a the circuit will require a total of n Ctrl-Add cir-
demonstration of these ideas, we present a series of cuits and Toffoli gate arrays. The Ctrl-Add circuits
quantum circuits designed for arithmetic operations and Toffoli gate arrays are placed such that the
such as addition, subtraction, and multiplication. shift operations are accomplished with no addi-
We draw these examples from recent results in the tional gates.
literature [106]–[108], [120]. Consider the multiplication of two n bit numbers a
and b stored in quantum registers |A​〉​and |B​〉​, respec-
Quantum circuit for addition tively. Furthermore, consider a quantum register |P​〉​
We show an example of a quantum ripple carry of size 2 · n + 1 initialized to z = 0. At the end of the
addition circuit with no input carry presented in computation, the quantum registers |A​〉​and |B​〉​keep
[107]. Consider the addition of two n-bit numbers the values a and b, respectively. At the end of the
a and b stored at quantum registers |A​〉​ and |B​〉​, computation, the quantum register locations |P0:2·n−1​〉​
respectively. Furthermore, let quantum register loca- will have the product of a and b. The quantum reg-
tion |An​〉​ be initialized with z = 0. At the end of the ister location |P2·n ​〉​ will be restored to the value 0.
computation, the quantum register |B​〉​ will have the
values sn −1:0 while the quantum register |A​〉​ keeps
the value a. The additional quantum register loca-
tion |An〉​ ​that initially stored the value z will have the
value sn at the end of the computation. Here, si is the
sum bit and is defined as

{​​c​ n​​​
​​ ​ i​  ​​  ⊕ ​ ​
s​ i​​  = ​ ​a bi​  ​​  ⊕ ​c​ i​ ​​​ ​if 0 ≤ i ≤ n − 1​​​​,​(7)
​if i = n​
where ci is the carry bit and is defined as

{​a​ i−1​​  ⋅ ​bi​  −1​​  ⊕ ​bi​  −1​​  ⋅ ​c​ i−1​​  ⊕ ​a​ i−1​​  ⋅ ​c​ i−1​​   if 1 ≤ i ≤ n .

0         ​if i = 0​ Figure 11. A four-qubit example of the
​​c​ i​​  = ​ ​ ​ ​​​
quantum ripple carry addition circuit with
(8) no input carry [107].

May/June 2019

Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
85
 Tutorial

­ igure 12 illustrates the quantum integer multiplica-

F
tion circuit for the case of two 4-bit inputs a and b.
Figure 13 illustrates the quantum Ctrl-Add circuit
used in the quantum multiplier for the case of two
4-bit inputs a and b. The operation of the quantum
Ctrl-Add circuit is conditioned on the value of the
qubit |ctrl​〉​. When |ctrl = 1​〉​, the circuit performs addi-
tion. The sum of a and b will appear on the quantum
register that originally holds the value b at the end
of the computation. The quantum register that origi- Figure 13. A four-qubit example of the
nally holds the value a will be restored to the value a. quantum Ctrl-Add circuit with no input
When |ctrl = 0​〉​, the quantum registers that initially carry [108].
hold the values a and b will be restored to the values
a and b at the end of the computation. used in scientific computing applications, such
as sin(x), ln(x), and ex. These functions have been
Application of quantum arithmetic circuits used in additional quantum algorithms besides
in Taylor series those for Hamiltonian evolution such as algo-
rithms for Pell’s equation and the principal ideal
In this section, we present an application of the
problem [123], [124]. The value of a given function
quantum multiplication and quantum addition circuits
f(x) near a point c is estimated by computing the
presented in the previous sections. For this example,
Taylor series equation shown below
we consider a quantum circuit implementation of the ∞ f​ ​​  i​(c )
Taylor series expansion. Taylor series are used in quan- ​ ​  ___
f (x ) ≈ ​  ∑ ​ ​ ​⋅ ​(x − c )​​  i​.​(9)
i=0 i!
tum algorithms for scientific computation such as to We will consider an example for a Taylor series
calculate the Hamiltonian evolution of a quantum limited to the first three terms, that is,
system or in the simulation of open quantum systems ​f ″ ​(c )
​f (x ) ≈ f (c ) + ​f ′ ​(c ) ⋅ (x − c ) + ​ ___
 ​  ⋅ ​(x − c )​​  2​  ,​ (10)
[121]–[123]. Furthermore, Taylor series polynomials 2
are used to approximate the functions frequently in which we compute this approximation for f (x)
centered at value c. Let c and x be n-bit values stored
in quantum registers |x​〉​ and |c​〉​, respectively. Fur-
thermore, let f (c), ​​f ′ ​​(c), and ​​f ″ ​​(c)/2 be represented
as n-bit numbers stored at quantum registers |f  (c)​〉​,
|​​f ′ ​​(c)​〉,​ and |​f ″ ​​(c)/2​〉​, respectively. Finally, consider
quantum registers |Y1​〉​, |Y2​〉​, |Y3​〉​, and |Y4​〉​that ­contain
ancillae set to 0. At the end of the computation,
quantum register |Y4​〉​ will have the first three terms
of the Taylor series expansion. The quantum registers
|c​〉​, |x​〉​, |f (c)​〉​, |​​f ′ ​​(c)​〉​, and |​f ″ ​​(c)/2​〉​ will be restored
to the values c, x, f (c), ​​f ′ ​​(c), and ​f ″ ​​(c)/2 at the end
of computation. The quantum registers |Y1​〉​, |Y2​〉​ and
|Y3​〉​that initially held ancillae will be restored to their
initial values.
The quantum Taylor series circuit is built from the
quantum addition circuit, the quantum subtraction
circuit, the quantum multiplication circuit and the
logical reverse of the quantum multiplication cir-
cuit. Figure 14 shows the graphical representation
of components used in the Taylor series circuit. We
Figure 12. A four-qubit example of will use a quantum subtraction circuit presented in
the quantum integer multiplication [120] that is based on the ­ripple carry adder shown
circuit [108]. in this article.

