Electrical control of quantum dot spin qubits

A dissertation presented
by

Edward Alexander Laird

to
The Department of Physics
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
in the subject of

Physics

Harvard University
Cambridge, Massachusetts

2009
c 2009 by Edward Alexander Laird
All rights reserved.
Dissertation Advisor: Professor Charles M. Marcus Author: Edward Alexander Laird

Electrical control of quantum dot spin qubits

Abstract

This thesis presents experiments exploring the interactions of electron spins with electric

fields in devices of up to four quantum dots. These experiments are particularly motivated

by the prospect of using electric fields to control spin qubits.

A novel hyperfine effect on a single spin in a quantum dot is presented in Chapter 2.

Fluctuations of the nuclear polarization allow single-spin resonance to be driven by an

oscillating electric field. Spin resonance spectroscopy revealed a nuclear polarization built

up inside the quantum dot device by driving the resonance.

The evolution of two coupled spins is controlled by the combination of hyperfine inter-

action, which tends to cause spin dephasing, and exchange, which tends to prevent it. In

Chapter 3, dephasing is studied in a device with tunable exchange, probing the crossover

between exchange-dominated and hyperfine-dominated regimes. In agreement with theoret-

ical predictions, oscillations of the spin conversion probability and saturation of dephasing

are observed.

Chapter 4 deals with a three-dot device, suggested as a potential qubit controlled entirely

by exchange. Preparation and readout of the qubit state are demonstrated, together with

one out of two coherent exchange operations needed for arbitrary manipulations. A new

readout technique allowing rapid device measurement is described.

In Chapter 5, an attempt to make a two-qubit gate using a four-dot device is presented.

Although spin qubit operation has not yet been possible, the electrostatic interaction be-

tween pairs of dots was measured to be sufficient in principle for coherent qubit coupling.

iii
Contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

1 Introduction 1
1.1 Organization of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Quantum computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 GaAs heterostructures, quantum dots and charge sensing . . . . . . . . . . 4
1.4 Double quantum dots and spin blockade . . . . . . . . . . . . . . . . . . . . 8
1.5 Electron spin resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 The singlet-triplet basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.7 Summary of contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 Gate-driven hyperfine-mediated electron spin resonance 16

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Device and measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 EDSR spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.1 Comparison with data . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5 Nuclear polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.6 Addressing individual spins . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.7 Open issues and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3 Effect of Exchange Interaction on Spin Dephasing in a Double Quantum

Dot 33
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Measurement of exchange energy, hyperfine interaction strength and readout
visibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5 Saturation singlet probability . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6 Time-dependence of the singlet probability . . . . . . . . . . . . . . . . . . 42
3.7 Summary and acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 43

4 Coherent operation of an exchange-only qubit 45

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Exchange in a three-spin system . . . . . . . . . . . . . . . . . . . . . . . . 47

iv
 4.3 Device and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Coherent exchange in a triple dot . . . . . . . . . . . . . . . . . . . . . . . . 52
4.5 Conclusion and acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 53

5 Towards a two-qubit gate in the singlet-triplet basis 54

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.2 Device and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3 Measuring the coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4 Summary and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

A Fabrication recipe 61
A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
A.2 Generic photolithography recipe . . . . . . . . . . . . . . . . . . . . . . . . 61
A.3 Complete fabrication procedure . . . . . . . . . . . . . . . . . . . . . . . . . 62
A.3.1 Mesa etch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
A.3.2 Ohmic contacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
A.3.3 Gate pads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
A.3.4 Fine gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

B Constructing and using a multiplexed reflectometry setup 65

B.1 Hardware inside the cryostat . . . . . . . . . . . . . . . . . . . . . . . . . . 65
B.2 Hardware outside the cryostat . . . . . . . . . . . . . . . . . . . . . . . . . . 69
B.2.1 The demodulation box . . . . . . . . . . . . . . . . . . . . . . . . . 71
B.3 Igor code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
B.3.1 do2dfast() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
B.3.2 tuneg() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
B.3.3 calsensor() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
B.4 Tuning hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

C Synchronizing two Tektronix AWG520s 80

v
List of Figures

1.1 Heterostructures and quantum dots . . . . . . . . . . . . . . . . . . . . . . . 5

1.2 Charge quantization and charge sensing . . . . . . . . . . . . . . . . . . . . 6
1.3 Double quantum dots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Spin blockade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Electron spin resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Bloch sphere in the singlet-triplet basis . . . . . . . . . . . . . . . . . . . . . 13

2.1 Measurement setup and pulse scheme . . . . . . . . . . . . . . . . . . . . . 18

2.2 Spin resonance signal as a function of magnetic field and microwave frequency 21
2.3 Spin resonance signal as a function of time, magnetic field and microwave
power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Nuclear polarization driven and detected via EDSR . . . . . . . . . . . . . . 28
2.5 EDSR signal in a device incorporating a micromagnet . . . . . . . . . . . . 30
2.6 An additional half-frequency EDSR line . . . . . . . . . . . . . . . . . . . . 31

3.1 Level diagram and pulse scheme . . . . . . . . . . . . . . . . . . . . . . . . 35

3.2 Exchange measured as a function of detuning . . . . . . . . . . . . . . . . . 37
3.3 Saturation singlet probability . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 Singlet probability as a function of time . . . . . . . . . . . . . . . . . . . . 43

4.1 An exchange-only qubit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.2 Electron states of a triple quantum dot . . . . . . . . . . . . . . . . . . . . . 48
4.3 Fast measurements with two charge sensors . . . . . . . . . . . . . . . . . . 50
4.4 Coherent spin exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5.1 A capacitive two-qubit interaction . . . . . . . . . . . . . . . . . . . . . . . 56

5.2 Charge stability diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3 Measuring the coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

B.1 Reflectometry cryostat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

B.2 Cold finger and sample PC board . . . . . . . . . . . . . . . . . . . . . . . . 69
B.3 Room temperature measurement schematic . . . . . . . . . . . . . . . . . . 70
B.4 Demodulation box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

vi
Acknowledgements
After seven years of excitement, confusion, frustration, fatigue, serendipity, and a fair few

occasions when I thought I would never finish1 it is a great pleasure to look back over my

graduate school experience and remember the people whose collective help was essential for

my PhD.

First I would like to thank my advisor, Charlie Marcus. Charlie’s high standards and

strong aesthetic sense were inspirational throughout my time working for him, and kept me

on a path of maximal PhD enjoyment globally no matter how disordered the topography

locally. As well as constantly supplying interesting problems to work on, Charlie provided

great practical help in resolving them through his endless inventiveness, willingness to advise

on almost any level of detail, and dedication to keeping the lab well-equipped and efficient.

I would also like to thank my committee members, Micha Lukin and Eugene Demler,

for asking pertinent, interesting questions, and for stimulating lectures.

Two theorists with whom I collaborated closely were Jake Taylor and Emmanuel Rashba.

I am grateful to both of them for generously sharing their knowledge with me and for tire-

lessly immersing themselves in the details of our ongoing projects.

The experiments described here relied heavily on the exceptionally quiet wafers from

Art Gossard’s MBE machines at Santa Barbara. I would like to thank the student who

grew them, Micah Hanson, collaborator on every one of my papers; perhaps we will get to

meet one day.

Within the Marcuslab, I have been enormously helped and educated by interacting with

knowledgeable, enthusiastic people. I am especially grateful to three senior students and

1
Usually in the McKay basement in the small hours of the morning.

vii
postdocs I worked with during my first four years. Dominik Zumbühl patiently taught me

fabrication and measurement. Jason Petta mentored me through my first successful experi-

ment and set an example of efficient research that I am still trying to emulate. David Reilly

brought not only the revolutionary technology of fast readout but also a wide knowledge of

physics and of life that made him great fun to work with.

For the spin resonance experiment I had the great pleasure of collaborating with Chris-

tian Barthel, whose forthright wit and example of shameless scavenging were frequent

sources of relief and sustenance as we painfully tuned our way through a long sequence

of not-quite-satisfactory devices. In the last few months I have enjoyed working with Teesa

Christian. Leaving the fridge in her care, I felt like a parent using a babysitter for the first

time. My fears for its safety proved unfounded; in fact, it is as though the babysitter has

taken the baby off to college and then eloped with it. All I can do is express my hope that

Teesa and the double-double experiment have a very happy future together.

Other lab members who particularly contributed to my happiness during my time here

include Alex Johnson, from whom I learnt much about coding and skiing; Leo DiCarlo, who

shared with Reilly a deep knowledge of electronics and a characteristic brand of Southern

Hemisphere toilet humor; Jimmy Williams, my companion on many laps of Memorial Drive,

who always managed to work hard while pretending to be unstressed; Ferdinand Kuemmeth,

who saw through all my facile explanations; and Jim Medford, who quickly became a

useful discussion partner on the spin qubit project. Many others generously taught me

what they knew and kept the lab an entertaining place to be, including Jeff Miller, Ron

Potok, Michael Biercuk, Nadya Mason, Douwe Monsma, Susan Watson, Nathaniel Craig,

Josh Parks, Will Koehl, Floris Zwanenberg, Abram Falk, Sang Chu, Jerry Chow, Doug

McClure, Slaven Garaj, Jacob Aptekar, Rob Barton, Hugh Churchill, Yongjie Hu, Maja

viii
Cassidy, Patrick Herring, Angela Kou, Bart Horn, Andrew Bestwick, Carolyn Stwertka,

Eli Levenson-Falk, Jennifer Harlow, Menyoung Lee, Alex Ogier, Michi Yamamoto, Reineer

Heeres, David Marcos, Shu Nakaharai, Max Lemme, Sandro Erni, Morten Kjaergaard and

more. From other groups, I benefited from discussions with David DiVincenzo, Sandra

Foletti, Hendrik Bluhm, Amir Yacoby, Mike Stopa, Jacob Krich and Mark Rudner among

others.

No matter how hard I worked to make devices and apparatus, I could have accomplished

nothing without the efforts of the CNS and DEAS technicians. I am grateful to Noah Clay,

Steve Sheppard, Ed Macomber, Jiangdong Deng and especially Yuan Lu in the cleanroom,

and to Louis Defeo and his colleagues in the machine shop. A succession of efficient admin-

istrators ensured all my orders got filled and helium was always on tap: James Reynolds,

James Houlahan, James Gotfredson, Danielle Reuter, and Jess Martin.

Outside work, I would like to thank David for the propaganda that got me into this, and

Tyler, Jack, Mary, and the Cambridge Running Club for helping me to get through it. I

acknowledge those teachers from my distant pre-PhD life who encouraged and developed my

interest in physics, especially Rodney Archard, Tchavdar Todorov, Tony Weidberg, Keith

Burnett and Armin Reichold.

Lastly I thank my parents and family for their encouragement, help and love during my

PhD and throughout my life.

ix
Chapter 1

Introduction

Recent advances in nanoscale fabrication and measurement make it possible to study

quantum behavior in individual electronic devices. Although quantum effects tend to be-

come more important for smaller devices, demonstrating coherence in an individual quantum

system remains difficult. This is because the system under study must satisfy two frequently

conflicting requirements. To prevent decoherence, it must be well isolated from its uncon-

trolled environment; but for measurement, it must be coupled to a detection apparatus.

The payoffs from being able to control and measure individual quantum systems are

threefold. Firstly, experiments beyond average measurements on ensembles become feasible,

allowing fundamental studies of quantum mechanics [1, 2, 3]. Secondly, delicate quantum

states are sensitive probes of their microscopic environment [4, 5]. Finally, information

encoded in quantum bits (‘qubits’) can be efficiently manipulated using a quantum com-

puter [6, 7].

This thesis focuses on controlling electron spins in GaAs quantum dots [8]. The electron

spin is the simplest non-trivial quantum system, described by a two-dimensional Hilbert

space. Since it couples only via its magnetic moment, it is often fairly well-isolated from

its environment (better isolated, at least, than the electron charge.) Because the well-

established techniques of spin resonance allow for any desired single-spin transformation,

1
the electron spin is a paradigmatic qubit [9].

1.1 Organization of this thesis

This thesis will present experiments on up to electron spins in systems of up to four coupled

GaAs quantum dots. In the rest of this chapter, I will first explain briefly why a quantum

computer is desirable and what the technical requirements are. I will then introduce the

material system, quantum dots in GaAs heterostructures, and describe charge sensing,

the measurement technique used throughout this thesis. Double quantum dots lead spin

and charge to be coupled through the phenomenon of spin blockade, allowing quantum

operations in the singlet-triplet basis of a pair of electrons. This chapter will conclude with

a brief summary of new contributions reported in this thesis.

In the second chapter, I will describe a novel mechanism of electric dipole spin resonance

mediated by an oscillating hyperfine coupling and driven by an electric field. Because time-

varying electric fields are easy to generate and localize, this technique for spin manipulation

can be technically easier than conventional magnetically driven spin resonance. Chapter 3

studies the dephasing of a pair of spins in a double quantum dot under the combined effects

of exchange and hyperfine interactions. Chapter 4 develops techniques to manipulate and

read out spin in a triple quantum dot, with the aim of realizing an exchange-only qubit.

Chapter 5 describes progress towards a two-spin-qubit gate in the singlet-triplet basis, using

four electrons in a pair of double dots.

Three appendices give technical information relating to the experiments. Appendix A

details my fabrication recipe. Appendix B describes the construction of a radio-frequency

reflectometry apparatus and gives a brief description of tuning. Appendix C explains how

to synchronize two Tektronix arbitrary waveform generators precisely.

2
1.2 Quantum computing

The process of computation involves encoding information in physical objects (‘bits’) and

manipulating those objects until they encode the desired solution to some problem. Present-

day classical computers restrict themselves to a subset of the manipulations allowed by the

laws of physics, namely those that map eigenstates of the measurement operator into each

other. A quantum computer does not obey this restriction and can carry out the general

unitary transformations allowed by quantum mechanics [6, 7].

Several problems are known for which a quantum computer is dramatically more efficient

than a quantum computer. The two most famous quantum algorithms are Shor’s algorithm

and Grover’s algorithm. Shor’s algorithm factorizes n-digit numbers in a time that scales
1/3
as n2 , rather than en classically [10], and Grover’s algorithm finds unique solutions to

mathematical functions of n bits in a time that scales as 2n/2 rather than 2n [11].

Actually making a quantum computer, a machine that can carry out arbitrary unitary

transformations on non-trivial inputs, seems at first sight impossible with any foreseeable

technology. Several theorems make it seem slightly less impossible. First, any unitary

transformation on a collection of quantum bits (‘qubits’) can be decomposed into a series of

one- and two-qubit unitary transformations [12]. In fact, any one of a large class of two-bit

transformations is sufficient; this class includes the so-called controlled-phase (CPHASE)

√
and square-root-of-swap ( SWAP) gates [13]. Secondly, it is not even necessary to perform

arbitrary one-qubit transformations; a discrete subset of transformations suffices for efficient

approximation [14, 6]. Finally, even imperfectly realized quantum gates can be corrected

at the price of a tolerable increase in computational complexity [15].

Any physical realization of a quantum computer must satisfy extremely demanding

requirements, but electron spins in semiconductor quantum dots may be suited to act as

3
a basis [16, 9]. Since the spin projection can take on two values, it naturally forms a

well-defined quantum bit, or qubit, comparatively insensitive to decoherence. At attainable

temperatures, thermal relaxation can be used to prepare spins in a known-state. Arbitrary

√
single-qubit gates are possible by electron spin resonance [17], and the universal SWAP

gate can be carried out via Heisenberg exchange between two spins. And spin-selective

electron tunnel rates allow for qubit readout [18]. Whether these elements can ever be

combined to make a practical computer is not known. Experiments on manipulating and

measuring single spins, including those reported here, are partly motivated by a desire to

answer this question.

1.3 GaAs heterostructures, quantum dots and charge sensing

The attraction GaAs/AlGaAs heterostructures as a substrate for quantum dots is the un-

rivaled tunability of the confinement potential. Individual parameters of the Hamiltonian

– electric and magnetic fields, energy levels and tunnel couplings – can be varied rapidly in

a single device. Electron spin states can be mapped to charge states and read out electron-

ically [18, 19]. Finally, well-established lithographic techniques can be applied, permitting

in principle devices of almost arbitrary size and complexity.

