Received 16 September 2020; revised 20 October 2020; accepted 22 October 2020.

Date of current version 7 January 2021.

Digital Object Identifier 10.1109/JMW.2020.3034071

Microwaves in Quantum Computing

JOSEPH C. BARDIN 1,2 (Senior Member, IEEE), DANIEL H. SLICHTER3 (Senior Member, IEEE),
AND DAVID J. REILLY4,5 (Member, IEEE)
(Invited Paper)
1
Department of Electrical and Computer Engineering, University of Massachusetts Amherst, Amherst, MA 01003 USA
2
Google LLC, Goleta, CA 93117 USA
3
Time and Frequency Division, National Institute of Standards and Technology, Boulder, CO 80305 USA
4
Microsoft Inc., Microsoft Quantum Sydney, The University of Sydney, Sydney, NSW 2050, Australia
5
ARC Centre of Excellence for Engineered Quantum Systems (EQuS), School of Physics, The University of Sydney, Sydney, NSW 2050, Australia
CORRESPONDING AUTHOR: Joseph C. Bardin (e-mail: [email protected]).
This work was supported in part by the National Science Foundation under Grant #1809114, and in part by the NIST Quantum Information Program.

ABSTRACT Quantum information processing systems rely on a broad range of microwave technologies and
have spurred development of microwave devices and methods in new operating regimes. Here we review
the use of microwave signals and systems in quantum computing, with specific reference to three leading
quantum computing platforms: trapped atomic ion qubits, spin qubits in semiconductors, and supercon-
ducting qubits. We highlight some key results and progress in quantum computing achieved through the
use of microwave systems, and discuss how quantum computing applications have pushed the frontiers
of microwave technology in some areas. We also describe open microwave engineering challenges for the
construction of large-scale, fault-tolerant quantum computers.

INDEX TERMS Semiconductor spin qubit, superconducting qubit, trapped ion qubit, quantum computing,
qubit control, qubit readout, quantum-classical interface.

I. INTRODUCTION quantum objects, such as the quantum two-level systems

Quantum computing and modern microwave engineering (called qubits) that form the building blocks of quantum
share a common ancestor in the pioneering work that led computers. Today, microwave technology is ubiquitous across
to the development of radar and related technologies in the many different quantum platforms, enabling the precise con-
1940’s [1]. Indeed, many of the fundamental mechanisms trol and readout of quantum states.
underpinning the generation, transmission, absorption, and Here, we review the use of microwaves in quantum comput-
detection of microwave energy were understood at that time ing. For the sake of brevity, rather than surveying the complete
to be governed by quantum mechanics, which describes the quantum computing landscape, we focus on three leading
light-matter interaction between microwaves and atoms or qubit technologies: trapped ion qubits, semiconductor spin
molecules [2], as well as their constituent charge and spin qubits, and superconducting circuit qubits. We chose these
states [3]. The leveraging of war-time radar technology and technologies not only because they span the representative
methods in the discovery of nuclear magnetic resonance in qubit technologies and are currently considered among the
solids [4], [5] provides an ideal example of the long-standing most promising of all qubit types, but also because they are
synergy between microwave engineering and quantum heavily reliant on microwave technologies (see Fig. 1). The
systems [1]. paper is organized as follows:
Fundamental to this light-matter interaction is the relation II) An overview of qubits and quantum processors
E = ω, which connects the angular frequency ω of mi- is presented; the analogy between a qubit and a
crowave photons to their energy E ( is the reduced Planck’s microwave resonator is explained, with implica-
constant). The quantization of microwave energy describes tions discussed; and physical qubit realizations are
how microwave photons can interact resonantly with other described.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 1, NO. 1, JANUARY 2021 403
BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

FIGURE 2. Qubit basics. (a) A qubit is composed of two quantum states

|0 and |1 with an energy difference E. (b) When excited resonantly at
ω01 = E/, the qubit state can be driven between |0 and |1, including
into linear combinations of |0 and |1. (c) Under cw resonant driving, the
qubit state undergoes so-called Rabi oscillations, where the probabilities
|α0 |2 and |α1 |2 evolve sinusoidally in time. (d) Coupling to a source of
dissipation at frequency ω01 (noise temperature Tb , impedance Zb ) with a
coupling quality factor Q causes qubit state transitions on the timescale
T1 . If kB Tb  ω01 , these transitions will be dominated by the |1 → |0
process. (e) If the qubit frequency can be shifted by an environmental
parameter λ (e.g. magnetic field), fluctuations in λ cause fluctuations in
FIGURE 1. Microwave electronics for operating a 50-qubit Google
ω01 , dephasing the qubit state on the timescale Tφ .
quantum processor. The system generates and receives signals in the
4–8 GHz band, and was used to execute a demonstration quantum
computing algorithm designed to be too complex for simulation by even
the largest classical supercomputers [7]. Four racks of microwave A. QUBIT BASICS
electronics are required to control and measure the quantum processor.
Abbreviations: BB–baseband, AWG–arbitrary waveform generator,
A qubit is a quantum mechanical system with two en-
SSB–single sideband. Photo credit: R. Ceselin. ergy eigenstates,1 which we label as |0 and |1, with
corresponding energy eigenvalues E0 and E1 , taking E0 < E1
(see Fig. 2(a)). The “textbook” qubit is a spin-1/2 particle
III) The coupling of a microwave source to various qubit (such as an electron) in a magnetic field, but as we shall see,
technologies is described and typical signal and noise there are many other possible ways to realize a qubit.
levels are compared and contrasted. The |0 and |1 states are used as computational basis states,
IV) The quantum gate abstraction is explained, mi- analogous to the use of “0” and “1” in classical computing.
crowave techniques for implementing single and two Just as the electromagnetic field can be decomposed into a
qubit quantum gates are described, and typical hard- linear combination of orthogonal modes (such as plane waves,
ware configurations are presented. spherical harmonics, or guided modes), the instantaneous
V) Microwave techniques for measuring the state of a state of a qubit |ψ can be written as a linear combination
qubit are described. of the two energy eigenstates, with complex amplitudes α0
VI) Additional microwave techniques required for the and α1 :
operation of trapped-ion qubits are described. |ψ = α0 |0 + α1 |1. (1)
VII) Microwave innovations in quantum-limited ampli-
fiers and non-reciprocal devices, driven by quantum The customary normalization of |ψ is that |α0 |2 + |α1 |2 = 1.
computing research, are presented. To meet this criterion, we introduce the suggestive parame-
VIII) Outstanding challenges related to microwave engi- terization α0 = cos(θ /2) and α1 = e jφ sin(θ /2), where φ and
neering that must be overcome to realize the full θ are real and j is the imaginary unit, for reasons that will
potential of quantum computing are described. become clear later. For a single qubit, one is free to choose
a global phase convention such that α0 is purely real, as seen
here, but in general both α0 and α1 are complex.
The quantum-mechanical nature of the state |ψ means that
II. QUBITS AND QUANTUM COMPUTING
the qubit can be in both states |0 and |1 simultaneously—a
The fundamental information carriers in a quantum computer
phenomenon known as superposition—in contrast to the be-
are quantum bits, or qubits, in analogy to the logical bits used
havior of classical bits, which can only be in one state at a
in a classical (non-quantum) computer. Here we describe the
time. When the state of the qubit is measured, however, the
basic characteristics and behavior of qubits, before discussing
qubit state is said to “collapse” to just one of its eigenstates.
particular physical implementations of qubits and their con-
The collapse is probabilistic, with the state being measured
nections with microwave technology. The interested reader
can find a much more detailed exposition of this topic, written
for microwave engineers, in Ref. [6]. 1 The notation “| ” denotes a quantum state.

404 VOLUME 1, NO. 1, JANUARY 2021

to be |0 with probability P|0 = |α0 |2 , or |1 with probability amplitudes {α00...00 , α00...01 , · · · , α11...11 } to increase the like-
P|1 = |α1 |2 ; for this reason, α0 and α1 are called probabil- lihood of measuring certain output bitstrings (ones which
ity amplitudes. Because of the collapse process, the post- yield the desired outcome of the computation). The action of
measurement qubit state no longer contains information about an N-qubit quantum algorithm is analogous to the scattering
α0 and α1 ; it has collapsed to either |0 or |1. The values matrix for a 2N -port passive lossless microwave device, where
of P|0 and P|1 can only be determined by many rounds of each quantum basis state (or bitstring) is mapped to a port.
preparing the same initial qubit state |ψ = α0 |0 + α1 |1 and An algorithm with speedup is equivalent to an S-matrix which
measuring it, to build statistics on the measurement collapse transforms uniform input excitation at all ports into nonzero
probabilities |α0 |2 and |α1 |2 . Although beyond the scope of output at only a small subset of ports, through constructive and
this article, the complex phase φ = arg(α1 ) − arg(α0 ) can be destructive interference, similarly to how a phased array can
determined using a procedure known as state tomography [8]. be used for beamforming, for example. Since the size of such
Quantum computers require more than just one qubit to a theoretical microwave device scales as 2N , it would become
perform useful computations. For N qubits, there are 2N basis impossible to realize for sufficiently large N. This highlights
states of the system, from |00 . . . 00 to |11 . . . 11. While N the potential for sufficiently large quantum computers to per-
classical bits can only be in one of the 2N basis states at form calculations which are classically intractable [7], [9].
a given time, the phenomenon of superposition means that Quantum algorithms create the desired interference be-
the quantum state |ψ of N qubits can be a linear combi- tween probability amplitudes by controlling the states of in-
nation of any—or even all—of the 2N basis states at the dividual qubits, and generating entanglement between qubits,
same time, with corresponding complex probability ampli- in the course of the algorithm. Experimentally, these tasks are
tudes {α00...00 , α00...01 , · · · , α11...11 }. Since these amplitudes usually carried out by microwave signals, or rely critically
have a physical interpretation in terms of probabilities of mea- on microwave techniques. In some qubit technologies, the
surement outcomes, they have the normalization condition measurement process is also carried out by microwave signals.
2N −1 These topics will be detailed in Sections IV and V.
k=0 |αk | = 1, where k indexes the 2 different bitstrings
2 N

corresponding to the basis states. It is useful to characterize the performance of qubit state
While it is possible to access a range of superposition preparation, control, and measurement operations using a
states by putting each qubit in its own independent super- metric known as the fidelity. For each of these three tasks,
position state, such an approach can only be used to reach there exist distinct methods to quantify the fidelity; a detailed
a small fraction of the basis states of the qubit state vector discussion is beyond the scope of this paper [8]. In general, the
|ψ, since most of the possible linear combinations exhibit fidelity can be thought of as characterizing how close the labo-
correlations between the qubits. For example, consider the ratory implementation of an operation is to its ideal theoretical
state |ψ = √1 |00 . . . 00 + √1 |11 . . . 11. It is not possible representation, with a fidelity of 1 representing a perfect im-
2 2 plementation and a fidelity of 0 indicating a complete failure.
to write this state as a product of separate, individual states of
The presence of noise, drifts, dissipation, or miscalibration
the constituent qubits; the state of each qubit is inextricably
can give rise to errors in the implementation and thus cause
correlated, or entangled, with all the others. The state |ψ
the fidelity to be less than 1. The error rate characterizes the
above is a superposition state, where the process of mea-
amount by which the fidelity of an operation is less than 1.
surement will cause a collapse to just one basis state. Let us
Techniques for efficiently and accurately estimating fidelities
consider what happens if we measure just one of the N qubits
and error rates, especially in larger quantum processors, are
in this entangled state. If the measured qubit collapses to |0,
an active area of research.
then this causes all the other qubits to collapse to |0 as well,
even though they were not measured directly, and the state
becomes |ψ = |00 . . . 00. Likewise, if the measured qubit B. QUBITS AS RESONATORS
happened to collapse to |1, then all the other qubits would to One way to think of a qubit is as a high-quality-factor electro-
collapse to |1, even though they were not measured directly, magnetic resonator, with a resonant frequency ω01 = (E1 −
giving |ψ = |11 . . . 11. This phenomenon of entanglement E0 )/ set by the energy difference between the qubit states,
is a defining feature of quantum mechanics, and an essential and a quality factor Q  1. Here Q = ω01 /γ , where γ is the
ingredient for quantum computing. decay rate of the energy in the qubit due to all sources of dis-
The 2N -dimensional state space of N qubits can hold ex- sipation. Good qubits typically have Q > 106 , and sometimes
ponentially more information than that of N classical bits, considerably higher.
offering the hope of greatly increased computing power. How- Unlike ordinary linear resonators, qubits are extremely an-
ever, the phenomenon of measurement collapse means that harmonic (or nonlinear). As shown in Fig. 2(b), a resonant cw
only N bits of information (a single bitstring representing drive tone can be used to excite the qubit from the “ground”
the state of the N qubits after measurement), selected prob- |0 state to the “excited” |1 state, but further excitation is not
abilistically by the measurement process, can be extracted possible since there are no higher energy levels resonant with
from the quantum state at the end of an algorithm. In or- the drive. Continued application of the drive tone can thus
der to realize a speedup over classical algorithms, quantum only return the qubit to |0 again. The principle of superpo-
algorithms generate interference between the 2N complex sition allows the qubit state to be driven to arbitrary linear
VOLUME 1, NO. 1, JANUARY 2021 405
BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

combinations of |0 and |1; accordingly, persistent resonant much longer than the duration of any algorithm using the
driving of a qubit causes sinusoidal oscillations of the prob- qubit.
abilities |α0 |2 and |α1 |2 in time, as seen in Fig. 2(c). These Qubit frequency fluctuations occur when ω01 is sensitive
oscillations are known as Rabi oscillations, and their angular to some external parameter λ. The parameter λ could be the
frequency (called the Rabi frequency ) is proportional to the local magnetic or electric field, for example (by analogy, for
amplitude of the resonant drive. a voltage-controlled oscillator circuit, λ could be the tun-
As with classical resonators, a qubit’s internal quality factor ing voltage). Small fluctuations δλ then lead to qubit fre-
Qi describes dissipation due to intrinsic loss mechanisms. quency fluctuations δω01 = ( ∂ω 01
∂λ )δλ, which causes dephas-
As we describe in detail in Section IV, the introduction of ing (Fig. 2(e)). Experimentally, T2 is maximized by reducing
external driving, interaction, and measurement ports—all nec- both the sensitivity ∂ω 01
∂λ of the qubit frequency to noise and
essary for quantum computing—will create additional loss the amount of environmental noise δλ(t )2  present.
channels which further damp the qubit resonance. Each loss The qubit properties ω01 , T1 , and T2 can vary widely be-
channel can be identified with its own quality factor: Qd for tween different qubit technologies, as we will see in the next
driving ports (where control signals are applied), Qc for cou- section. We will expand further on the analogy between qubits
pling ports (enabling interactions with other qubits), and Qm and high-quality-factor resonators in Section III.
for measurement ports. The total quality factor of the qubit is
then given by the inverse sum:
1 1 1 1 1 C. PHYSICAL REALIZATION OF QUBITS
= + + + . (2) Just as classical bits can have many different physical
Q Qi Qd Qc Qm
realizations—the voltage on the gate of a transistor, the spin
The characteristic time scale T1 over which a qubit initially orientation of a small magnetic domain on a hard disk, the
in the |1 state will spontaneously transition to the |0 state2 reflectivity of a small region of an optical storage medium—
is given by Q/ω01 . A qubit’s T1 —analogous to the ring-down qubits can have different physical implementations as well. In
time for a high-Q resonator—is an important metric for mea- this paper, we focus on three leading physical implementa-
suring qubit performance, and should ideally be much longer tions of qubits: trapped atomic ions, spins in semiconductors,
than the duration of any algorithm using the qubit. and superconducting circuits. Below, we briefly explain the
These loss channels can be thought of as arising from cou- fundamentals and properties of these different types of qubits,
pling to different sources of dissipation, or “baths” (for exam- all of which can have ω01 in the microwave region of the
ple, the real impedance of a control line), each with some ef- spectrum, as seen in Fig. 3.
fective noise temperature that is usually (but not always) close
to the physical temperature (see Fig. 2(d)). The qubit will
thermalize to these baths on the timescale T1 . If kB Tb  ω01 , 1) TRAPPED ION QUBITS
where Tb is the coupling-weighted average temperature of the Qubits realized in the quantum states of atomic ions trapped
baths, then the qubit will “reset” thermally to the |0 state in ultra-high vacuum are one of the most mature and high-
by itself. For qubit frequencies of 5 GHz, this corresponds fidelity quantum technologies [10]–[12]. Because ions have a
to bath temperatures in the  100 mK range. Otherwise, the net charge, they can be readily trapped and held in isolation in
qubit will thermalize to some combination of |0 and |1, and vacuum using electromagnetic fields. For quantum computing
must be actively reset before it is used in a computation. applications, the ions are usually trapped using so-called lin-
In addition to loss or damping, there is a second type of ear rf Paul traps [11], which confine charged particles—for
decoherence we must consider. Even with a completely loss- hours to months, depending on parameters—using a combi-
less resonator, the resonance frequency itself can still fluctuate nation of static electric fields and oscillating radio-frequency
randomly, causing the excitation in the resonator to lose phase (typically between 20 MHz and 150 MHz) electric fields.
coherence over time relative to a stable reference oscillator. These fields are generated by applying dc and/or rf poten-
If a qubit exhibits fluctuations in its resonance frequency ω01 , tials to sets of trapping electrodes; some example ion traps
the phase information φ in the complex amplitudes α0 and are shown in Fig. 4. The largest traps have centimeter-scale
α1 will be lost on a timescale Tφ (assuming infinite T1 ). Real electrodes made in a machine shop, usually held together
qubits (and real resonators) experience damping too, which with insulating ceramic parts in a three-dimensional geom-
also causes loss of phase information. We can define a charac- etry (Fig. 4(a)). Intermediate-scale three-dimensional traps,
teristic total dephasing time T2 over which phase information with electrode dimensions down to hundreds of microns, can
is lost as 1/T2 = 1/(2T1 ) + 1/Tφ . The loss of phase informa- be made by depositing patterned metal films on laser-cut
tion destroys the interference between probability amplitudes or etched insulating substrates (Fig. 4(b)). Two-dimensional
α on which quantum algorithms rely, so T2 should also be traps, known as surface-electrode ion traps [13], are made
on planar substrates using microfabrication techniques, and
2 The direction of the transitions depends on the effective temperature T of
b have typical electrode dimensions from ∼ 100 μm down to a
the loss channel. If kB Tb  ω01 , the loss channel will cause relaxation from
|1 to |0. However, if kB Tb  ω01 , the loss channel will induce transitions few μm (Fig. 4(c)). Numerous groups, including commercial
in both directions. quantum computing entities, are pursuing surface-electrode