86
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
IEEE Design&Test
Figure 14. Quantum circuit diagrams for
the gates used in computing the Taylor
series circuit. (a) Quantum addition circuit.
(b) Quantum subtraction circuit. (c) Quantum
multiplication circuit. (d) Logical reverse of
quantum multiplication circuit.

Figure 15 illustrates an example of the quantum

subtraction circuit based on the design in [120].
_
The quantum circuit shown calculates b​
_
​​‾
+ a​​ where
‾
​ + a​ = b − a. Given two 2's complement inputs, the
​b​ ​ Figure 16. Generation of the quantum
quantum subtractor will not experience overflow. circuit for the calculation of the first three
Therefore, the circuitry used to calculate the sum terms of the Taylor series of f(x): Steps 1
bit sn is removed from the quantum adder because and 2. (a) After Step 1. (b) After Step 2.
_
the circuitry is not needed to calculate b​​‾
​ + a​. The
steps to design the quantum Taylor series circuit are
­uantum subtraction circuit in this step. The
q
explained below. Figure 16 illustrates Steps 1 and 2.
result of the quantum subtraction circuit x − c is
• Step 1: Calculate​f (x ) ≈ f (c ) + ​f ′ ​(c ) ⋅ (x − c ) + copied to ancillae using an array of n CNOT gates.
​f ″ ​(c )
___ • Step 2: Remove garbage output. At the end of
​   ​  ⋅ ​(x − c )​​  2​ ​
. We use the quantum multipli-
2 Step 1, three quantum registers (|Y1​〉​, |Y2​〉​, and |Y3​〉​)
cation circuit, quantum addition circuit, and
that initially held ancillae are tran sf ormed to
f (c) + ​​f ′ ​​(c) · (x − c), (x − c)2 and (x − c). Further-
more, at t he end of the computation, qu antum
register |x​〉​that initially held the value x has been
transformed to t he value x − c. These outputs
are garbage outputs. We use the logical reverse
of the quantum multiplication circuit, the quan-
tum adder, t he q u antum subtraction ci rc uit,
and an ar r ay o f C NOT gates to remove th ese
garbage outputs.

Quantum circuit synthesis

Quantum circuit synthesis is the generation of a
Figure 15. A four-qubit example of quantum circuit derived from a given input definition
the conversion of a quantum addition and any necessary constraints. While this process
circuit into a subtraction circuit via the may be either manual or automated, a typical generic
procedure in [120]. quantum circuit synthesis design flow is shown in

May/June 2019
87
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

Figure 17. The first step of presynthesis optimization quickly become too large for optimal ­synthesis tech-
includes the addition of ancilla lines and output order- niques. Heuristic algorithms, such as decision dia-
ing to satisfy reversibility. The synthesis step then grams (see [127] and [129]) and the search of circuit
transforms any irreversible operation into a reversible databases (see [125]), are widely used in these lim-
operation—this step may be performed either opti- its. Such heuristic synthesis methods are often sub-
mally or heuristically. There are many different opti- optimal and therefore further improvements may be
mal methods for quantum circuit synthesis including achieved by local optimization methods. In general,
those proposed in [126]–[128]. However, many cir- synthesis methods can optimize resource metrics by
cuits grow non­­linearly with input size and they can reducing gate count, reducing the number of ancillae
qubits, reducing the overall circuit depth, and improv-
ing locality. Postsynthesis optimization techniques
such as template matching [125], [127] can also be
employed to further achieve better resource usage.
Finally, during the technology mapping stage, the
quantum circuit is mapped (decomposed) into the
intrinsic gates available within the target technology.
The growing literature on the topic of quantum cir-
cuit synthesis for various levels of abstraction hints
at many ­important ­avenues of research [125]–[127],
[129]–[133].

