Quantum firmware and the quantum computing stack

Harrison Ball, Michael J. Biercuk, and Michael R. Hush

Citation: Physics Today 74, 3, 28 (2021); doi: 10.1063/PT.3.4698

View online: https://doi.org/10.1063/PT.3.4698
View Table of Contents: https://physicstoday.scitation.org/toc/pto/74/3
Published by the American Institute of Physics

ARTICLES YOU MAY BE INTERESTED IN

Quantum computing ramps up in private sector

Physics Today 73, 22 (2020); https://doi.org/10.1063/PT.3.4612

Magnetic fields for modulating the nervous system

Physics Today 74, 28 (2021); https://doi.org/10.1063/PT.3.4677

Stressed? Depressed? You are not alone

Physics Today 74, 20 (2021); https://doi.org/10.1063/PT.3.4696

Liquid metals at room temperature

Physics Today 74, 30 (2021); https://doi.org/10.1063/PT.3.4723

A researcher’s how-to manual

Physics Today 74, 50 (2021); https://doi.org/10.1063/PT.3.4702

The math behind epidemics

Physics Today 73, 28 (2020); https://doi.org/10.1063/PT.3.4614
 PAINTING BY GERALDINE COX

28 PHYSICS TODAY | MARCH 2021

 Harrison Ball is the lead quantum research scientist at the quantum-
technology company Q-CTRL in Sydney, Australia. Michael Biercuk is the
CEO and founder of Q-CTRL. He is a professor of quantum physics and
quantum technology and is a chief investigator in the Australian Research
Council Centre of Excellence for Engineered Quantum Systems at the
University of Sydney. Michael Hush is the chief scientific officer at Q-CTRL.

Harrison Ball, Michael J. Biercuk, and Michael R. Hush

Integrated quantum-control protocols could bridge

the gap between abstract algorithms and the physical
manipulation of imperfect hardware.

Q
uantum computers have rapidly advanced from lab- the Achilles’ heel of quantum comput-
ers and ultimately limits the algo-
oratory curiosities to full-fledged systems operating rithms they can run. Researchers are
with dozens of interacting information carriers called working to improve their devices’ per-
qubits. In 2019, researchers at Google became the first formance through passive means like
circuit design, but they’re also pursu-
to demonstrate quantum supremacy1—a quantum ing active measures; mitigating hard-
computer capable of calculations that are impossible for conventional ware errors through quantum error
correction (QEC) has driven research
devices—by using just over 50 qubits. for decades. The complexity and re-
source intensity of QEC—the set of al-
Researchers are now on the threshold of being able to de- gorithmic protocols necessary to ensure errors are identified
ploy quantum computers to solve a host of critical problems and corrected—has motivated consideration of complemen-
ranging from pharmaceutical drug discovery and industrial tary techniques that enable augmented performance without
chemistry to codebreaking and information security. (For more that computational overhead.
on quantum cryptography, see the article by Marcos Curty, Koji Quantum firmware is a generalized designation for a set
Azuma, and Hoi-Kwong Lo on page 36 of this issue.) Because of protocols that connect quantum hardware with higher,
of ongoing developments in computational heuristics and ap- more abstract levels in the quantum computing stack (see fig-
proximate quantum algorithms, quantum computers may well ure 1). More specifically, quantum firmware stipulates how
be able to solve commercially relevant problems with some physical hardware should be manipulated to improve stabil-
computational benefit, reaching what’s known as quantum ad- ity and reduce various error processes—in essence, “virtual-
vantage, within the next decade. izing” the underlying imperfect hardware. Higher abstrac-
Realizing useful computations using quantum systems re- tion layers in the quantum computing stack then interact with
quires scientists to recognize that performance is limited pre- qubits whose performance is different than that of the qubits
dominantly by hardware imperfections and failures rather in the bare hardware.2 (For more on quantum computing ar-
than just system size. Susceptibility to noise and error remains chitectures, see the article by Anne Matsuura, Sonika Johri,

MARCH 2021 | PHYSICS TODAY 29

QUANTUM FIRMWARE

and Justin Hogaboam, PHYSICS TODAY, March 2019, page 40.)