Fabrication of a quantum dot device begins with a heterostructure grown by molecular

beam epitaxy [22]. A layer of dopant atoms (usually Si substituting Ga) donates elec-

trons to a quantum well formed at the heterointerface, creating a two-dimensional electron

gas (2DEG) located ∼ 100 nm below the surface (Fig. 1.1(a)). Electrical contact to the

2DEG is achieved through PtAuGe contact pads deposited on the surface of the wafer and

annealed to diffuse AuGe down to the 2DEG. To create nanostructures, TiAu topgates are

patterned using electron beam lithography. Voltages of order ∼ −300 mV locally deplete

4
 - * ) + )&$ *$ &%,+&) * %# +  )+ $% ',#+ &% % &%+)&#
& #+)&% .- ,%+ &%*  * * , +& &+ + #) #+)&%  .-#%+ % $&)
$'&)+%+#0 +&  # " *+),+,) % .  $+# +* * +,+  )+#0 &- +
 ) # +& '#+ +  ,)  *&.*  *  - . & + +)&*+),+,) . + &+
&%++* +& +  %  '#+ &% +  % +  * +0' ##0  +&  %$ #&.
+ *,) & + )0*+# % + +* +) *  #+)# )*&#,+ &% # $ + &% +  # +0 +&
IinIin
a) b) (b)
&%+)&# % '++)% (b)+ #+)&% * % ')+  &% % &%+)&# + .-,%+ &% &%#0 +&
10 nm GaAs . + %  . .-#%+* %  &$'#+#0 ) +))0 .0
60 nm Al0.3Ga0.7As VV
LL VRVR
CONTACT GATE CONTACT
4 x 1012 cm-2 Si doping
2DEG
2DEG
2DEG
40 nm Al0.3Ga0.7As GaAs Iout
Iout

2DEG
800 nm GaAs

30-period superlattice 2DEG

(c)
(c) 2DEG
3 nm Al0.3Ga0.7As, 3 nm GaAs
I (pA)

 ,)    % ')*'+ - - . & + (,%+,$

I (pA) &+ *+),+,)
500!m GaAs substrate I (pA)

')*'+ - &+ &)$ ,* % '#+ &% +* * *&.% % + &++&$ &  ,)  
 # * !,*+ ,%) + *,) .& &'% %* ##&. ,))%+ +& '** +)&, + - 
*,) % + &%,+% +)&, + -  * + ,%$%+# $%* 0 .  + *
Figure 1.1: Heterostructures and ')&
quantum dots.
% 
)%+ (a)& Schematic
) $* of )*&#-
&')+ &% &% % GaAs/AlGaAs
%&)$+ &% &,+ het-
%)0 .-
erostructure and potential profile of,%+
the&%conduction band,+)!+&)0
%+%* +0 % )+ showing%+))%
formation of the 2DEG
. + &%,+% $*,)$%+*
at the interface. The specific parameters
 ,) shown
 *&.*here are*+),+,)
+ % those . ofthe wafers
*,''&)+* + 031104B and
 #+)&% * +)+ % +
050329A grown by the Gossard group + .)
at *,)
Santa+Barbara
 * ' '**
and-+*
used+ *,)
in the)&$ &/ + &% + %&'
experiments de- ) &%
')&- * )) )* +& &,'0 + (,%+,$ .## % + # * *') "'* + &%&) * +*
scribed in this thesis. (b) In a quantum point contact device, topgate voltages deplete
.0 )&$ + .## %% % $& # +0   +*# * 
% 0 + (,%+,$ .##
the 2DEG, creating a one-dimensional channel for electrical transport [20]. (c) A ring of
&)$ + + %+) & + # * % + ,#"  * *,')#++  * ,*,##0 
topgates creates a quantum dot [21].

the 2DEG, forming two kinds of structure important for this thesis: quantum point contacts

(QPCs) and quantum dots [23, 24]. A QPC is formed when two topgates define a narrow

channel through which electrical current can flow 1.1. Changes in the electrical potential

in the channel, caused by varying the topgate voltage or by the motion of nearby charges,

lead to changes in the electrical resistance of the current path. A quantum dot is formed in

a potential minimum defined by a ring of topgates. The topgate voltages control the size,

shape and depth of the minimum, and hence the equilibrium electron occupation of the

quantum dot. The potential barrier(s) between the dot and the rest of the 2DEG, which

control the electron tunnel rate to and from the leads, are also tunable. Measurements are

typically performed in a dilution refrigerator at an electron temperature ∼ 150 mK.

As well as high mobility, a heterostructure for well-tuned quantum dots must have

two other properties. Good control over the potential at the 2DEG requires that the

5
Figure 1.2: (a) The chemical potential µ(N ) of a quantum dot for various electron occu-
pancies N . The dot will be filled up to the chemical potential of the leads µs . Reducing
a gate voltage raises the chemical potential in the dot, decreasing the electron number.
(b) Conductance of a charge sensing point contact as a gate voltage VG is swept. Steps
correspond to changes in the dot occupancyND , marked along the bottom axis. The overall
background slope reflects the direct electrostatic coupling of VG to the point contact. Inset:
The device, with locations of dot and charge sensor indicated.

heterointerface be located close to the surface, and a stable device potential implies that

charges do not switch between donors in the dopant layer [25]. The heterostructures used

in this thesis, although exceptionally stable, do not have especially good properties in the

other respects; this makes device tunability a serious difficulty and the most time-consuming

obstacle overcome in the experiments described here.

Although the gate voltages that control the charge on the dot vary continuously, the

charge itself is quantized when the potential barrier to the leads is made large enough. The

number of electrons ND on the dot is determined by background charges, gate voltages and

the electrochemical potentials µS and µD of the leads. With no bias applied across the dot,

so that µS = µD , ND satisfies:

µ(ND ) < µs < µ(ND + 1) (1.1)

6
 a) b) -5 0 5 dgs/dVL (a.u.)

500 nm

gs -500
(1.0)
(1.1)
(1.2)

VL (mV)
NL NR
VR (0.0) (0.1)
(0.2)

-550

VL VR
-450 -400
VR (mV)

Figure 1.3: (a) A few-electron double quantum dot device with a charge sensor of conduc-
tance gs . (b) Charge stability diagram of the device, measured using the charge sensor
as a function of VL and VR [28]. Transition lines running approximately vertical (horizon-
tal) separate regions of separate right (left) dot occupancy. Capacitive coupling between
dots leads to anticrossings that show up as bright diagonal lines. The equilibrium electron
occupancy for each gate configuration is indicated.

where µ(N ) is the chemical potential of a dot containing N electrons.

Sweeping the voltage VG on a gate with capacitance CG to the dot changes the chemical

potential and therefore ND . It is usually valid to separate the energy of a single dot into a

charging term, parameterized by a constant total dot capacitance C, and a much smaller

confinement term. In this approximation, changing VG by

∆VG = e/CG (1.2)

changes ND by one [26].

The occupation can be measured using a nearby charge sensor QPC, as shown in

Fig. 1.2(b) [27]. As gate voltage VG is swept, the sensor conductance gs has a step every

time ND changes. The regularity of the steps in Fig. 1.2(b) validates the approximations

in the previous paragraph.

7
a)
0 5 10 15 0 10
T J02 |Id | (pA) g s (10-3 e2/h)

c) e)
eVsd Vsd = +0.5 mV Vsd = +0.5 mV
S
-1.025 -1.025
(1,2) (1,2)
(1,1) (1,1)

V6 (V)
V6 (V)
-1.035 (0,1) (0,2) -1.035 (0,1) (0,2)

d) f)
b) Singlet Triplet Vsd = -0.5 mV Vsd = -0.5 mV

-1.03 -1.03

V6 (V)

V6 (V)
-1.04 -1.04
-0.96 -0.95 -0.96 -0.95
V2 (V) V2 (V)

Figure 1.4: Spin blockade in a double quantum dot. (a) Chemical potential of dots and leads
under positive bias Vsd . The triplet state of the right dot is higher in energy by the exchange
J02 , but this does not prevent electron transport through the device. (b) Under negative
bias, exchange makes electron tunneling spin-selective, suppressing transport. (c) and (d)
Current Id through the device for positive and negative bias, showing strong asymmetry.
(e) and (f) Charge sensing signal, also showing asymmetry [32].

1.4 Double quantum dots and spin blockade

A richer and more tunable spectrum of electron states can be achieved in double quantum

dots (Fig. 1.3(a)) [29]. To some extent, the properties of each dot can be tuned separately.

The occupations (NL , NR ) of left and right dots respectively are controlled mainly by the

gate voltages VL and VR , resulting in the charge stability diagram shown in Fig. 1.3(b) [30,

28, 31]. Small enough devices can be completely emptied, allowing precise control of the

number of electrons in each dot.

One of the most important features of double dots from the point of view of spin physics

is that the exchange between two electrons occupying the device can be tuned over a very

wide range. The exchange, defined as the energy difference between the lowest ms = 0spin-

triplet and spin-singlet levels, arises because Pauli exclusion requires overall antisymmetry

8
of the wavefunction under electron interchange. The ground state, with symmetric spa-

tial wavefunction at zero magnetic field, must therefore be spin-antisymmetric, i. e. a

singlet [33].

The magnitude of the exchange depends on the degree to which the two spatial wave-

functions overlap. When the electrons occupy separate dots, the exchange can be arbitrarily

small, but for electrons occupying the same dot, the exchange is as large as several hundred

microelectron volts, leading to a strong coupling of spin and charge degrees of freedom.

The most dramatic manifestation of this coupling is an asymmetry of electron tunneling

through the double dot, known as spin blockade [34, 32]. Spin blockade can be observed

when the device is configured close to the degeneracy of (1,1) and (0,2) occupations. An

electrical bias applied across the device introduces a chemical potential difference between

left and right leads, driving electrons to tunnel through the device by occupying states of

successively lower chemical potential. Under positive bias, the dots’ chemical potentials

can be tuned as shown in Fig. 1.4(a) and an electrical current flows (Fig. 1.4(c)). However,

under negative bias, if the electron loaded from the left forms a triplet state with the electron

already present on the right, it cannot escape to either side (Fig. 1.4(b)). Transport stops

until one of the spins is flipped or exchanged via higher-order tunneling with the leads, so

that the average current Id through the double dot is strongly suppressed (Fig. 1.4(d)).

The same asymmetry is also evident in charge sensing [32]. For the configuration

of Fig. 1.4(a), the device shuttles rapidly between the charge states (0,2), (1,1) and (0,1),

spending approximately equal time in each. The time-averaged sensor conductance will

therefore be an average of the values corresponding to these three charge states, as seen in

the lower triangle of (Fig. 1.4(e)). (In the upper triangle, the chemical potentials are such

that transport occurs via the sequence of charge transitions (0, 2) → (1, 1) → (1, 2) → (0, 2),

9
 b)

a) z'̂ 1
|↓〉
y'̂
x'̂
0

c)

B̂ y'

B̂ x'

d
|↑〉

Figure 1.5: (a) Bloch sphere of a single spin in the rotating frame. ESR bursts drive
rotations about two independent axes, allowing arbitrary spin rotations. (b) A device used
to demonstrate ESR in a single-spin quantum dot [17]. The ESR field B1 is driven with a
microwave current ICPS through a stripline. (c) Coherent oscillations of the electron spin
state, detected via the spin-blocked current Id [17].

and the average sensor conductance is again an average of values corresponding to these

states.) However, for the spin-blocked configuration, the (1,1) charge configuration domi-

nates, as seen in Fig. 1.4(f).

1.5 Electron spin resonance

To see how electron spin resonance allows arbitrary single-qubit operations, consider the

evolution of an electron spin under the influence of a static field B0 defining the z-axis, and

an oscillating field 2B1 along the x-axis [35]. Schrödinger’s equation is:

d|ψi
i~ = −gµB (B0 Sz + 2B1 Sx cos(ωt + φ))|ψi, (1.3)
dt

10
where ψi is the spin wavefunction, g the electron g-factor, µB the Bohr magneton, S =

(Sx , Sy , Sz ) the electron spin, and φ the phase of the oscillating field. This equation has

a simpler form in a frame rotating about the z axis at frequency ω. To transform to this

frame, we make the substitutions:

S = eiωtSz S0 e−iωtSz (1.4)

|ψi = e−iωtSz |ψ 0 i. (1.5)

The Schrödinger equation becomes:

d|ψ 0 i  
i~ = ~ (ω − ωL )Sz 0 + Ω(Sx0 cos φ + Sy0 sin φ) + terms oscillating with frequency 2ω
dt
(1.6)

where

ωL ≡ gµB B0 /~ (1.7)

Ω ≡ gµB B1 /2~ (1.8)

are the Larmor and Rabi frequencies. Under the approximation (almost always justified)

Ω  ω, the effect of the oscillating terms averages away, and so in the rotating frame the

spin evolves as though under the influence of a static magnetic field

 
 Ω cos φ 
 
0 ~  
B =  Ω sin φ . (1.9)
gµB 



 
ω − ωL

Usually the oscillating magnetic field is applied with angular frequency ω = ωL , so that

by applying ESR bursts with appropriate phase, the electron spin can be rotated around

any axis in the (x0 , y 0 ) plane (Fig. 1.5). By combining up to three rotations around x0 and

y 0 axes, arbitrary rotations on the Bloch sphere can be performed, and therefore arbitrary

unitary transformations of the spin [36].

11
 ESR was first demonstrated in a quantum dot in Ref. [17]. To create large enough

oscillating magnetic fields, a microwave stripline was deposited over the double-dot de-

vice (Fig. 1.5(b). Coherent spin rotations were detected via the breaking of spin blockade

when microwaves were applied (Fig. 1.5)(c).

There are several technological reasons why one may wish to avoid using magnetically

driven ESR as a means of spin manipulation. Generation of local time-varying fields implies

on-chip ohmic heating, whereas quantum dot operation requires dilution refrigerator tem-

peratures. Since the power density needed to generate a given B1 scales with the inverse

square of the device dimension, this limitation becomes more severe for nanoscale devices.

(Using superconducting striplines is of limited help, as the current density used in Ref. [17]

approaches the critical current of Nb [37]) Apart from ohmic heating, the stripline design

shown in Fig. 1.5(b) would also disturb nearby devices or a charge sensor.

One could hope to circumvent these difficulties by applying a global B1 [38]. However,

this creates electric fields and eddy current heating throughout the chip, as well as posing

the problem of how to address individual spins. These difficulties provide motivation for

seeking to manipulate spins electrically.

1.6 The singlet-triplet basis

In the singlet-triplet basis, two electron spins are used to encode each qubit [39, 40]. In

return, all logic operations become possible by using gate voltage pulses rather than time-

varying magnetic fields.

The two states of the qubit are represented by the singlet S and ms = 0 triplet T

states of a double quantum dot in the (1, 1) charge configuration. The corresponding Bloch

sphere is shown in Fig. 1.6. Single-qubit manipulations are carried out by making use of

12
 |S〉
J

|↓↑〉 |↑↓〉

∆Bz

|T〉

Figure 1.6: Bloch sphere in the singlet-triplet basis. The two logic states are represented
by the singlet S and ms = 0 triplet T states of the (1, 1) charge configuration. The two
hyperfine-split states | ↑↓i and | ↓↑i are represented by points on the equator of the sphere.
Single-qubt rotations are accomplished via rotations about two independent axes driven by
exchange J and a magnetic field or hyperfine difference between dots ∆Bz .

two interactions, whose combination allows for any unitary transformation in this basis:

Exchange J, which drives rotations about the z-axis, and a magnetic field difference ∆Bz

between the dots, which drives rotations about the x-axis.

The qubit can be initialized and read out as follows [41, 40] . The device is configured

in (0, 2), where tunneling to the leads causes rapid relaxation to the singlet state. Rapidly

pulsing the gate voltages (in ∼ 1 ns) to configure the device in (1, 1) preserves the electron

spin, preparing the state S. To read out the qubit, this sequence is reversed. For state S,

the two electrons can recombine as a singlet in (0, 2); however, for state T , the electrons

remain on separate dots. This difference in the charge state, detected with a charge sensor,

allows electrical readout of the qubit.

All the elements for single-qubit operation have now been realized, although not in the

same device. Tuning the wavefunction overlap between electrons using gate voltages allows

rapid control of the exchange interaction, demonstrated by coherent qubit rotations [41].

Controlled rotations about an orthogonal axis can be achieved by establishing an energy

13
difference between | ↑↓i and | ↓↑i states using a controlled hyperfine interaction [42]. Opti-

mized charge sensing permits single-shot qubit readout [19]. And coherence times over 1 µs

can be achieved using a spin echo sequence to suppress hyperfine dephasing [41].

1.7 Summary of contributions

The key contributions reported here are as follows:

• In Chapter 2, electrically driven spin resonance of a single electron is demonstrated.