We have summarized the basic features and require-

ments for quantum computing devices. This includes
the fundamental criteria that a quantum computing
device must implement as well as the principles of
operation for performing computation within the cir-
cuit model. We have reviewed the state of the art in
three specific technologies currently being developed
for quantum computing devices. Silicon spins, trapped
ions, and superconducting transmons represent three
of the leading approaches for quantum computing
but these devices still face fundamental research chal-
lenges. Therefore, methods to accurately characterize
and benchmark the behavior of quantum computing
devices play an important role in design and testing.
We have emphasized the necessity of statistical anal-
ysis to infer the operation of quantum devices. We
have also discussed the design of quantum circuits
for the case of arithmetic operations, which represent
an important use case for future quantum computing
devices. These circuits were designed to minimize the
occurrence of a specific instruction, the T gate, due to
the expected complexity of fault-tolerant implemen-
tation. These designs are expected to play a critical
role in future device operation as tradeoffs in gate
and device complexity become more sophisticated.
While we have not discussed logic test for quantum
Figure 17. Generic automatic quantum circuits, there are existing methods for fault detection
circuit synthesis design flow. Derived in classical reversible circuits that might be applica-
from representation in [125]. ble to quantum ­circuits [134]–[137]. As an example,

88 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
quantum circuits with noisy gates would require  References
novel fault-models and novel methods for test-vector [1] A. Montanaro, “Quantum algorithms: An overview,” npj
generation [138]. Therefore, we anticipate that the Quant. Inf., vol. 2, p. 15023, 2016.
fault-testing of quantum circuits will be an important [2] C. Degen, F. Reinhard, and P. Cappellaro, “Quantum
future research area. sensing,” Rev. Modern Phy., vol. 89, no. 3, p. 035002,
The design and testing of early quantum comput- 2017.
ing devices face many near-term challenges. We have [3] M. Krenn et al., “Quantum communication with
emphasized a small subset of the technologies currently photons,” in Optics in Our Time, M. D. Al-Amri et al.,
under investigation for developing quantum computing
Eds. Cham, Switzerland: Springer, 2010, pp. 455–482.
devices. However, there are many more approaches to
[4] M. A. Nielsen and I. L. Chuang, Quantum Computation
be considered, each with their own nuanced physics.
and Quantum Information. Cambridge, U.K.:
This suggests that variations in the physics of each
Cambridge University Press, 2010.
quantum computing technology may lead to different
[5] N. M. Linke et al., “Experimental comparison of two
implementations for design and testing. Comparison
quantum computing architectures,” in Proc. Nat. Acad.
across technologies will require standard calibration
Sci., vol. 114, no. 13, pp. 3305–3310, 2017.
techniques that have yet to be developed. In addition,
[6] K. A. Britt and T. S. Humble, “High-performance computing
methods for quantifying well-defined metrics will be
with quantum processing units,” ACM J. Emerg.
important for evaluating device performance. Current
Technol. Comput. Syst., vol. 13, no. 3, p. 39, 2017.
testing is focused on meeting the minimal criteria for
[7] K. A. Britt, F. A. Mohiyaddin, and T. S. Humble,
functionality in the regime of noisy, error-prone, and
“Quantum accelerators for high-performance
faulty devices. Finally, we note that the current state
computing systems,” in Proc. IEEE Int. Conf. Reboot.
of quantum computing remains focused on relatively
Comput., Washington, DC, Nov. 2017, pp. 1–7.
small scale devices. Future devices, or networks of
[8] T. Humble, “Consumer applications of quantum
devices, are likely to include quantum registers with
computing: A promising approach for secure
millions of elements and sequences with millions of
computation, trusted data storage, and efficient
highly parallelized instructions. Those devices and
applications,” IEEE Consum. Electron. Mag., vol. 7,
circuits will require more sophisticated methods for
design and testing.  no. 6, pp. 8–14, Nov. 2018.
[9] J. J. Sakurai et al., “Modern quantum mechanics, revised
edition,” Am. J. Phy., vol. 63, no. 1, pp. 93–95, 1995.
Acknowledgments
[10] A. S. Holevo, Probabilistic and Statistical Aspects of
We thank Dr. Y.-M. Shue for valuable insights into
Quantum Theory. Pisa, Italy: Springer Science and
superconducting fabrication tools and Dr. R. Pooser
for guidance on quantum device development. This Business Media, vol. 1, 2011.

material is based on the work supported by the U.S. [11] R. Horodecki et al., “Quantum entanglement,” Rev.

Department of Energy, Office of Science, Office of Mod. Phys., vol. 81, pp. 865–942, Jun. 2009.

Advanced Scientific Computing Research, and Oak [12] W. K. Wootters and W. H. Zurek, “A single quantum

Ridge National Laboratory Directed Research and cannot be cloned,” Nature, vol. 299, no. 5886,
Development. This article has been authored by pp. 802–803, 1982.
UT-Battelle, LLC, under Contract DE-AC0500OR22725 [13] V. Scarani et al., “Quantum cloning,” Rev. Mod. Phys.,
with the U.S. Department of Energy. The U.S. Govern- vol. 77, pp. 1225–1256, Nov. 2005. DOI: 10.1103/
ment and the publisher, by accepting the article for RevModPhys.77.1225
publication, acknowledge that the U.S. Government [14] M. H. S. Amin, “Consistency of the adiabatic theorem,”
retains a nonexclusive, paid-up, irrevocable, world- Phys. Rev. Lett., vol. 102, p. 220401, Jun. 2009.
wide license to publish or reproduce the published DOI: 10.1103/PhysRevLett.102.220401
form of this manuscript, or allow others to do so, [15] M. W. Johnson et al., “Quantum annealing with
for the U.S. Government purposes. The Department manufactured spins,” Nature, vol. 473, no. 7346,
of Energy will provide public access to these results p. 194, 2011.
of federally sponsored research in accordance with [16] T. Albash and D. A. Lidar, “Adiabatic quantum
the DOE Public Access Plan (http://energy.gov/­ computation,” Rev. Mod. Phys., vol. 90, p. 015002,
downloads/doe-public-accessplan). Jan. 2018. DOI: 10.1103/RevModPhys.90.015002