406 VOLUME 1, NO. 1, JANUARY 2021

FIGURE 3. Relevant ranges of the electromagnetic spectrum for quantum computing. We indicate the regions of the spectrum typically used by different
qubit technologies for control and readout. The carrier frequency for trapped ions can be as low as a few MHz. The narrow trapped ion carrier frequency
bands at ∼ 10 GHz and ∼ 12.6 GHz correspond to 133 Ba+ and 171 Yb+ , respectively. Signals at the carrier frequencies are modulated with instantaneous
IF/baseband bandwidths shown in the right panel. These technologies also all use direct baseband control signals with bandwidths as shown in the right
panel.

in Fig. 4(d). The motion of the ions in this configuration is

strongly coupled by the Coulomb force, and is described as a
set of collective normal modes of ion motion, whose resonant
frequencies are typically between 500 kHz and 10 MHz. Traps
can have multiple spatially separated potential wells, each
containing such a Coulomb crystal.
Trapped ions used in quantum computing applications typi-
cally have a single valence electron, with all other electrons in
closed shells. Among the ion species most commonly used for
quantum computing applications are 9 Be+ , 25 Mg+ , 40 Ca+ ,
43 Ca+ , 88 Sr+ , 133 Ba+ , 138 Ba+ , and 171 Yb+ , although numer-

ous others have been employed as well. Below, we describe

several possible choices of qubit levels from among the many
quantum states of these ions.
When the ion species has no nuclear spin, a qubit can be
realized using the two spin states of the ground state va-
lence electron in the presence of an external magnetic field.
FIGURE 4. Trapped ion qubits. (a) Machined 3D trap, (b) laser-machined This type of trapped ion qubit is known as a Zeeman qubit,
3D trap, and (c) microfabricated surface-electrode trap with integrated and is shown in Fig. 4(e). The qubit resonance frequency
microwave antenna structures for qubit control (blue). Ions (shown
is proportional to the external magnetic field B0 according
schematically as red spheres in (c), each in a separate trapping potential
well) are held at reconfigurable positions along the red dotted lines, to ω01 /2π = (γe /2π )|B0 |, where γe /2π ≈ 28 GHz/T is the
between ((a) and (b)) or above (c) the trap electrodes. (d) Camera image of electron gyromagnetic ratio. Trapped ion Zeeman qubits are
chain of trapped ions, fluorescing from laser excitation. Ion-ion spacing is
typically operated in magnetic fields of less than 1 mT, giving
a few μm. (e) Energy level diagram of generic trapped ion Zeeman qubit.
(f) Energy level diagram of 9 Be+ hyperfine qubit with B0 > 0 (the levels are qubit frequencies of ∼ 10 MHz. For ion species with nonzero
degenerate at B0 = 0, shown as grey dotted lines), and one particular nuclear spin, such as 9 Be+ or 171 Yb+ , the hyperfine interac-
choice of qubit levels. In 9 Be+ , ωHFS /2π = 1.25 GHz. In (e) and (f), the
tion between the nuclear spin and the valence electron spin
optical transitions to excited electronic states (blue arrows, not to scale)
are used for laser cooling, qubit state preparation, and qubit readout. gives rise to two manifolds of hyperfine states in the ground
Photo credits: Ion Storage Group/NIST. electronic state, as shown in Fig. 4(f). At low magnetic fields
(B0  50 mT), these manifolds are separated by the hyperfine
splitting ωHFS /2π , which ranges from 1.25 GHz for 9 Be+ to
traps as a path toward large-scale trapped-ion quantum com- 12.6 GHz for 171 Yb+ , with higher-mass ions having larger
puting, due to the ability to make complex trap designs with splittings. A small magnetic field B0 (typically < 1 mT, but
many different trapping regions to hold large numbers of sometimes up to ∼ tens of mT) is applied such that each
ions [14], [15]. hyperfine state has a unique energy. A trapped ion hyperfine
Ions are typically loaded into traps by electron impact ion- qubit consists of two such states, usually chosen to be in
ization or resonant photoionization of a flux of neutral atoms separate hyperfine manifolds. The qubit resonance frequency
from an in-vacuum thermal oven or laser ablation target [12]. ω01 may differ by up to several hundred MHz from ωHFS for
When more than one ion is held in a trap, the mutual Coulomb the range of magnetic fields listed above, depending on the
repulsion between ions keeps them spatially separated. When choice of qubit states. Zeeman and hyperfine qubits can be
laser-cooled [10], [16], the ions form a static Coulomb crystal manipulated directly using rf or microwave magnetic fields,
with ion-ion spacings on the order of a few μm, as seen either launched in free space by distant antennas or horns, or

VOLUME 1, NO. 1, JANUARY 2021 407

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

from local antenna structures fabricated in the trap (as seen in slow quantum gate times, potentially cancelling any advan-
Fig. 4(c)). tage afforded by the long coherence times. For this reason
Trapped ions in ultra-high vacuum are isolated from the many different ‘flavors’ of spin qubit have been devised, with
nearest surfaces and bulk materials by tens to hundreds of μm. trade-offs between controllability, device complexity, and sen-
As a result, electric and magnetic field noise at the ion are sitivity to charge or voltage noise. In this review, we will limit
orders of magnitude lower than typically seen inside or on the our discussion to spin qubits based on confined electrons or
surface of solids; combined with the relatively weak coupling holes in semiconductors.
of the quantum states of the ion to external fields, this means Modern nanofabrication makes it possible to confine and
that trapped ion qubits do not thermalize to the environment detect single electron spins in ‘zero-dimensional’ nanostruc-
rapidly. Typical T1 values for Zeeman and hyperfine qubits tures referred to as quantum dots (QDs) [26]. The potential
are years. As a result, optical pumping is used to initialize the that confines the electron (or hole) is produced electrostati-
internal states of the ions, and laser cooling is used to bring cally via gate electrodes on the surface of a semiconductor,
the ion motion near its quantum mechanical ground state [10], enabling the number of electrons on a dot and their coupling
[12]. This very slow thermalization also means that the tem- to the neighboring dots and reservoirs to be tuned by varying
perature of the trap electrodes and the vacuum chamber need gate voltages. The ability to confine, manipulate, and detect
not satisfy kB T  ω01 to achieve quantum behavior. How- single spin states on quantum dots is largely a consequence of
ever, cryogenic operation of ion traps (in the 4 K to 10 K the Coulomb blockade of charge, an electrostatic phenomenon
range) can be useful for increasing ion lifetime in the trap by arising when the energy to charge a capacitor C by a single
cryopumping background gas, and for reducing electric field electron charge e, E = e2 /2C, is larger than thermal energy
noise that heats and decoheres the ion motion [12]. kB T . For sub-micron devices with self-capacitance in the at-
Trapped ion qubits can be dephased by magnetic field fluc- tofarad range, the energy scale for Coulomb blockade requires
tuations. Zeeman qubits are directly sensitive to magnetic field temperatures below a few kelvin. This necessitates the use
fluctuations, with T2 ∼ tens of ms, but T2 values of up to of dilution refrigerators for operating spin qubit systems, al-
300 ms have been achieved with appropriate magnetic field though work is underway to operate at elevated temperatures
shielding [17]. For hyperfine qubits, it is possible to choose [27], [28].
B0 such that a particular hyperfine transition is insensitive to The initial proposal by Loss and DiVincenzo [29] for a
magnetic field noise to first order. Such a qubit is known as a spin-based quantum computer assumed arrays of coupled
“clock” qubit, so named because field-insensitive transitions quantum dots, each hosting a single electron spin. A large
generally have very long dephasing times and are therefore external magnetic field B0 then sets the energy difference
ideal for realizing microwave-frequency atomic clocks. Bare between the two spin states aligned or anti-aligned with B0 ,
clock qubit T2 values are usually  1 s, but values as high much as with a trapped ion Zeeman qubit. The qubit resonance
as 50 s have been reported [18]; performing a type of qubit frequency is given by ω01 /2π = (γe /2π )|B0 |. Single electron
“chopping” (called dynamical decoupling) to counteract 1/ f spin qubits are typically operated in tesla-scale magnetic fields
magnetic field noise can yield T2 in excess of an hour [19]. in order to ensure ω01  kB T , with ω01 /2π ∼ 1 − 50 GHz.
Measurements of qubit coherence on these timescales are Kane proposed exploiting the exceedingly long coherence
generally limited by the frequency stability and drift of the of phosphorous donors in isotopically purified 28 Si by cou-
microwave reference oscillator to which the qubit is com- pling their nuclei to localized electron spins for single qubit
pared [19]–[21]. addressing, two-qubit coupling, and readout [22]. Any such
The interested reader is referred to Ref. [12] for further qubit architecture requires methods for the precise placement
details on trapped ion quantum computing. of single atomic donors in a solid. Despite significant experi-
mental progress since Kane’s original proposal [30], the MHz
(rather than GHz) resonance frequencies of nuclei present a
2) SEMICONDUCTOR SPIN QUBITS major challenge, leading to kHz clock rates for a quantum
The spin degree of freedom in solids provides another poten- computer. The potential for nuclear spins to be used as quan-
tial platform for scalable quantum computing systems. Nu- tum memories, however, appears more promising [30].
clear spins in silicon, for instance, can exhibit hours-long The challenge of requiring GHz-frequency magnetic fields
T1 times [22]. In contrast to the vacuum of a trapped ion, for single spin manipulation can be overcome at the expense
spin qubits are embedded in solids and surrounded by other of requiring a two-electron system for a single qubit. Here,
atoms, many of which may interact with the spin qubit in double quantum dots are used to host two tunnel-coupled
uncontrolled ways [3]. Fortunately, these interactions are rel- electrons, and the qubit is created by the energy splitting of
atively weak in materials that contain few nuclear spins such the spin singlet state |S and one of the three spin triplet3
as silicon [23], silicon-germanium [24], and carbon materials configurations (|T+ , |T0 , |T− ) [31]. Coupling singlet-triplet
such as diamond [25].
Highly isolated qubits that are only weakly interacting with 3 For two spins, each with basis states |↑ and |↓, these states
their environment are also, in general, weakly coupled to any are |S = √1 (|↓↑ − |↑↓), |T+  = |↑↑, |T0  = √1 (|↓↑ + |↑↓),
2 2
means of control. Weak coupling to control fields results in |T−  = |↓↓.

408 VOLUME 1, NO. 1, JANUARY 2021

 FIGURE 6. The transmon qubit. (a) Schematic diagram. The “X” symbol
represents a Josephson junction. (b) Energy diagram. The unequal energy
spacings are due to the qubit nonlinearity.

Nakamura et al. a little over two decades ago [37]. Here, we

focus on one particular type of superconducting qubit, the
transmon [38], currently used in commercial quantum com-
puting efforts [7]. For a detailed review of superconducting
qubit technology, we refer the reader to [39].
FIGURE 5. A five quantum dot device in GaAs with additional quantum A transmon qubit is a nonlinear microwave LC resonator,
dot charge sensors either side. White arrows indicate microwave currents
used in readout. Blue and orange ovals are electrons. The pairs of blue
constructed by shunting a JJ with a capacitance CQ , as shown
ovals represent two S-T qubits. schematically in Fig. 6(a). The nonlinearity arises from the
JJ, which
 behaves as a current-dependent inductance LJ =
LJ0 / 1 − IJ2 /IC2 , where LJ0 = 0 /2π IC is the zero-bias in-
(S-T) spin qubits is challenging, and work is underway to de- ductance of the JJ, 0 = π /e is the magnetic flux quantum,
vise various architectures to facilitate two-qubit gates. A pair e is the elementary charge, IJ is the current through the JJ, and
of S-T qubits, each requiring a double quantum dot, is shown IC is the critical current of the JJ. The value of IC is determined
in Fig. 5. A multi-electron dot (shown in orange) provides a by the JJ geometry and is typically about 40 nA for transmons,
means of coupling the qubits [32]. corresponding to LJ0 ≈ 8 nH [40].
One can extend the idea of using multiple spins to define The nonlinearity causes the transmon’s resonant frequency
a qubit by implementing three exchange-coupled electrons. to decrease in proportion to the energy stored in the resonator,
This approach creates a qubit via the relative orientation of leading to an energy level diagram of the form shown in
spins, rather than the alignment of spins to an external mag- Fig. 6(b). The non-uniform nature of the energy spacings is
netic field. Requiring three quantum dots as host platform, referred to as anharmonicity and, while quantum engineers
the exchange-only (E-O) qubit [33] (and its ac variant, the typically refer to the transmon qubit as weakly anharmonic, it
resonant exchange qubit [34]) can be controlled entirely with is actually very non-linear in comparison to typical microwave
baseband gate voltages rather than microwave magnetic fields. components. The resonant frequency shifts by an amount η =
The coherence times of semiconductor spin qubits are T1 ∼ −e2 /2CQ for each microwave photon added to the energy
0.1–1 ms and T2 of tens to hundreds of μs for all of the in the transmon, such that all transitions between neighboring
variants described above. For a detailed review of spin qubit transmon energy levels are at different frequencies. For typical
technology, we refer the reader to [35]. component values, ω01 /2π and |η/2π | are in the range of
4–8 GHz and 150–300 MHz, respectively. To take an example,
3) SUPERCONDUCTING QUBITS for ω01 /2π = 6 GHz and |η/2π | = 200 MHz, the addition of
Unlike the qubit variants described above, whose degrees of a single microwave photon of energy 4 × 10−24 J causes a
freedom are those of single electrons and atomic nuclei, su- drop in the transmon’s resonant frequency of over 3%.
perconducting qubits are macroscopic devices that are defined Transitioning from energy Ei to energy Ej and vice-versa
at the circuit level and implemented using nominally lossless requires coupling energy into the transmon at frequency ωi j ,
capacitors, inductors, and Josephson junctions (JJs). When so one can treat the transmon as an ideal two-level qubit if
operated at a low enough physical temperature—typically in caution is taken never to excite it in a way which results in
the low tens of millikelvins—these circuits display coherent leaving the (|0, |1) manifold—that is, one must not drive
quantum mechanical behavior [36], as necessary for use in a the device at ω12 . Notably, the transmon anharmonicity can be
quantum processor. Since they are constructed using lumped engineered through the choice of CQ (while selecting the ap-
and/or distributed circuit elements, the properties of super- propriate JJ sizing to obtain a desired ω01 ). However, smaller
conducting qubits can be engineered similarly to classical CQ results in an increased sensitivity to 1/ f charge noise,
circuit structures. As such, a rich family of quantum devices resulting in a practical upper limit on |η|. As we will see, this
can be realized in this technology platform. Owing to the en- places a constraint on the spectral content of the signals used
gineerable nature of these monolithically fabricated devices, to drive the ω01 transition.
the field of superconducting qubit technology has attracted The relaxation time constant T1 of a transmon is limited
considerable attention since the first device was realized by by materials losses; recent work has demonstrated values of

VOLUME 1, NO. 1, JANUARY 2021 409

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

T1 as high as 300 μs for isolated planar qubits fabricated

on a sapphire substrate [41]. Unfortunately, qubits used in a
quantum processor tend to be coupled to additional loss chan-
nels, and the best values of T1 reported for a ≥ 25 qubit quan-
tum processor are about a factor of three lower [42]. These
losses can occur through couplings to local defects that behave
as two level systems (TLSs), among other mechanisms. The
transition frequencies of such TLSs can be time dependent,
causing T1 to vary with time [43]. Mitigating such effects is
an active area of research.
The dephasing time constant Tφ of single-JJ transmons is
typically limited in part by fluctuations in IC , resulting in
time-dependent variations in LJ0 . Transmons used in some
of today’s 25 qubit quantum processors have T2 as high as
100 μs [42]. However, many contemporary quantum proces-
sor architectures leverage frequency tunable transmons (de-
scribed below), and these devices have considerably lower T2 ,
due to coupling to magnetic flux noise.