By studying the magnetic field dependence of the resonance strength, it is shown that

a novel mechanism couples the electric field to the electron spin, namely a fluctuating

hyperfine field. Driving the resonance is found to create a nuclear polarization in the

quantum dot. Using a micromagnet to create a magnetic field gradient across the

device, a technique to address individual spins in a multi-electron device is presented.

• In Chapter 3, the exchange between a pair of electrons in a double quantum dot

is carefully adjusted and measured, and its effect on spin dephasing is studied. An

exchange comparable to or stronger than the hyperfine coupling was found to lead

to a saturation of dephasing, in agreement with a prediction from quasistatic theory.

The spin dephasing was also measured as a function of time, and oscillations of the

spin-flip probability were observed, again in agreement with theory.

• In Chapter 4, a triple quantum dot device incorporating two point contact charge

sensors was fabricated. The technique of multiplexed reflectometry, not previously

applied to point contacts, was used to monitor both charge sensors at MHz frequencies.

The device was tuned to a charge configuration where it could be operated as a qubit,

and preparation, readout, and coherent control of the electron spins was demonstrated.

14
• In Chapter 5, a prototype two-qubit device was fabricated and measured. Although

spin-qubit operation proved impossible in this device, the coupling strength was mea-

sured for the first time in the qubit charge configuration and found to be in principle

sufficient for two-qubit gates.

15
Chapter 2

Gate-driven hyperfine-mediated
electron spin resonance

E. A. Laird, C. Barthel, E. I. Rashba, C. M. Marcus

Department of Physics, Harvard University, Cambridge, Massachusetts 02138

M. P. Hanson, A. C. Gossard
Department of Materials, University of California, Santa Barbara, California 93106

A recently discovered mechanism of electric dipole spin resonance, mediated by the

hyperfine interaction, is investigated experimentally and theoretically. The effect is studied

using a spin-selective transition in a GaAs double quantum dot. The resonant frequency

is sensitive to the instantaneous hyperfine effective field, revealing a nuclear polarization

created by driving the resonance. A device incorporating a micromagnet exhibits a magnetic

field difference between dots, allowing electrons in either dot to be addressed selectively.

An unexplained additional signal at half the resonant frequency is presented.2

2
This chapter is adapted from Refs. [43, 44] with permission, c (2007) by the American
Physical Society].

16
2.1 Introduction

Electric dipole spin resonance (EDSR) is a method to electrically manipulate electron spins.

In this technique, two fields are applied; a static magnetic field B and an oscillating elec-

tric field Ẽ(t) resonant with the electron precession (Larmor) frequency [45, 46, 47, 48].

Spin resonance techniques are of interest for quantum computing schemes based on single

electron spins, because they allow arbitrary one-qubit operations [9]. Single-spin EDSR

is a particularly desirable experimental tool because it allows spin manipulation with-

out time-dependent magnetic fields, which are difficult to generate and localize at the

nanoscale [49, 50, 51, 17].

Achieving EDSR requires a mechanism to couple Ẽ to the electron spin σ. This coupling

can be achieved by the traditional spin-orbit interaction, which couples σ to the electron

momentum k, or by an inhomogeneous Zeeman interaction, which couples σ to the electron

coordinate r [52, 48, 53, 54, 55]. Single-spin EDSR has recently been achieved in quantum

dots using both techniques [56, 57].

Recently, we presented an experimental and theoretical study of a novel EDSR effect

mediated by the spatial inhomogeneity of the hyperfine nuclear field [43]. An electron

moving under the influence of the electric field Ẽ(t) experiences this inhomogeneity as an

oscillating hyperfine coupling which drives spin transitions. In this paper, we illuminate the

underlying physics and present new experimental data on a still unexplained phenomenon

at half the resonant frequency.

This EDSR effect is observed via spin-blocked transitions in a few-electron GaAs double

quantum dot [58]. As expected for a hyperfine mechanism, but in contrast to k−σ-coupling

mediated EDSR, the resonance strength is independent of B at low field and shows, when

averaged over nuclear configurations, no Rabi oscillations as a function of time. We find

17
 a) b) -3
gs(10 e /h)
2
Vsd
500 nm 0 2 4
-410 C
B (1,1)

VL (mV)
(1,2)
VQPC
[110]
gs M
[110]
-420 (0,1)
(0,2)
L R
-430 -420 -410
VR (mV)

c)
T
Vsd J02
S T±

S, T0

Prepare Manipulate Measure

(at M) (at C) (at M)

Figure 2.1: (a) Micrograph of a device lithographically identical to the one measured,
with schematic of the measurement circuit. Quantum dot locations are shown by dashed
circles, and a bias Vsd drives sequential tunneling in the direction marked by black arrows.
The conductance gs of the QPC on the right is sensitive to the dot occupation. The di-
rection of the magnetic field B and the crystal axes are indicated. (b) QPC conductance
gs measured at Vsd ∼ 600 µeV near the (1,1)-(0,2) transition. Equilibrium occupations for
different gate voltages are shown, as are gate voltage configurations during the measure-
ment/reinitialization (M) and manipulation (C) pulses. The two white dashed triangles
outline regions where transport is not Coulomb blocked; the solid black line outlines where
spin blockade is active. A plane background has been subtracted. (c) Energy levels of the
double dot during the pulse cycle (See text).

that at large B driving the resonance creates a nuclear polarization, which we interpret as

the backaction of EDSR on the nuclei [59, 60, 17, 61, 62]. Finally, we demonstrate that

spins can be individually addressed in each dot by creating a local field gradient.

2.2 Device and measurement

The device for which most data is presented (Figure 1(a)) was fabricated on a GaAs/Al0.3 Ga0.7 As

heterostructure with two-dimensional electron gas (2DEG) of density 2 × 1015 m−2 and mo-

18
bility 20 m2 /Vs located 110 nm below the surface. Voltages applied to Ti/Au top gates

locally deplete the 2DEG, defining a few-electron double quantum dot. A nearby charge

sensing quantum point contact (QPC) is sensitive to the electron occupation (NL , NR ) of

the left (NL ) and right (NR ) dots [27, 30]. The voltages VL and VR on gates L and R can be

rapidly pulsed; in addition, L is coupled to a microwave source. The static magnetic field

B was applied in the plane of the heterostructure, and measurements were performed in a

dilution refrigerator at 150 mK electron temperature.

The characteristic feature of tunnel-coupled quantum dots is a discrete electron energy

spectrum. An overall shift to the spectrum, proportional to the electron occupation, is in-

duced by VL and VR , which therefore determine which occupation is energetically favoured.

Figure 1(b) shows the QPC conductance gs as a function of VL and VR ; different conduc-

tances correspond to different (NL , NR ). For most VL , VR configurations, only one value of

(NL , NR ) is energetically accessible; these correspond in Figure 1(b) to regions of uniform

gs .

A bias Vsd applied across the device drives electron transport via sequential tunneling

subject to two constraints [8]. The first constraint, Coulomb blockade, arises because for

most gate configurations electrostatic repulsion prevents additional electrons from tunneling

onto either dot. This constraint inhibits transport except when VL , VR are tuned so that

three occupation configurations are near-degenerate. The energy cost of an extra electron

tunneling through the device is then small enough to be provided by the bias voltage.

Values of VL and VR satisfying this condition correspond to the two white dashed triangular

regions marked in Figure 1(b), for which transport is permitted via the transition sequences

(0, 2) → (0, 1) → (1, 1) → (0, 2) or (0, 2) → (1, 2) → (1, 1) → (0, 2).

A second constraint, spin blockade, is caused by the Pauli exclusion principle, which

19
leads to an intra-dot exchange energy J02 in the right dot [34, 32]. As shown in the first panel

of Figure 1(c), the effect of this exchange is to make the (1, 1) → (0, 2) transition selective

in the two-electron spin state, inhibited for triplet states but allowed for the singlet. The

hyperfine field difference between dots converts the ms = 0 component T0 of the blocked

triplet T to an unblocked singlet S within ∼ 10 ns, as we have confirmed by the technique

of [41]. However, decay of ms = ±1 components T± requires a spin flip and therefore

proceeds much more slowly. This spin flip becomes the rate-limiting step in transport, and

so the time-averaged occupation is dominated by the (1,1) portion of the transport sequence

[32]. Gate configurations where spin blockade applies correspond to the black solid outlined

region of Figure 1(b); inside this region, gs has the value corresponding to (1,1). Any process

that induces spin flips will partially break spin blockade and lead to a decrease in gs .

Unless stated otherwise, EDSR is detected via changes in gs while the following cycle of

voltage pulses VL and VR [17] is applied to L and R (Figure 1(c)). The cycle begins inside

the spin blockade region (M in Figure 1(b)), so that the two-electron state is initialized

to (1, 1)T± with high probability. A ∼1 µs pulse to point C prevents electron tunneling

regardless of spin state. Towards the end of this pulse, a microwave burst of duration τEDSR

at frequency f is applied to gate L. Finally the system is brought back to M for ∼3 µs for

readout/reinitialization. If and only if a spin (on either dot) was flipped during the pulse,

the transition (1, 1) → (0, 2) occurs, leading to a change in average occupation and in gs .

If this transition occurs, subsequent electron transitions reinitialize the state to (1, 1)T± by

the end of this step, after which the pulse cycle is repeated. This pulsed EDSR scheme has

the advantage of separating spin manipulation from readout.

Changes in gs are monitored via the voltage VQPC across the QPC sensor biased at

5 nA. For increased sensitivity, the microwaves are chopped at 227 Hz and the change in

20
 !VQPC (nV)
0 50 100

0.6

f (GHz)
0.4

0.2

0.0
-100 -50 0 50 100
B (mT)

Figure 2.2: Signal of spin resonance δVQPC as a function of magnetic field B and microwave
frequency f . EDSR induces a breaking of spin blockade, which appears as a peak in the
voltage across the charge sensor δVQPC at the Larmor frequency. Field- and frequency-
independent backgrounds have been subtracted. Inset: Jitter of resonant frequency due to
random Overhauser shifts.

voltage δVQPC is synchronously detected using a lock-in amplifier. We interpret δVQPC as

proportional to the spin-flip probability during a microwave burst, averaged over the 100 ms

lock-in time constant.

2.3 EDSR spectroscopy

Resonant response is seen clearly as B and f are varied for constant τEDSR = 1 µs (Figure 2.)

A peak in δVQPC , corresponding to a spin transition, is seen at a frequency proportional

to B. This is the key signature of spin resonance. (A feature corresponding to lifted

spin blockade around B=0 is not seen or expected, because this measurement technique

is sensitive only to the differential effect of the microwaves [17].) From the slope of the

resonant line in Figure 2 a g-factor |g| = 0.39 ± 0.01 is found, typical of similar GaAs

21
 a)
560
!VQPC (nV)
100 550
2.9 GHz 60

B (mT)
80 40
50
20

!VQPC (nV)
40
0.17 GHz 0
60

peak
0.0 0.2 0.4 0.6 0.8
"EDSR (µs)
40

20
0.17 GHz, 44 mT
0 2.9 GHz, 550 mT

0.0 0.2 0.4 0.6 0.8

"EDSR (µs)
b)
!V QPC (nV)

30
20
peak

10
#R (10 s )
6 -1

1 0

!V QPC (nV)
9 30
0 1 2
8 "EDSR (µs) 20

peak
7 10
0
6
0.91 GHz 0 1
185 mT "EDSR (µs)
5
9 2 3 4 5 6 7 8 9
0.1 PMW (mW) 1

peak
Figure 2.3: (a) Measured EDSR peak strength δVQPC (symbols) versus microwave pulse du-
ration τEDSR for two frequencies, along with theoretical fits (curves) obtained by numerically
evaluating and scaling Equation (4) (see text). Both the applied power (PMW ∼ 0.6 mW)
and the calibrated power at the device are equal at these two frequencies (see footnote to
Section 4.1). Inset: Raw data from which the points in the main figure are extracted. Each
vertical cut corresponds to one point in the main figure. Jitter in the field position of the
resonance reflects time-dependent Overhauser shifts. (b) Spin-flip rate √ ΩR as a function of
applied microwave power PMW , along with a fit to the form ΩR ∝ PMW (dashed line).
peak
Insets: δVQPC versus τEDSR for two values of the microwave power, showing the fits from
which points in the main figure are derived.

devices [63, 64]. We attribute fluctuations of the resonance frequency (Figure 2 inset) to

Overhauser shift caused by the time-varying hyperfine field acting on the electron spin.

Their range is ∼ ±22 MHz, corresponding to a field of ∼ 4 mT, consistent with Overhauser

fields in similar devices [65, 66, 41].

Information about the EDSR mechanism can be obtained by studying the peak height as

a function of duration, strength, and frequency of the microwave burst (Figure 3). To reduce

the effects of the shifting Overhauser field, the microwave source is frequency modulated at

22
3 kHz in a sawtooth pattern with depth 36 MHz about a central frequency f . The resonance

line as a function of τEDSR is shown in the inset of Figure 3(a). For equal microwave power
peak
at two different frequencies f , the peak heights δVQPC are plotted in Figure 3(a) (main

panel). The two data sets are similar in turn-on time and saturation value; this is the case

for frequencies up to f = 6 GHz. From similar data (insets of Figure 3(b)), using theory to

be described, we extract the dependence of the spin-flip rate ΩR on microwave power PMW
peak
shown in the main panel of Figure 3(b). Coherent Rabi-type oscillations in δVQPC (τEDSR )

are not observed for any microwave power or magnetic field over the range measured.

The B-independence of the EDSR strength rules out spin-orbit mediated EDSR of the

k − σ type (either Dresselhaus or Rashba), for which the Rabi frequency is proportional to

B [48, 55, 56]. This is in contrast to the results of [56], where the spin-orbit effect was found

to dominate in a similar device to ours. A possible explanation is the device orientation

relative to B and the crystal axes. In both our experiment and [56], the gate geometry

suggests a dominant Ẽ(t) oriented along one of the diagonal axes ([110] or [110]), leading

to an in-plane spin-orbit effective field BSO

eff perpendicular to Ẽ(t). In our geometry (see

Figure 1(a)), this orientation of BSO

eff is parallel to B, and therefore ineffective at driving

spin transitions. In the geometry of [56], B is perpendicular to BSO

eff , so that the k − σ

spin-orbit mechanism becomes more efficient.

Although the strength of the EDSR line is field-independent, the hyperfine-induced

jitter becomes more pronounced with increasing field. As seen from the upper inset to

Figure 3(a), repeated scans over the resonance at high field display larger fluctuations in

the position of the peak center. This difference presumably reflects slower nuclear spin

diffusion [4] as well as incipient polarization (see Section 5). In none of the data was any

periodicity of the jitter detectible [67].

23
2.4 Theory

peak
A theoretical description of δVQPC (τEDSR ) and its dependence on B and PMW can be

obtained by modeling EDSR as arising from the coupling of an electron in a single dot to

an oscillating electric field Ẽ(t) and the hyperfine field of an ensemble of nuclei 3 [69, 70].

Then the center of the dot oscillates as R(t) = −eẼ(t)/mω02 , where m is the electron

effective mass, and ω0 is its confinement frequency in a parabolic dot. As a result, the

Hamiltonian of the hyperfine coupling of the electron spin S = σ/2 with spatial coordinate

r to nuclear spins Ij located at rj becomes time dependent, Hhf = AΣj δ(r+R(t)−rj )(Ij ·S).

Here A is the hyperfine coupling constant and the summation over j runs over all nuclear

spins. After expanding Hhf in R(t) (assumed small compared to the dot size) and averaging

over the orbital ground-state wave function ψ0 (r) of the dot, the time dependent part of

Hhf becomes Hhf (t) = J(t) · σ, where J(t) is an operator in all Ij . Choosing the z-axis in

spin space along B, the components of J(t) are Jz = 21 A ψ02 (rj )Ijz and
P
j

eA X
J± (t) = ψ0 (rj )Ẽ(t) · ∇ψ0 (rj )Ij± . (2.1)
mω02 j

The time-dependent off-diagonal components J± (t) drive EDSR, while the quasi-static

diagonal component Jz describes detuning of EDSR from the Larmor frequency ωL by an

√
amount ωz randomly distributed as ρ(ωz ) = exp(−ωz2 /∆2 )/(∆ π) [71]. The dispersions ∆

of the detuning and ΩR of the Rabi frequency are the root-mean-square values of Jz and

J± respectively. Whereas Jz is dominated by fluctuations of Ij symmetric about the dot

3
There exists some physical similarity between the hyperfine mechanism of EDSR described in this paper

and EDSR due to the coupling of electron spin to a random exchange field in semimagnetic semiconduc-

tors [68].