May/June 2019
89
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

[17] D. P. DiVincenzo et al., “The physical implementation [34] T. S. Humble et al., “A computational workflow for
of quantum computation,” arXiv preprint designing silicon donor qubits,” Nanotechnology,
quant-ph/0002077, 2000. vol. 27, no. 42, p. 424002, 2016.
[18] C. M. Dawson and M. A. Nielsen, “The Solovay-Kitaev [35] C. P. Slichter, Principles of Magnetic Resonance.
algorithm,” arXiv preprint quant-ph/0505030, 2005. Berlin, Germany: Springer Science and Business
[19] K. A. Britt and T. S. Humble, “Instruction set architectures Media, vol. 1, 2013.
for quantum processing units,” in Proc. Int. Conf. High [36] J. P. Dehollain et al., “Nanoscale broadband
Performance Comput., Frankfurt, Germany, Springer, transmission lines for spin qubit control,”
2017, pp. 98–105. Nanotechnology, vol. 24, no. 1, p. 015202, Jan. 2013.
[20] M. Schlosshauer, “Decoherence, the measurement [37] E. Kawakami et al., “Gate fidelity and coherence
problem, and interpretations of quantum mechanics,” of an electron spin in an Si/SiGe quantum dot with
Rev. Mod. Phys., vol. 76, pp. 1267–1305, Feb. 2005. micromagnet,” in Proc. Nat. Acad. Sci., vol. 113,
DOI: 10. 1103/RevModPhys.76.1267 no. 42, pp. 11738–11743, 2016.
[21] A. Streltsov, G. Adesso, and M. B. Plenio, “Colloquium,” [38] B. E. Kane, “A silicon-based nuclear spin quantum
Rev. Mod. Phys., vol. 89, p. 041003, Oct. 2017. computer,” Nature, vol. 393, no. 6681, pp. 133–137,
DOI: 10.1103/RevModPhys.89.041003 May 1998.
[22] E. T. Campbell, B. M. Terhal, and C. Vuillot, “Roads
[39] Y.-P. Shim and C. Tahan, “Barrier versus tilt exchange
towards faulttolerant universal quantum computation,”
gate operations in spin-based quantum computing,”
Nature, vol. 549, no. 7671, pp. 172–179, 2017.
Phys. Rev. B, vol. 97, no. 15, p. 155402, 2018.
[23] I. Buluta, S. Ashhab, and F. Nori, “Natural and artificial
[40] F. A. Zwanenburg et al., “Silicon quantum electronics,”
atoms for quantum computation,” Rep. Prog. Phys.,
Rev. Modern Phy., vol. 85, no. 3, p. 961, 2013.
vol. 74, no. 10, p. 104401, 2011.
[41] W. M. Witzel et al., “Electron spin decoherence in
[24] M. Veldhorst et al., “A two-qubit logic gate in silicon,”
isotope-enriched silicon,” Phys. Rev. Lett., vol. 105,
Nature, vol. 526, pp. 410–414, Oct. 2015.
no. 18, p. 187602, Oct. 2010.
[25] D. M. Zajac et al., “Resonantly driven cnot gate for
[42] B. Koiller, X. Hu, and S. Das Sarma, “Exchange in
electron spins,” Science, vol. 359, no. 6374,
silicon-based quantum computer architecture,” Phys.
pp. 439–442, 2018.
Rev. Lett., vol. 88, p. 027903, Dec. 2001.
[26] A. Laucht et al., “A dressed spin qubit in silicon,”
[43] Y. Song and S. Das Sarma, “Statistical exchange-
Nature Nanotechnol., vol. 12, no. 1, pp. 61–66, 2017.
coupling errors and the practicality of scalable silicon
[27] J. T. Muhonen et al., “Storing quantum information
donor qubits,” Appl. Phys. Lett., vol. 109, no. 25,
for 30 seconds in a nanoelectronic device,” Nature
p. 253113, 2016.
Nanotechnol., vol. 9, no. 12, pp. 986–991, Oct. 2014.
[44] M. Fuechsle et al., “A single-atom transistor,” Nature
[28] G. Feher, “Electron spin resonance experiments on
Nanotech., vol. 7, no. 4, pp. 242–246, Feb. 2012.
donors in silicon. I. Electronic structure of donors by
[45] G. Tosi et al., “Silicon quantum processor with robust
the electron nuclear double resonance technique,”
long-distance qubit couplings,” Nature Commun.,
Phys. Rev., vol. 114, no. 5, pp. 1219–1244, Jun. 1959.
[29] J. J. Pla et al., “A single-atom electron spin qubit vol. 8, no. 1, p. 450, 2017.

in silicon,” Nature, vol. 489, no. 7417, pp. 541–545, [46] C. D. Hill et al., “A surface code quantum computer

Sep. 2012. in silicon,” Sci. Adv., vol. 1, no. 9, p. e1500707, 2015.