III. INTERFACING A MICROWAVE SOURCE TO A QUBIT FIGURE 7. Engineering control and measurement ports into the transmon
circuit. (a) An XY control port that is capacitively coupled to the qubit
Transitioning between the |0 and |1 states of trapped ion, permits microwave drive of the |0 ↔ |1 transition. (b) Replacing the
single-electron spin, and transmon qubits requires exciting the single JJ with a flux-biased SQUID permits control of ω01 via a current bias.
qubit on resonance, which means that we need some mecha- (c) A linear resonator is coupled to the qubit for dispersive qubit state
readout. (d) Photograph showing each of these circuit techniques being
nism to couple microwave energy to the device. This can be simultaneously employed. Capacitances CD and Cg are those between the
done either electrically or magnetically, and since a qubit can coplanar waveguide lines marked “XY control” and “readout resonator”
be thought of as a microwave resonator, we can also think of and the qubit, respectively. In addition, the inductive coupling between the
Z control line and the qubit is visible in the expanded view.
coupling to it just as one would couple to any other microwave
resonator.
In the next sections, we describe some of the considerations
related to microwave drive and deterministic state control of coupling capacitance of 30 aF [40]. Fortunately, this coupling
qubits in a quantum processor. capacitor—which limits the T1 of a 6 GHz qubit to just over
1 ms—can be readily designed using modern electromagnetic
A. DRIVE COUPLING QUALITY FACTOR, Qd design tools.
The qubit-drive coupling can be quantified in terms of a drive The coupling of microwave fields to semiconducting spin
coupling quality factor Qd , defined as the contribution to the qubits, and to trapped ion hyperfine and Zeeman qubits, can be
loaded quality factor of the qubit resonance due to dissipation thought of as an inductive coupling: the microwave magnetic
in the impedance of the drive source. The value of Qd sets an field couples to the electron spin, whose magnetic moment
upper limit on the qubit’s relaxation time constant, is fixed by nature and is “atom-sized”. By contrast, super-
conducting qubits use an electric field coupling to the qubit
Qd
T1 ≤ = T1,d . (3) circuit, where the effective electric dipole moment can be
ω01 engineered to be much larger than “atom-sized” by increasing
Qubits are very under-coupled to the drive source so that the dimensions of the qubit circuit. As an analogy, one can
T1,d  T1 . However, the degree to which the drive is under- think of the drive as coupling either to an extremely small
coupled to the qubit varies drastically from technology to loop antenna or to a large dipole antenna.
technology. This difference means that if one puts the different types
In superconducting qubits, where the qubit is engineered of qubits at the same distance from a propagating electro-
into a circuit environment, the coupling quality factor associ- magnetic wave on a drive line, the Qd would be roughly 108
ated with the microwave drive can be engineered just like in times higher for semiconductor spin qubits and trapped ion
any passive planar circuit. For example, with the capacitive hyperfine and Zeeman qubits than for superconducting qubits,
coupling of Fig. 7(a), Qd ≈ CQ /(CD2 Z0 ω01 ); thus, Qd can be given typical superconducting qubit parameters. The Qd can
set by choosing the coupling capacitor CD . Materials prop- also be made larger or smaller by increasing or decreasing
erties limit the internal quality factor Qi of today’s state-of- (respectively) the distance between the qubit and the drive
the-art transmons to about 4 × 106 [41], so Qd is typically line, because of the fall-off of drive field strengths.
designed to be about an order of magnitude larger such that For semiconductor spin qubits, values of Qd in the 1013 to
it does not limit the qubit T1 . For the typical case of ca- 1015 range have been reported [44], [45]; these qubits were
pacitive coupling (Fig. 7(a)), Qd ≈ 4 × 107 corresponds to a located within ∼ 100 nm of the driving transmission line. For

410 VOLUME 1, NO. 1, JANUARY 2021

trapped ions, the smallest reported experimental values of Qd In general, we do not want noise on the drive line to limit
are ∼ 1019 [46], using surface-electrode traps with integrated coherence, so it is useful to put the required noise levels in
near-field antenna structures where the ions are ∼ 30 μm from context for each of the technologies under consideration. To
the antenna (see Fig. 4(c)). Many experiments use large three- begin, let us consider the limit in which relaxation through
dimensional traps with microwave horns outside the vacuum the drive line produces the same amount of decoherence as
chamber, giving substantially weaker coupling and Qd values noise emitted from the drive line—that is, R↑↓,d = 1/T1,d .
up to ∼ 1024 [19]. Remarkably, keeping the rate of decoherence for these two
mechanisms equal requires the effective noise temperature
B. RABI OSCILLATION FREQUENCY presented by the drive line (Te,d ) be kept at the single photon
Once the value of Qd is known, we can readily determine the level (Te,d = ω01 /kB ), independent of the coupling strength
Rabi oscillation frequency as a function of average available from the drive line to the qubit.
power during the pulse Pav , referenced to the qubit drive port For the case of superconducting qubits, the rate of relax-
 ation through the drive line is typically within an order of
Pav magnitude of the qubit’s intrinsic relaxation rate, and it is
=2 . (4)
Qd essential that noise on the drive line not further decohere the
qubit. Therefore, the limit described above is relevant, which
The Rabi frequency is an important metric, in that π / is means that it is essential to attenuate the thermal noise floor
the time required to flip between the |0 and |1 states. From well below the single-photon noise temperature (≈ 300 mK at
(4) we see that the required available power referenced to the 6 GHz). As such, the microwave control lines typically feature
drive terminal to achieve a given Rabi frequency is directly heavy attenuation, with the final 20–30 dB of loss thermalized
proportional to the coupling quality factor. It is also important to the lowest temperature stage.
to emphasize that the Rabi frequency is proportional to signal Since semiconductor spin qubits are coupled less strongly
amplitude (as opposed to power) since the qubit is a coherent to the microwave drive source in comparison to superconduct-
device. ing qubits, they are also considerably less sensitive to noise.
Since the qubit to drive line coupling is strongest for super- For the range of Qd described in Section III.III-A, the value of
conducting qubits, the available power required to achieve a T1,d for ω01 /2π = 5 GHz is between 5 minutes and 9 hours.
given Rabi frequency is lowest. For instance, achieving a Rabi Since the coherence of semiconductor spin qubits is limited
frequency of 50 MHz with a superconducting qubit having Qd by other mechanisms to timescales considerably shorter than
of 4 × 107 requires an available power of about −70 dBm. this, the single photon limit described above is not applicable.
Compare this to a trapped ion qubit with Qd ≈ 1019 , where Limiting the rate of qubit transitions due to noise on the
roughly +40 dBm is required to achieve Rabi frequencies in microwave drive line to one per second requires limiting the
this range (the highest reported Rabi frequency for a trapped effective noise temperature of the drive line to between 75 K
ion qubit using microwaves is 26 MHz [46]). Most microwave and 7,500 K, depending upon the qubit coupling to the drive
single-qubit gates for trapped ions are implemented with Rabi line.
frequencies below ∼ 100 kHz. For semiconductor spin qubits, Trapped ion qubits are so weakly coupled to the microwave
achieving a Rabi frequency of 50 MHz requires an available drive that a considerable amount of noise (in absolute terms)
power in the range of −16 dBm to +4 dBm [44], [45], refer- can be tolerated on the drive line. For instance, in a worst-
enced to the drive line, with the exact level depending upon case scenario using the lowest reported values of Qd ≈ 1019
the mode of inductive coupling to the qubit. This increased and requiring the highest reported Rabi frequency of ≈ 25
sensitivity compared to trapped ions is simply due to the closer MHz, a typical effective drive line noise would be Sav = −120
proximity of the feed structure to the qubit. dBm/Hz, or a noise temperature of ≈ 7 × 107 K. This would
limit the qubit T1 to approximately 1 second. However, since
C. EFFECT OF NOISE COUPLED THROUGH MICROWAVE values of Qd are typically orders of magnitude higher, and
DRIVE LINE substantially lower Rabi frequencies are used, in practice the
One must also consider the effect of thermal noise coupled limit on T1 due to the drive line noise would be minutes or
to the qubit through the microwave drive line. Since this port hours.
is used to drive transitions between |0 and |1, noise at ω01 The noise requirements are drastically different for each of
injected to the qubit through this channel will also induce the technologies, so it is worth asking whether there is any
transitions, leading to a transition rate [47]: common ground between the signal and noise requirements. It
Sav turns out that it is in fact possible to relate the signal-to-noise
R↑↓,d = , (5) ratio on the drive line to the Rabi frequency and transition
Qd rates:
where Sav is the spectral density of the noise power at the qubit
frequency that is available at the qubit drive port. For a fixed
value of Sav , the transition rate is inversely proportional to the Pav 2
= . (6)
coupling quality factor. Sav 4R↑↓

VOLUME 1, NO. 1, JANUARY 2021 411

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

FIGURE 8. Diagram of a simple quantum algorithm. The horizontal axis represents time, positive towards the right. Each horizontal black line represents
a qubit (labeled with subscripts as “a” and “b”). Each green box represents a control operation on a single qubit, while the orange box represents a
control operation that entangles two qubits. The blue boxes at far right indicate measurement of the qubits at the end of the algorithm. By preparing the
identical initial state, running the algorithm, measuring, and then repeating this cycle many times, one can build up statistics about P|00 and the
probabilities of the other three possible outcomes. These probabilities represent the result of the algorithm.

Thus, once a desired Rabi frequency and Rabi frequency to A Hadamard gate is then applied to the second qubit, which
transition rate ratio (effectively the average number of transi- produces constructive and destructive interference between
tions which can be coherently driven before a noise-induced α00 and α01 , as well as α11 and α10 , resulting in the production
transition occurs) are determined, the required signal to noise of the desired entangled Bell state. Finally, a measurement is
per unit bandwidth is easily calculated using this universal carried out on both qubits.
relationship. Shaped pulses have a peak to average ratio which While the algorithm described above only involves a pair
is greater than unity, so additional margin is required if using of qubits, it turns out that a library of gate operations giving
such control waveforms. full state control of a single qubit (described by 2 × 2 unitary
matrices), combined with a single two-qubit entangling gate
IV. COHERENT CONTROL OF QUANTUM PROCESSORS (described by a 4 × 4 unitary matrix), is sufficient to imple-
USING MICROWAVE TECHNIQUES ment a universal quantum algorithm—that is, one can decom-
Thus far, we have described how one might drive a single pose any arbitrary 2N × 2N unitary operator into a sequence
qubit between the |0 and |1 states, but to perform quantum of these basic operations, each of which is applied to either
computing, we need coherent control of the full multi-qubit a single qubit or a pair of qubits [8]. This so-called universal
complex state vector. This section describes the role of mi- gate set can be thought of in analogy to how all digital logic
crowaves in this process. operations in a classical computer can be constructed from
To get a sense of the type of control needed to implement a NAND gates, for example. Just as NAND gates are only one
quantum algorithm, we will begin by considering the simple of many possible choices of a universal gate for classical com-
quantum algorithm shown in Fig. 8, which is used to generate puting, there are many choices for the universal gate set used
an entangled state known as a Bell state: |ψ = √1 (|00 + in quantum computing; the particular choice of the universal
2
|11). The quantum state can be written as a four-element gate set varies from technology to technology, since each tech-
vector representing the four complex amplitudes α00 , α01 , α10 , nology has its own particularly convenient set of “native” gate
and α11 . Control operations, called gates, can be represented operations. However, in contrast with classical computing,
by matrices that act on the state vector, as shown in Fig. 8. where a gate is thought of as a physical object implemented
The algorithm begins by resetting both qubits (labeled “a” with transistors to which bits are brought to carry out logical
and “b”) to the |0 state. Then a series of quantum gate operations, a quantum gate is an operation applied directly to
operations are carried out. First, each qubit is placed in a a qubit or a pair of qubits in situ. Quantum gate operations are
superposition state by applying a so-called Hadamard gate H often carried out using microwave techniques.
to each of the qubits. After this step, the two qubits are in an When thinking about qubit control and measurement (to
equal superposition of each of the four possible basis states. be discussed in Section V), a natural question to ask is: how
Next, the qubits are entangled via a controlled Z (CZ) gate, good must our control and measurement be? To what extent
which inverts the sign of α11 while leaving the other ampli- can we tolerate errors in either? The answer depends on a
tudes unaltered. While it is not obvious from the measurement variety of factors, but in general, the lower the errors, the
statistics (which are unaltered by the application of the CZ larger the algorithm that can be run successfully, so striving
gate), this is an entangled state, since it is no longer possible to for lower error is important. State-of-the-art error rates for
describe the joint state |ψab  as a product of single qubit states. control and measurement are currently in the range of 10−2

412 VOLUME 1, NO. 1, JANUARY 2021

to 10−6 per operation, depending on the type of operation
and the physical qubit implementation. These are much higher
than typical error rates in classical computing hardware. The
primary reason for this difference is that analog fluctuations
on a digital signal in classical computing must have a certain
minimum amplitude before the signal crosses a digital logic
level and gives rise to a logical error, affording high intrinsic
noise immunity. By contrast, any amount of noise or miscal-
ibration in the operations of a quantum computer can affect
the continuously-variable amplitudes α of the computational
basis states, potentially altering the result of the computation.4
FIGURE 9. Bloch-sphere representation of a single-qubit state.
Since it is never possible to eliminate errors completely,
error correction techniques must be used for large-scale com-
putations. Quantum error correction (QEC) is a large field
of active research, and we will not attempt to give details representation, the |0 and |1 states map to the north and
in this paper, directing the reader instead to [8], [48], [49] south poles, whereas all other points correspond to unique
and references therein. However, a general rule of thumb is superposition states. In the Bloch sphere picture, single-qubit
that control errors below a threshold of roughly 10−4 per gate quantum gate operations can be thought of as rotations of the
are low enough for most QEC protocols [50], and some can qubit state vector. To construct a universal single-qubit gate
tolerate errors as high as roughly 1% per gate [51]. Another set, we must be able to perform deterministic rotations of the
general rule of thumb is that the larger the gate error rate, qubit state about the X, Y, and Z axes.
the more resources (in terms of qubits, gate operations, and
measurements) are required to implement the QEC protocol, 1) XY GATES
an overhead that can become very cumbersome as error rates Single-qubit gate operations are typically classified into two
near the threshold. A practical target for average control error types of gates, XY and Z, which each typically have their
rates is 10−4 per gate. Since control errors are an aggrega- own physical implementation. As the name suggests, XY gates
tion of many different physical error mechanisms, including produce rotations about an axis in the XY plane of the Bloch
decoherence mechanisms intrinsic to the qubits themselves, sphere. Since resonant microwave excitation of the qubit pro-
meeting this goal means that each error contribution should be duces oscillation between the |0 and |1 states, microwave
significantly lower; here we consider the goal of a maximum pulses can be used to mediate XY gates. As the Rabi frequency
error contribution of 10−5 for each of the control error mech- is proportional to the drive amplitude, we can rotate the state
anisms. Readout errors should generally be at similar levels, by a deterministic amount simply by controlling the integrated
although certain error correcting codes can tolerate readout envelope amplitude of the microwave pulse. This can be ac-
errors as high as ∼ 1%. complished by controlling the envelope amplitude and pulse
duration. The axis of rotation is set by the microwave carrier
A. SINGLE QUBIT GATES phase, so adding control of this degree of freedom allows
While Rabi oscillations describe the response of a qubit to access to an arbitrary set of XY gates. Additionally, detuning
the amplitude of a resonant drive signal, a qubit is a coherent the drive from ω01 causes the axis of rotation to tilt away
device, and just like the IQ outputs of a direct conversion from the XY plane, so drive frequency can be used as an
receiver depend upon the relative phase relationship between additional degree of freedom. Achieving a 180◦ rotation (also
the LO and RF signals, the response of a qubit also de- called a ‘π pulse’) in a duration τg nanoseconds while limiting
pends upon the phase of the drive signal. To help understand the impact of each of the error mechanisms to below 10−5
this relationship, we consider the single-qubit state vector, requires control of the integrated envelope amplitude, carrier
|ψ = cos(θ /2)|0 + e jφ sin(θ /2)|1. Since the qubit state is phase, and carrier frequency offset to better than 0.25%, 0.22◦ ,
completely described by angles θ and φ, it can be thought and (2/τg ) MHz, respectively.
of as a vector terminating on the surface of a unit sphere, In general, it is desirable that gate operations be carried out
which is referred to as the Bloch sphere (Fig. 9).5 In this as quickly as possible to limit the impact of decoherence on
the achieved gate fidelities. However, the spectral width of a
4 We stress that quantum computing is not analog computing, in the sense of control pulse is inversely proportional to pulse duration, so it
circuit-based analog classical computers. A quantum computer uses a discrete is essential to design these pulses properly to avoid driving un-
set of operations performed on a discrete set of basis states, and gives a digital desired off-resonant transitions. This is especially true when
output; control and measurement errors can be thought of as digital errors
appearing with some probability, a fact which underlies the ability to perform working with transmon qubits, where the typical separation of
quantum error correction [8]. ω01 and ω12 is about 2π ×200 MHz, on the order of . The
5 While the Bloch sphere representation of a single-qubit state is a valuable
leakage rate to an undesired transition ωi j can be estimated as
tool for gaining intuition into single-qubit gate operations, once two qubits
are entangled, we can no longer separate their states and the Bloch sphere the relative energy at ωi j to that at ω01 [52], so pulse shaping
picture fails to be meaningful. techniques are typically employed when driving transmons to