24
centre, J± is dominated by fluctuations antisymmetric in the Ẽ direction because Ẽ·∇ψ0 (r)

is odd with respect to the Ẽ projection of r. Finally,

r r
A I(I + 1)mω0 n0 eẼA I(I + 1)n0
∆= , ΩR = 2 , (2.2)
2~ 2π~d ~ ω0 8πd

with I = 3/2 for GaAs, n0 the nuclear concentration, and d the vertical confinement. It is

seen that ΩR is independent of B; this is in contrast to EDSR mediated by the conventional

k − σ spin-orbit coupling, where Kramers’ theorem requires that the Rabi frequency vanish

linearly as B → 0 [48, 72, 55].

In an instantaneous nuclear spin configuration with detuning δω = 2πf − (ωL + ωz ) and

Rabi frequency Ω, the spin-flip probability from an initial ↑ spin state is [73]:

Ω2
q 
p↓ (τEDSR ) = 2 sin2 (δω/2)2 + Ω2 τEDSR . (2.3)
(δω/2) + Ω 2

(We neglect the electron spin relaxation and nuclear-spin dynamics, which remain slow

compared with the Rabi frequency even in the EDSR regime [41, 70].) To compare with the

time-averaged data of Figure 3, we average Equation (2.3) over ωz with weight ρ(ωz ) and

over Ω with weight ρ(Ω) = 2Ω exp(−Ω2 /Ω2R )/Ω2R . This latter distribution arises because

the J± acquire Gaussian-distributed contributions from both Ijx and Ijy components of the

nuclear spins, hence it is two-dimensional. Averaging over ωz and Ω results in a mean-field

theory of the hyperfine-mediated EDSR. The resulting spin-flip probability

Z +∞ Z +∞
p↓ (τEDSR ; ∆, ΩR ) = dωz ρ(ωz ) dΩ ρ(Ω)p↓ (τEDSR ) (2.4)
−∞ 0

shows only a remnant of Rabi oscillations as a weak overshoot at τEDSR ∼ Ω−1

R . The ab-

sence of Rabi oscillations is characteristic of hyperfine-driven EDSR when the measurement

integration time exceeds the nuclear evolution time [4], and arises because J± average to

zero.

25
2.4.1 Comparison with data

To compare theory and experiment, the probability p↓ (τEDSR ; ∆, ΩR ) is scaled by a QPC

0 peak
sensitivity VQPC to convert to a voltage δVQPC . After scaling, numerical evaluation of

Equation (4) gives the theoretical curves shown in Figure 3(a). The parameters that de-

termine these curves are as follows: The Larmor frequency spread, ∆ = 2π × 28 MHz, is

taken as the quadrature sum of the jitter amplitude seen in Figure 2 and half the frequency

0
modulation depth, whereas ΩR and VQPC are numerical fit parameters. The 44 mT data

(green curve in Figure 3(a)) give ΩR = 1.7 × 106 s−1 and VQPC
0 0
= 2.4 µV. Holding VQPC to

this value, the 550 mT data give ΩR = 1.8 × 106 s−1 (blue curve in Figure 3(a)) and the

185 mT data give the dependence of ΩR on microwave power PMW shown in Figure 3(b).
√
The Rabi frequency ΩR increases as PMW (Figure 3(b)) and is independent of B, both

consistent with Equation (1). The B-independence of ΩR — also evident in the EDSR

intensity in Figure 2—and the absence of Rabi oscillations support our interpretation of

hyperfine-mediated EDSR in the parameter range investigated 4

Estimating ~ω0 ∼ 1 meV [64], Ẽ ∼ 3 × 103 Vm−1 at maximum applied power 5 , d ∼ 5

4
Although ΩR is found to be substantially smaller than the inhomogeneous dephasing rate 1/T2∗ ∼

100 MHz, oscillations would still be expected from a coherent process. Quasistatic dephasing processes, such

as the hyperfine process dominant here, allow Rabi oscillations to persist even when ΩT2∗ is considerably

less than unity [56, 74].

5
The power at the device is calibrated separately at each frequency from the threshold for non-resonant

lifting of spin blockade, which we take to indicate a microwave amplitude large enough to configure the dot

temporarily in a different charge state. This amplitude corresponds in Figure 1(b) to the 4mV distance from

point C to the nearest charge transition. The data in Figure 3(a) and the last data point in Figure 3(b) use

power 2 ± 1 dB below this threshold, corresponding to 3.2 mV. Dropped uniformly across the 500 nm width

of the device this voltage gives a field Ẽ ∼ 3 × 103 Vm−1 .

26
nm, and using values from the literature n0 = 4 × 1028 m−3 and An0 =90 µeV [75] we

calculate ΩR ∼ 11 × 106 s−1 , an order of magnitude larger than measured. The discrepancy

may reflect uncertainty in our estimate of Ẽ.

We have neglected any effect of residual exchange in (1,1) during the ESR burst. From

the width of the (1,1)-(0,2) charge transition, the interdot tunnel rate tc is deduced to be

much smaller than Boltzmann’s constant multiplied by the electron temperature [76]. From

the known (1,1)-(0,2) energy detuning  with gate voltages configured at C, we estimate

an upper bound on the (1,1) exchange t2c /  80 neV, of the same order as the hyperfine

coupling. Since different choices of point C give qualitatively similar results, we conclude

that (1,1) exchange is negligible.

Above, we generalized a mean-field description of the hyperfine interaction [77, 71] to

the resonance regime. Justification for this procedure was provided recently in [70]. A

distinctive feature of the mean-field theory is a weak overshoot, about 10 - 15%, that is
peak
expected in the data of Fig. 3(a) before δVQPC (τEDSR ) reaches its asymptotic value at

τEDSR → ∞. No overshoot is observed in the 550 mT data (blue symbols in Figure 3(a)),

which was taken in a parameter range where an instability of the nuclear polarization begins

to develop; see Section 5. For the 44 mT data (green symbols in Figure 3(a)), a considerable

spread of experimental points does not allow a specific conclusion regarding the presence

or absence of an overshoot. The theory of [70] suggests that the existence of the overshoot

is a quite general property of the mean-field theory. However, after passing the maximum,
2 2
the signal decays to its saturation value vary fast, with Gaussian exponent e−ΩR τEDSR . By

contrast, the first correction to the mean-field theory decays slowly, as 1/(N Ω2R τEDSR
2 ),

where N is the number of nuclei in the dot. As a result, the two terms become comparable
√
at τEDSR ∼ ln N /ΩR , which should make the maximum less pronounced. Because for

27
Figure 2.4: (a) Shift of the resonance frequency with time at constant B = 2390 mT, showing
build-up of nuclear polarization over ∼ 200 s. (b) A scheme to allow larger polarizations:
the microwave frequency is repeatedly scanned over the resonance while B is swept upwards
at 6 mT/min. Nuclear polarization partly counteracts B, moving the resonance away from
its equilibrium position (black diagonal line) by up to 840 mT. (c) Similar data taken at
lower microwave power and opposite frequency sweep direction, showing approximately the
equilibrium resonance position. (Grey scale as in (b)). (d) Similar data as in (b), with faster
sweep rate (22 mT/min), showing more clearly the displacement and subsequent return to
equilibrium of the resonance. ♦ marks the escape of the resonance from the swept frequency
window. In all plots, arrows denote frequency sweep direction.

√
N ∼ 105 the factor ln N ∼ 3, the corrections to the mean-field theory manifest themselves

surprisingly early, at times only about τEDSR ≈ 3/ΩR , making the overshoot difficult to

observe.

2.5 Nuclear polarization

Consistent with a hyperfine mechanism, this EDSR effect can create a non-equilibrium

nuclear polarization [61]. If f is scanned repeatedly over the resonance at high power, a shift

of the resonance develops (Figure 4(a)), corresponding to a nuclear spin alignment parallel to

B. The effect is stronger at higher B, and saturates over a timescale ∼ 200 s. In Figure 4(b),

28
we show how to build up a substantial polarization: While slowly increasing B, we scan

f repeatedly downwards, i. e., in the direction which tracks the moving resonance. The

resonance frequency remains approximately fixed, showing that the developing polarization

compensates the increase in B. From the maximum line displacement from equilibrium, an

effective hyperfine field of 840 mT can be read off, corresponding to a nuclear polarization

of ∼ 16%. Figure 4(c) shows similar data for lower power and opposite frequency sweep

direction, indicating the approximate equilibrium line position. Figure 4(d), similar to

Figure 4(b) but with a faster sweep rate, makes the displacement and eventual escape of

the resonance clearer although the maximum polarization is less.

The resonance shift is observed to be towards lower frequency, corresponding to a nuclear

polarization parallel to B. This can be understood if the pulse cycle preferentially prepares

the electron ground state T+ over T− , either because it is more efficiently loaded or because

of electron spin relaxation. EDSR then transfers this electron polarization to the nuclei [62].

We emphasize that the line shift is opposite to what is given by the usual Overhauser

mechanism for inducing nuclear polarization via electron resonance [78, 59].

2.6 Addressing individual spins

In quantum information applications, it is desirable to address individual spins selec-

tively [9]. A scheme to allow this is presented in Figure 5. In an otherwise similar de-

vice (Figure 5(a)), we incorporated a 100 nm thick micron-scale permalloy (84% Ni, 16%

Fe) magnet over 35 nm of atomic-layer-deposited alumina [54, 79]. This device was measured

with external field B normal to the heterostructure plane. A finite-element simulation of the

field Bmag due to the micromagnet, assuming complete permalloy magnetization along B,

yields the field profiles shown in Figure 5(b). The difference in total field B tot = |B + Bmag |

29
 a) 2!m c)
!VQPC (nV)
0 30
y

Frequency (GHz)
1.4

1.2
x
1.0

180 200 220

Magnetic Field (mT)

b)
component (mT) 0 mag
Bz
200

B (mT)
tot
tot
B
-50
180
mag

mag
Bx
B

-200 0 200 400

x (nm)

Figure 2.5: (a) A device similar to that of Figure 1, incorporating a micromagnet. (b)
Total field magnitude B tot (right axis) and the x and z components of the micromagnet
contribution Bmag (left axis), simulated at y = 0 for external field B = 200 mT along ẑ (out
of the plane). Bymag vanishes by symmetry. The gate layout is shown in the background.
(c) The associated split EDSR line. The lower resonance is stronger, as expected if the left
electron is confined close to the minimum of Btot .

between dots is ∼ 5 mT. As expected, the EDSR line measured in this device is frequently

split (Figure 5(c)). The splitting, 10 − 20 mT depending on precise gate voltage and pulse

parameters, is not observed without the magnet and presumably reflects the field difference

between dots. Since this splitting is considerably larger than the Overhauser field fluctua-

tions, spins in left and right dots can be separately addressed by matching f to the local

resonance condition [57].

The observation of a field difference between dots raises the possibility of EDSR driven

by a field gradient [57]. We cannot exclude a contribution from this effect to the signal in

Figure 5(c); however we did not observe the Rabi oscillations which would be expected if

the field gradient were the primary EDSR mechanism.

30
 -3 2
gs (10 e /h)
0.0 1.0

2.0 |g|=0.45

1.5

f (GHz) 1.0

0.5

0.0
0 100 200 300
B (mT)

Figure 2.6: Spin resonance signal (measured in conductance) in the device of Figure 5(a).
The EDSR signal shows up as a decrease in conductance as expected at frequency corre-
sponding to |g| = 0.45 (marked with dashed line.) An additional signal of opposite sign
appears at exactly half this frequency (dotted line). The larger splitting of both signals
below 100mT is consistent with a greater contribution of Bxmag to the total field difference
between dots. The horizontal features at 0.5 and 1.5 GHz result from resonances of the
microwave circuit. As in Figure 2, field- and frequency-independent backgrounds have been
subtracted, including any signal due to spin blockade lifting around B = 0 [17].

2.7 Open issues and discussion

Finally, we discuss unexplained behavior observed only in the device of Figure 5(a). For

the data described in this section, a simplified measurement scheme is used: Rather than

applying gate pulses, the device is configured in the spin blockade region (point M in Figure

1(a)) throughout. Microwaves are applied continuously, and spin resonance is detected by

directly measuring the QPC conductance gs .

As well as the EDSR signal at full frequency f = gµB B/h, an unexpected half-frequency

signal is sometimes seen (Figure 6.) Furthermore, depending on the exact gate configuration,

both full-frequency and half-frequency signals can have either sign; the change in gs at full

31
frequency is usually negative as expected, but sometimes positive close to degeneracy of

(1,1) and (0,2) charge configurations, where spin blockade is weakest [65]; by contrast, the

change in gs at half frequency is usually positive but sometimes negative far from degeneracy.

For most gate configurations, full-frequency and half-frequency signals have opposite sign,

as seen in Figure 6.

A half-frequency response is as far as we know unprecedented in spin resonance, and

suggests second harmonic generation (SHG) from the microwave field. SHG is generally a

non-linear phenomenon; it occurs for example in optical materials with non-linear polariz-

ability [80] and in non-linear electronic components. For hydrogenic donors in a semicon-

ductor, the nonlinear dependence of g-tensor on electric field has been predicted to drive

EDSR at subharmonics of the Larmor frequency [81]. In our system, a hyperfine field at a

harmonic of the microwave frequency arises if the confinement potential is non-parabolic.

However, SHG alone does not explain the sign of the conductance change seen at half-

frequency in Figure 6. The positive signal would be consistent with a reduced admixture

of (0,2), corresponding to a unexpected enhancement of the spin lifetime by microwaves.

Alternatively, a positive signal could be caused by an admixture of the (0,1) charge state;

but it is observed even for the gate configurations where (0,1) is energetically inaccessible

(in the top right of the spin blockade region of Figure 1(b)). Also, there is no reason why

(0,1) should be admixed for one resonance but not the other. These anomalous behaviours

are therefore left unexplained.

We acknowledge useful discussions with Al. L. Efros, H.-A. Engel, F. H. L. Koppens,

J. R. Petta, D. J. Reilly, M. S. Rudner, J. M. Taylor, and L. M. K. Vandersypen. We

acknowledge support from the DTO and from DARPA. E. I. R. was supported in part by

a Rutherford Professorship at Loughborough University, U.K.

32
Chapter 3

Effect of Exchange Interaction on

Spin Dephasing in a Double
Quantum Dot

E. A. Laird, J. R. Petta, A. C. Johnson, C. M. Marcus

Department of Physics, Harvard University, Cambridge, Massachusetts 02138

A. Yacoby
Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot 76100,
Israel

M. P. Hanson, A. C. Gossard
Department of Materials, University of California, Santa Barbara, California 93106

We measure singlet-triplet dephasing in a two-electron double quantum dot in the pres-

ence of an exchange interaction which can be electrically tuned from much smaller to much

larger than the hyperfine energy. Saturation of dephasing and damped oscillations of the

spin correlator as a function of time are observed when the two interaction strengths are

comparable. Both features of the data are compared with predictions from a quasistatic

model of the hyperfine field.2

2
This chapter is adapted from Ref. [69] [with permission, c (2006) by the American
Physical Society].

33
3.1 Introduction

Implementing quantum information processing in solid-state circuitry is an enticing ex-

perimental goal, offering the possibility of tunable device parameters and straightforward

scaling. However, realization will require control over the strong environmental decoher-

ence typical of solid-state systems. An attractive candidate system uses electron spin as

the holder of quantum information [9, 40]. In III-V semiconductor quantum dots, where

the highest degree of spin control has been achieved [82, 18, 83, 84, 85, 66, 41], the dom-

inant decoherence mechanism is hyperfine interaction with the lattice nuclei [86, 71, 77].

A recent experiment [41] studied this decoherence in a qubit encoded in a pair of spins

[39]. In this situation, the dynamics are governed by two competing effects: the hyperfine

interaction, which tends to mix the singlet and triplet basis states, and exchange, which

tends to preserve them.

The interplay of hyperfine and exchange effects has been studied recently via spin-

blockaded transport in two double-dot systems [67, 65]. Oscillations and bistability [67]

of the leakage current, as well as suppression of mixing with stronger exchange [65] were

observed. The topic also has a long history in physical chemistry: recombination of a radical

pair created in a triplet state proceeds significantly faster for radicals containing isotopes

whose nuclei carry spin [87, 88]. By lifting the singlet-triplet degeneracy, the exchange

interaction suppresses spin transitions; its strength can be deduced from the magnetic field

dependence of the recombination rate [89, 90]. However, exchange is difficult to tune in situ

in chemical systems.