[30] J. J. Pla et al., “Highfidelity readout and control of [47] C. H. Yang et al., “Spin-valley lifetimes in a silicon
a nuclear spin qubit in silicon,” Nature, vol. 496, quantum dot with tunable valley splitting,” Nature
no. 7445, pp. 334–338, Apr. 2013. Commun., vol. 4, p. 2069, Jun. 2013.
[31] A. Laucht et al., “Electrically controlling single-spin [48] T. Thorbeck and N. M. Zimmerman, “Formation of
qubits in a continuous microwave field,” Sci. Adv., strain-induced quantum dots in gated semiconductor
vol. 1, no. 3, p. e1500022, Apr. 2015. nanostructures,” AIP Adv., vol. 5, no. 8, p. 087107, 2015.
[32] M. Veldhorst et al., “Spin-orbit coupling and operation [49] X. Hu, Y.-X. Liu, and F. Nori, “Strong coupling of a
of multivalley spin qubits,” Phys. Rev. B, vol. 92, spin qubit to a superconducting stripline cavity,” Phy.
no. 20, p. 201401, 2015. Rev. B, vol. 86, no. 3, p. 035314, 2012.
[33] A. Morello et al., “Singleshot readout of an electron [50] N. Samkharadze et al., “Strong spinphoton coupling in
spin in silicon,” Nature, vol. 467, no. 7316, silicon,” Science, vol. 359, no. 6380, pp. 1123–1127,
pp. 687–691, Oct. 2010. 2018.

90 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
[51] X. Mi et al., “Strong coupling of a single electron in [67] J. Koch et al., “Chargeinsensitive qubit design derived
silicon to a microwave photon,” Science, vol. 355, from the cooper pair box,” Phys. Rev. A, vol. 76, no. 4,
no. 6321, pp. 156–158, 2017. p. 042319, 2007.
[52] R. Blatt and D. Wineland, “Entangled states of [68] S. M. Girvin, “Circuit qed: Superconducting qubits
trapped atomic ions,” Nature, vol. 453, no. 7198, coupled to microwave photons,” in Proc. 2011 Les
pp. 1008–1015, 2008. Houches Summer School, Les Houches, France.
[53] E. Mount et al., “Single qubit manipulation in a [69] G. Wendin, “Quantum information processing with
microfabricated surface electrode ion trap,” New superconducting circuits: A review,” Rep. Progress
J. Phy., vol. 15, no. 9, p. 093018, 2013. Phy., vol. 80, no. 10, p. 106001, 2017.
[54] R. Blatt et al., “Ion trap quantum computing with ca+ [70] A. Blais et al., “Cavity quantum electrodynamics for
ions,” Quant. Inf. Process., vol. 3, no. 1–5, pp. 61–73, superconducting electrical circuits: An architecture
2004. for quantum computation,” Phy. Rev. A, vol. 69, no. 6,
[55] W. Paul, “Electromagnetic traps for charged and p. 062320, 2004.
neutral particles,” Rev. Modern Phys., vol. 62, [71] X. Gu et al., “Microwave photonics with superconducting
no. 3, p. 531, 1990. quantum circuits,” Phy. Rep., vol. 718, pp. 1–102, 2017.
[56] D. Leibfried et al., “Quantum dynamics of single [72] J. Majer et al., “Coupling superconducting qubits via
trapped ions,” Rev. Modern Phy., vol. 75, no. 1, p. 281, a cavity bus,” Nature, vol. 449, no. 7161, pp. 443–447,
2003. 2007.
[57] B. B. Blinov et al., “Quantum computing with trapped [73] J. M. Martinis and A. Megrant, “Ucsb final report for the
ion hyperfine qubits,” Quant. Inf. Process., vol. 3, csq program: Review of decoherence and materials
no. 1–5, pp. 45–59, 2004. physics for superconducting qubits,” arXiv preprint
[58] P. T. Fisk et al., “Accurate measurement of the 12.6 arXiv:1410.5793, 2014.
GHz clock transition in trapped/sup 171/yb/sup+/ [74] O. Dial et al., “Bulk and surface loss in
ions,” IEEE Trans. Ultrason. Ferroelectr. Freq. Control, superconducting transmon qubits,” Supercond.
vol. 44, no. 2, pp. 344–354, 1997. Sci. Technol., vol. 29, no. 4, p. 044001, 2016.
[59] A. D. Ludlow et al., “Optical atomic clocks,” Rev. [75] V. Ambegaokar and A. Baratoff, “Tunneling between
Modern Phys., vol. 87, no. 2, p. 637, 2015. superconductors,” Phy. Rev. Lett., vol. 10, no. 11,
[60] S. Debnath, “A Programmable Five Qubit Quantum p. 486, 1963.
Computer Using Trapped Atomic Ions,” PhD [76] S. J. Bosman et al., “Multi-mode ultra-strong coupling
dissertation, Univ. of Maryland, College Park, 2016. in circuit quantum electrodynamics,” npj Quant. Inf.,
[61] C. Monroe and J. Kim, “Scaling the ion trap quantum vol. 3, no. 1, p. 46, 2017.
processor,” Science, vol. 339, no. 6124, [77] S. T. Flammia and Y.-K. Liu, “Direct fidelity estimation
pp. 1164–1169, 2013. from few Pauli measurements,” Phy. Rev. Lett.,
[62] H. Haffner, C. F. Roos, and R. Blatt, “Quantum vol. 106, no. 23, Jun. 2011.
computing with trapped ions,” Phys. Rep., vol. 469, [78] M. P. da Silva, O. Landon-Cardinal, and D. Poulin,
no. 4, pp. 155–203, 2008. “Practical characterization of quantum devices without
[63] R. Blume-Kohout et al., “Demonstration of qubit tomography,” Phy. Rev. Lett., vol. 107, no. 21, Nov. 2011.
operations below a rigorous fault tolerance threshold [79] E. Magesan, J. M. Gambetta, and J. Emerson,
with gate set tomography,” Nature Commun., vol. 8, “Characterizing quantum gates via randomized
2017. benchmarking,” Phy. Rev. A, vol. 85, no. 4, p. 042311,
[64] K. R. Brown, J. Kim, and C. Monroe, “Co-designing a Apr. 2012.
scalable quantum computer with trapped atomic ions,” [80] E. Magesan et al., “Efficient measurement of quantum
npj Quant. Inf., vol. 2, p. 16034, 2016. gate error by interleaved randomized benchmarking,”
[65] D. J. Wineland et al., “Experimental issues in coherent Phy. Rev. Lett., vol. 109, no. 8, Aug. 2012.
quantum-state manipulation of trapped atomic ions,” [80] S. Kimmel et al., “Robust extraction of tomographic
J. Res. Nat. Inst. Stand. Technol., vol. 103, no. 3, information via randomized benchmarking,” Phy.
p. 259, 1998. Rev. X, vol. 4, no. 1, p. 011050, Mar. 2014.
[66] D. Kielpinski, C. Monroe, and D. J. Wineland, [82] J. P. Gaebler et al., “Randomized benchmarking of
“Architecture for a large-scale ion-trap quantum multiqubit gates,” Phys. Rev. Lett., vol. 108, no. 26,
computer,” Nature, vol. 417, no. 6890, p. 709, 2002. p. 260503, Jun. 2012.