VOLUME 1, NO. 1, JANUARY 2021 413

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

optimize the tradeoff between pulse duration and ω12 drive. individual addressing by creating differential Rabi frequencies
Common envelope waveforms include Gaussian and raised or differential qubit frequencies on multiple ions, among other
cosine shapes, which have much reduced frequency-domain techniques [58]–[61]. These latter methods can be used on
sidelobes in comparison to a rectangular envelope. These sim- both field-sensitive and field-insensitive (clock) qubits. The
ple envelopes are sufficient to achieve pulse durations as short full literature for trapped ion qubit individual addressing tech-
as about 20 ns, but reaching shorter gate durations with trans- niques, including individual addressing using laser beams, is
mon qubits requires further waveform optimization. First, to extensive and is not referenced here.
further suppress drive of ω12 , one can employ the derivative Implementing XY control of single semiconductor spin
removal by adiabatic gate (DRAG) technique, in which a qubits requires microwave magnetic fields B1 , applied or-
notch is generated at ω12 by adding a quadrature derivative thogonal to the direction of B0 , and typically generated from
term to the baseband envelope [53], [54]. While this takes dedicated on-chip antenna structures near each qubit. An al-
care of the ω12 leakage term, moving to shorter gate durations ternate approach uses a global cw microwave magnetic field
requires larger amplitudes and, due to the ac Stark effect, on all qubits in the array [22]. Individual qubits are then tuned
the effective value of ω01 becomes amplitude-dependent. As in and out of resonance with the global field via local gate
such, a time-varying detuning also must be applied to the electrode voltages, effectively pulling or pushing the electron
microwave carrier signal [55]. wavefunction towards an interface to modify the g-factor [22],
In trapped ion qubits, XY gate fidelity is typically limited by or away from the nuclear spin of a donor atom to vary the
errors in the microwave pulse parameters as described above, hyperfine coupling [22], [30].
rather than qubit decoherence during the gate, especially for For S-T qubits, static magnetic field gradients between the
clock qubits. For this reason, high-fidelity microwave gate two dots that make up the qubit can drive XY rotations. For
pulses are generally of longer duration, ∼ 1 μs to ∼ 100 μs, control of these rotations, nanosecond rectangular pulses are
which allows more fine-grained control of pulse durations and used to separate the two electrons for a time such that they ex-
thus integrated pulse amplitude. The highest fidelity single- perience different magnetic fields. These gradients are either
qubit XY gates reported to date in trapped ion qubits have produced naturally by hyperfine magnetic fields from neigh-
infidelities of 1.0(3) × 10−6 per gate, with a π -pulse duration boring nuclear spins, or engineered using micro-magnets on
of 24 μs [18]. Because of their longer duration, the spectral the surface of the semiconductor (the latter is better con-
content of these control pulses is fairly narrow. Off-resonant trolled and gives reduced gate error). Resonant microwave
coupling to other states is generally negligible, given typical driving at the frequency corresponding to the exchange energy
separations to the nearest neighboring hyperfine transitions (1 GHz) can also be used for manipulation of S-T qubits,
of  5 MHz, and sometimes ∼ 100 MHz, depending on the with the advantage that the drive frequency is much lower than
ion species and the magnetic field. Zeeman qubits have no for single-spin qubits [62].
additional neighboring levels and so off-resonant excitation The need for magnetic fields can be alleviated altogether
of other levels is not a concern. using the three-electron E-O spin qubit. Here, qubit XY con-
For simplicity, pulses are generally rectangular, without trol is implemented using modulation of the exchange energy
shaped rise and fall times, because even the broader spectral between two of the electrons, proportional to the overlap of
content from the sharp pulse edges is far detuned from other their wavefunctions and controllable using time-dependent
transitions. Pulses with durations much longer than ∼ 100 μs voltages applied to surface gates, similar to S-T qubits. Ex-
can become problematic for field-sensitive qubits, if the fluc- change between the right-most spin pair drives qubit rotations
tuations of the qubit frequency due to environmental magnetic about the Z axis, whereas exchange between the left-most pair
field noise start to become comparable to the Rabi frequency drives rotations about an axis tilted by 120 degrees from the
/2π . Z axis. Concatenating up to four pulses produces single qubit
Unlike with superconducting qubits, where each qubit has rotations around any axis. Again, these pulses are typically
a dedicated drive line, microwave control fields are rela- baseband square pulses with nanosecond rise and fall times.
tively uniform over an array of trapped ions, providing only The use of the exchange interaction for qubit XY control
global control. This arises because the ion-ion spacing is reduces microwave complexity, but does so at the cost of
much smaller than either the distance to the driving an- increased sensitivity to voltage noise on the control line and
tenna/horn or the free-space wavelength of the control fields. charge noise in the material system.
In surface-electrode traps, “regional” control can be achieved
through “beamforming” from multiple antennas spaced a suit-
able distance apart [56]. Individually addressed XY gates for 2) Z GATES
closely-spaced ions can be realized in one of several ways. In addition to XY gates, a universal gate set requires rotations
For magnetic-field-sensitive qubits, an applied magnetic field about the Z axis, which are referred to as Z or phase gates.
gradient along the ion string causes each ion to have a separate Referring to Fig. 9, Z gates only affect the qubit phase φ. One
qubit frequency, so that frequency domain addressing is pos- can perform these gates either virtually, by applying a phase
sible [57]. Alternatively, focused laser beams, or microwave jump to the RF carrier used for subsequent XY gates [63],
field gradients from near-field electrodes, can be used for or physically, either by applying a sequence of two XY gates

414 VOLUME 1, NO. 1, JANUARY 2021

whose combined phase values yield a prescribed Z rotation
or by detuning the qubit frequency by δω01 for a controlled
duration τ , so as to accumulate phase (similar to how phase is
adjusted in a phased locked loop during acquisition). Virtual Z
gates are attractive since they can be instantaneously applied
to baseband IQ envelopes and their accuracy just requires
a stable system clock, which is necessary to generate high
fidelity XY gates in the first place. However, for systems with
more than one qubit, control of the relative phases of qubits is
often necessary, which requires physical Z gates.
Minimizing error rates requires carrying out physical Z
gates quickly, so it is often preferable that these gates be me-
diated via a frequency tuning mechanism rather than via mul-
FIGURE 10. (a) Coupling of linear resonators via a switched capacitance.
tiple XY gates. For superconducting transmon qubits, control (b) Coupling of transmon qubits via a tunable coupler. Two qubit gates can
of ω01 can be enabled by replacing the single JJ of Fig. 6(a) be mediated using the qubit and coupler bias currents (iZ1 , iZ2 , and iCPL ).
with a two-JJ loop, known as a superconducting quantum
interference device (SQUID) (Fig. 7(b)). In this context, the
SQUID behaves as a magnetic-flux-tunable nonlinear induc- only specific ions, or by using a gradient of an rf/microwave
tor, so the qubit frequency can be externally controlled via a magnetic field [60], [64]. These techniques have the added
current bias, analogous to how a voltage controlled oscillator benefit that they can be applied to hyperfine clock qubits as
is tuned via a control voltage. Coupling is kept very weak to well.
minimize frequency fluctuations due to noise on the flux-bias
tuning line, with typical mutual inductances in the range of B. MULTI-QUBIT GATES
just 2 pH. This weak coupling leads to control currents of a While a library of single qubit gates allowing arbitrary con-
hundred microamps or more, and nanoamp-level resolution trol on the Bloch sphere is required for universal quantum
is required to minimize errors. Digital-to-analog converters computing, this alone is not sufficient. As described earlier,
(DACs) with at least 14 bits of resolution are typically used we also need to create entangled states if we are going to
to generate these currents. exploit the exponentially large computational space. To do
For gate-defined semiconductor spin qubits, the specific this, we must carry out operations involving multiple qubits,
method of producing Z rotations depends on the flavor of which requires entangling the qubits in a deterministic man-
qubit, but generally amounts to applying a rectangular pulse ner. To understand how one might interact qubits, we can
to modulate the energy detuning or the exchange interaction begin by considering the circuit of Fig. 10(a), in which two
between electrons. For single spins, square-shaped pulses can linear resonators are coupled through a switched capacitor
be applied to vary the position of the electron wavefunction, network. Assuming an ideal switch in the open position, the
modulating the Zeeman energy for the duration of the pulse. resonators are isolated and the natural frequencies of the√ cir-
For S-T qubits, turning on the exchange coupling swaps the cuit are just those
√ of the individual resonators: ω1 = 1/ L1C1
two spins in the presence of the magnetic gradient, effectively and ω2 = 1/ L2C2 . However, when the switch is closed, the
performing a rotation about the Z axis. Similarly, for the three resonators pull each other and the natural frequencies of the
electron E-O qubit, rotations about X,Y, or Z are produced circuit shift. If each resonator was initially oscillating at its
via application of concatenated rectangular pulses to modulate respective resonant frequency, the oscillation dynamics would
the exchange interaction between left and middle electrons, or shift from a distinct frequency at each node to the joint modes
right and middle electrons. of the coupled resonator system (which will be shifted from
Physical Z rotations in trapped ion Zeeman qubits and the bare frequencies. In the case where the coupling is weak,
magnetic-field-sensitive hyperfine qubits (where ∂ω ∂B0 = 0)
01
we can think of the coupling as a perturbation applied to
could be implemented by changing the magnetic field B0 to each of the isolated resonance frequencies. As such, if we
shift the qubit frequency by δω01 for a specified duration energize one of the resonators and pulse the switch closed
τ , yielding a Z rotation by δω01 τ . However, the static mag- for a short period of time, the oscillation in the resonator will
netic fields in trapped ion experiments are usually generated acquire a phase shift due to the temporary shift in resonance
by highly stabilized current sources or permanent magnets, frequency.
which are not amenable to rapid field shifting and would Now, imagine both of these resonators have a nonlinear
provide only global Z rotations. Therefore, physical Z ro- relationship between their isolated natural frequencies and the
tations are typically implemented by compiling them to XY amplitude of oscillation (similar to a qubit). In this case, if we
gates, applying ac Stark shifts to the ions using detuned laser perform the same experiment where we energize one of the
beams, or applying ac Zeeman shifts to the ions using detuned resonators, pulse the coupling on, and look at the acquired
microwave or rf magnetic fields. Individually addressed Z phase, the result will depend upon the state of the second
rotations can be applied using focused laser beams to shift

VOLUME 1, NO. 1, JANUARY 2021 415

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

resonator, since the degree to which it pulls the first resonator major technical challenge for exchange coupled qubits is the
is state dependent. This behavior describes one way the ba- sensitivity of the tunnel rate to gate voltage and the require-
sic interactions required to perform entanglement generating ment that electrons must be brought within nanometers of
two-qubit gates can be carried out. each other for coupling. This latter aspect leads to crowding of
Two-qubit gates require a mechanism for interacting qubits gate electrodes and challenges for crosstalk mitigation. Alter-
in a deterministic manner, and many different approaches native coupling schemes making use of intermediate electron
have been proposed and demonstrated. For instance, inter- states [32] are presently an active area of research. A further
actions between superconducting qubits can be engineered fruitful direction is to couple remote S-T qubits via a cavity
at the circuit level by introducing either static [65] or tun- resonator, following similar approaches to superconducting
able couplings [7], which may be implemented inductively qubits [70]. To date, two-qubit gates using E-O qubits have
or capacitively. One approach is shown in Fig. 10(b), where not been demonstrated, although qubit coupling schemes are
a tunable coupler allows interactions between a pair of fre- likely to be similar to S-T qubits.
quency tunable transmons [66]. The coupler itself consists Trapped ion hyperfine and Zeeman qubits have negligible
of a transmon—which serves as a frequency tunable LC direct interaction with each other due to very weak spin-spin
resonator—that is capacitively coupled to each of the qubits. coupling. However, the motion of multiple ions in a single
With this structure, it is possible to realize coupling strengths trapping potential is very strongly coupled. As a result, almost
ranging from completely off to tens of MHz. One can engineer all entangling gates carried out between trapped ion qubits
a wide range of two-qubit gates through proper design of the are realized using the quantum motion of the trapped ions
three current bias waveforms for this circuit [67]. For instance, as an intermediary “bus”. An effective ion-ion interaction is
if the qubits are tuned into resonance and the coupling is realized by coupling the trapped ion spin to its motion in an
enabled, a single excitation will oscillate back and forth be- appropriate way, using external control fields. Typically this
tween the qubits (that is, oscillation will occur between the is done with laser beams, but it is also possible to do using
probability amplitudes of the |01 and |10 states), and by rf and/or microwave fields. Crucially, spin-motion coupling
properly setting the gate duration it is possible to engineer a requires a spatial gradient of the control field over the spatial
gate where α01 and α10 are swapped. extent of the ions’ quantum mechanical zero-point motion
While two-qubit gates for superconducting qubits can be in the trap, typical ∼ 10 nm. The magnetic field gradient of
carried out through control of the qubit frequencies as de- free-space microwaves near ω01 (usually a few GHz) is very
scribed above, they can also be performed without changing small over this length scale. However, microwave magnetic
the qubit frequency, using an additional microwave drive tone fields with negligible gradient strength can be combined with
instead. Such approaches are necessary in architectures that additional static [71], [72] or few-MHz [64], [73] magnetic
use fixed-frequency qubits (see for example Ref. [42]). One field gradients to produce the desired spin-motion coupling.
example of an all-microwave two qubit gate is the cross- Magnetic field gradients at microwave frequencies near ω01 ,
resonance (CR) gate [68], in which one of a pair of reactively- or near resonance with the ion motional frequency, can also be
coupled qubits is driven by a microwave tone at the qubit used for spin-motion coupling and entangling gates [10], [74].
frequency of the other qubit. With the appropriate drive am- These gate protocols are often carried out in surface-electrode
plitude and duration along with the addition of a single qubit traps, where larger near-field magnetic field gradients can be
gate applied to each of the qubits, the CR gate can be used generated [46], [75], [76].
to implement a CNOT operation, which swaps α10 and α11 A number of experimental demonstrations of high-fidelity
while leaving α00 and α01 unchanged. The gates described microwave-based entangling gates between ions have been
here are just two of a wide assortment of gates that can be carried out [61], [77]–[80]. In general, these gates are slower
applied to superconducting qubits; for further discussion of (∼ ms duration) than laser-based entangling gates, which can
two-qubit gates in superconducting qubits, we refer the reader be performed in tens or hundreds of μs (and some as fast as a
to Ref. [39]. few μs [81]). However, the fidelities reported for microwave-
Two-qubit entangling gates in semiconductor spin qubits based gates, with errors in the few 10−3 range per gate, are
are generally implemented via the Heisenberg exchange in- competitive with the fidelities of laser-based gates. In addi-
teraction between electrons on neighbouring dots of an array tion, laser-based gates have fundamental fidelity limitations
[29]. The charge dipole associated with the two-electron sys- due to off-resonant scattering from excited electronic states
tem opens the prospect of coupling S-T qubits capacitively, in the ion [82], making microwave-based gates, which do not
since the relative spin orientation of one qubit can lead to have this fundamental limit, an appealing alternative.
charge rearrangement that effectively gates another qubit [69].
The advantage of exchange coupling is its controllability, C. HARDWARE FOR QUANTUM STATE CONTROL
modulating the tunnel coupling between two adjacent quan- High-level control requirements for each of the three qubit
tum dots for a controlled amount of time. The evolution of the technologies are summarized in Fig. 11. While the details
two-qubit system then depends on the wave-function overlap vary greatly among the technologies, some form of pulsed RF
of the electron states, resulting in√
the physical exchange of the waveforms and/or baseband control signals are required to run
electron positions to execute a SWAP entangling gate. A a quantum processor. An exemplary control system for a pair

416 VOLUME 1, NO. 1, JANUARY 2021

 the quantum computing community. Part of the reason that
the field of quantum computing has experienced rapid growth
over the past ten years is that components required to build
control systems such as this have become commodity items,
thanks to the wireless communications revolution.