The choice of the term “firmware” reflects the fact that al-
though the routines are usually software defined, they reside
just above the physical layer in the stack and are effectively in-
visible to higher layers of abstraction. That approach to low-
level control resembles other forms of firmware in computer
engineering, such as DRAM (dynamic random-access mem-
ory) refresh protocols that stabilize classical storage hardware
against degradation caused by charge leakage. Such protocols
are responsible for scheduling, defining relevant control and
measurement operations, executing logic for actuation, and the
like. A user employing DRAM has little awareness of its pres-
ence or activity except in the small effects its execution has on,
say, memory access latency.
So that’s the “what.” But what about the “how?”

The underlying technology

Underpinning quantum firmware’s functionality is quantum
control,3 a discipline that addresses the question, How can sys-
tems that obey the laws of quantum mechanics be efficiently
manipulated to create desired behaviors? Ultimately, quantum
control is concerned with how the classical world interacts
with quantum devices. It guides researchers in gaining infor-
mation about system dynamics through measurements and en-
ables useful performance in computing, sensing, and metrol-
ogy. (For more on quantum control see the article by Ian FIGURE 1. THE STACK in a fault-tolerant quantum computer is
Walmsley and Herschel Rabitz, PHYSICS TODAY, August 2003, made of layers that correspond to levels of software abstraction.
page 43.) At the top sits Quantum As A Service (QAAS), which represents
The field of quantum control largely owes its existence to functions a user might interact with through, for instance, a cloud
decades of research in nuclear magnetic resonance and electron service. Below that are quantum algorithms and applications that
paramagnetic resonance, in which semiclassical magnetiza- are coded using developer tools that permit high-level abstraction.
tions formed from nuclear or electronic spins are manipulated The algorithms and applications are compiled on the third level to
by pulses of resonant RF or microwave radiation. In those dis- enact circuits on encoded blocks. In fault-tolerant computing, that
ciplines, hardware imperfections limited the ability to spectro- enaction is performed on logical qubits encoded using quantum
scopically characterize molecules. Then, in 1950, Erwin Hahn error correction (QEC), although realizing the QEC code and other
demonstrated that a dynamic-control protocol now known as associated tasks occupies a dedicated layer. Physical connectivity
the Hahn echo could mitigate the impacts of magnetic field in- between devices and compensation for any stray couplings are
homogeneities on spectroscopic resolution. His discovery led accounted for in a hardware-aware compiler. The quantum firmware
to the development of average Hamiltonian theory, which is layer, which is responsible for minimizing hardware error, resides
used to analyze the temporal evolution of spin systems, and of between that layer and the physical hardware. It handles all tasks
dynamical decoupling, a technique for canceling the impacts necessary for hardware calibration, tune-up, characterization,
stabilization, and automation. (Image courtesy of Q-CTRL.)
of unwanted spin interactions in molecules.4
Beginning in the 1980s, a parallel research discipline
emerged that sought to adapt the concepts and numeric tools Gaussian pulse of microwaves on resonance with a super-
from control engineering to the strictures of quantum- conducting qubit can act as an X operation, the quantum equiv-
coherent devices. That included both the treatment of linear alent of a NOT gate on a single qubit, whereas another pulse
systems, such as quantum harmonic oscillators,5 and the de- may implement a controlled-NOT operation on a pair of
velopment of numeric techniques for using imperfect hard- qubits, similar to a classical exclusive OR. An appropriately
ware to effectively manipulate spin systems.6 More recently, constructed temporal and spatial composition of such electro-
quantum optimal-control methods have been extended to magnetic signals makes up a quantum algorithm.
more general Hilbert spaces and Hamiltonians7 and have be- The physical correspondence between spin-½ systems and
come powerful tools for optimizing quantum experimental qubits builds a natural bridge for transferring quantum-control
system performance.8 techniques into quantum information in order to improve al-
Much like NMR, quantum computing hardware—whether gorithmic success despite hardware imperfections and ambi-
trapped ions, neutral atoms, superconducting circuits, or an- ent decoherence. One of the clearest efforts to explicitly incor-
other technology—generally relies on precisely engineered porate quantum control into quantum computers was
light–matter interactions to enact quantum logic. (For more on articulated2 by N. Cody Jones and coworkers at Stanford Uni-
qubit technology, see the article by Lieven Vandersypen and versity in 2012. They introduced a so-called virtual layer that
Mark Eriksson, PHYSICS TODAY, August 2019, page 38.) Those sat between quantum hardware and higher-level algorithmic
operations constitute the native machine language; a timed abstractions in the quantum computing stack and leveraged
30 PHYSICS TODAY | MARCH 2021
NMR-inspired composite pulsing.4 Their foun-
dational work inspired the quantum firmware
layer described here. Researchers now have
greater clarity about both the utility of quantum
control and the structure of higher-level soft-
ware abstractions with which the quantum
firmware layer interacts.