In this Letter, singlet correlations between two separated electrons in a GaAs double dot

system are measured as a function of a gate-voltage tunable exchange J and as a function

of time τS following the preparation of an initial singlet. This study gives insight into

34
 a) T b) T-
EZ
T0
T+
L R gs 2tc

c) 0 15 J
δgS(10-3e2/h)
-455
(1,2) S (0,2)S
S
P

VL (mV)
-458 (1,1) P'
P′,M

time
τS S
(0,2)
-461 (0,1) M
P

-335 -332 -329 0 ε

VR (mV)

Figure 3.1: (a) Micrograph of a device with the same gate design as the one measured (Scale
bar = 500 nm.) Voltages applied to gates L and R adjust the double dot detuning, . Gate
T sets the inter-dot tunnel coupling. The conductance gs of a nearby sensor quantum point
contact monitors the average occupation of each dot. (b) Upper panel: Level diagram for
the double dot near the (1,1)-(0,2) transition ( = 0) plotted versus . Exchange (J) and
Zeeman (EZ ) energies are indicated. • denotes the S-T+ degeneracy. Labels (m, n) denote
the occupancies of the left and right dot respectively. Lower panel: The prepare (P, P0 ) -
separate (S) - measure (M) pulse scheme. ∼90% of the cycle is spent in M. (c) gs close to
the (1,1)-(0,2) transition during application of pulses, showing the pulse triangle (marked)
and the positions of points P, P0 , S and M. A background plane has been subtracted.

the interplay of local hyperfine interactions and exchange in a highly controllable quantum

system. We measure the probability PS (τS ) that an initial singlet will be detected as a

singlet after time τS for J ranging from much smaller than to much greater than the rms

hyperfine interaction strength in each dot, Enuc . When J  Enuc , we find that PS decays

on a timescale T2∗ ≡ ~/Enuc = 14 ns. In the opposite limit where exchange dominates,

J  Enuc , we find that singlet correlations are substantially preserved over hundreds of ns.

In the intermediate regime, where J ∼ Enuc , we observe oscillations in PS that depend on

the ratio Enuc /J. Our results show that a finite exchange energy can be used to extend

spin correlations for times well beyond T2∗ .

These observations are in reasonable agreement with recent theory, which predicts a

singlet probability (assuming perfect readout) PS0 (τS ) that exhibits damped oscillations as

35
a function of time and a long-time saturation that depends solely on the ratio Enuc /J [91].

To compare experiment and theory quantitatively we introduce an empirical visibility, V ,

to account for readout inefficiency, PS (τS ) = 1 − V (1 − PS0 (τS )).

3.2 Device

The device used in the experiment, shown in Fig. 1(a), is fabricated on a GaAs/Al0.3 Ga0.7 As

heterostructure with a two-dimensional electron gas (density 2 × 1015 m−2 , mobility 20

m2 /Vs) 100 nm below the surface. Ti/Au top gates define a few-electron double quantum

dot. The inter-dot tunnel coupling tc and (0,2)-(1,1) detuning  are also separately tun-

able. A charge-sensing quantum point contact with conductance gs ∼ 0.2e2 /h allows the

occupancy of each dot to be separately measured [27, 30]. We monitor gs using a lock-in

amplifier with a 1 nA current bias at 335 Hz, with a 30 ms time constant.

3.3 Methods

Measurements were made in a dilution refrigerator at electron temperature Te ≈ 100 mK

measured from the width of the (1,1)-(0,2) transition [76]. Gates L and R (see Fig. 1)

were connected via filtered coaxial lines to the outputs of a Tektronix AWG520. We report

measurements for two settings of tunneling strength, controlled using voltages on gate T

and measured from the width of the (1,1)-(0,2) transition: tc ≈ 23 µeV (“large tc ”) and tc <

9 µeV (“small tc ”) [76]. Except where stated, measurements were made in a perpendicular

magnetic field of 200 mT, corresponding to a Zeeman energy EZ = 5 µeV  Enuc .

Figure 1(b) shows the relevant energy levels near the (1,1)-(0,2) charge transition, where

measurements are carried out, as a function of energy detuning  between these two charge

36
 a) c) 0.8 1.0 P (τ =200ns)
200 small tc S S

large tc

tR(ns)
0.6
100
20

B (mT)
-2.0 -1.5 -1.0 -0.5 0.4
ε (mV)

J(µeV)
b)
0.1 10
0.2

J(µeV)
6
5
4
3 small tc
large tc
2

2 9 8 7 6 5 4
0 0.0
1 -0.8 -0.4 0.0
-ε (mV) ε (mV)

Figure 3.2: (a) Period tR of first Rabi oscillation versus exchange point detuning for small
and large tunnel coupling. (b) Exchange energy as a function of detuning, deduced from
the data in (a), together with empirical power-law fits J ∝ ||−1.4±0.1 . tR corresponding
to the fits is shown as curves in (a). (c) Color scale plot of PS as a function of S-point
detuning and magnetic field B obtained using the pulse sequence in Fig. 1(b). The bright
band indicates rapid decoherence where J = gµB B. The white points and the dashed line
are the same data and fits plotted in (b).

states. With tc =0, the (1,1) singlet S and ms = 0 triplet T0 are degenerate; the ms = ±1

triplets T± are split off in energy from T0 by ∓EZ . Finite tc leads to hybridization of the

(0,2) and (1,1) singlets, inducing an exchange splitting J between S and T0 . The (0,2)

triplet (not shown) is split off by the much larger intra-dot exchange energy J(0,2) ∼ 600

µeV [32] and is inaccessible. Rapid mixing due to hyperfine interaction occurs between

states whose energies differ by less than Enuc . This occurs at large negative  (lower left of

Fig. 1(b)), where S and T0 mix, and at J() = EZ (black dot in Fig. 1(b)), where S and T+

mix.

A cycle of gate configurations is used to prepare and measure two-electron spin states

[41], as illustrated in Fig. 1(b). A 200 ns preparation step (denoted P in Fig. 1) configures

the dot in (0,2) at a position where the series (0,2)T→(0,1)→(0,2)S is energetically allowed

and occurs rapidly, giving efficient initialization to a singlet. The gates then shift (waiting

200 ns at P0 to reduce pulse overshoot) to a separation point (S) in (1,1) for a time τS

37
during which singlet-triplet evolution occurs. Finally, the gates are set to the measurement

point (M) for τM = 5 µs, for spin-to-charge conversion. Inside the pulse triangle marked

in Fig. 1(c), the triplet states will remain in (1,1) over the measurement time τM [66, 92].

Since ∼90% of the pulse cycle is spent at M, the relatively slow measurement of the sensor

gs gives a time-averaged charge configuration at the M point. The time-averaged gs signal

is calibrated to give a singlet state probability PS (τS ) by comparing the signal in the pulse

triangle with the values measured in the (1,1) and (0,2) regions of the charge stability

diagram. When the gates are configured so that M is outside the pulse triangle in (0,2),

both singlet and triplet relax rapidly to (0,2); gs in this region defines PS = 1. When M is

in (1,1), the value of gs defines PS = 0.

3.4 Measurement of exchange energy, hyperfine interaction

strength and readout visibility

We first measure J(), Enuc , and V at two values of tc , allowing the saturation probability

PS (∞) to be measured as a function of J. This saturation probability is found to depend

on the ratio Enuc /J approximately as predicted by theory [91]. We then measure the

time evolution PS (τS ), which shows damped oscillations, also in reasonable agreement with

theory [91]. J() is measured using the Rabi (or Larmor) sequence described in Ref. [41],

in which an adiabatic (compared with Enuc ) ramp over 1 µs to (1,1) is used to prepare

and measure the electron spin state in the {|↑↓i , |↓↑i} basis. An exchange pulse produces

coherent rotations with a period tR (shown in Fig. 2(a)) from which we deduce the exchange

38
coupling J() = h/tR 3 . Values of J() for small and large tc are shown in Fig. 2(b), along

with a fit to an empirical power-law form J ∝ −α , giving α ∼ 1.4 4 . In Fig. 2(c), these values

of J() are compared with the results of an alternative method in which rapid dephasing

at the S-T+ degeneracy produces a dip in PS when the value of  at the S point satisfies

J() = EZ . J() can then be measured from a knowledge of the field, using EZ = gµB B

where µB is the Bohr magneton, and taking the value g = −0.44, measured (using an

in-plane field) in a different quantum dot device on made from the same wafer [93]. J()

measured by this technique is in qualitative agreement with the power-law derived from

Fig. 2(b); discrepancies may be due to an anisotropic g-factor, nuclear polarization effects,

or may indicate a dependence of J() on field. Since the first method more closely matches

the conditions under which data in the rest of the paper was taken and is more precise in

the range of J of interest, we henceforth take the function J() from Fig. 2(b).

Parameters Enuc and V are extracted from PS (τS ) measured for the S-point at large

negative , where J  Enuc . In this regime the initial singlet evolves into an equal mixture

of singlet and triplet with characteristic time h/Enuc . PS (τS ) for small and large tc (shown

in the insets of Fig. 3) are fit to the form for PS0 (τS ) given in [91], with fit parameters

Enuc = 45 ± 3 neV (47 ± 4 neV) and V = 0.53 ± 0.06 (0.46 ± 0.06) for small (large) tc 5 .

These hyperfine energies correspond to an effective hyperfine field of 1.8 mT, similar to the

3
When J . Enuc , J must be corrected downwards slightly because precession in the nuclear field enhances

the average Rabi frequency. The correction to J never exceeds 13%.

4
A simple level anticrossing with tc and  independent would give J ∝ −1 . The discrepancy may be due

to a detuning-dependent tc

5
The dependence of V on tc is not understood. The effect of alternative visibility parameters on the

predictions of Figs. 3 and 4 is simply to scale them towards or away from PS = 1

39
value measured previously on this device [41]. The fit function PS0 (τS ) depends on J at this

detuning, which is too small to measure directly. To perform the fit, we choose a value J()

extrapolated using the power-law from Fig. 2b; however, since J is so small at this large ||,

the best-fit parameters are essentially independent of details of the extrapolation, and, for

example, are within the error bars for the alternative extrapolations J ∝ ||−1 and J = 0.

3.5 Saturation singlet probability

The variance of the hyperfine field arises either from a quantum superposition of nuclear-

field eigenstates or through dynamics of the nuclear system on timescales faster than the

measurement averaging time [91]. PS0 (τS ) is calculated by integrating Schrödinger’s equation

from the initial singlet for given nuclear field and averaging the resulting singlet probability

over nuclear fields. The resulting PS0 (τS ) shows a range of interesting behavior depending

on the relative magnitudes of J and Enuc [91]: In the limit J = 0, PS0 (τS → ∞) rapidly

saturates to 1/2. As J is increased, hyperfine dephasing becomes less effective, with PS0 (∞)

saturating at progressively higher values, approaching unity when J  Enuc , and following

a universal function of Enuc /J. As a function of τS , PS0 (τS ) is predicted to undergo damped

oscillations, which when plotted versus τS J follow another universal function of Enuc /J and

exhibit a universal phase shift of 3π/4 at large τS J.

Knowing J() and Enuc allows the long-time (τS  h/J) saturation of the measured PS

to be compared with theory [91]. We set τS = 400 ns and sweep the position of the S-point.

For small and large tc , PS (400 ns) is plotted in Fig. 3 as a function of Enuc /J, where Enuc is

obtained from the fits described above and J() are taken from Fig. 2. At the most negative

detunings (in the regions marked by gray bars in Fig. 3) J is too small to be measured by

either Rabi period or S-T+ degeneracy methods; instead, J() is found by extrapolating

40
 a)
-ε(mV)
0.1 0.2 0.5 1 2 3 4
1.0

small tc

PS
1.0

PS
0.8
0.8

0 50 τ (ns) 100
S

0.1 0.2 0.5 1 2 3 4 5 6 78

Enuc/J
b) -ε(mV)
0.2 0.5 1 2 3 4 5 6
1.0

large tc
PS

1.0
PS

0.8
0.8

0 50 τ (ns) 100
S

0.1 0.2 0.5 1 2 3 4 5 6 78

Enuc/J

Figure 3.3: (a) Inset: PS (τS ) for small tc and = -5.5 mV, with fit (see text) giving
Enuc =45± 3 neV and V =0.53±0.06. Main panel: Measured PS (τS =400 ns) (points) plotted
against Enuc /J. Open symbols correspond to PS in the traces of Fig. 4(a) at the largest
τS measured for each . Curve shows theoretical dependence (from [91]) of PS (τS → ∞) on
Enuc /J, taking into account the measurement fidelity deduced from the inset. The gray bar
along the top axis indicates the region where J() is extrapolated (see text). Dashed lines
indicate the theoretical predictions (plotted as functions of ) if an alternative extrapolation
J ∝ ||−1 is chosen in this region. (b)Large tc data. The fit to the inset gives Enuc = 47 ± 4
neV and V = 0.46 ± 0.06, from which the theoretical saturation PS (curve in main panel) is
calculated. Open symbols correspond to the large-τS values in Fig. 4(b). Error bars on the
filled symbols shows the uncertainty in PS arising from charge noise in the sensing point
contact.

the power-law fits (Fig. 2.) Agreement with theory (discussed below) is insensitive to the

details of the extrapolation, as shown by the dashed lines in Fig. 2.

The long-τS PS data shown in Fig. 3 agrees fairly well with the saturation values pre-

dicted from [91], taking into account the visibility (assumed independent of ) obtained

from the insets. In particular, PS has the same dependence on Enuc /J at both values of tc

measured, even though the function J() depends on tc . PS is up to ∼ 0.06 smaller than

predicted at the largest detunings; both cotunneling and nuclear decorrelation over the du-

ration of the separation pulse tend to equalize singlet and triplet occupations, although it

41
is unclear whether they are the cause of this effect.

3.6 Time-dependence of the singlet probability

We next investigate the time dependence of PS (τS ) at finite J. For five (two) S-point

detunings at small (large) tc , PS (τS ) was measured out to τS J/~ ≈ 15. The results are

shown in Fig. 4, together with the predicted time evolution from [91] with values for V and

Enuc taken from fits shown in the insets of Fig. 3. Because PS remains close to unity, these

data are particularly sensitive to calibration imperfections caused by quantum point contact

nonlinearities and noise in the calibration data, whose effect to lowest order is to shift the

data vertically. Traces in Fig. 4 are therefore shifted vertically to satisfy the constraint

PS (τS = 0) = 1. In no case was this greater than ±0.05. Here and in Fig. 3, the error bars

reflect uncertainty in PS from charge noise in the sensing point contact; additional scatter

in the data may be due to long nuclear correlation times [41, 65].

Damped oscillations are observed as predicted in [91]; however, even after taking account

the empirical visibility factor, the amplitude of the oscillations is less than expected. This is

likely due to the finite rise time of the separation pulse and to switching noise, which make

each trace effectively an average over a range of J values. Where the amplitude is large

enough for the period and phase of the oscillations to be made out, these approximately

match the predictions of [91], although with two significant departures: The topmost trace,

with smallest Enuc /J, does not show clear oscillations, and the expected shift of the first

minimum to smaller τs J at intermediate J is not observed. We do not understand the origin

of these effects. The amplitude of the oscillations falls off too rapidly for the expected 3π/4

phase shift at large τS J to be visible. Similar oscillations of PS are predicted close to the

S-T+ degeneracy with a characteristic frequency ∼ ∆ = J − EZ . We have searched for these

42
 a)
1.0
small tc

J =131 neV (Enuc/J = 0.36)

J = 60 neV (Enuc/J = 0.77)

0.8

PS
J = 42 neV (Enuc/J = 1.1)
J = 25 neV (Enuc/J = 1.8)

0.6 J ≈ 8 neV (Enuc/J ≈ 5.6)

b)1.0
large tc

J = 118 neV (Enuc/J = 0.40)

PS

J = 52 neV (Enuc/J = 0.91)

0.8
0 5 10 15
τ J/h
S

Figure 3.4: (a) Symbols: Experimental PS (τS ) at small tc for various J, plotted as a function
of τS J/~. Curves: Predictions from [91] using Enuc and V fit from Fig. 3(a). Adjacent traces
after the first are offset by 0.05 for clarity. (b) Corresponding data and theory for large tc .
Lower trace is offset by 0.05 for clarity. Error bars reflect the contribution of sensor charge
noise.

oscillations but do not observe them. We believe the reason for this is that ∆ varies much

more rapidly with  in this region than J does at the S-T0 near-degeneracy; the oscillations

are therefore washed out by switching noise and pulse overshoot.