V. MEASURING THE STATE OF A QUBIT

To behave quantum mechanically, as required for quantum
computation, qubits must be well-isolated from sources of
noise or dissipation in their environment. We have described
this requirement thus far by stating that Qd , Qi , and Qc must
be large (> 106 , and ideally even higher). However, at some
point in a quantum algorithm, the state of the qubit must be
measured, a task which necessarily requires coupling the qubit
to the outside world strongly enough that the qubit can influ-
ence the state of the measurement apparatus and thus allow us
to determine the qubit state. This coupling is characterized by
a quality factor Qm which describes the effective dissipation
FIGURE 11. Summary of microwave/baseband control requirements for bath seen by the qubit due to coupling with the measurement
each of the qubit technologies. Abbreviations: Q–qubit, Pav –available apparatus.
power at qubit drive port.
Because the energy difference between the two qubit states
ω01 is very small for microwave-frequency qubits, some
form of amplification must be used to convert it into a
classical-level signal strong enough to be digitized and an-
alyzed to read out the state of the qubit. Multiple stages of
amplification are used; as with standard low-noise microwave
receivers, the overall performance is most sensitive to the first
amplification stages.
Superconducting qubits, and increasingly semiconducting
qubits as well, typically use a technique called dispersive
readout to measure the qubit state [85]–[89]. This approach
relies on coupling the qubit to a microwave-frequency super-
conducting resonator. The resonator frequency ωr is detuned
from ω01 by an amount   g, where g is the qubit-resonator
coupling strength. In this so-called “dispersive regime” of the
coupling, the qubit and resonator modes do not exchange
energy, but the frequency of the resonator is shifted by an
FIGURE 12. Simplified control hardware for a two-qubit transmon-based amount 2χ , called the “dispersive shift”, depending on the
quantum processor with tunable coupler.
qubit state. The resonator is overcoupled to a feedline, giv-
ing a resonator linewidth κ which is ideally ∼ 2χ (typically
κ ∼ 2χ ∼ 2π × 5 MHz). The resonator is typically probed
of superconducting qubits connected via a tunable coupler in reflection, as shown in Fig. 7(c), with a probe tone near
appears in Fig. 12. Each XY control signal is generated using ωr . The reflected signal will depend strongly on whether the
single-sideband mixing, with the complex envelope generated resonance frequency has been shifted, which in turn indicates
using a pair of high-speed DACs. Alternatively, the RF XY the qubit state. Thus the measurement of the qubit state can
signals are sometimes directly generated using high speed be realized by measuring S11 of the readout resonator at a sin-
DACs [83], [84], obviating the need for analog mixing. The gle frequency near ωr . For strongly overcoupled resonators,
XY signals are heavily attenuated to suppress thermal noise |S11 | ≈ 1, but arg(S11 ) will be qubit-state-dependent near ωr .
at the qubit drive port. An additional three DACs generate One can think of this as a binary phase-shift-keyed signal,
the broadband control signals required to drive the Z control whose symbols (typically separated by ≈ 180◦ , but poten-
lines and coupler bias port. These DACs feature 14 bits of tially separated by a smaller angle) correspond to the two
resolution, as required for the frequency control of the qubits qubit states. We note that this process also constitutes a kind of
and coupler. The digital waveforms are generated using a amplification: the single-photon qubit energy ω01 has been
field-programmable gate array (FPGA), which is configured turned into a phase shift on many cw probe photons.
to orchestrate quantum algorithms. Architectures such as this The challenge here lies in the fact that the probe tone must
are extensible to hundreds of qubits and widely in use among be very weak, so as not to scramble the qubit state after

VOLUME 1, NO. 1, JANUARY 2021 417

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

initially collapsing it to |0 or |1 [90], [91]. Otherwise, the If the qubit were in state |1, it would decay to |0 and
resonator frequency would jump back and forth along with emit a photon of energy ω01 into the photon counter. It is
the qubit state, and the readout signal would stop providing challenging in practice to realize a microwave photon counter
information about the state to which the qubit initially col- because the single photon energy is so small in the few-GHz
lapsed. The limit on the probe tone amplitude depends on the regime, but such a device (which is essentially a modified
vector difference in the IQ plane between the reflected probe superconducting phase qubit) has been demonstrated with
tone corresponding to state |0 and the reflected probe tone readout fidelity as high as 98.4 % [96], [97].
corresponding to state |1—in other words, the size of the For trapped ion qubits, the very weak coupling of the qubit
phase-shift-keyed signal. For superconducting qubits this dif- state to microwave fields (much higher Qd than the other tech-
ference signal should generally be kept below ∼ 300 nVpeak , nologies) means that it would be very difficult and inefficient
which corresponds to a signal power of around −120 dBm. to extract information from the ion at microwave frequencies.
For spins, probe powers up to −80 dBm have been used. However, trapped ions possess optical transitions, allowing
This weak probe tone must be amplified, but the amplifier easy extraction of optical photons, which have the added ben-
noise must be small enough not to drown out the signal. efit of being easily detected with low background noise by
While one can average in time (in other words, decrease the room-temperature single-photon counters.6
resolution bandwidth) to remove amplifier noise and recover Trapped ions are generally read out using the so-called
the weak signal, we must perform our measurement in a time electron shelving technique [98], where the probability am-
 T1 , so that the qubit state does not decay during readout plitudes in states |0 and |1 are mapped to two suitable states
and thus corrupt the readout result. Practically speaking, this |b and |d using coherent control pulses of the same types
means that readout must be performed in several hundred used for qubit manipulation (microwave pulses and/or laser
nanoseconds, or equivalently that the resolution bandwidth pulses). When illuminated with a laser beam of appropriate
must be  10 MHz. The duration of measurement also poses wavelength and polarization, an ion in state |b will fluoresce,
a limit to the clock rate of a quantum algorithm or error absorbing photons and re-emitting them in all directions. The
correcting code. laser beam is chosen to drive a so-called cycling transition,
The solution is to use an ultra-low-noise amplifier as the where the ion is excited from the state |b and then emits
first stage of the receiver, such that even the weak readout a photon, always returning to |b after the emission. This
signal can be amplified with signal-to-noise ratio well above enables repeated rounds of excitation and emission. It is some-
1. Superconducting parametric amplifiers based on Josephson times necessary to use multiple laser beams to “close” this
junctions as nonlinear elements [92] can reach the quantum cycle. In contrast, an ion “shelved” in the state |d (chosen
limit for noise, where the amplifier noise temperature TN at such that all transitions out of |d are far off resonance with
frequency ω is equal to ω/2. For 6 GHz signals, this cor- the readout laser) will not interact with the laser beam and
responds to TN = 144 mK. Dispersive qubit readout using thus will not give off fluorescence photons. By collecting a
superconducting parametric amplifiers was first implemented fraction of the fluorescence photons with an imaging objec-
a decade ago [93], and readout fidelities of ∼ 99 % have tive and counting them with a single photon counter, it is
been reported [94]. Progress in superconducting parametric possible to distinguish between a fluorescing “bright” ion in
amplifiers in the past 15 years has been driven by quantum state |b and a “dark” ion in state |d, as long as the mean
computing applications; we discuss this in more detail in number of photons counted for bright and dark ions is suf-
Section VII.A. ficiently different. In practice, a readout duration of several
As discussed above, it is important that Qm be high, so hundred microseconds typically gives tens of counts for a
that the measurement process not couple the qubit to a lossy bright ion, and ∼ 1 − 4 counts for a dark ion, although the
environment that would cause T1 decay. The readout resonator duration and count rates can vary by an order of magnitude
acts as a bandpass filter between the qubit and the lossy (real) or more depending on the specifics of the setup. Readout
impedance of the readout transmission line; since the res- fidelities as high as 99.99% have been demonstrated using this
onator is far detuned from the qubit, this suppresses coupling technique [99].
between the qubit and this source of loss. However, for some Electron shelving readout implements two forms of amplifi-
qubits this coupling is still the dominant source of loss. A cation. First, mapping into the states |b and |d and scattering
standard way to mitigate this loss is to place a bandpass filter a single photon can be thought of as turning the qubit energy
(usually a second resonator) between the readout resonator difference ω01 into the energy of an emitted UV or visible
and the transmission line [94], [95]. This type of filter is photon, which is typically between 105 and 108 times larger
known as a Purcell filter and can be used to boost Qm by an for hyperfine or Zeeman qubits. Secondly, the cycling transi-
additional two orders of magnitude or more, depending on the tion allows up to ∼ 106 such photons to be scattered during
design parameters. The bandwidth of the Purcell filter can be the readout operation, giving further gain.
much larger than κ, so that it does not affect the speed of the
readout.
It is also possible to read out a superconducting qubit by 6 This is primarily because of their high energy relative to the available
turning on a strong coupling to a microwave photon counter. thermal energy (ω  kB T for optical photons at T = 300 K).

418 VOLUME 1, NO. 1, JANUARY 2021

VI. OTHER APPLICATIONS OF MICROWAVE TECHNOLOGY than a part in 105 . In a typical linear rf Paul trap, 1/3 of
Beyond the direct use of microwave signals and techniques the motional modes (called “axial” modes) have frequencies
for qubit state control and measurement, there are many other determined purely by static potentials applied to trap elec-
critical applications of microwave and rf technology in quan- trodes and 2/3 of the modes (called “radial” modes) have
tum computing. This section details several of the essential frequencies that depend linearly on the rf voltage applied
supporting roles played by microwave technology, beyond to the trap electrodes. Many multi-qubit gate schemes use
direct control and measurement of qubits. radial modes, so achieving the required motional frequency
stability of ∼ 10 ppm or better means that the rf voltage on the
trap electrodes must be stable at the ∼ 10 ppm level as well.
A. RF FOR ION TRAPPING This requires high stability of the resonator frequency and
Trapped ion qubits for quantum computing applications are al- quality factor, high gain stability for the rf amplifier driving
most always confined using a combination of radio-frequency the resonator, and low amplitude and phase noise for the
and static electric fields in a so-called Paul trap [10] (Penning rf generator. If the resonator Q is too high, this can cause
traps, which use a combination of static magnetic and electric increased voltage fluctuations at the trap through FM-to-AM
fields to confine ions, can also be used for quantum computing conversion, giving some intuition for why trap rf resonators
but are more typically employed for quantum simulation or are not designed simply for maximum Q. Recent work in the
precision spectroscopy experiments [100]–[102]). The rf elec- field has demonstrated rf limiting amplifiers [109] and active
tric fields for trapping are generated by applying rf voltages to feedback methods for rf amplitude stabilization [109], [110]
specific electrodes of the trap, such that an rf electric field which can meet the desired ∼ 10 ppm amplitude stability of
quadrupole is formed in vacuum at a distance of typically the rf voltage at the trap.
30 μm to several mm from the electrodes. The oscillating
quadrupolar rf fields provide an effective confining potential
for ions. We refer the reader to Refs. [10], [11] for further B. APPLICATIONS TO LASER SYSTEMS FOR QC
details on Paul traps. Numerous qubit technologies, including trapped ion qubits,
The frequency and amplitude of the applied rf necessary for neutral atom qubits [111], [112], and optically-active defect
stable trapping of ions depend on the physical dimensions of centers in solids [113], [114], rely on laser-based control and
the trap, the species of ion, and the desired strength of the con- readout methods. These laser systems use microwave tech-
finement. Larger traps, heavier ions, and stronger confinement nology for a number of tasks. Most laser beams for quan-
all require larger rf voltages on the trap electrodes. Reducing tum information experiments need to be pulsed on and off
the rf frequency enables trapping with lower rf voltages, but with microsecond rise/fall times, too fast for a mechanical
if the frequency is reduced too much then the trap will be- shutter. In addition, sometimes a laser beam with amplitude
come unstable and the ions cannot be held reliably. Typical rf modulation, rapidly tunable frequency, or multiple frequency
Paul traps for quantum computing applications operate at rf components in a single beam is desired. For these tasks,
frequencies between 20 MHz and ∼ 100 MHz, and with peak acousto-optic modulators (AOMs) [115] and/or electro-optic
rf voltages on the trap electrodes between 10 V and ∼ 500 V. modulators (EOMs) [116] are used. An EOM can be thought
These combinations of frequency and amplitude are gen- of like a microwave mixer, except the LO port uses an optical
erally quite difficult to achieve with direct driving, so the signal while the IF port accepts an rf or microwave signal;
rf electrodes are typically incorporated into an rf resonant signals at the IF port modulate the optical LO and give rise
structure, which provides rf voltage step-up and filters noise to sidebands on the optical output. AOMs are more akin to
from the rf drive electronics that could excite the ion motion. single-sideband mixers. They perform frequency-shifting of
Many different classes of resonators have been used, including the laser light passing through them, but can also be used for
helical resonators [103], [104], coaxial resonators [105], and amplitude modulation (including on/off switching) of laser
lumped element resonators [106], [107]. These resonators typ- beams. The “carrier leakage” and “spurious sideband” laser
ically have loaded quality factors of several hundred, although beams have different directions of propagation from the de-
some cryogenic resonators can approach 104 . Toroidal trans- sired “sideband” at the AOM output, enabling them to be
formers on ferrite cores, operating as flux-coupled impedance- filtered out spatially. Both AOMs and EOMs require rf or
transforming ununs, can also be used to provide voltage step- microwave signals at around +30 dBm to function; generally
up [108]. AOMs are driven at frequencies from ∼ 50 to ∼ 500 MHz,
Qubit-qubit coupling for trapped ions relies on the use of with 10s of MHz bandwidth, while EOMs can accept drive
the ions’ shared motional degree of freedom as a coupling tones from near dc up to many GHz, and have widely varying
“bus”. Multi-qubit gates typically have durations τg from ∼ 10 bandwidths depending on the choice of resonant rf/microwave
μs up to several ms [12], with high-fidelity microwave-based enhancement circuits inside.
gates typically in the ∼ms range. To achieve high gate fidelity, In addition to modulation of laser light, microwave technol-
the motional frequency must be stable to  τ1g . In practice, ogy is also important for stabilization of the lasers themselves.
this means that the ion motional frequencies, typically a few A famous method for locking a microwave oscillator to a
MHz, should be stable at the ∼ 10 Hz level or below, or better stable reference cavity using modulation sidebands, due to

VOLUME 1, NO. 1, JANUARY 2021 419

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

Pound [117], was extended to operation with laser oscilla- and bandwidths of ∼ 10 MHz [122]–[125]. The viability of
tors and stable optical references (cavities or spectral lines) these amplifiers for high-fidelity qubit readout was demon-
by Drever, Hall, and coworkers [118], and is known in the strated shortly thereafter [93], and they were rapidly adopted
laser community as Pound-Drever-Hall (PDH) locking. PDH as the state of the art [94], [126]. Subsequent work has used
lock circuits, which operate at rf/microwave frequencies, are impedance engineering to increase bandwidths to the ∼ GHz
ubiquitous in laser systems. There are also a wide variety range [127]–[129], and more complex designs with many
of servo loops for laser amplitude stabilization which rely Josephson junctions have shown substantial improvement in
on modulating the amplitude of the rf drive to an AOM as saturation powers [129], [130]. Most designs operate in reflec-
the feedback signal. Mode-locked lasers, which emit periodic tion, typically requiring bulky circulators to separate the out-
short ( ps) pulses of light with an extremely stable repetition put signal from the input signal, and to protect the qubit circuit
rate, forming a frequency comb, are used in trapped ion appli- from the strong pump tone, so transmission-mode parametric
cations to drive stimulated Raman transitions between states amplifiers providing directional, non-reciprocal amplification
with very large detuning, such as the 12.6 GHz hyperfine have been developed [130]–[133]. A considerably more ex-
qubit states in 171 Yb+ [119]. For optimum performance, the tensive review of the literature than is provided here can be
repetition rate of the mode locked laser (typically ∼ 100 MHz, found in Ref. [92].
but potentially as high as a few GHz) must be stabilized, The bandwidth, saturation power, and noise performance
which requires implementing a phase-locked loop between a of current state-of-the-art parametric amplifiers allows simul-
stable reference oscillator at the desired repetition rate and the taneous readout of multiple qubits by frequency multiplexing
signal from the laser pulse train on a fast photodiode. their readout resonators, which share a common feedline to
the parametric amplifier. The gain and noise performance of
VII. MICROWAVE INNOVATIONS FROM QUANTUM these parametric amplifiers are such that they set the overall
COMPUTING receiver noise for the amplification chain (the rest of which
Quantum computing relies heavily on microwave technolo- consists of ultra-low-noise cryogenic and room-temperature
gies already developed for other applications. However, quan- microwave transistor amplifiers) out to the room temperature
tum computing also requires operation in new performance demodulation and digitization circuitry.
regimes, and has inspired the development of novel mi-
crowave technologies and systems to meet those challenges.
We describe two major advances in microwave technology
that have arisen from quantum computing research: quantum- B. NON-RECIPROCAL DEVICES
limited microwave amplifiers, and cryogenic non-reciprocal Microwave circuit elements exhibiting non-reciprocity are
microwave devices, including chip-scale non-reciprocal de- currently used heavily in quantum computing, mostly in the
vices. amplification of readout signals from qubits. In this context
they serve two main purposes. First, they function as isolation
A. QUANTUM-LIMITED AMPLIFIERS devices that prevent noise originating in the readout ampli-
As described in Section V, the readout of superconducting fication chain from impinging on the qubits. Their second
qubits and semiconducting qubits involves microwave signals use is in the context of parametric amplifiers (see above) that
so weak that they are not far above the quantum noise floor operate in reflective mode. Here, circulators are configured
of half a photon per unit bandwidth, S(ω) = ω/2. Faithful to separate input and output signals as well as to isolate the
amplification of these signals requires amplifiers with noise qubits from the pump tones that supply energy to the amplifier.
performance at or near the quantum limit, which at 6 GHz To achieve non-reciprocity, traditional microwave circulators
corresponds to a noise temperature of 144 mK. exploit ferromagnetic materials. These devices are necessarily
To achieve this noise performance, the field has turned to large and bulky components, since they make use of inter-
superconducting parametric amplifiers. Parametric amplifiers ference effects that occur over a length scale comparable to
rely on a nonlinear element or elements whose parameters the microwave wavelength. Further limitations include their
(inductance or capacitance) are modulated in time by a strong typically narrow-band performance, insertion loss, limited
pump. This modulation transfers power from the pump into non-reciprocity, and variation in specifications or failure when
other modes (known as the signal and the idler), coherently operated at cryogenic temperatures or in large magnetic fields.
amplifying the energy in those modes [92], [120]. From the perspective of scaling up to the number of readout
Before the advent of high-electron-mobility transistor channels needed for a large quantum system, the footprint
(HEMT) amplifiers, microwave amplification was sometimes alone of conventional circulators is a significant barrier to the
accomplished using parametric amplifiers based on varactor development of tightly integrated systems.
diodes or inductors with saturable cores [121]. Parametric am- Recently, efforts have focused on realising miniaturized
plifiers based on the Josephson effect in superconductors have devices that exhibit nonreciprocity. One avenue uses ac-
been studied intermittently since the 1960s, but the field was tive devices, specifically Josephson parametric amplifiers de-
revitalized by work in the late 2000s on Josephson paramet- signed to achieve directional gain and reverse isolation while
ric amplifiers with near-quantum-limited noise performance maintaining quantum-limited noise performance [130]–[132],