Technical aims
Contemporary quantum firmware is charged
with implementing the following functionality:
‣Error-robust quantum logic operations
that are supported by measurement-free open-
loop control.
‣Measurement-based closed-loop feedback
stabilization at the hardware level.
‣Microscopic hardware characterization for
calibration, noise identification, and Hamilto-
nian parameter estimation.
‣Machine learning–inspired approaches to
realize autonomy for the above tasks in large
systems.
Open-loop control refers to feedback-free ac-
tuation akin to a timed irrigation system that
maintains a healthy lawn without information
on soil moisture or rainfall. It’s resource effi-
cient and has proved to be remarkably effective
in stabilizing quantum devices, both during
free evolution and during nontrivial logic oper-
ations.9 When open-loop error suppression is
used in quantum computers, the instructions
for quantum hardware manipulation are re-
defined such that they execute the same math-
ematical transformation, but in a way that is ro-
bust against error-inducing noise, such as
fluctuations in ambient magnetic fields. The
suppression is typically realized by temporally
modulating the incident control fields that ma-
nipulate the physical devices (see the box on
page 32), and the modulation patterns may be
derived from Hamiltonian models or even FIGURE 2. QUANTUM FIRMWARE is an abstract layer of the computing stack
machine-learning techniques. Thus the control whose actions are orchestrated by an embedded microprocessor. The microprocessor
solutions defined by quantum firmware consti- accesses cloud-computing resources for computationally intense tasks such as open-
tute an effective error-robust machine language loop-control optimization and virtualizes the hardware for its interaction with higher
for manipulating quantum hardware. layers of the software stack. In the conception shown here, the microprocessor sends
In closed-loop feedback control, actuation commands to programmable logic devices, such as field-programmable gate arrays.
is determined by measurements of the system. Those devices are responsible for processing measurement results in real time for
physical-layer feedback stabilization, measurement-based decision making, and other
Its use is constrained by the destructive nature
tasks. They also provide instructions to other hardware elements such as direct digital
of projective measurement in quantum me-
synthesizers and arbitrary waveform generators. Arrows indicate communication
chanics. Several strategies may nevertheless be
pathways between elements. (Image courtesy of Q-CTRL.)
employed for hardware-level feedback-based
stabilization; they all are designed to gain suf-
ficient information about the underlying system without de- tion in the control-theoretic literature, has benefited from a
stroying encoded information needed in a computation. In large body of experimental and theoretical developments.11
fact, QEC—the gold-standard approach for large-scale quan- The underlying techniques complement external bench-
tum computers—is a form of closed-loop feedback that em- marking routines that quantify the hardware’s overall per-
ploys indirect measurement through ancilla qubits. The direct formance by focusing on the determination of actionable mi-
integration of hardware-level feedback stabilization remains croscopic information for system optimization and tune-up.
an ongoing area of exploration with some exciting results.10 Noise spectroscopy, which is widely used as a complemen-
Hardware characterization, known as system identifica- tary capability to noise suppression, provides information
MARCH 2021 | PHYSICS TODAY 31
QUANTUM FIRMWARE