3.7 Summary and acknowledgements

In summary, after including the measured readout efficiency, we find that the singlet corre-

lator shows damped oscillations as a function of time and saturates at a value that depends

only on Enuc /J. Both these features are qualitatively as expected from theory [91]; some of

the departures from expected behavior may be qualitatively accounted for by cotunneling

and nuclear decorrelation (which tend to equalize singlet and triplet probabilities at long

times), and charge noise (which tends to smear out the oscillations seen in Fig. 4.)

We acknowledge useful discussions with W. Coish, H. A. Engel, D. Loss, M. Lukin, J. M.

43
Taylor. This work was supported by DARPA-QuIST and the ARO/ARDA STIC program.

44
Chapter 4

Coherent operation of an
exchange-only qubit

E. A. Laird, C. M. Marcus
Department of Physics, Harvard University, Cambridge, Massachusetts 02138

J. M. Taylor
National Institute of Standards and Technology, Gaithersburg, Maryland 20899

D. P. DiVincenzo
IBM T. J. Watson Research Center, Yorktown Heights, New York 10598

M. P. Hanson, A. C. Gossard
Department of Materials, University of California, Santa Barbara, California 93106

A triplet quantum dot is tuned to a charge configuration useful for operation as an

exchange-controlled spin qubit. For qubit readout, we apply the technique of frequency-

multiplexed reflectometry to a pair of charge sensing quantum point contacts. With a series

of gate pulses applied to the device, we demonstrate qubit initialization and readout and

use the exchange interaction to carry out coherent spin manipulation.

45
4.1 Introduction

Electron spins confined in semiconductor quantum dots are an attractive physical basis for

quantum computing because of their long coherence times and potential for straightforward

scaling [9, 94, 40]. In the simplest proposal [9], the single-spin states | ↑i and | ↓i form

the logical basis, with single-qubit operations via electron spin resonance (ESR) and two-

qubit operations via exchange. The technical difficulty of single-spin ESR [17] inspired

an alternative scheme in which the logical basis is formed from the singlet and triplet

states of a spin pair [39, 41, 40]. No time-varying ESR field is then necessary, but an

inhomogeneous static magnetic field is required for full single-qubit control [42]. Using three

electron spins to represent each qubit removes the need for an inhomogeneous field; exchange

between adjacent spins suffices for one and two-qubit operations as well as state preparation

and readout [94]. Here we present a three-spin qubit defined in a triple quantum dot.

A pair of proximal charge sensors, monitored independently using frequency-multiplexed

reflectometry, allows rapid readout of the electron configuration. Pulsing gate voltages to

tune the energy levels of the system, we demonstrate initialization, coherent operation, and

measurement of the qubit.

The interactions of three spins have been explored experimentally [95] and theoreti-

cally [96] in the context of physical chemistry, where the recombination of two radicals,

originally in an unreactive triplet state, can be catalyzed by exchange with a third spin.

Few-electron triple quantum dots [97, 98, 99]have been used to realize charge reconfigura-

tions corresponding to the elementary operations of quantum cellular automata have been

measured [100], however, tunable spin interactions have not yet been demonstrated [101].

46
a) c) "D'±1/2 œ
J12
J12 J23 "D±1/2 œ
b)
" œ "Q -3/2 œ "Q -1/2 œ "Q +1/2 œ "Q +3/2 œ
120º

"œ" œ "D'-1/2œ "D'+1/2 œ

"D -1/2 œ "D +1/2 œ

J23
"œ" œ

" œ"œ
"D'±1/2 œ
"D'-1/2 œ "D'+1/2 œ

" œ"œ "D -1/2 œ "D +1/2 œ

"D±1/2œ

Figure 4.1: An exchange-only qubit. a, Electron spins in three adjacent quantum dots are
coupled by nearest-neighbour exchange. b, The eight states of the system form a quadruplet
Q and two doublets D0 and D characterized by the multiplicity (singlet or triplet) of the
leftmost pair of spins. Alternatively the doublets can be classified according the multiplicity
of the rightmost pair and labelled D and D0 . c, Choosing an element from each doublet as
the qubit basis, arbitrary unitary transformations are equivalent to rotations on the Bloch
sphere shown. Doublet states |D±1/2 0 i and |D±1/2 i are denoted by the north and south
poles, and states |D0 ±1/2 i and |D±1/2 i by points on an axis rotated by 120◦ . Exchange
between left and middle dots drives rotations about the D − D0 axis, whereas exchange
between middle and right dots drives rotations about the D − D0 axis. In combination, any
rotation can be accomplished.

4.2 Exchange in a three-spin system

To see how exchange interaction allows arbitrary qubit operations, consider three electron

spins coupled by nearest-neighbour exchange strengths J12 and J23 (Fig. 4.1(a)) [94]. The

eight spin states can be classified by both overall multiplicity and multiplicity of the leftmost

spin pair, and comprise a quadruplet of states |QSz i and two doublets of states |DS0 z i and

|DSz i, where Sz denotes the z-component of total spin and takes values Sz = ±1/2 or

±3/2 for the quadruplet and Sz = ±1/2 for the doublets (Fig. 4.1(b)) [96]. Whereas for

|DS0 z i states, the leftmost pair of spins is a singlet, for |DSz i states, the leftmost pair is a

mixture of triplet states. An alternative doublet classification is based on the multiplicity

of the rightmost pair: States |D0 Sz i correspond to singlets on the right whereas states |DSz i

correspond to triplet states.

47
 a) (2,0,1) (1,1,1) (1,0,2)
Q-3/2
EZ Q-1/2
D-1/2 D-1/2 Q+1/2
D+1/2 D+1/2 Q+3/2

Energy
J12 J23
D'-1/2 D'-1/2

D'+1/2
D'+1/2
ε− 0 ε+
ε
b)
(2,1,1)

(2,0,1)

ε−
(1,1,1)
(1,1,2)
VL

ε 0

ε+
(1,0,1)
(1,0,2)

VR

Figure 4.2: Electron states of a triple quantum dot. a, Energy levels as a function of
detuning , showing Zeeman and exchange splitting. Near zero detuning the device is
configured in (1,1,1) with negligible exchange; increasing (decreasing)  lowers the energy of
the D0 (D0 ) doublet by exchange J23 (J12 ). For  > + ( < − ), states in doublet D0 (D0 )
correspond to a predominant (1,0,2) ((2,0,1)) configuration. b, Ground-state configuration
of a triple dot as a function of gate voltages VL and VR coupled to left and right dots. The
detuning axis is shown.

The logical basis is formed from two states with equal Sz , one taken from each doublet

|DS0 z i and |DSz i. States of the qubit correspond to points on the surface of the Bloch sphere

shown in Fig. 4.1(c). Whereas exchange J12 between the leftmost spin pair drives qubit

rotations about the vertical axis, exchange J23 between the rightmost pair drives rotations

about an axis tilted by 120◦ and connecting elements from doublets |D0 Sz i and |DSz i.

Arbitrary single-qubit operations can be achieved by concatenating up to four exchange

pulses [94]. Implementation of this scheme is simplified by the fact that valid qubits can

be formed using either Sz = +1/2 or Sz = −1/2 doublet components; it is therefore only

necessary to prepare and read out two of the three spins to operate the qubit.

48
 The energy levels of the triple quantum dot are tuned with an external magnetic field

B and by using gate voltages to adjust the energy differences between different charge

configurations (NL , NM , NR ), where NL , NM and NR denote electron occupancies of left,

middle and right dots respectively. Defining the detuning  as the energy difference between

(2,0,1) and (1,0,2) configurations, three regimes are accessible (Fig. 4.2(a)). Near  = 0,

the device is in the (1,1,1) configuration with negligible exchange. Neglecting hyperfine

coupling, spin states are split only by the Zeeman energy EZ = gµB B, where g is the electron

g-factor and µB is the Bohr magneton. As  is decreased, hybridization between (1,1,1) and

(2,0,1) configurations lowers the energy of |DS0 z i states, until for  < − , the ground state

configuration becomes predominantly (2,0,1). Because Pauli exclusion prevents occupation

of the (2,0,1) configuration with |QSz i and |DSz i spin states, an exchange splitting J12

results for  < 0. Similarly, with increasing  the energy of |DSz i states is lowered by

an amount J23 , and above  = + the ground state configuration becomes predominantly

(1,0,2). The various configurations are accessed by tuning gate voltages VL and VR coupled

predominantly to left and right dots respectively. The lowest-energy configurations of three

capacitively coupled dots are modeled in Fig. 4.2(b), which also illustrates the detuning

axis in gate space.

4.3 Device and methods

The measured device (Fig. 4.3(a)) was fabricated by patterning Ti/Au topgates on a

GaAs/AlGaAs heterostructure incorporating a two-dimensional electron gas 100 nm be-

neath the surface. Depletion gate voltages create a triple quantum dot together with a

pair of charge sensing quantum point contacts (QPCs) [27]. Four of the gates, including

those marked L and R, were connected to coaxial lines allowing rapid voltage pulses to be

49
Figure 4.3: Fast measurements with two charge sensors a, Device and measurement circuit.
Patterned topgates define three quantum dots and QPC charge sensors on left and right;
voltages applied to gates L and R control the energy levels of the device, while voltages VLQP C
and VRQP C tune QPC conductances gL and gR . The QPCs are connected on one side to
100 pF grounding capacitors and on the other to resonant tank circuits comprising chip
inductors LL and LR combined with parasitic capacitances CLP and CR P ; bias tees allow the

QPCs to be measured both at DC and via RF reflectometry. An RF carrier, generated

by combining signals at resonant frequencies fL and fR , is applied to the device via a
directional coupler; the reflected signal, after amplification, is demodulated by mixing with
the original carrier frequencies to yield voltages RFL and RFR sensitive predominantly
to left and right QPCs respectively. b, Reflectance S21 of the combined tank circuits as
the QPCs are pinched off, showing separate resonances corresponding to left and right. c
and d, QPC pinchoff measured simultaneously in reflectometry and DC conductance. e,
Reflectometry signal for the right sensor measured as a function of VL and VR , showing
steps corresponding to charge transitions. Electron configurations for each gate setting are
indicated.

50
applied. The device was measured at 150 mK electron temperature in a dilution refrigerator

equipped with an in-plane magnetic field.

A frequency-multiplexed radio-frequency (RF) reflectometry circuit [102, 103] was devel-

oped to allow both QPCs to be measured independently with MHz bandwidth (Fig. 4.3(a)).

Parallel resonant tank circuits incorporating left and right QPCs were formed from proximal

chip inductors LL = 910 nH and LR =750 nH together with the parasitic capacitances CLP

P of the bond wires. Bias tees coupled to each tank circuit allowed the DC conduc-
and CR

tances gL , gR of left and right QPCs to be measured simultaneously with the reflectance

of the RF circuit. As each QPC was pinched off, a separate dip developed in the reflected
q
P
signal at corresponding resonant frequency fL,R ∼ 1/2π LL,R CL,R (Fig. 4.3(b)). To mon-

itor the charge sensors, a carrier wave with components at fL and fR was injected into the

refrigerator (Fig. 4.3(a)). After amplification both cryogenically and at room temperature,

the reflected signal was demodulated by mixing with local oscillators to yield intermediate-

frequency voltages RFL and RFR sensitive predominantly to gL and gR (Fig. 4.3(c) and

(d)). To suppress backaction and reduce pulse coupling into the readout circuit, the RF

carrier was blanked on both signal and return paths when not needed for readout.

Tuning gR to the point of maximum charge sensitivity gR ∼ 0.4e2 /h, the configuration

of the triple dot could be montored via RFR [103]. Sweeping voltages VL and VR on gates L

and R, the charge stability diagram of the triple dot is mapped out as shown in Fig. 4.3(e)).

Dark transition lines are seen to run with three different slopes, corresponding to changes

of electron number in each of the three dots [97, 98]. For the most negative gate voltages,

no more transitions are seen; this indicates that the device has been completely emptied,

allowing absolute electron occupancies to be assigned to each region of the diagram.

51
 a)
6
0.6

ε (mV)

PD'
0.3

4
0

0 20 τE (ns) 40 60
b)
0.5 ε= 5.6 mV
PD' J23= 250 neV
0.3
ε= 5.1 mV
0.5 J23= 200 neV
PD'

0.3
ε= 4.5 mV
0.4 J23 = 140 neV
PD'

0.2
0 20 τE (ns) 40 60

Figure 4.4: Coherent spin exchange a, Probability PD0 to return to the initial |D0 Sz i state
following an exchange pulse sequence, measured as a function of  during the exchange
pulse and pulse duration τE . Dark and bright regions respectively indicate odd and even
numbers of complete spin exchanges. b, Points: Measured PD0 as a function of τE for values
of  indicated by horizontal lines in a. Lines: Fits to exponentially damped phase-shifted
cosines. The fitted exchange J23 () for each curve is shown.

4.4 Coherent exchange in a triple dot

We demonstrate state preparation, coherent exchange, and readout using the following cy-

cle of voltage pulses [41] applied to gates L and R to rapidly tune . Beginning at  > +

configures the device in (1,0,2) where tunneling to the leads initializes the qubit within the

doublet |D0 Sz i. The detuning is then decreased to  ∼ 0 over 1 µs, configuring the device

in (1,1,1). Because this ramp time is adiabatic compared to the characteristic hyperfine in-

teraction strength, it causes the spin system to enter a ground state defined by the random

instantaneous configuration of the lattice nuclei, such as |↑↓↑i [41, 104]. Pulsing the detun-

ing close to + , where J23 is large, for a time τE leads to coherent exchange of spins between

the right-hand dots. Finally, the detuning is ramped back to its original value  > + . The

charge configuration is now determined by the outcome of the exchange pulse: Whereas the

52
hyperfine ground state reenters the |D0 Sz i doublet in the (1,0,2) configuration, a swapped

state such as | ↑↑↓i evolves into a superposition of |DSz i and |Q±1/2 i states, causing the

device to remain in (1,1,1). At the end of this final ramp, the carrier is unblanked for

readout of the charge sensor.

Averaged over many pulse cycles, the resulting voltage RFR is converted to a spin state

probability by calibrating it against the known values corresponding to (1,1,1) and (1,0,2)

configurations. The probability PD0 to return to the initial spin state is shown in Fig. 4.4(a)

as a function of τE and  during the exchange pulse. As a function of τE , PD0 oscillates

corresponding to coherent rotation between spin states, and as expected from Fig. 4.2(a),

the oscillation frequency, set by J23 (), increases with . The measured PD0 (τE ) is fitted

for three values of  with an exponentially damped cosine, corresponding to dephasing by

electric fields with a white noise spectrum [41, 104]; the resulting values of J23 are shown in

Fig. 4.4(b). The contrast of the oscillations, here taken as a fit parameter, is below unity;

this can be qualitatively explained by pulse imperfections [42], which also cause the small

phase shift observed.

4.5 Conclusion and acknowledgements

The exchange interaction is attractive for qubit control because it acts on ns timescales, can

be tuned with gate voltage pulses, and does not rely on any particular property of the host

material [41, 40]. The techniques demonstrated here to control and read out a three-spin

system open the way to performing arbitrary qubit operations using this single physical

interaction.

We acknowledge C. Barthel and D. J. Reilly for discussions.

53
Chapter 5

Towards a two-qubit gate in the

singlet-triplet basis

This chapter presents an attempt to demonstrate a two-qubit gate in the singlet-triplet

basis, mediated by capacitive coupling between a pair of double dots. By measuring the

shift in the charge stability diagram of each double dot when the configuration of the other

double dot changes, the coupling is estimated at ∼ 60 µeV, large enough in principle for ns-

timescale operations.

54
5.1 Introduction

Although many experiments on electron spins in semiconductor quantum dots have shown

their potential as a quantum computer basis, no two-qubit operation has yet been demon-

strated [9, 18, 41, 17]. In the simplest approach, with each qubit represented by a single

spin [9], single-qubit operations are achieved by pulsed electron spin resonance and two-

qubit operations using exchange between electrons in adjacent dots. Although both these

elements have been demonstrated independently [41, 17], the separate technical demands

they place on the qubit device have so far prevented their use in combination.

An alternative basis encodes each qubit using two electron spins in a double dot; the

logical subspace is defined by the singlet S and ms = 0 triplet T states [39, 40]. The qubit

can be initialized by preparing two spins in a single dot, where Pauli exclusion favors the

singlet state; arbitrary single-qubit operations can be accomplished through the combination

of controlled inter-dot exchange and magnetic field gradients [39, 40, 41]. For readout, the

device is configured so that the two electrons occupy the same dot if they form a singlet

but not if they form a triplet, so that the two states can be distinguished using a nearby

charge sensor [27, 41, 19].