420 VOLUME 1, NO. 1, JANUARY 2021

 of error-prone physical qubits to implement “logical qubits,”
which have reduced rates of error in comparison to their
constituent physical qubits. The degree to which errors are
suppressed depends both upon the error rates of the physical
qubits and the degree of redundancy. However, for such a
protocol to work in the first place, the physical qubit error
rates must be on average below a threshold, which depends
on the specifics of the QEC protocol. There are many differ-
ent QEC schemes [48], [49], with physical error thresholds
ranging from ∼ 10−4 up to about 1 %. Once the physical
error rates are below the threshold, the degree of redundancy
FIGURE 13. On-chip microwave circulator based on the quantum Hall
effect. Microwave excitations travel at a velocity 1000 times slower than in (the number of physical qubits) required to achieve a specific
vacuum, enabling an equivalent reduction in the signal wavelength and logical error rate scales logarithmically with further reduction
device footprint [134]. in the physical error rates. A practical target of exceeding the
threshold by an order of magnitude (0.1% physical qubit error
rates for some codes) appears feasible, but still requires about
[135], [136]. Ref. [133] provides an in-depth review of nonre- 1,000 physical qubits per logical qubit. Thus, it is estimated
ciprocal active circuits with quantum-limited noise. Such de- that realizing a practical system with 1,000 logical qubits
vices have recently been demonstrated for direct, high-fidelity (far below the number required to break RSA encryption)
readout of superconducting qubits without external circula- will require building a million-qubit-scale system [51]. There
tors [137]–[139]. are numerous microwave-related challenges associated with
Miniaturized passive circulators also hold promise for building such a computer. Here we provide a discussion of
quantum computing applications. For example, Fig. 13 some of these challenges.
shows a 3-port circulator device based on the inherent non- Today’s quantum processors are controlled and measured
reciprocity of the quantum Hall effect [134]. Here, the size using systems that mostly comprise room temperature elec-
of the device is 1/1000th of the free-space wavelength of tronics, with high-performance microwave interconnects used
the microwave signals it handles. This small size is achieved to interface to the quantum processor. Let us begin by con-
by exploiting the chiral, “slow-light” response of a two- sidering the feasibility of scaling this approach to the mil-
dimensional electron gas in the quantum Hall regime. For an lion qubit level. It will be helpful to start by quantifying the
integrated GaAs device with 330 μm diameter and 1 GHz number of control and measurement channels required for a
center frequency, a nonreciprocity of 25 dB is observed over brute force scaling approach. For a large-scale quantum pro-
a 50 MHz bandwidth. Furthermore, the nonreciprocity can be cessor implemented using superconducting transmon qubits
dynamically tuned by varying the voltage at the port, an aspect arranged in the architecture of [7], processor operation re-
that may enable reconfigurable passive routing of microwave quires one XY channel and three Z control channels per qubit.
signals on chip. The forward transmission of this particular In addition, one readout channel is required for every ∼10
device was limited to −20 dB due to the impedance mismatch qubits. In the case of semiconductor spin qubits a handful of
between 50  environment and the device impedance, which dc bias wires per qubit are needed to define tunnel barriers
is set by the resistance quantum RK ≈ 25 k. However, im- and chemical potentials. Depending on the flavor of spin qubit
proved impedance matching design could be used to increase at least one high bandwidth channel is required per qubit,
the forward transmission [140]. Although the quantum Hall with additional channels, shifted in phase, often used to cancel
device requires a large external magnetic field, magnetically- crosstalk. Readout of spin qubits based on rf reflectometry re-
doped topological insulator materials offer a further route to quires a single wideband line per ∼10 qubits using frequency
realizing non-reciprocal devices without the need for such ex- multiplexing. Trapped ion systems generally rely on their long
ternal magnetic fields [141]. Beyond non-reciprocal behavior, qubit coherence times, as well as the ability to reconfigure
the underlying physics of these systems suggests their use as ion and/or laser beam positions in the trap, to perform logic
compact and tunable delay lines, microwave interferometers, operations on different subsets of qubits in series, rather than
and low-loss, high impedance transmission lines [142]. performing simultaneous operations on all qubits. This means
For future quantum computing systems, miniaturized, on- that the scaling of microwave control hardware does not have
chip non-reciprocal microwave devices of the kinds described a precise relationship to the scaling of qubit numbers. How-
above are likely to play a key role. ever, to avoid excessive slowing of computations, the serial-
ization would ideally be at most a factor of tens, not hundreds
VIII. MICROWAVE CHALLENGES IN REACHING THE FULL or larger.
POTENTIAL OF QUANTUM COMPUTING Given these constraints, is does not appear feasible to scale
In order to realize the full promise of quantum computing, existing technology to the level demanded by large-scale
it will be necessary to build systems capable of executing error-corrected quantum computers. For instance, if one were
QEC protocols. In general, these protocols use ensembles to build a quantum control and measurement system for a

VOLUME 1, NO. 1, JANUARY 2021 421

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

million transmon qubits using present Google technology [7], Two distinct approaches are currently being pursued to ad-
it would occupy over 15,000 m2 of floor space and dissipate dress this trade-off between the I/O bottleneck and qubit tem-
about 40 MW (not including the dissipation of the amplifiers perature. The first involves relocating the spin qubit platform
that would be required to compensate for loss in the cable and integrated control circuits to higher temperatures (near 1
runs). K), where substantially more cooling power is available by
using pumped helium-4 [27], [28]. However, this approach
A. SCALING OF INTERCONNECTS causes reductions in qubit fidelity, eventually requiring orders
Beyond the sheer volume of the quantum control and mea- of magnitude more noisy physical qubits to encode a logical
surement electronics, one must also consider the feasibility qubit. The increased qubit count then requires additional I/O
of connecting the electronics to a quantum processor. Here, and control electronics that may effectively cancel the gains
the fundamental constraint is going to be the ratio of the from operating at higher temperature. An alternate approach
required number of interconnects to the chip surface area, involves operating the control sub-systems at the same mil-
which sets the required interconnect density. For large-scale likelvin temperature as the qubits, but on a separate chip
quantum processors, this simplifies to the ratio of the number that is thermally decoupled [148]. Taking advantage of the
of I/O signals required per qubit to the qubit pitch squared high impedance (open circuit) nature of gate electrodes and
(assuming a 2D array). For superconducting qubits, which leveraging lithographically defined chip-to-chip interconnect
are arranged on a pitch of about 1 mm and require about strategies, this millikelvin approach alleviates the power dis-
4.2 lines per qubit, an interconnect density of about 4.2 lines sipation otherwise required to drive low impedance cables
per square millimeter is required; while this is believed to between temperature stages.
be feasible, it will require significant microwave engineering Large-scale trapped-ion quantum computers will likely op-
to deliver these signals while maintaining low crosstalk and erate cryogenically, in the 4 K to 10 K temperature range,
avoiding dissipation. For instance, a million qubit quantum because of superior vacuum pressure (enabling ions to remain
processor will require about 3 million Z control lines, and to trapped for much longer times) and reduced electric field noise
keep heating due to the ≈ 50 μA static Z currents to below (improving motional coherence and entangling gate fidelities).
10 μW (necessary because of the limited available cooling The much larger cooling powers available at these tempera-
power at mK temperatures), the contact resistance between tures make the interconnect problem less formidable, although
the qubit package and the cable assembly must be kept to the still not easy. Because many ions can be addressed with a
m level. Developing qubit packaging techniques that enable single microwave control line, a relatively small number of
high coherence and low crosstalk while achieving this speci- control lines will be needed compared to the other qubit tech-
fication will require significant research. Superconducting in- nologies. However, because the control signals have much
terconnects, which offer near-lossless and near-dispersionless higher power, on-chip dissipation may present an issue. Ion
electrical propagation [143] while providing very low thermal traps rely on a very large number of static or slowly-varying
conductivity, can be used to transport signals from 10 mK control voltages for defining trap potentials and moving ions
to 4 K. However, achieving the required interconnect density around the trap, with roughly 10 times as many control volt-
between 4 K and room temperature (where superconducting ages as potential wells (each of which may hold single or
interconnects are not an option) may be challenging due to multiple ions) in the trap. However, these lines draw zero or
dispersion and losses, both of which distort control wave- minimal current and do not require high bandwidth (< 1 MHz
forms. As such, the control and measurement system may is generally ample), so they can be made from very thin, low-
have to reside partially or entirely at 4 K. Regardless, consid- thermal-conductivity wires. Research into generating these
erable work is required to develop interconnect systems be- voltages with DACs fabricated in the trap substrate itself is
tween 4 K and 10 mK which offer reproducible performance being pursued by some groups [149].
and the stringent cross-talk performance that is required for a
large-scale quantum computer. B. SCALING OF CONTROL SYSTEMS
For semiconductor spin qubits, the nanoscale qubit dimen- Today’s quantum computers are essentially research devices
sions make interfacing a large scale quantum processor to an with individual qubit performance just approaching the edge
external control system impractical without a large degree of of what is required to implement QEC protocols. Developing
multiplexing [144], [145]. Consider, for instance, that as many control protocols for use on these prototype systems has ne-
as 10 wires are required within the 100 nm × 100 nm footprint cessitated the use of flexible, high-speed arbitrary waveform
of a spin qubit. This geometric I/O bottleneck motivates the generators so that a researcher can quickly test new control
integration of cryogenic electronics with the qubit platform paradigms without developing new hardware. While this is a
to handle signal generation and multiplexing [146] without logical approach for operating today’s relatively small-scale
needing large cable assemblies that carry signals to room quantum processors, the cost, size, and power associated with
temperature electronics [147]. The power dissipation of these scaling this approach to run a million-qubit quantum proces-
classical electronics can be significant, however, leading to sor motivate the development of a more optimized approach.
heating of the qubits and a degradation in fidelity if tightly Several research groups are currently investigating the in-
integrated in a monolithic configuration. tegration of quantum control circuits, targeting operation

422 VOLUME 1, NO. 1, JANUARY 2021

at a physical temperature of 4 K [150]–[155]. However, currently at room temperature can be directly integrated with
the requirements for these systems are stringent and co- the low noise amplifier and thermalized to 4 K. While SiGe
optimization of the classical controller and the quantum pro- processes appear to be a promising technology for meeting
cessor will likely be required. For a device thermalized at these goals [159], [160], research is still required to deter-
4 K, the power consumption will be limited to well below mine if it is possible to achieve both the performance and
1 mW per control channel, and this tight power budget must repeatability required using SiGe technologies. Additionally,
be met without increasing control errors. For XY controllers, work is required to minimize the volume of the microwave
strategies must be developed to minimize the number of mi- electronics that must be thermalized to the 10 mK stage (i.e.
crowave carriers required, mitigate crosstalk, optimize pulse the parametric amplifiers and circulators).
waveforms, and maintain phase coherence among the ∼1
million XY control channels. The same considerations, albeit
D. OTHER MICROWAVE CHALLENGES
with somewhat relaxed power restrictions, are also relevant
Beyond the classical to quantum interface, there are signif-
for room-temperature control electronics.
icant microwave challenges in the design of the quantum
Similar challenges are present for the development of the
processor itself. For instance, considering superconducting
scalable baseband waveform generators required for Z control.
qubits, as the dimensions of the quantum processor grow,
In particular, losses and reflections along the interconnects
designers will have to rely more heavily on electromagnetic
between the Z controllers and the qubit/coupler control ports
simulation tools to predict and avoid undesired moding in
cause distortion in the same way that non-return-to-zero sig-
what will eventually become wafer-scale devices. Modeling
nals experience distortion when traveling over a backplane.
these effects requires incorporating superconducting physics
In today’s systems, the long cables between the room temper-
and the cryogenic properties of dielectrics into electromag-
ature electronics and the quantum processor lead to settling
netic tools while developing methods to efficiently solve for
times that can be on the order of microseconds [147], [156].
high-Q resonances in large structures. Microwave expertise
To compensate for these long settling times, this response is
will be required, both in developing efficient simulation tools
often characterized and deconvolved from the transmitted Z
tailored to this purpose as well as in developing techniques to
signal, similar to the use of pre-emphasis in wireline applica-
mitigate these undesired modes.
tions. While the use of bipolar Z pulses reduces the impact of
As the system size grows, it may also become important
long-term settling, the short time-scale system response must
to develop techniques to predict system performance from
still be compensated [157]. As such, research will be required
within a single design environment (similar to the infrastruc-
to architect Z control systems that are both precise and low
ture that has been developed for digital design). This may
power.
involve developing qubit models which are compatible with
commercial circuit solvers (e.g. [161]) and tying the classical
C. SCALING OF READOUT
and quantum systems together in a circuit solver tool. For
Scaling of readout is also of critical importance and mi-
systems such as superconducting quantum computers, which
crowave innovations will certaintly be needed. For super-
are spatially distributed, it may also be necessary to develop
conducting and semiconductor spin qubits using dispersive
techniques to accurately model the interconnects between the
readout, scalable amplification chains achieving near-
classical control system and the quantum processor.
quantum-limited noise performance are required. Assuming a
frequency domain multiplexing factor of 10×, a million-qubit
quantum controller will require 100,000 readout channels. In IX. CONCLUSION
today’s systems, each readout channel contains a paramet- Microwave technology has played a key role in the rise of
ric amplifier and several circulators (up to 5) thermalized to quantum computing, and the two fields will continue to in-
10 mK, a HEMT-based low noise amplifier thermalized to teract synergistically in the years ahead. The pursuit of an
4 K, and further amplification and digitization at room tem- error-corrected quantum computer has been enabled by mi-
perature. In order to scale to the levels required to implement crowave technologies made available by the explosive growth
a practical error-corrected quantum computer, each of these of the wireless communication industry. However, reaching
technologies will have to be optimized for manufacturability, the performance currently achieved by today’s state-of-the-art
cost, performance, and size. For instance, the semiconductor quantum processors has also necessitated the development of
low noise amplifiers used at 4 K (e.g., [158]) achieve excel- new microwave technologies, such as quantum-limited para-
lent noise temperatures, but are hand assembled, hand tested, metric amplifiers.
and dissipate about an order of magnitude too much power Today, there is a growing race to implement a fault-tolerant
for use in a large-scale system. As with the case of control quantum computer. As described in this article, numerous
systems, it is necessary that scalable readout systems be in- microwave-related challenges must be overcome to build such
tegrated to the maximum extent possible. Amplifiers that can a device, and continued microwave innovation will certainly
be mass manufactured in a silicon technology platform and be required. Critical areas for further research range from the
achieve similar performance to today’s HEMT LNAs while development of new ultra-low-loss interconnect systems to
dissipating  1 mW should be developed so that electronics the design of highly efficient quantum control systems. The