OPTIMIZED GATES FOR SINGLE-QUBIT LOGIC OPERATIONS

Existing medium-scale superconducting quantum computers provide
an ideal platform for studies of quantum control because they allow
[ ]
σx = ∣0〉〈1∣ + ∣1〉〈0∣ = 0 1 ,
10
for cloud access to advanced hardware. Using one such platform—a which is the quantum mechanical analog of a classical NOT gate.
cloud-based IBM quantum computer—the research team at Q-CTRL, In each of the Bloch spheres shown, the quantum state—
a quantum computing startup with facilities in Sydney, Australia, and represented by the locus of a unit-length vector on the sphere’s
Los Angeles, California, has explored the efficacy of quantum-control surface—follows a path from the sphere’s north pole to its south
optimization in real systems. They employed specialized analog-layer pole. However, the qubit subject to a default implementation
programming that permits direct control of physical signals. (panel a) takes a substantially different path from those of the two
As an example, we demonstrate how to make an effective ma- qubits subject to error-robust pulses (panels b and c).
chine language that defines quantum logic operations that are re- The Pauli X operation is implemented using a pulse of mi-
silient against the typical sources of hardware instability. In the il- crowave radiation that enacts a control Hamiltonian
lustration here, we show different techniques to produce gates that
Hc(t) = ½ (Ω(t)eiφ(t)â).
perform a Pauli X “spin-flip” operation,
a Here â is a function of a and a†, the lowering and raising operators for
IBM DEFAULT
0〉 the state of the superconducting qubit. (We treat only the two lowest
7
AMPLITUDE (MHz)

I(t) Q(t) z levels as an effective qubit.) The coupling term Ω(t)eiφ(t) ≡ I(t) + iQ(t)
represents the control-pulse waveform, and the functions I(t) and Q(t)
y x therein represent user-adjustable controls.
0
In the default implementation of the X operation (panel a), I(t) is
composed of two sequential pulses that are approximately Gauss-
284 ns ian, with only a small component in Q(t). Those pulses largely drive
−7 the qubit state along a meridian of the Bloch sphere. In a quantum
0 165 330 1〉
firmware protocol, the simple physical definition for X is replaced
TIME (ns)
with a new one that parameterizes the gate in terms of I(t) and Q(t).
b DEPHASING-ROBUST
Numeric optimization is used to minimize a cost function that en-
0〉
7
AMPLITUDE (MHz)

I(t) Q(t) z sures that the quantum logic gate is implemented correctly, even in
the presence of noise. The controls applied in panels b and c are de-
y x rived using Hc(t) and take into account smoothing functions that en-
0 sure pulses can be faithfully transmitted from room-temperature
electronics into a dilution refrigerator where the qubits are housed.
239 ns Because of the error reduction incorporated into the control wave-
−7 forms, qubit states subject to optimized controls take paths along the
0 165 330 1〉
Bloch sphere that are more complex than those taken by the default
TIME (ns) qubit. Enacting those complex paths often requires a longer pulse.
c AMPLITUDE-ROBUST The optimized controls described here are designed to imple-
0〉
ment the X operation in a manner that is robust against either errors
AMPLITUDE (MHz)

7 z
I(t) Q(t)
in the amplitude Ω(t) or in the driving frequency that implements the
y x pulses. To test the control’s performance, it is repeatedly applied in
0 the presence of either quasistatic pulse-amplitude errors or pulse-
frequency (detuning) errors, and the gate’s infidelity—the probabil-
ity that a qubit state evolves to the wrong target—is measured (see
333 ns panels d and e). Even when large amplitude or detuning errors
−7
0 165 330 1〉 were added to the applied pulse, each optimized solution shows a
TIME (ns) flat response, which is a signature of robustness. The
e control designed to be robust against dephasing
d
25 25 (panel b) is indeed flat in the presence of dephasing
errors (panel e). Likewise, the amplitude-robust con-
INFIDELITY (×10−2)

IBM default
trol response (panel c) is flat despite amplitude errors
ROBUSTNESS

Dephasing-robust
(panel d).
Amplitude-robust
15 15 The shaded areas in panels d and e indicate the
overall improvement achieved through the choice
of appropriate optimized pulses. Additional controls
designed to exhibit robustness to both error
processes simultaneously have also been demon-
5 5
strated. (For a full explanation of the experiment de-
−1 0 1 −1 0 1 scribed here, see reference 9.)
AMPLITUDE ERROR (%) DETUNING ERROR (MHz)