Proposed physical mechanisms for two-qubit operations in this basis include exchange

and Coulomb interaction [39, 40], of which the latter is appealing because it keeps both

qubits within the logical subspace [40]. To see how a logic operation can be carried out,

consider the capacitively coupled pair of qubits shown in Figure 5.1(a). A tilt in the electro-

static potential of each double dot leads to a spin-dependent probability of double occupancy

because singlet-state spatial wavefunctions can overlap in the same dot. The resulting spin-

55
Figure 5.1: (a) Protocol for two-qubit coupling. Because the wavefunctions of two electrons
can overlap only if they are in a singlet spin state, tilting the potential in a pair of double dots
induces a spin-selective dipole dipole coupling. This coupling mediates a controlled-phase
gate between qubits [40]. (b) and (c) Two devices designed to demonstrate this interaction.
Charge sensor conductances gA and gB are sensitive to the occupancies of upper and lower
double dots respectively.

selective dipole-dipole interaction leads to a term in the effective Hamiltonian [40]:

HAB = Ecc sin4 θ|SiA |SiB hS|A hS|B , (5.1)

where sin2 θ is the probability of double occupancy for the singlet, and

Ecc ≡ U (0, 2, 2, 0) − U (1, 1, 2, 0) − (U (0, 2, 1, 1) − U (1, 1, 1, 1)) (5.2)

is the differential cross-capacitance energy between the two double dots [40]. Here U (NLA , NRA , NLB , NRB )

is the energy of the system with charge configuration (NLA , NRA ) in double dot A and

(NLB , NRB ) in double dot B. If both qubits are in state S, their overall wavefunction ac-

quires a phase π after a time ~π/Ecc sin4 θ, thus executing a controlled-phase gate, which

56
 a) (2 b)
,0)
(2
,1)
1 (2
-1.2 ,0)
(1 1
,0) (2
(2 ,1) (2
,2) ,2)
-1.80 (1
,0)
(1

dgB/dV13 (arb)
V13 (mV)
V5 (mV)

dIF1/dV5 (arb)
,1) (1
-1.3 ,1) (1
,2)
0 (1 0
(1 ,3)
(0 ,2)
,0) (0
,0)
(0
,1) (0, (
(0 -1.4 2) 0, (0
,1) 3) ,4)
-1.90
(0
,2) -1
-1

-1.5
-1.80 -1.70 -1.60 -0.70 -0.60 -0.50 -0.40
V3 (mV) V11 (mV)

Figure 5.2: Charge stability diagrams of upper (a) and lower (b) double dots, measured
using the charge sensors as a function of their respective plunger gate voltages. Both double
dots can be tuned into the few-electron regime. The two broad near-vertical lines in (a) are
resonances of the charge sensor.

is universal for quantum computation [13].

Previous work has measured the capacitive coupling between a pair of double dots [105,

106]. Recently, a pair of few-electron double quantum dots was realized, although it was not

possible to measure the coupling in a charge configuration useful for qubit operation [107].

In this chapter I configure both double dots to the desired charge transitions and measure

their coupling, finding that it is large enough in principle for rapid two-qubit gates.

5.2 Device and methods

To demonstrate two-qubit operation, we fabricated the devices shown in Figure 5.1. De-

pletion gates 1-16 defined a pair of double quantum dots and four charge sensing point

contacts, two of which were measured by lock-in conductance measurements and frequency-

multiplexed reflectometry. The conductance gA of the upper sensor, also monitored via

output voltage RF1 of the reflectometry circuit, was sensitive predominantly to the upper

double dot; the conductance gB of the lower sensor, monitored via reflectometry voltage

57
 a) (2,1) b) -1.60 (1,2)
(1,1)
(1,1)
4
-1.32 2 -1.61
2

δgA (10 e /h)

δgB (10 e /h)

V13 (V)
V5 (V)
0 0
-1.33 -1.62

-3 2

-3 2
Di
ag
-2

Di
-2 B
(1,0) (0,1)

ag A
-4
-1.34 -1.63
(2,0) (0,2)

-2.30 -2.29 -2.28 -1.58 -1.57 -1.56 -1.55

V3 (V) V11 (V)
c) d)

4
-2.28 2 -2.28 2

δg A(10 e /h)

δgB (10 e /h)

DiagA (mV)

DiagA (mV)
0 0
1.05 mV

-3 2

-3 2
-2
-2
-4
4.0 mV
-2.29 -2.29
-1.57 -1.56 -1.57 -1.56
DiagB (mV) DiagB (mV)

Figure 5.3: Measuring the electrostatic coupling between double dots. (a) and (b) Charge
stability diagrams of upper and lower quantum dots around the charge transitions of in-
terest. White lines define the detuning axes DiagA and DiagB . (c) and (d) The charge
sensor conductances as a function of DiagA and DiagB . Shifts in the location of the charge
transition allow measurement of the inter-qubit coupling.

RF2 , was sensitive predominantly to the lower double dot.

Data presented in this chapter is measured entirely using Device I. The occupation

states of upper and lower double dots were mapped out by monitoring the corresponding

charge sensors as the plunger gates V3 , V5 , V11 and V13 were swept. As expected from the

geometry of Figure 5.1, V3 and V5 primarily couple to the upper double dot whereas V11

and V13 couple to the lower double dot. The respective charge stability diagrams, presented

in Figure 5.2, show that the device could be configured with only a few electrons in each

dot.

58
5.3 Measuring the coupling

The dipole-dipole coupling was investigated with the device configured in the qubit regime,

close to the degeneracy of (1,1) and (2,0) in the upper double dot and (1,1) and (0,2) in

the lower double dot. Detailed charge stability diagrams in this regime are shown in Fig-

ure 5.3 (a) and (b). The coupling between double dots is investigated by measuring the

shift of the stability diagram in one double dot when the other is swept through a charge

transition [105, 106]. A pair of gate voltage detuning axes, DiagA and DiagB , were defined

as shown in Figure 5.3 (a) and (b). Measuring the charge sensors as a function of both de-

tunings, the location of the (1,1)-(2,0) transition in the upper double dot is seen to change

around the value of DiagB corresponding to the (1,1)-(0,2) transition in the lower double

dot (Figure 5.3(c)). A corresponding shift in the location of the lower double dot transition

is seen (less clearly) in (Figure 5.3(d)).

The magnitude of the shift in gate voltage is measured as 1.1 ± 0.2 × 10−3 V in the

upper double dot and 4.0 ± 2 × 10−3 V in the lower dot. The constant of proportionality

between gate voltage to energy is deduced separately for each double dot by measuring the

shift of the charge transitions (not shown) when a known voltage bias is applied across the

device [29]. When converted to an energy, the measured gate voltage shift corresponds to

an energy Ecc = 60 ± 15 µeV or Ecc = 90 ± 50 µeV, measured from Figure 5.3(c) and

Figure 5.3(d) respectively.

5.4 Summary and outlook

I have configured a pair of capacitively coupled double quantum dots in the few-electron

regime, and measured the coupling at an occupancy of interest for qubit operation. The

59
measured coupling is in principle sufficient for ns-timescale two-qubit operations [40].

Attempts to actually realize a pair of qubits in Device I proved unsuccessful, and several

variations to the design produced no improvement. The most common failure mode was for

each double dot to break up into three or more dots as the electron number was reduced.

(Incipient breakup can be seen in Figure 5.2(c),where adjacent charge transition lines have

different slopes.) Another common problem was very low tunnel rates between quantum

dots and to the leads, whereas fairly rapid tunneling is needed for qubit initialization and

inter-dot exchange. What prevented qubit operation in Device I was a lack of spin blockade,

making readout impossible.

Apart from the larger number of elements that must work together in a four-dot rather

than a two-dot device, I hypothesize that the common difficulty is the longitudinal ar-

rangement of quantum dots. In contrast to devices shown in previous chapters, the tunnel

barriers in Figure 5.1 are located on opposite sides of each quantum dot. In order to be

tunnel-coupled to both sides, each dot must take on an elongated shape, prone to being

broken up by disorder. The elongated shape also leads to a low intra-dot exchange [108]

and therefore inhibits spin blockade.

In an attempt to circumvent this problem, much smaller devices were fabricated, in-

cluding Device II (Figure 5.1(c)). The upper part of this device was fairly easy configured

as a few-electron double dot; however, the lower part of the device could not be properly

configured as a double dot at all, presumably because of disorder. The difficulty of realizing

two qubits in the same device illustrates the severe design and materials challenges involved

in scaling a spin-based quantum computer.

60
Appendix A

Fabrication recipe

A.1 Introduction

Most of the fabrication techniques for the devices described in this thesis are well-established

in the Marcus group [21, 109, 110, 111], although incremental improvements have been made

during my time in the group. In this Appendix I summarize my recipe and key parameters.

Over several runs my lithographic yield has averaged over 70% even for complex multi-dot

devices.

A.2 Generic photolithography recipe

• Sonicate for 5 min in trichloroethylene, acetone and isopropanol.

• Bake for 5 min at 200 ◦ C.

• Spin Shipley S1813 resist for 60 s at 4000 rpm.

• Bake for 2 min at 115 ◦ C.

• Expose corner bead removal pattern for 40 s in AB-M mask aligner

• Develop for 15 s in CD-26 developer, rinse in water.

61
 • Expose main pattern for 4 s in AB-M mask aligner, making sure to press chip firmly

against mask.

• Develop for 60 s in CD-26 developer, rinse in water.

A.3 Complete fabrication procedure

A.3.1 Mesa etch

• Define mesa pattern photolithographically as described above, omitting corner bead

removal.

• Etch chip in a solution of 1:8:240 H2 O: H2 O2 : H2 SO4 to a depth 20 nm below the

2DEG (A typical etch rate is 8 nm/s.)

A.3.2 Ohmic contacts

• Define ohmic pattern photolithographically as described above.

• Oxygen plasma clean in Technics 220 series micro-stripper at 160 mT pressure and

60 W power for 60 s.

• Dip in 10 % ammonium hydroxide solution for 5 s.

• Evaporate in Sharon electron beam evaporator

50 Å Pt at 1.0 Å/s

1200 Å Au at 3.0 Å/s

800 Å Ge at 2.0 Å/s

500 Å Pt at 1.5 Å/s

500 Å Au at 3.0 Å/s.

62
 • Liftoff in acetone.

• Anneal for 100 s at 510 ◦ C.

A.3.3 Gate pads

• Define gate pad pattern photolithographically as described above.

• Oxygen plasma clean in Technics 220 series micro-stripper at 160 mT pressure and

60 W power for 60 s.

• Evaporate in Sharon thermal evaporator

250 Å Ti at 1.0 Å/s

1200 Å Au at 3.0 Å/s.

• Liftoff in acetone.

A.3.4 Fine gates

• Sonicate chip for 5 min in trichloroethylene, acetone and isopropanol

• Bake for 2 min at 170 ◦ C.

• Spin 950 PMMA A4 for 60 s at 4000 rpm.

• Bake for 15 min at 170 ◦ C.

• Expose fine features using 10 µm aperture at 6 nm step size. Set area dose to

900 µAs/cm2 , line dose to 2700 pAs/cm. If possible design your finest features as

single-pixel lines.

• Expose connector features using 120 µm aperture at 56 nm step size. Set area dose

to 1080 µAs/cm2 .

63
• Develop for 60 s in a solution of 1:3 methyl isobutyl ketone: isopropanol cooled to 0 ◦ C,

rinse in isopropanol.

• Evaporate in Sharon thermal evaporator

50 Å Ti at 0.5 Å/s

150 Å Au at 1.0 Å/s.

• Liftoff in acetone.

64
Appendix B

Constructing and using a

multiplexed reflectometry setup

This Appendix describes the fast readout system I created and used for the measure-

ments described in Chapters 4 and 5. As well as details of the hardware, I explain how to

use the setup efficiently. The Appendix has four sections. Sections B.1 and B.2 describe

the hardware inside and outside the cryostat respectively. Section B.3 describes the code I

adapted to take data fast. Section B.4 describes the tuning process.

B.1 Hardware inside the cryostat

The fast readout system I constructed, based on an Oxford Instruments Kelvinox 100 di-

lution refrigerator, was partly copied from the one built by David Reilly [103]. I extended

Reilly’s design to allow for multiplexed readout, more pulse lines and heat sinking, and

easier sample exchange. Photographs and schematics of the modified cryostat are shown in

Figures B.1 and B.2.

The design must fulfill two opposing criteria: good electrical access to the device, with-

out unacceptable heat load. This must be achieved for three separate types of connection

to the device: ∼ 25 dc wires, with bandwidth up to ∼ 20 kHz; four pulse lines, with

bandwidth up to ∼ 10 GHz; and a multiplexed reflectometry line, operating at frequencies

65
 Reflectometry
a) b) TX RX Pulse lines DC lines

300 K
85 mil SS/SS 85 mil SS/SS 85 mil SS/BeCu
Cryogenic amplifier

4K 43 dB
Copper spools
85 mil SS/SS 85 mil Nb/Nb 85 mil SS/SS
Attenuator
30 dB 20 dB
1.5 K (pot)

85 mil SS/SS 85 mil Nb/Nb 85 mil SS/SS

Sapphire stripline
10 dB 10 dB
700 mK (still)

(above mixing chamber)

Constantan loom
85 mil SS/SS 85 mil Nb/Nb 85 mil SS/SS
Directional coupler

~200 mK (cold plate)

85 mil Nb/Nb 85 mil Nb/Nb
Bulkhead feedthrough
3 dB
70 mK (mixing chamber)

(below mixing chamber)

47 mil Cu/Cu
85 mil Cu/Cu

Copper loom
47 mil Cu/Cu AC 10 kΩ

SUM

DC
100 pF 100 pF
pulsed gate voltages
100 nH 5 kΩ 5 kΩ 100 nH
LL Left Sample Right L
dc QPC sensor sensor R dc QPC
DC gate voltages
bias bias Cold resistors
(left) 100 pF 100 pF (right) (2 × 442 Ω
each line)

Figure B.1: Dilution refrigerator equipped with radio-frequency reflectometry setup. (a)
Photograph of the cryostat. (b) Schematic of the electronics inside the cryostat. The grey
background indicates components mounted on the PC board. Key components are: Cryo-
genic amplifier: Quinstar QCA-U230, custom-built; Sapphire stripline: Home-built [112];
Directional coupler: Mini-circuits ZEDC-15-2B; Bias tees: Anritsu K251.

∼ 190 − 250 MHz.

For the DC wiring, long (∼ 3 m) sections of constantan loom are used with a total

resistance ∼ 200 Ω from room temperature to the mixing chamber. At each stage of the

fridge, the loom is wound around one or more copper spools, and between stages a generous

amount of slack is left to minimize the heat load down the fridge. Below the mixing chamber,

copper loom is used to maximize cooling of the sample. As a final stage of filtering, two

66
banks of resistors are fitted tightly into brass plates at the cold finger and fixed with Stycast.

For the high-frequency wiring, it is necessary to use semi-rigid coaxial cable. The po-

tential for large heat loads is here much greater, because the cables can transmit black-body

microwave radiation down the fridge, and because the inner conductor is quite poorly ther-

malized through the dielectric. Increasing the cable length does reduce the heat load, but

also creates frequency-dependent attenuation that is undesirable for pulse-gate experiments.

Instead, we rely mainly on broadband attenuators mounted at each refrigerator stage. To

allow gate voltage offsets to be applied without ohmic heating, DC connections are made

through bias tees mounted at the final cooling stage.

The choice of diameter and material for the coaxial cables was determined by the length

of each coax segment and the temperature and cooling power at each refrigerator stage. Over

most of the length of the fridge, .085” diameter stainless steel (SS) coax was used because

it combines comparatively weak attenuation with low thermal conductivity. From the top

of the fridge to the pot, coax with BeCu inner conductor was used to reduce attenuation

over the longest length of cable, and because the pot has enough cooling power to handle

the extra heat load. From the cold plate to the mixing chamber superconducting Nb coax

is used to take advantage of its lower thermal conductivity (below ∼ 90 mK) and weaker

attenuation [113]. Below the last set of attenuators, thermal conductivity is no longer an

issue, and copper is used for its low attenuation, with 47 mil diameter cable chosen to

save space in the cold finger. Over the whole length of the fridge, the attenuation of the

pulse lines is found to increase smoothly with frequency up to at least 6 GHz, with roll-off
√
∼ 5dB/ GHz at room temperature.