VOLUME 1, NO. 1, JANUARY 2021 423

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

active engagement of microwave engineers in this exciting [24] A. Sigillito et al., “Site-selective quantum control in an isotopically
and important effort will be essential to its success. enriched 28 Si/Si0.7 Ge0.3 quadruple quantum dot,” Phys. Rev. Appl.,
vol. 11, Jun. 2019, Art. no. 061006.
[25] J. Wrachtrup and F. Jelezko, “Processing quantum information in di-
amond,” J. Phys., Condens. Matter, vol. 18, no. 21, pp. S807–S824,
ACKNOWLEDGMENT May 2006.
The authors would like to thank J. Aumentado, K. Beloy, O. [26] L. P. Kouwenhoven et al., Electron. Transport Quantum Dots. Dor-
drecht, The Netherlands: Springer Netherlands, 1997, pp. 105–214.
Naaman, H. Neven, and L. J. Stephenson for a careful reading [27] C. H. Yang et al., “Operation of a silicon quantum processor unit cell
of the manuscript, and M. Foss-Feig, D. Sank, and K. C. above one kelvin,” Nature, vol. 580, no. 7803, pp. 350–354, 2020.
Young for helpful discussions. [28] L. Petit et al., “Universal quantum logic in hot silicon qubits,” Nature,
vol. 580, no. 7803, pp. 355–359, 2020.
[29] D. Loss and D. P. DiVincenzo, “Quantum computation with quantum
dots,” Phys. Rev. A, vol. 57, pp. 120–126, Jan. 1998.
REFERENCES [30] A. Morello et al., “Single-shot readout of an electron spin in silicon,”
[1] R. Buderi, The Invention That Changed the World: How a Small Nature, vol. 467, no. 7316, pp. 687–91, 2010.
Group of Radar Pioneers Won the Second World War and Launched a [31] J. Levy, “Universal quantum computation with spin-1/2 pairs and
Technological Revolution (Sloan Technology Series). New York, NY, Heisenberg exchange,” Phys. Rev. Lett., vol. 89, Sep. 2002, Art.
USA: Simon & Schuster, 1996. no. 147902.
[2] C. H. Townes and A. L. Schawlow, Microwave Spectroscopy. Mineola, [32] X. Croot et al., “Device architecture for coupling spin qubits via an
NY, USA: Dover, 1975. intermediate quantum state,” Phys. Rev. Appl., vol. 10, Oct. 2018, Art.
[3] C. P. Slichter, Principles of Magnetic Resonance (Springer Series in no. 044058.
Solid-State Sciences). Berlin, Germany: Springer, 1996. [33] D. P. DiVincenzo, D. Bacon, J. Kempe, G. Burkard, and K. B. Wha-
[4] E. M. Purcell, H. C. Torrey, and R. V. Pound, “Resonance absorp- ley, “Universal quantum computation with the exchange interaction,”
tion by nuclear magnetic moments in a solid,” Phys. Rev., vol. 69, Nature, vol. 408, no. 6810, pp. 339–342, 2000.
pp. 37–38, 1946. [34] J. Medford et al., “Quantum-dot-based resonant exchange qubit,”
[5] F. Bloch, W. W. Hansen, and M. Packard, “Nuclear induction,” Phys. Phys. Rev. Lett., vol. 111, Jul. 2013, Art. no. 050501.
Rev., vol. 69, pp. 127–127, 1946. [35] R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K.
[6] J. C. Bardin, D. Sank, O. Naaman, and E. Jeffrey, “Quantum comput- Vandersypen, “Spins in few-electron quantum dots,” Rev. Mod. Phys.,
ing: An introduction for microwave engineers,” IEEE Microw. Mag., vol. 79, pp. 1217–1265, Oct. 2007.
vol. 21, no. 8, pp. 24–44, Aug. 2020. [36] J. M. Martinis, M. H. Devoret, and J. Clarke, “Energy-level quantiza-
[7] F. Arute et al., “Quantum supremacy using a programmable su- tion in the zero-voltage state of a current-biased Josephson junction,”
perconducting processor,” Nature, vol. 574, no. 7779, pp. 505–510, Phys. Rev. Lett., vol. 55, no. 15, pp. 1543–1546, 1985.
2019. [37] Y. Nakamura, Y. A. Pashkin, and J. S. Tsai, “Coherent control of
[8] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum macroscopic quantum states in a single-Cooper-pair box,” Nature,
Information. Cambridge, U.K.: Cambridge Univ. Press, 2000. vol. 398, no. 6730, pp. 786–788, 1999.
[9] P. W. Shor, “Polynomial-time algorithms for prime factorization and [38] J. Koch et al., “Charge-insensitive qubit design derived from the
discrete logarithms on a quantum computer,” SIAM Rev., vol. 41, no. 2, Cooper pair box,” Phys. Rev. A, vol. 76, no. 4, 2007, Art. no. 042319.
p. 303, 1999. [39] P. Krantz et al., “A quantum engineer’s guide to superconducting
[10] D. J. Wineland et al., “Experimental issues in coherent quantum-state qubits,” Appl. Phys. Rev., vol. 6, no. 2, 2019, Art. no. 021318.
manipulation of trapped atomic ions,” J. Res. Nat. Inst. Stand. Tech- [40] Z. Chen, “Metrology of quantum control and measurement in super-
nol., vol. 103, no. 3, pp. 259–328, 1998. conducting qubits,” Ph.D. dissertation, UC Santa Barbara, 2018.
[11] D. Leibfried, R. Blatt, C. Monroe, and D. Wineland, “Quantum dy- [41] A. P. Place et al., “New material platform for superconducting trans-
namics of single trapped ions,” Rev. Mod. Phys., vol. 75, no. 1, mon qubits with coherence times exceeding 0.3 milliseconds,” 2020,
pp. 281–324, 2003. arXiv:2003.00024.
[12] C. D. Bruzewicz, J. Chiaverini, R. McConnell, and J. M. Sage, [42] P. Jurcevic et al., “Demonstration of quantum volume 64 on a super-
“Trapped-ion quantum computing: Progress and challenges,” Appl. conducting quantum computing system,” 2020, arXiv:2008.08571.
Phys. Rev., vol. 6, no. 2, 2019, Art. no. 021314. [43] P. Klimov et al., “Fluctuations of energy-relaxation times in supercon-
[13] S. Seidelin et al., “Microfabricated surface-electrode ion trap for scal- ducting qubits,” Phys. Rev. Lett., vol. 121, no. 9, 2018, Art. no. 090502.
able quantum information processing,” Phys. Rev. Lett., vol. 96, no. 25, [44] F. H. L. Koppens et al., “Driven coherent oscillations of a single elec-
2006, Art. no. 253003. tron spin in a quantum dot,” Nature, vol. 442, no. 7104, pp. 766–771,
[14] K. Wright et al., “Benchmarking an 11-qubit quantum computer,” 2006.
Nature Commun., vol. 10, no. 1, 2019, Art. no. 5464. [45] J. van Dijk et al., “Impact of classical control electronics on qubit
[15] J. M. Pino et al., “Demonstration of the QCCD trapped-ion quantum fidelity,” Phys. Rev. Appl., vol. 12, no. 4, 2019, Art. no. 044054.
computer architecture,” 2020, arXiv:2003.01293. [46] C. Ospelkaus et al., “Microwave quantum logic gates for trapped ions,”
[16] J. Eschner, G. Morigi, F. Schmidt-Kaler, and R. Blatt, “Laser cooling Nature, vol. 476, no. 7359, pp. 181–184, 2011.
of trapped ions,” J. Opt. Soc. Amer. B, vol. 20, no. 5, pp. 1003–1015, [47] A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Marquardt, and R.
2003. J. Schoelkopf, “Introduction to quantum noise, measurement, and
[17] T. Ruster et al., “A long-lived Zeeman trapped-ion qubit,” Appl. Phys. amplification,” Rev. Modern Phys., vol. 82, no. 2, pp. 1155–1208,
B, vol. 122, no. 10, 2016, Art. no. 254. 2010.
[18] T. P. Harty et al., “High-fidelity preparation, gates, memory, and read- [48] T. A. Brun, “Quantum error correction,” 2020, arXiv:1910.03672.
out of a trapped-ion quantum bit,” Phys. Rev. Lett., vol. 113, no. 22, [49] J. Roffe, “Quantum error correction: an introductory guide,” Contem-
2014, Art. no. 220501. porary Phys., vol. 60, no. 3, pp. 226–245, 2019.
[19] P. Wang et al., “Single ion-qubit exceeding one hour coherence time,” [50] E. Knill, “Quantum computing with realistically noisy devices,” Na-
2020, arXiv:2008.00251. ture, vol. 434, no. 7029, pp. 39–44, 2005.
[20] H. Ball, W. D. Oliver, and M. J. Biercuk, “The role of master clock [51] A. G. Fowler, M. Mariantoni, J. M. Martinis, and A. N. Cleland,
stability in quantum information processing,” npj Quantum Inf., vol. 2, “Surface codes: Towards practical large-scale quantum computation,”
no. 1, 2016, Art. no. 16033. Phys. Rev. A, vol. 86, no. 3, 2012, Art. no. 032324.
[21] M. A. Sepiol et al., “Probing qubit memory errors at the part-per- [52] M. Steffen, J. M. Martinis, and I. L. Chuang, “Accurate control of
million level,” Phys. Rev. Lett., vol. 123, no. 11, 2019, Art. no. 110503. Josephson phase qubits,” Phys. Rev. B, vol. 68, no. 22, 2003, Art.
[22] B. E. Kane, “A silicon-based nuclear spin quantum computer,” Nature, no. 224518.
vol. 393, no. 6681, pp. 133–137, 1998. [53] F. Motzoi, J. M. Gambetta, P. Rebentrost, and F. K. Wilhelm, “Simple
[23] F. A. Zwanenburg et al., “Silicon quantum electronics,” Rev. Mod. pulses for elimination of leakage in weakly nonlinear qubits,” Phys.
Phys., vol. 85, pp. 961–1019, Jul. 2013. Rev. Lett., vol. 103, no. 11, 2009, Art. no. 110501.

424 VOLUME 1, NO. 1, JANUARY 2021

[54] J. M. Chow et al., “Optimized driving of superconducting artificial [80] G. Zarantonello et al., “Robust and resource-efficient microwave near-
atoms for improved single-qubit gates,” Phys. Rev. A, vol. 82, no. 4, field entangling 9 Be+ gate,” Phys. Rev. Lett., vol. 123, no. 26, 2019,
2010, Art. no. 040305. Art. no. 260503.
[55] Z. Chen et al., “Measuring and suppressing quantum state leakage in [81] V. M. Schäfer et al., “Fast quantum logic gates with trapped-ion
a superconducting qubit,” Phys. Rev. Lett., vol. 116, no. 2, 2016, Art. qubits,” Nature, vol. 555, no. 7694, pp. 75–78, 2018.
no. 020501. [82] R. Ozeri et al., “Errors in trapped-ion quantum gates due to spon-
[56] D. P. L. Aude Craik et al., “High-fidelity spatial and polarization taneous photon scattering,” Phys. Rev. A, vol. 75, no. 4, 2007, Art.
addressing of 43 Ca+ qubits using near-field microwave control,” Phys. no. 042329.
Rev. A, vol. 95, no. 2, 2017, Art. no. 022337. [83] J. Raftery et al., “Direct digital synthesis of microwave waveforms for
[57] C. Piltz, T. Sriarunothai, A. F. Varón, and C. Wunderlich, “A trapped- quantum computing,” 2017, arXiv:1703.00942.
ion-based quantum byte with 10−5 next-neighbour cross-talk,” Nature [84] W. D. Kalfus et al., “High-fidelity control of superconducting qubits
Commun., vol. 5, no. 1, 2014, Art. no. 4679. using direct microwave synthesis in higher Nyquist zones,” 2020,
[58] D. Leibfried, “Individual addressing and state readout of trapped arXiv:2008.02873.
ions utilizing RF micromotion,” Phys. Rev. A, vol. 60, no. 5, [85] A. Blais, R.-S. Huang, A. Wallraff, S. M. Girvin, and R. J. Schoelkopf,
pp. R3335–R3338, 1999. “Cavity quantum electrodynamics for superconducting electrical cir-
[59] P. Staanum, and M. Drewsen, “Trapped-ion quantum logic utilizing cuits: An architecture for quantum computation,” Phys. Rev. A, vol. 69,
position-dependent ac Stark shifts,” Phys. Rev. A, vol. 66, no. 4, 2002, no. 6, 2004, Art. no. 062320.
Art. no. 040302. [86] A. Wallraff et al., “Approaching unit visibility for control of a super-
[60] U. Warring et al., “Individual-ion addressing with microwave field conducting qubit with dispersive readout,” Phys. Rev. Lett., vol. 95,
gradients,” Phys. Rev. Lett., vol. 110, no. 17, 2013, Art. no. 173002. no. 6, 2005, Art. no. 060501.
[61] R. Srinivas, “Laser-free trapped-ion quantum logic with a radiofre- [87] X. Mi, J. Cady, D. Zajac, P. Deelman, and J. Petta, “Strong coupling of
quency magnetic field gradient,” Ph.D. dissertation, Univ. Colorado, a single electron in silicon to a microwave photon,” Science, vol. 355,
Boulder, 2020. no. 6321, pp. 156–158, 2017.
[62] K. Takeda, A. Noiri, J. Yoneda, T. Nakajima, and S. Tarucha, “Res- [88] A. Stockklauser et al., “Strong coupling cavity QED with gate-defined
onantly driven singlet-triplet spin qubit in silicon,” Phys. Rev. Lett., double quantum dots enabled by a high impedance resonator,” Phys.
vol. 124, Mar. 2020, Art. no. 117701. Rev. X, vol. 7, no. 1, 2017, Art. no. 011030.
[63] D. C. McKay, C. J. Wood, S. Sheldon, J. M. Chow, and J. M. Gam- [89] A. West et al., “Gate-based single-shot readout of spins in silicon,”
betta, “Efficient Z gates for quantum computing,” Phys. Rev. A, vol. 96, Nature Nanotechnol., vol. 14, no. 5, pp. 437–441, 2019.
no. 2, 2017, Art. no. 022330. [90] M. Boissonneault, J. M. Gambetta, and A. Blais, “Dispersive regime of
[64] R. Srinivas et al., “Trapped-ion spin-motion coupling with microwaves circuit QED: Photon-dependent qubit dephasing and relaxation rates,”
and a near-motional oscillating magnetic field gradient,” Phys. Rev. Phys. Rev. A, vol. 79, no. 1, 2009, Art. no. 013819.
Lett., vol. 122, no. 16, 2019, Art. no. 163201. [91] D. H. Slichter et al., “Measurement-induced qubit state mixing in
[65] R. Barends et al., “Superconducting quantum circuits at the sur- circuit QED from up-converted dephasing noise,” Phys. Rev. Lett.,
face code threshold for fault tolerance,” Nature, vol. 508, no. 7497, vol. 109, no. 15, 2012, Art. no. 153601.
pp. 500–503, 2014. [92] J. Aumentado, “Superconducting parametric amplifiers: The state of
[66] F. Yan et al., “Tunable coupling scheme for implementing high- the art in Josephson parametric amplifiers,” IEEE Microw. Mag.,
fidelity two-qubit gates,” Phys. Rev. App., vol. 10, no. 5, 2018, vol. 21, no. 8, pp. 45–59, Aug. 2020.
Art. no. 054062. [93] R. Vijay, D. H. Slichter, and I. Siddiqi, “Observation of quantum jumps
[67] B. Foxen et al., “Demonstrating a continuous set of two-qubit gates in a superconducting artificial atom,” Phys. Rev. Lett., vol. 106, no. 11,
for near-term quantum algorithms,” 2020, arXiv:2001.08343. 2011, Art. no. 110502.
[68] C. Rigetti and M. Devoret, “Fully microwave-tunable universal gates [94] E. Jeffrey et al., “Fast accurate state measurement with superconduct-
in superconducting qubits with linear couplings and fixed transition ing qubits,” Phys. Rev. Lett., vol. 112, no. 19, 2014, Art. no. 190504.
frequencies,” Phys. Rev. B, vol. 81, no. 13, 2010, Art. no. 134507. [95] M. D. Reed et al., “Fast reset and suppressing spontaneous emission
[69] M. D. Shulman et al., “Demonstration of entanglement of electro- of a superconducting qubit,” Appl. Phys. Lett., vol. 96, no. 20, 2010,
statically coupled singlet-triplet qubits,” Science, vol. 336, no. 6078, Art. no. 203110.
pp. 202–205, 2012. [96] A. Opremcak et al., “Measurement of a superconducting qubit
[70] A. A. Clerk, K. W. Lehnert, P. Bertet, J. R. Petta, and Y. Nakamura, with a microwave photon counter,” Science, vol. 361, no. 6408,
“Hybrid quantum systems with circuit quantum electrodynamics,” Na- pp. 1239–1242, 2018.
ture Phys., vol. 16, no. 3, pp. 257–267, 2020. [97] A. Opremcak et al., “High-fidelity measurement of a supercon-
[71] F. Mintert and C. Wunderlich, “Ion-trap quantum logic using long- ducting qubit using an on-chip microwave photon counter,” 2020,
wavelength radiation,” Phys. Rev. Lett., vol. 87, no. 25, 2001, Art. arXiv:2008.02346.
no. 257904. [98] H. G. Dehmelt “Monoion oscillator as potential ultimate laser fre-
[72] M. Johanning et al., “Individual addressing of trapped ions and cou- quency standard,” IEEE Trans. Instrum. Meas., vol. IM-31, no. 2,
pling of motional and spin states using RF radiation,” Phys. Rev. Lett., pp. 83–87, Jun. 1982.
vol. 102, no. 7, 2009, Art. no. 073004. [99] A. H. Myerson et al., “High-fidelity readout of trapped-ion qubits,”
[73] R. T. Sutherland et al., “Versatile laser-free trapped-ion entangling Phys. Rev. Lett., vol. 100, no. 20, 2008, Art. no. 200502.
gates,” New J. Phys., vol. 21, no. 3, 2019, Art. no. 033033. [100] J. W. Britton et al., “Engineered two-dimensional Ising interactions
[74] C. Ospelkaus et al., “Trapped-ion quantum logic gates based on os- in a trapped-ion quantum simulator with hundreds of spins,” Nature,
cillating magnetic fields,” Phys. Rev. Lett., vol. 101, no. 9, 2008, Art. vol. 484, no. 7395, pp. 489–492, 2012.
no. 090502. [101] J. Dilling, K. Blaum, M. Brodeur, and S. Eliseev, “Penning-trap mass
[75] U. Warring et al., “Techniques for microwave near-field quantum measurements in atomic and nuclear physics,” Annu. Rev. Nucl. Parti-
control of trapped ions,” Phys. Rev. A, vol. 87, no. 1, 2013, Art. cle Sci., vol. 68, no. 1, pp. 45–74, 2018.
no. 013437. [102] S. Jain, J. Alonso, M. Grau, and J. P. Home, “Scalable arrays of micro-
[76] J. Welzel, F. Stopp, and F. Schmidt-Kaler, “Spin and motion dynamics penning traps for quantum computing and simulation,” Phys. Rev. X,
with zigzag ion crystals in transverse magnetic gradients,” J. Phys. B, vol. 10, no. 3, 2020, Art. no. 031027.
Atomic Mol. Opt. Phys., vol. 52, no. 2, 2019, Art. no. 025301. [103] W. Macalpine and R. Schildknecht, “Coaxial resonators with he-
[77] T. P. Harty et al., “High-fidelity trapped-ion quantum logic using lical inner conductor,” Proc. IRE, vol. 47, no. 12, pp. 2099–2105,
near-field microwaves,” Phys. Rev. Lett., vol. 117, no. 14, 2016, Art. 1959.
no. 140501. [104] J. D. Siverns, L. R. Simkins, S. Weidt, and W. K. Hensinger, “On
[78] S. Weidt et al., “Trapped-ion quantum logic with global radiation the application of radio frequency voltages to ion traps via helical
fields,” Phys. Rev. Lett., vol. 117, no. 22, 2016, Art. no. 220501. resonators,” Appl. Phys. B, vol. 107, no. 4, pp. 921–934, 2012.
[79] H. Hahn et al., “Integrated 9 Be+ multi-qubit gate device for the ion- [105] S. R. Jefferts, C. Monroe, E. W. Bell, and D. J. Wineland, “Coaxial-
trap quantum computer,” npj Quantum Inf., vol. 5, no. 1, 2019, Art. resonator-driven RF (Paul) trap for strong confinement,” Phys. Rev. A,
no. 70. vol. 51, no. 4, pp. 3112–3116, 1995.