32 PHYSICS TODAY | MARCH 2021

used in the design of open-loop controls. To implement it, a velop more diverse technical solutions for efficient hardware
qubit can serve as a sensor that when subjected to appropri- manipulation.
ately designed time-domain control, probes noise at different A dedicated firmware layer could autonomously orches-
frequencies. trate quantum-control tasks that span different classical com-
Hardware imperfections can be identified through various putational hardware. Those processes could exploit local pro-
techniques generally classified as Hamiltonian parameter esti- cessing to support automated scheduling and unsupervised
mation. A simple implementation might allow for the determi- stabilization, distributed computing infrastructure to execute
nation of the coupling rates between physical devices and con- computationally intensive optimization tasks, and low-latency
trol fields and reveal transmission losses experienced by programmable logic to conduct real-time processing. For ex-
signals en route to the qubits. More complex routines are com- ample, a local microcontroller can, on a schedule, initiate a
monly employed to determine the frequencies of unwanted en- cloud-based numeric optimization of a multiqubit gate (see fig-
ergy levels in a device or to characterize unknown couplings ure 2). That solution can be used to seed a hardware-executed
between qubits. The information obtained from such protocols tune-up of the final control waveform, which is then written to
informs open-loop error-suppressing controls and dynamic embedded memory. Slaved to the microcontroller is a field-
models for feedback-based stabilization. programmable gate array that both directs signal-synthesis
Artificial intelligence‒enabled autonomy, the final class of hardware to output the waveform used to manipulate the
techniques in quantum firmware operation, represents one of qubits and also processes measurement results from the qubits.
the most exciting exploratory areas of research in quantum in- (A logically distinct “embedded operating system” always re-
formation. Scaled-up systems will require high-efficiency rou- mains that defines and enables the functionality of the classical
tines that can tune up, calibrate, optimize, and characterize the electronics in use.)
underlying hardware with minimal user intervention. An in- For the near term, researchers are exploring how the dis-
terdisciplinary effort integrating machine learning, robotic tinctions between layers in the emerging stack could be blurred
control, and data inference is showing how adaptive measure- to deliver maximum performance. For example, it’s possible to
ment routines may be deployed to reduce the number of de- pursue a hardware–firmware co-design strategy to directly in-
structive measurements required1,12 and to enable rapid au- tegrate certain critical tasks into the classical electronics13 while
tonomous system bring-up and operation. others remain in the experimental software.
Opportunities may also arise to fundamentally rethink the
Integration strategies organization of quantum computer software stacks based on
Any practical implementation of quantum control must be tai- the functionality provided through exploitation of quantum
lored to the needs of a hardware system; each scheme will re- control. The potential value of such approaches is evident in
quire a particular subset of the functions described above. But hardware-aware compilation, in which optimal control is
the control techniques have much in common, and they are in- used to efficiently produce high-fidelity hardware-optimized
creasingly being used in state-of-the-art experimental and logical blocks. Quantum algorithms may then be compiled
commercial-grade quantum computers. (See PHYSICS TODAY, into a library of numerically optimized “analog” control se-
November 2020, page 22, for more about the commercializa- quences that would replace a smaller but more general set of
tion of quantum computers.) Researchers have thus been mo- universal gates.14
tivated to organize the relevant functionalities into an identifi-
able abstraction layer. Creating that organizational structure, System-level impacts
however, is distinct from determining how quantum control Regardless of how quantum firmware is realized, recent exper-
should be integrated into real systems. iments have made clear that the quantum-control functionality
One approach involves integrating the control functions encompassed therein could affect or even reshape higher ab-
into their own encapsulated layer in the quantum computing straction layers. That’s because the virtualization produced by
stack. In that conception, quantum firmware is responsible quantum firmware fundamentally transforms the behavior of
for defining and executing all actions that bridge the gap be- the underlying hardware, especially as it pertains to the char-
tween high-level abstractions, such as compilation or appli- acteristics of hardware errors.
cation programming, and the many low-level quantum- Open-loop control strategies are broadly used to suppress
control routines customized for particular hardware systems. errors in state-of-the-art quantum computer hardware; for ex-
Firmware can be embedded into appropriate computational ample, in certain settings, DRAG (derivative removal by adia-
hardware to virtualize the underlying quantum-coherent batic gate) pulses—an example of open-loop control—have
hardware. That is, the firmware changes the behavior and been shown to reduce gate errors in superconducting qubits by
performance of the hardware such that high-level abstraction approximately an order of magnitude compared with conven-
layers have no visibility into the “bare” performance of the tional Gaussian pulses. More recent results demonstrate that
underlying hardware. numerically optimized gates can mitigate the effects of hard-
The implementation of a dedicated quantum firmware layer ware imperfections in cloud quantum computers, thereby sup-
brings several potential benefits. First, the development of effi- pressing pulse-amplitude, off-resonance, and cross-talk errors.
cient high-level programming frameworks such as Cirq, Quil, Those demonstrations are particularly powerful because the
and Qiskit has led to an explosion of capability at the applica- error processes effectively addressed by quantum firmware
tion and algorithmic level. Building a framework to standard- often generate far larger effects than one would expect from
ize quantum-control integration may also encourage the best-case-scenario benchmark routines.
quantum-control and machine-learning communities to de- In both research-grade systems and publicly available cloud
MARCH 2021 | PHYSICS TODAY 33
QUANTUM FIRMWARE