May/June 2019
91
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

[83] T. Chasseur and F. K. Wilhelm, “Complete randomized [99] C. E. Granade et al., “Robust online Hamiltonian
benchmarking protocol accounting for leakage errors,” learning,” New J. Phy., vol. 14, no. 10, p. 103013,
Phy. Rev. A, vol. 92, no. 4, p. 042333, Oct. 2015. Oct. 2012.
[84] J. M. Gambetta et al., “Characterization of [100] G. Carleo and M. Troyer, “Solving the quantum
addressability by simultaneous randomized many-body problem with artificial neural networks,”
benchmarking,” Phy. Rev. Lett., vol. 109, no. 24, Science, vol. 355, no. 6325, pp. 602–606, Feb. 2017.
p. 240504, Dec. 2012. [100] D.-L. Deng, X. Li, and S. D. Sarma, “Quantum
[85] H. Ball et al., “The effect of noise correlations on ran- entanglement in neural network states,” Phy. Rev. X,
domized benchmarking,” Phy. Rev. A, vol. 93, no. 2, vol. 7, no. 2, p. 021021, May 2017.
p. 022303, Feb. 2016. [102] N. S. Yanofsky and M. A. Mannucci, Quantum
[86] T. Proctor et al., “What randomized benchmarking Computing for Computer Scientists, 1st ed. New York,
actually measures,” Phy. Rev. Lett., vol. 119, no. 13, NY: Cambridge University Press, 2008.
p. 130502, Sep. 2017. [103] F. Johansson, “Efficient implementation of elementary
[87] S. T. Merkel et al., “Self-consistent quantum process functions in the medium-precision range,” in Proc.
tomography,” Phys. Rev. A, vol. 87, no. 6, p. 062119, 2015 IEEE 22nd Symp. Comput. Arithmet., Lyon,
Jun. 2013. France, pp. 83–89.
[88] R. Blume-Kohout et al., “Demonstration of qubit [104] M. Moise, “A Fixed Point Arithmetic Library for
operations below a rigorous fault tolerance threshold Spinnaker,” master’s thesis, The Univ. of Manchester,
with gate set tomography,” Nature Commun., vol. 8, Manchester, 2012.
Article No. 14485, Feb. 2017. [105] M. K. Bhaskar et al., “Quantum algorithms and
[89] J. P. Dehollain et al., “Optimization of a solid-state circuits for scientific computing,” Quant. Inf. Comput.,
electron spin qubit using gate set tomography,” New vol. 16, no. 3–4, pp. 197–236, Mar. 2016.
J. Phy., vol. 18, no. 10, p. 103018, Oct. 2016. [106] C.-C. Lin, A. Chakrabarti, and N. K. Jha, “Qlib:
[90] C. Macchiavello and M. Rossi, “Quantum channel Quantum module library,” J. Emerg. Technol.
detection,” Phy. Rev. A, vol. 88, no. 4, p. 042335, Comput. Syst., vol. 11, no. 1, pp. 7:1–7:20, Oct. 2014.
Oct. 2013. DOI: 10.1145/2629430
[91] C. Granade, C. Ferrie, and D. G. Cory, “Accelerated [107] H. Thapliyal and N. Ranganathan, “Design of efficient
randomized benchmarking,” New J. Phy., vol. 17, reversible logic-based binary and bcd adder circuits,”
no. 1, p. 013042, Jan. 2015. J. Emerg. Technol. Comput. Syst., vol. 9, no. 3,
[92] F. Huszar and N. M. T. Houlsby, “Adaptive Bayesian pp. 17:1–17:31, Oct. 2013. DOI: 10.1145/2491682
quantum tomography,” Phys. Rev. A, vol. 85, no. 5, [108] E. Munoz-Coreas and H. Thapliyal, “T-count
Article No. 052120, May 2012. optimized design of quantum integer multiplication,”
[93] D. H. Mahler et al., “Adaptive quantum state ArXiv e-prints, Jun. 2017.
tomography improves accuracy quadratically,” [109] T. Toffoli, “Reversible computing,” in Proc. 7th
Phy. Rev. Lett., vol. 111, no. 18, p. 183601, Colloquium Autom. Lang. Program, London,
Oct. 2013. U.K., 1980, pp. 632–644. [Online]. Available: http://
[94] J. Chen et al., “Uniqueness of quantum states dl.acm.org/citation.cfm?id=646234.682540
compatible with given measurement results,” [110] E. Fredkin and T. Toffoli, “Conservative logic,” Int.
Phy. Rev. A, vol. 88, no. 1, p. 012109, Jul. 2013. J. Theor. Phy., vol. 21, no. 3, pp. 219–253, Apr. 1982.
[95] G. Toth et al., “Permutationally invariant quantum DOI: 10.1007/BF01857727
tomography,” Phy. Rev. Lett., vol. 105, no. 25, [111] M. Amy et al., “A meet-in-the middle algorithm for fast
p. 250403, Dec. 2010. synthesis of depth-optimal quantum circuits,” IEEE
[96] D. Gross et al., “Quantum state tomography via Trans. Comput.-Aided Design Integr. Circ. Syst.,
compressed sensing,” Phy. Rev. Lett., vol. 105, no. 15, vol. 32, no. 6, pp. 818–830, Jun. 2013.
p. 150401, Oct. 2010. [111] S. J. Devitt et al., “Requirements for fault-tolerant
[97] S. Boixo et al., “Characterizing quantum supremacy in factoring on an atom-optics quantum computer,”
near-term devices,” ArXiv e-prints, Jul. 2016. Nature Commun., vol. 4, no. 4, p. 2524, Oct. 2013.
[98] O. Landon-Cardinal and D. Poulin, “Practical learning [113] M. B. Hastings and J. Haah, “Distillation with
method for multi-scale entangled states,” New J. Phy., sublogarithmic overhead,” ArXiv e-prints,
vol. 14, no. 8, p. 085004, Aug. 2012. Sep. 2017.