VOLUME 1, NO. 1, JANUARY 2021 425

BARDIN ET AL.: MICROWAVES IN QUANTUM COMPUTING

[106] D. Gandolfi, M. Niedermayr, M. Kumph, M. Brownnutt, and R. Blatt, [134] A. C. Mahoney et al., “On-chip microwave quantum hall circulator,”
“Compact radio-frequency resonator for cryogenic ion traps,” Rev. Sci. Phys. Rev. X, vol. 7, no. 1, 2017, Art. no. 011007.
Instrum., vol. 83, no. 8, 2012, Art. no. 084705. [135] K. M. Sliwa et al., “Reconfigurable Josephson circulator / directional
[107] M. F. Brandl, P. Schindler, T. Monz, and R. Blatt, “Cryogenic res- amplifier,” Phys. Rev. X, vol. 5, Aug. 2015, Art. no. 041020.
onator design for trapped ion experiments in Paul traps,” Appl. Phys. [136] F. Lecocq et al., “Microwave measurement beyond the quantum limit
B, vol. 122, no. 6, 2016, Art. no. 157. with a nonreciprocal amplifier,” Phys. Rev. Appl., vol. 13, no. 4, 2020,
[108] D. T. C. Allcock, “Surface-electrode ion traps for scalable quantum Art. no. 044005.
computing,” Ph.D. dissertation, Oxford Univ., 2011. [137] B. Abdo, O. Jinka, N. T. Bronn, S. Olivadese, and M. Brink, “On-chip
[109] T. P. Harty, “High-fidelity quantum logic in intermediate-field 43Ca+,” single-pump interferometric Josephson isolator for quantum measure-
Ph.D. dissertation, Oxford Univ., 2013. ments,” 2020, arXiv:2006.01918.
[110] K. G. Johnson et al., “Active stabilization of ion trap radiofrequency [138] E. I. Rosenthal et al., “Efficient and low-backaction quantum measure-
potentials,” Rev. Sci. Instrum., vol. 87, no. 5, 2016, Art. no. 053110. ment using a chip-scale detector,” 2020, arXiv:2008.03805.
[111] M. Saffman, “Quantum computing with atomic qubits and Rydberg [139] F. Lecocq et al., “Efficient qubit measurement with a nonreciprocal
interactions: Progress and challenges,” J. Phys. B, Atomic Mol. Opt. microwave amplifier,” 2020, arXiv:2009.08863.
Phys., vol. 49, no. 20, 2016, Art. no. 202001. [140] S. Bosco, F. Haupt, and D. P. DiVincenzo, “Self-impedance-
[112] L. Henriet et al., “Quantum computing with neutral atoms,” Quantum, matched hall-effect gyrators and circulators,” Phys. Rev. Appl., vol. 7,
vol. 4, 2020, Art. no. 327. Feb. 2017, Art. no. 024030.
[113] V. V. Dobrovitski, G. D. Fuchs, A. L. Falk, C. Santori, and D. D. [141] A. C. Mahoney et al., “Zero-field edge plasmons in a magnetic topo-
Awschalom, “Quantum control over single spins in diamond,” Annu. logical insulator,” Nature Commun., vol. 8, no. 1, 2017, Art. no. 1836.
Rev. Condens. Matter Phys., vol. 4, no. 1, pp. 23–50, 2013. [142] S. Bosco, D. DiVincenzo, and D. Reilly, “Transmission lines and
[114] M. W. Dohertyet al., “The nitrogen-vacancy colour centre in dia- metamaterials based on quantum hall plasmonics,” Phys. Rev. Appl.,
mond,” Phys. Rep., vol. 528, no. 1, pp. 1–45, 2013. vol. 12, Jul. 2019, Art. no. 014030.
[115] A. Korpel, “Acousto-optics–A review of fundamentals,” Proc. IEEE, [143] Q. P. Herr, A. D. Smith, and M. S. Wire, “High speed data link be-
vol. 69, no. 1, pp. 48–53, 1981. tween digital superconductor chips,” Appl. Phys. Lett., vol. 80, no. 17,
[116] L. Desmarais, Applied Electro Optics. Upper Saddle River, NJ, USA: pp. 3210–3212, 2002.
Prentice-Hall, 1998. [144] D. P. Franke, J. S. Clarke, L. M. Vandersypen, and M. Veldhorst,
[117] R. V. Pound, “Electronic frequency stabilization of microwave oscilla- “Rents rule and extensibility in quantum computing,” Microprocessors
tors,” Rev. Sci. Instrum., vol. 17, no. 11, pp. 490–505, 1946. Microsyst., vol. 67, pp. 1–7, 2019.
[118] R. W. P. Drever et al., “Laser phase and frequency stabilization using [145] J. Hornibrook et al., “Cryogenic control architecture for large-scale
an optical resonator,” Appl. Phys. B, Photophys. Laser Chem., vol. 31, quantum computing,” Phys. Rev. Appl., vol. 3, no. 2, 2015, Art.
no. 2, pp. 97–105, 1983. no. 024010.
[119] J. Mizrahi et al., “Quantum control of qubits and atomic mo- [146] S. Pauka et al., “Characterizing quantum devices at scale with custom
tion using ultrafast laser pulses,” Appl. Phys. B, vol. 114, no. 1-2, cryo-CMOS,” Phys. Rev. Appl., vol. 13, no. 5, 2020, Art. no. 054072.
pp. 45–61, 2014. [147] D. Reilly “Challenges in scaling-up the control interface of a quan-
[120] W. H. Louisell, Coupled Mode and Parametric Electronics. New York, tum computer,” in Proc. IEEE Int. Electron Devices Meet., 2019,
NY, USA: Wiley, 1960. pp. 31–7.
[121] W. W. Mumford, “Some notes on the history of parametric transduc- [148] S. Pauka et al., “A cryogenic interface for controlling many qubits,”
ers,” Proc. IRE, vol. 48, no. 5, pp. 848–853, 1960. 2019, arXiv:1912.01299.
[122] M. A. Castellanos-Beltran and K. W. Lehnert, “Widely tunable para- [149] J. Stuart et al., “Chip-integrated voltage sources for control of trapped
metric amplifier based on a superconducting quantum interference ions,” Phys. Rev. Appl., vol. 11, no. 2, 2019, Art. no. 024010.
device array resonator,” Appl. Phys. Lett., vol. 91, no. 8, 2007, Art. [150] B. Patra et al., “Cryo-CMOS circuits and systems for quantum com-
no. 083509. puting applications,” IEEE J. Solid-State Circuits, vol. 53, no. 1,
[123] T. Yamamoto et al., “Flux-driven Josephson parametric amplifier,” pp. 309–321, Jan. 2017.
Appl. Phys. Lett., vol. 93, no. 4, 2008, Art. no. 042510. [151] B. Patra et al., “A scalable cryo-CMOS 2-to-20 GHz digitally-
[124] R. Vijay, “Josephson bifurcation amplifier: Amplifying quantum sig- intensive controller for 4 × 32 frequency multiplexed spin
nals using a dynamical bifurcation,” Ph.D. dissertation, Yale Univ., qubits/transmons in 22-nm FinFET technology for quantum comput-
2008. ers,” in Proc. Int. Solid-State Circuits Conf., 2020, pp. 304–306.
[125] N. Bergeal et al., “Phase-preserving amplification near the quantum [152] J. C. Bardin et al., “A 28 nm bulk-CMOS 4-to-8 GHz 2 mW cryogenic
limit with a Josephson ring modulator,” Nature, vol. 465, no. 7294, pulse modulator for scalable quantum computing,” in Proc. IEEE Int.
pp. 64–68, 2010. Solid-State Circuits Conf., 2019, pp. 456–458.
[126] M. Hatridge et al., “Quantum back-action of an individual variable- [153] J. C. Bardin et al., “Design and characterization of a 28-nm bulk-
strength measurement,” Science, vol. 339, no. 6116, pp. 178–181, cmos cryogenic quantum controller dissipating less than 2 mW at
2013. 3 K,” IEEE J. Solid-State Circuits, vol. 54, no. 11, pp. 3043–3060,
[127] J. Y. Mutus et al., “Strong environmental coupling in a Josephson 2019.
parametric amplifier,” Appl. Phys. Lett., vol. 104, no. 26, 2014, Art. [154] I. Bashir et al., “A mixed-signal control core for a fully in-
no. 263513. tegrated semiconductor quantum computer system-on-chip,” in
[128] T. Roy et al., “Broadband parametric amplification with impedance Proc. ESSCIRC IEEE 45th Eur. Solid State Circuits Conf., 2019,
engineering: Beyond the gain-bandwidth product,” Appl. Phys. Lett., pp. 125–128.
vol. 107, no. 26, 2015, Art. no. 262601. [155] C. Degenhardt et al., “Systems engineering of cryogenic CMOS elec-
[129] O. Naaman et al., “High saturation power Josephson parametric ampli- tronics for scalable quantum computers,” in Proc. IEEE Int. Symp. Cir-
fier with GHz bandwidth,” in Proc. IEEE MTT-S Int. Microw. Symp., cuits Syst., 2019, pp. 1–5.
vol. 2019, Jun. 2019, pp. 259–262. [156] N. Langford et al., “Experimentally simulating the dynamics of quan-
[130] C. Macklin et al., “A near-quantum-limited Josephson traveling-wave tum light and matter at deep-strong coupling,” Nature Commun.,
parametric amplifier,” Science, vol. 350, no. 6258, pp. 307–310, 2015. vol. 8, no. 1, 2017, Art. no. 1715.
[131] B. Abdo, K. Sliwa, L. Frunzio, and M. Devoret, “Directional amplifi- [157] M. Rol et al., “Fast, high-fidelity conditional-phase gate exploiting
cation with a Josephson circuit,” Phys. Rev. X, vol. 3, Feb. 2013, Art. leakage interference in weakly anharmonic superconducting qubits,”
no. 031001. Phys. Rev. Lett., vol. 123, no. 12, 2019, Art. no. 120502.
[132] F. Lecocq et al., “Nonreciprocal microwave signal processing with [158] LNF-LNC4_8C 4–8 GHz Cryogenic Low Noise Amplifier, Low
a field-programmable Josephson amplifier,” Phys. Rev. Appl., vol. 7, Noise Factory, Göteborg, Sweden. [Online]. Available: https://www.
no. 2, 2017, Art. no. 024028. lownoisefactory.com/files/7015/7825/6000/LNF-LNC4_8C.pdf
[133] L. Ranzani and J. Aumentado, “Circulators at the quantum limit: Re- [159] S. Montazeri, W.-T. Wong, A. H. Coskun, and J. C. Bardin, “Ultra-low-
cent realizations of quantum-limited superconducting circulators and power cryogenic SiGe low-noise amplifiers: Theory and demonstra-
related approaches,” IEEE Microw. Mag., vol. 20, no. 4, pp. 112–122, tion,” IEEE Trans. Microw. Theory Techn., vol. 64, no. 1, pp. 178–187,
Apr. 2019. Jan. 2016.

426 VOLUME 1, NO. 1, JANUARY 2021

[160] W.-T. Wong, M. Hosseini, H. Rücker, and J. C. Bardin, “A 1 mW DAVID J. REILLY (Member, IEEE) received the
cryogenic LNA exploiting optimized SiGe HBTs to achieve an average B. App. Sci. (H1) degree in applied physics from
noise temperature of 3.2 K from 4-8 GHz,” in Proc. IEEE IMS, 2020, the University of Technology, Sydney (UTS) in
pp. 181–184. 1998. From 1995 to 1998, he held a Technical Staff
[161] J. v. Dijk, A. Vladimirescu, M. Babaie, E. Charbon, and F. Sebas- Position with CSIRO (Applied Physics), charac-
tiano, “Spine (spin emulator) - A quantum-electronics interface simu- terising high Tc superconducting films for RF and
lator,” in Proc. IEEE 8th Int. Workshop Adv. Sensors Interfaces, 2019, microwave applications. In 2003, he was awarded
pp. 23–28. the Ph.D. degree in physics from the University
of New South Wales (UNSW), for work related
JOSEPH C. BARDIN (Senior Member, IEEE) re- to quantum transport in mesoscale semiconductor
ceived the B.S. degree from the University of Cal- devices.
ifornia, Santa Barbara, in 2003, the M.S. degree From 2003 to 2005, he was a Hewlett-Packard Fellow in quantum com-
from the University of California, Los Angeles, puting with UNSW, employing high frequency charge sensing measurements
in 2005, and the Ph.D. degree from the California for the readout of superconducting qubits and silicon spin qubits. From 2005
Institute of Technology, in 2009, all in electrical to 2008, he continued this research theme at Harvard University, developing
engineering. microwave approaches for detecting the state of spin qubits in a single shot
From 2003 to 2005, he was with the Jet Propul- as well as probing spin dynamics. Since 2008, he has been a Faculty member
sion Laboratory, California Institute of Technol- of the School of Physics at the University of Sydney, where he is currently
ogy, where he was involved in the demonstration of Professor. Dr. Reilly is also a Chief Investigator in the ARC Centre of Excel-
an array-based downlink for the NASA deep-space lence for Engineered Quantum Systems (EQuS), where his research focuses
network. From 2009 to 2010, he was a Postdoctoral Researcher with the on developing quantum technology that spans computing, sensing, and new
Caltech High-Speed Integrated Circuits Group, where he focused on selfheal- biomedical imaging modalities. In 2017, he joined Microsoft Corp. where he
ing microwave receivers. In 2010, he joined the University of Massachusetts is a Partner Research Manager and Director of Microsoft Quantum Sydney.
Amherst, where he is currently a Professor of Electrical and Computer En- As a leader in Microsoft’s efforts in quantum computing, much of his work is
gineering. Since 2017, he has also been with Google Quantum AI, where at the quantum-classical interface and the scale-up of quantum technology.
he focuses on the development of novel integrated electronics for quantum
computing.
Dr. Bardin currently is an Associate Editor and Steering Committee Mem-
ber for IEEE TRANSACTIONS ON QUANTUM ENGINEERING and a Track Editor
for IEEE JOURNAL OF MICROWAVES. He was the recipient of the 2011 DARPA
Young Faculty Award, the 2014 NSF CAREER Award, the 2015 ONR YIP
Award, the 2016 UMass College of Engineering Outstanding Junior Faculty
Award, the 2016 UMass Convocation Award for Excellence in Research and
Creative Activities, and the 2020 IEEE MTT-S Outstanding Young Engineer
Award.

DANIEL H. SLICHTER (Senior Member, IEEE)

received the A.B. (magna cum laude) degree from
Harvard University in 2004, the M.A. degree from
the University of California, Berkeley, in 2007, and
the Ph.D. degree from the University of California,
in 2011, all in physics.
From 2007 to 2011, he was a researcher with the
University of California, Berkeley, Quantum Na-
noelectronics Laboratory, where he demonstrated
the first continuous high-fidelity state readout of a
superconducting qubit by using a near-quantum-
limited microwave-frequency superconducting parametric amplifier, and
studied quantum measurement backaction and quantum feedback. Since
2012, he has been with the National Institute of Standards and Technol-
ogy (NIST), Boulder, CO, first as a Postdoctoral Research Fellow and cur-
rently as a Staff Physicist, where his research has focused on high-fidelity
microwave-based control and entanglement of trapped ion qubits, as well
as trap-integrated superconducting photon detectors for trapped ion qubit
readout.
Dr. Slichter was the recipient of a Hertz Foundation Fellowship (2006–
2011), the Hertz Foundation Thesis Prize (2012), an NRC Postdoctoral Fel-
lowship (2012–2014), and a Kavli Fellowship (2016). He currently is an
Associate Editor for IEEE TRANSACTIONS ON QUANTUM ENGINEERING.

VOLUME 1, NO. 1, JANUARY 2021 427