systems, best- and worst-case qubit error rates across a device quantum volume, a metric that accounts for architectural fea-
often differ by more than an order of magnitude. Those errors tures, including hardware connectivity, and device-level pa-
can arise from fabrication variances and coupling inhomo- rameters, such as the one- and two-qubit error rates across the
geneities between qubits and the ambient electromagnetic en- device.18 Honeywell has claimed a quantum volume of 128
vironment. Quantum firmware homogenizes hardware per- with just a handful of qubits compared with approximately 64
formance in space and time. Optimized quantum-control from IBM’s larger systems; the results demonstrate that hard-
operations implemented in real systems have brought error ware performance is the primary bottleneck.
rates for all qubits close to the best-case performance; likewise, Improving both one- and two-qubit error rates by more than
drift-robust controls can extend typical calibration windows on a factor of 10, as has been demonstrated experimentally, would
cloud and laboratory hardware from 6–12 hours to more than have a massive impact on system-level performance. Those im-
five days.9 pacts would be largest in devices with weak connectivity, where
Why do those improvements matter? To start, current algo- the spatial rearrangement of qubit data requires many multi-
rithmic compilers can improve performance by trading an in- qubit swap operations. Device sizes are increasing rapidly—
crease in compiled-circuit complexity for the ability to avoid Google and IBM have each released a road map to 1000-qubit
poorly performing devices. But in large-scale systems with systems—and quantum control provides a means to ensure sys-
substantial performance variation, that compilation process tem utility at the algorithmic level tracks with system size.
can become quite complicated, and shuttling information Ultimately, we believe that building and operating large-
around the worst-performing devices may require many more scale quantum computers is effectively impossible without in-
gates and time steps. By homogenizing device performance in tegrating advanced quantum-control techniques into a quan-
space and time, quantum firmware can simplify higher-level tum firmware abstraction layer. Autonomous vehicles, walking
compilation protocols,15 thereby reducing the complexity and robots, and advanced avionics systems have all demonstrated
duration of the implemented algorithm. the importance of dynamic control and automation. Similarly,
Quantum control will also have a long-term impact on the in quantum computing, advanced control-theoretic strategies
performance of QEC. Both the hardware-level feedback stabi- were instrumental in the calibration and tune-up of devices
lization and open-loop control found in firmware exploit the used to achieve quantum supremacy. Many techniques from
fact that noise processes often vary slowly in space and time; the fields of machine learning and robotic control are likely to
those methods provide little benefit for truly stochastic errors. improve performance and increase autonomy, thereby allow-
On the other hand, QEC formulations generally assume statis- ing future quantum developers to confidently abstract away
tically independent error models. Thus quantum firmware the details of a computer’s underlying hardware.
works in concert with QEC to correct for a broad range of error