92 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
[114] M. Amy, D. Maslov, and M. Mosca, “Polynomial-time eprints, p. arXiv:1803.01022, Mar. 2018. [Online].
t-depth optimization of clifford+t circuits via matroid Available: https://arxiv.org/abs/1803.01022
partitioning,” IEEE Trans. Comput.-Aided Design [128] T. Haener et al., “Quantum circuits for floating-point
Integr. Circ. Syst., vol. 33, no. 10, pp. 1476–1489, arithmetic,” in Reversible Computation, J. Kari and I.
Oct. 2014. Ulidowski, eds. Cham, Switzerland: Springer
[115] E. Knill, “Fault-tolerant postselected quantum International Publishing, 2018, pp. 162–174.
computation: Schemes,” eprint arXiv:quant- [129] M. Khan, H. Thapliyal, and E. Munoz-Coreas,
ph/0402171, Feb. 2004. “Automatic synthesis of quaternary quantum
[116] D. Gosset et al., “An algorithm for the t-count,” Quant. circuits,” J. Supercomput., vol. 73, no. 5,
Inf. Comput., vol. 14, no. 15–16, pp. 1261–1276, pp. 1733–1759, 2017.
2014. [Online]. Available: http://www.rintonpress.com/ [130] Y. Nam et al., “Automated optimization of large
xxqic14/qic-14-1516/1261-1276.pdf quantum circuits with continuous parameters,” npj
[117] D. Lu et al., “Enhancing quantum control by Quant. Inf., vol. 4, no. 1, p. 23, 2018.
bootstrapping a quantum processor of 12 qubits,” npj [131] A. Chia-Chun Lin, N. K. Chakrabarti, and N. K. Jha,
Quant. Inf., vol. 3, no. 1, p. 45, Oct. 2017. “Ftqls: Faulttolerant quantum logic synthesis,” in Proc.
[118] C. Song et al., “10-qubit entanglement and IEEE Trans. Very Large Scale Integr. Syst., vol. 22,
parallel logic operations with a superconducting no. 6, pp. 1350–1363, 2014.
circuit,” Phy. Rev. Lett., vol. 119, no. 18, p. 180511, [132] M. Soeken et al., “Logic synthesis for quantum
Nov. 2017. computing,” arXiv eprints, p. arXiv:1706.02721,
[119] C. H. Bennett, “Logical reversibility of computation,” Jun. 2017. [Online]. Available: https://arxiv.org/
IBM J. Res. Dev., vol. 17, no. 6, pp. 525–532, abs/1706.02721
Nov. 1973. DOI: 10.1147/rd.176.0525 [133] A. Zulehner and R. Wille, “One-pass design of
[120] H. Thapliyal, “Mapping of subtractor and reversible circuits: Combining embedding and
addersubtractor circuits on reversible quantum synthesis for reversible logic,” IEEE Trans. Comput.-
gates,” in Transactions on Computational Science Aided Design Integr. Circ. Syst., vol. 37, no. 5,
XXVII, M. L. Gavrilova and C. J. K. Tan, Eds. Berlin, pp. 996–1008, 2018.
Germany: Springer-Verlag, 2016, pp. 10–34. [134] K. N. Patel, J. P. Hayes, and I. L. Markov, “Fault testing
[121] S.-J. Wei and G.-L. Long, “Efficient simulation of open for reversible circuits,” IEEE Trans. Comput.-Aided
quantum system in duality quantum computing,” Design Integr. Circ. Syst., vol. 23, no. 8,
in Proc. SPIE, Beijing, China, 2016, vol. 10029, pp. 1220–1230, 2004.