types and effectively preconditions the properties of the resid-
ual errors to be compatible with QEC.16 But more than that, the REFERENCES
way in which quantum firmware closes the gap between the 1. F. Arute et al., Nature 574, 505 (2019).
best and worst performing qubits and reduces statistical corre- 2. N. Cody Jones et al., Phys. Rev. X 2, 031007 (2012).
lations in the residual errors17 actually reduces QEC’s resource 3. H. M. Wiseman, G. J. Milburn, Quantum Measurement and Control,
Cambridge U. Press (2011).
intensity. It’s a win–win combination. 4. U. Haeberlen, J. S. Waugh, Phys. Rev. 175, 453 (1968).
5. T. J. Tarn, G. Huang, J. W. Clark, Math. Modell. 1, 109 (1980); J. W.
The future of quantum firmware Clark, D. G. Lucarelli, T.-J. Tarn, Int. J. Mod. Phys. B 17, 5397
Quantum computing is complex, so algorithm designers and (2003); D. Dong, I. R. Petersen, IET Control Theory Appl. 4, 2651
(2010); L. Bouten, R. Van Handel, M. R. James, SIAM J. Control
end users need a framework through which they can effi- Optim. 46, 2199 (2007); H. I. Nurdin, M. R. James, I. R. Petersen,
ciently exploit quantum computers without having detailed Automatica 45, 1837 (2009).
technical knowledge of the underlying hardware. They ex- 6. N. Khaneja et al., J. Magn. Reson. 172, 296 (2005).
pect high-performance quantum hardware to be stable and 7. S. Chaudhury et al., Phys. Rev. Lett. 99, 163002 (2007); A. Smith et
provide consistent outputs irrespective of small changes in al., Phys. Rev. Lett. 111, 170502 (2013); R. W. Heeres et al., Nat.
Commun. 8, 94 (2017).
an algorithm’s structure. Quantum firmware enables those 8. A. D. Tranter et al., Nat. Commun. 9, 4360 (2018); P. B. Wigley et
capabilities. al., Sci. Rep. 6, 25890 (2016); B. M. Henson et al., Proc. Natl. Acad.
Quantum-control demonstrations have confirmed im- Sci. USA 115, 13216 (2018).
provements of about a factor of 10 in the performance of quan- 9. A. R. R. Carvalho et al., https://arxiv.org/abs/2010.08057.
10. N. Ofek et al., Nature 536, 441 (2016).
tum logic operations relative to naive gate implementations.9 11. H. Ball et al., https://arxiv.org/abs/2001.04060.
Similarly, dynamic memory stabilization has extended qubit 12. R. S. Gupta et al., NPJ Quantum Inf. 6, 53 (2020).
lifetimes to time scales measured in minutes. In those settings, 13. H. Ball et al., Phys. Rev. Appl. 6, 064009 (2016); S. J. Pauka et al.,
the performance gains have been limited by either incoherent https://arxiv.org/abs/1912.01299; V. Negnevitsky et al., Nature
processes or the capabilities of classical electronics, but both 563, 527 (2018).
14. Y. Shi et al., in ASPLOS '19: Proceedings of the Twenty-Fourth
are showing steady gains with time and specialization for the International Conference on Architectural Support for Programming
quantum computing market. We therefore expect that control Languages and Operating Systems, Association for Computing
systems and device performance will improve in parallel with Machinery (2019), p. 1031.
quantum firmware protocols. 15. S. Nishio et al., https://arxiv.org/abs/1903.10963; P. Murali et al.,
https://arxiv.org/abs/1901.11054.
The effect of using quantum-control technologies such as 16. J. P. Barnes et al., Phys. Rev. A 95, 062338 (2017).
error-robust open-loop control on algorithmic performance 17. C. L. Edmunds et al., Phys. Rev. Res. 2, 013156 (2020).
can be quantified with benchmarks. One such benchmark is 18. A. W. Cross et al., Phys. Rev. A. 100, 032328 (2019). PT

34 PHYSICS TODAY | MARCH 2021