p. 1002915. [135] Y.-H. Chou, I.-M. Tsai, and S.-Y. Kuo, “Quantum

[122] L. Novo and D. Berry, “Improved Hamiltonian Boolean circuits are 1-testable,” IEEE Trans.

simulation via a truncated Taylor series and Nanotechnol., vol. 7, no. 4, pp. 484–492, 2008.

corrections,” Quant. Inf. Comput., vol. 17, no. 7–8, [136] H. Thapliyal and N. Ranganathan, “Reversible logic

pp. 623–635, 2017. based concurrent error detection methodology

[123] D. W. Berry et al., “Simulating Hamiltonian dynamics for emerging nanocircuits,” in Proc. 10th IEEE Int.

with a truncated Taylor series,” Phys. Rev. Lett., Conf. Nanotechnol., Seoul, South Korea, Aug. 2010,

vol. 114, p. 090502, Mar. 2015. DOI: 10.1103/ pp. 217–222.

PhysRevLett.114.090502 [137] A. Nagamani et al., “A genetic algorithm-based

[124] S. Hallgren, “Polynomial-time quantum algorithms heuristic method for test set generation in reversible
circuits,” IEEE Trans. Comput.-Aided Design Integr.
for Pell’s equation and the principal ideal problem,”
Circ. Syst., vol. 37, no. 2, pp. 324–336, 2018.
JACM, vol. 54, no. 1, pp. 1–19, 2007.
[138] D. Bera, “Detection and diagnosis of single faults
[125] M. Saeedi and I. Markov, “Synthesis and optimization
in quantum circuits,” IEEE Trans. Comput.-Aided
of reversible circuits: A survey,” ACM Comput.
Design Integr. Circ. Syst., vol. 37, no. 3,
Surveys, vol. 45, no. 2, pp. 1–34, 2013.
pp. 587–600, 2018.
[126] D. Maslov, “Optimal and asymptotically optimal nct
reversible circuits by the gate types,” Quant. Inf. Travis S. Humble is a Distinguished Scientist
Comput., vol. 16, no. 13–14, pp. 1096–1112, 2016. with the Oak Ridge National Laboratory, Oak Ridge,
[127] M. Soeken, T. Haner, and M. Roetteler, “Programming TN, and holds a Joint Faculty Appointment at the
quantum computers using design automation,” arXiv Bredesen Center for Interdisciplinary Research

May/June 2019
93
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.
 Tutorial

and Graduate Education, University of Tennessee, Fahd A. Mohiyaddin is a Research and Deve­
Knoxville. His research interests include the design lopment Engineer with the Interuniversity Micro-
and development of quantum computing systems for Electronics Center (IMEC), Leuven, Belgium. His
scientific discovery and innovation. Humble has a PhD research interests include silicon quantum computation,
from the University of Oregon, Eugene, OR (2005). nanoelectronic design, and semiconduc­ tor modeling
at the atomic scale. Mohiyaddin has a PhD in electrical
Himanshu Thapliyal is an Assistant Professor engineering from the University of New South Wales
and Endowed Robley D. Evans Faculty Fellow (UNSW Sydney), Sydney, NSW, Australia.
with the Department of Electrical and Computer
Engineering, University of Kentucky, Lexington, KY.
Ryan S. Bennink is a Senior Scientist with the
His research interests include quantum circuits,
Oak Ridge National Laboratory, Oak Ridge, TN.
emerging technologies, and hardware security.
His research interests include the verification and
Thapliyal has a PhD from the University of South
validation of quantum information systems. Bennink
Florida, Tampa, FL (2011).
has a PhD from the University of Rochester, New
York, NY (2004).
Edgard Muñoz-Coreas is currently pursuing
the PhD degree in electrical engineering from the
University of Kentucky, Lexington, KY. His research  Direct questions and comments about this article
interests include the circuit design of quantum to Travis Humble, Oak Ridge National Laboratory,
computing and emerging technologies Oak Ridge, TN 37830 USA; [email protected].

94 IEEE Design&Test
Authorized licensed use limited to: Helsingin Yliopisto. Downloaded on December 09,2024 at 09:03:48 UTC from IEEE Xplore. Restrictions apply.