Wiring the reflectometry lines is in some ways easier because the highest operating

frequency is much lower (although definitely high enough to require semi-rigid coax). How-

67
ever, attenuators cannot be used on the measurement path because the signal would also

be reduced. The solution that Reilly developed was to heat sink the signal path at the still

with a sapphire stripline and at the cold plate through the directional coupler (which acts

as a short to ground at DC). At other refrigerator stages, the outer conductor was thermal-

ized using bulkheads. Coaxial segments in both reflectometry and pulse lines were heavily

looped1 to increase the cable length and thermal load between stages by a factor ∼ 2.

My cold finger is shown in Figure B.2, and is fairly similar to previous Marcuslab

designs [20]. The design was made easier by Jason Petta’s earlier expansion of the cryostat’s

inner vacuum chamber. The main novel feature is the 1/4” × 1/2” copper spine brazed

to the top plate of the cold finger and running down to the sample holder. This provides

good thermal contact down the center and allows various heat sinking elements (bias tees,

bulkheads and resistor plates) to be mounted accessibly on the outside of the cold finger.

The PC board sample holder (Figure B.2(c)) follows Reilly’s design. Connection to the

fridge wiring is through a Cristek connector for the DC wiring and through SMM connectors

for the coaxial cables. With the exception of the reflectometry connection, which is via a

microwave launcher, all cables are soldered directly to the PC board. Strain relief for the

dc wires is provided by a drop of Stycast on the back of the board (although Ferdinand

reckons dental floss is preferable).

1
It has since become fashionable to make coax segments as short as possible, for the following reason. For
√
a normal-metal coax of length L, the frequency-dependent attenuation α is proportional to ρL/D, where

ρ is its resistivity and D is the diameter [114]. By the Wiedemann-Franz law, the thermal conductance due

to the conduction electrons is Θ ∝ D2 /ρL ∝ L/α2 . So for given attenuation, the heat load is minimized

by using a short length of thin, high-resistance coax. Of course, coax is available in only a few discrete

diameters, so ideal parameters are often not realizable in practice.

68
a) b) c)
Pulse lines
Top plate bolted to mixing chamber

SMA feedthroughs /
attenuators

Bias tees

Coaxial Cold resistor

feedthroughs feedthroughs

dc QPC
bias
100 nH
Radiation shield
100 nH
Spine 5 kΩ 5 kΩ
LL LR 100 pF
100 pF
Sample

PC board

Sample

Reflectometry 100 pF grounding

1΄΄ port capacitors

Figure B.2: Cold finger and sample board. (a) Cross-section of cold finger, showing key
components. The structure is entirely of brass except for the top plate and spine, which
are of copper. The feedthroughs and radiation shield are designed to provide an almost
completely sealed sample space. (b) Photograph of cold finger. To show the structure more
clearly, two out of four brass plates screening the feedthroughs have been removed and the
main radiation shield has been unscrewed. (c) PC board holding sample. Key components
are indicated, including the electronic components making up the QPC bias tee and the
tank circuit inductors LL and LR . An LED allowed illumination of the cold sample but
this was never found to have a beneficial effect.

B.2 Hardware outside the cryostat

The experimental configuration must allow for quasistatic and pulsed gate voltages to be

applied to the device, and for three kinds of measurement: Lock-in and reflectometry mea-

surements of the charge sensors, and dc transport measurements of the dots. A simplified

version of the setup is shown in Figure B.3. Quasistatic gate voltages are supplied by a 20-

channel digital-to-analog converter (DecaDAC) from the Harvard Physics electronics shop

supplemented by a battery box. For a few channels, a divider/adder allows rapid ramping

of the gate voltage with an HP33210A function generator. Gate pulses are provided by a

69
 Shielded SR560
room wall ×10 ×10

HP8654B

LPF

LPF
OUT

C1 IF1 IF2
HP8654B OUT C2

TTLR
Demodulation box
TTLT
RX TX

Reflectometry measurement
M1 M2 M3
CH1
AWG520
CH2

Pulse lines RX TX
CH1
AWG520 SMA ports

Gates
CH2 RC Battery
TRIG DAC filters box

Breakout box
Divider/adder
DC OUT

Contacts
12 kΩ
48 kΩ 3 kΩ
AC

OUT
HP33210A
SYNC

I in
Ithaco V
out

50/50
OUT ÷ 1000 Cryostat
HP33210A

Transport measurement

OUT ÷ 1000
IN1 IN2 IN3 TRIG HP33210A 1 MΩ
SYNC
HP54845A
oscilloscope
TRIG
Lockin A in
V
out

DMM V in

Sensor conductance measurement

Figure B.3: Schematic of the room temperature measurement setup. For a description, see
the text.

pair of Tektronix AWG520s ganged together (see Appendix C).

The usual purpose of the lock-in measurement is to ensure that the charge sensor is

tuned for optimum sensitivity, usually corresponding to a conductance around 0.4e2 /h.

The conductance is measured using a 1 nA current bias at of 137 Hz and digitized via a

DMM at ∼ 10 Hz. The bias, supplied by an HP33210A, is blanked during reflectometry

measurements.

70
 For reflectometry, carrier signals generated by a pair of HP8654Bs are fed into the fridge.

The returned signal is mixed with the carriers inside a home-built demodulation box (see

Section B.2.1), generating two IF voltages which are filtered, amplified and finally digitized

using an oscilloscope. Low-frequency ground loops are suppressed with inside/outside DC

blocks (Midwest Microwave DCB-3537-IO-SMA-02) on Rx and Tx lines.

Transport measurement of the dots is performed in voltage bias using a DC excitation2

of ∼ 100µV. The current is detected using an Ithaco 1211 current amplifier set to its

highest bandwidth, with its output again digitized using an HP54845A oscilloscope. A

ground loop between Ithaco and oscilloscope is suppressed by 50 Ω resistors on inner and

outer conductors of the connecting BNC cable.

To map reflectometry and charge sensing signals to gate voltages, the oscilloscope is

triggered from the synchronization signal of the HP33210 that ramps the gate. Typically

the gate is ramped at 47 Hz in a sawtooth pattern, and the oscilloscope averages 16 ramps

to generate each slice of data.

B.2.1 The demodulation box

The demodulation box is shown in Figure B.4. Carrier signals supplied through ports C1

and C2 are combined and transmitted to the fridge via an optional RF switch for blanking.

The signal from the fridge is amplified, fed through a second switch, and then mixed with

the carriers to generate voltages IF1 and IF2. Voltages PS1 and PS2 tune the relative phase

2
Supplying the DC bias with a function generator, as shown in Figure B.3, is presumably unnecessary.

The reason for this extravagance is historical; until last year the usual method to measure transport was

with a lock-in amplifier. I have found that measuring with the scope is just as useful and much faster, but

I never got round to replacing the HP33210A by a DAC channel.

71
 c) OUT
RF switch
a) TTL

IN

TX Combiner
Phase shifter
Directional coupler
C1 φ
C2 φ
PS1
PS2

IF1 IF LO

RF
IF2 IF LO

TO MIXERS RF

b) OUT
Mixer
TTL

IN

AMP OUT
LPF HPF

55 dB
RX
LPF HPF
To fridge From fridge

Figure B.4: Room temperature demodulation box. (a) and (b) Top and front view (cover
removed). (c) Circuit diagram. Dashed lines denote cabling outside the box that allows
input and output signals to be fed into optional blanking switches. Power wiring for switches
and amplifier (BNC connectors labelled ± 5 V and 15 V in (b)) is not shown. Components
are: RF switch: Mini-circuits ZASWA-2-50DR+; Combiner/splitter: Mini-circuits ZESC-
2-11; Directional coupler: ZEDC-15-2B; Phase shifter: Pulsar SO-06-411; Mixer: Mini-
circuits ZP-3MH; Filters: Mini-circuits ZHP-100 and ZLP-450.

between LO and RF inputs; these are set by hand to maximize IF1 and IF2. For tidiness

and shielding, the entire setup is assembled inside a Lansing aluminum enclosure.

B.3 Igor code

Data acquisition was performed using Wavemetrics’ Igor Pro, which as usual was more than

up to the task. The main acquisition code I based on Alex Johnson’s library of routines [20]

and modified in a fairly obvious way to acquire data from the oscilloscope as well as from

DMMs. A few supplementary tasks need to be performed before each scan, including

72
optimizing the charge sensor conductance and setting the gate voltage ramp appropriately.

To carry these out efficiently and automatically, I wrote the function do2dfast(). The

functions tuneg() and calsensor(), which automatically optimize the sensor conductance,

are also given below. The functions WC stream(), WC QPCname() and WC idloc() are not

given here; they are look-up functions for which sensor associates with which swept gate

and are on the MarcusLab Wiki3 .

B.3.1 do2dfast()
function do2dfast(idstr1,start1,stop1,numdivs1,delay1,idstr2, start2,stop2)
//Function to do fast 2D sweeps
string idstr1 // id of outer loop (a code for the variable scanned - see description
// of Alex Johnson’s code in his thesis)
variable start1 // outer loop starting value
variable stop1 // ending value
variable numdivs1 // number of points minus 1
variable delay1 // seconds of delay between sweeps
string idstr2 // same for inner loop
variable start2,stop2

variable hpnum=hpfromidstr(idstr2) // Address of function generator on gate

variable QPCstimulus=3 // Address of function generator for QPC bias
variable calsteps=3 // Number of QPC calibration steps

// Begin by calibrating the charge sensor

setval(idstr2,(start2+stop2)/2)
setfuncHP(hpnum,"DC") // Make sure we’re not ramping the gate ...
setfuncHP(QPCstimulus,"SIN") // ... and that the lock-in excitation is unblanked.
wait(1)
strswitch(idstr1)
case "b": // Scan parameters other than gates.
case "ts":
case "te":
tuneg(WC_stream(idstr1), WC_QPCname(idstr1)) // Stepping these variables barely changes
// the sensor conductance, so we need only optimize it once.
default:
if (WC_idloc(idstr1)>=0) // If the stepped parameter is a gate, we must optimize the
// sensor across the full range of the scan.
calsensor(idstr1,start1,stop1,calsteps-1)
// Optimize the charge sensor across the swept range.
else
print "do2dfast(): idstr1 not recognized"
endif
endswitch

setfuncHP(QPCstimulus,"DC") // Blank the lock-in excitation.

oscilforwallwall() // Set up oscilloscope.
if (start2<stop2)
setfuncHP(hpnum,"RAMPUP") // Start ramping the gate.
osciltrigslope(1)
else
setfuncHP(hpnum,"RAMPDOWN")

3
https://qhall.fas.harvard.edu

73
 osciltrigslope(0)
endif
setval("vac"+num2str(hpnum),abs(start2-stop2)*1000) // Set gate ramp amplitude.
wait(2)

//The next seven lines handle bookkeeping for do2d() so the wave ends up displaying correctly.
variable channum
sscanf idstr2, "c%g", channum
if ((channum>0)||stringmatch(idstr2,"c0"))
idstr2="f"+idstr2
else
idstr2="nscope"
endif

do2d(idstr1,start1,stop1,numdivs1,delay1,idstr2, start2,stop2,0,0)
setfuncHP(hpnum,"DC") // Stop ramping the gate.
setfuncHP(QPCstimulus,"SIN") // Unblank lock-in excitation.
end

B.3.2 tuneg()
function tuneg(stream,idstr)
// Optimizes QPC conductance for charge sensing by tuning a nearby gate.
// Two simple search algorithms for the optimal point are applied. The first algorithm measures the
// conductance and adjusts the gate by an amount proportional to the mismatch from the target value. It
// repeats until the conductance gets close to the ideal value or a counter is exceeded.
// After the first algorithm completes, the second algorithm fine-tunes the conductance by scanning once in
// and then out with the gate until the target value is passed.
string idstr // QPC gate
variable stream // Acquisition stream for QPC conductance (see Alex’s thesis for details)

variable /G targetcon // Optimal conductance value to aim for.

variable target=targetcon
variable toplimit=400 // Limits within gate voltage can safely be swept (in mV).
variable bottomlimit=-1950
variable waittime=0.1 // Settling time after changing gate (in s).

wave DAC=DAC
variable QPCnum
sscanf idstr, "c%g", QPCnum
if (numtype(getdata(stream))||numtype(QPCnum)) // Input checking.
printf "tuneg(): Invalid number %g %g", stream, QPCnum
AbortOnValue 1, 1
return -1
endif
variable returnval

setval(idstr, max(bottomlimit, min(toplimit, DAC[QPCnum])))

// Make sure we start off in the allowed range .
wait(4*waittime)

// Run the first search algorithm

variable maxstep=20, minstep=0.1 // mV
variable gradient=0.2 // How much to change gate voltage for a given conductance
// mismatch. The best value depends on how steep the QPC pinchoff is. Too large, and the search
// is slow; too small, and iteration may not converge.
variable counter=200 //Maximum number of iterations.
do
wait(2*waittime)
variable stepsize=-(getdata(stream)-target)/gradient
if (stepsize < 0)
stepsize=-max(min(-stepsize,maxstep),minstep)
else
stepsize=max(min(stepsize,maxstep),minstep)
endif

74
 if (((DAC[QPCnum]+stepsize)>bottomlimit)&&((DAC[QPCnum]+stepsize)<toplimit))
returnval = setval(idstr, DAC[QPCnum]+stepsize)
if (((returnval == -1)&&(stepsize<0))||((returnval==-2)&&(stepsize>0)))
printf "tuneg() part 1: Exceeded limit, %g %g", stepsize, returnval
counter=0
endif
endif
doupdate
counter-=1
while ((abs(stepsize)>minstep)&&(counter>0))

wait(2*waittime)

// Run the second search algorithm

variable downstep=2, upstep=0.3 //mV
do //Go down until conductance is below target.
returnval = setval(idstr, DAC[QPCnum]-downstep)
if (returnval==-1)
printf "tuneg() crawl down: Exceeded limit, %g %g", downstep, returnval
return NaN
endif
doupdate
wait(2*waittime)
while((DAC[QPCnum]-downstep>bottomlimit)&&(getdata(stream)>target))
if (DAC[QPCnum]-downstep<=bottomlimit)
print "tuneg():hit bottomlimit"
AbortOnValue 1, -2
endif
do //Now go up until conductance is above target.
returnval = setval(idstr, DAC[QPCnum]+upstep)
if (returnval==-2)
printf "tuneg() crawl up: Exceeded limit, %g %g", upstep, returnval
return NaN
endif
doupdate
wait(waittime)
while((DAC[QPCnum]+upstep<toplimit)&&(getdata(stream)<target))
if (DAC[QPCnum]+upstep>=toplimit)
print "tuneg():hit toplimit"
AbortOnValue 1, -3
endif
return DAC[QPCnum]
end

B.3.3 calsensor()
function calsensor(idstr,start,stop,numdivs)
// Over a range of values of gate idstr, find the optimal QPC gate settings, and fit those settings with a
// parabola to allow optimized sensing across the entire range.
string idstr // Gate voltage to be swept
variable start, stop, numdivs // Sweep range, and number of points to tune QPC at.

newdatafolder /O root:util:calsensor
wave wKK0 = root:util:calsensor:KK0 // These waves store estimates (derived from previous runs
// of this function) for the parabola fit coefficients.
wave wKK1 = root:util:calsensor:KK1
wave wKK2 = root:util:calsensor:KK2

variable KK0=wKK0[WC_idloc(idstr)]
variable KK1=wKK1[WC_idloc(idstr)]
variable KK2=wKK2[WC_idloc(idstr)]

// If no valid estimates for the fit parameters are found, then guess.
if (numtype(KK0))
KK0=-1000
endif

75
 if (numtype(KK1))
KK1=0
endif
if (numtype(KK2))
KK2=0
endif

strswitch (idstr) // Look up the QPC gate and data stream associated with
// gate idstr.
default:
if (WC_idloc(idstr)>=0)
string QPCstr=WC_QPCname(idstr)
variable stream=WC_stream(idstr)
else
print "calsensor(): wall not recognized."
return 0
endif
endswitch

//Make waves to store the gate settings in.

make /O /N=(numdivs+1) calx=start+(stop-start)/numdivs * p
duplicate /o calx caly
caly=NaN

// Now search for the optimal QPC setting at each point.

variable i
for (i=0; i<=numdivs; i+=1)
setval(idstr,calx[i])
try
variable tunegreturn=tuneg(stream,QPCstr)
catch
if (V_Abortcode==1)
abort
endif
if ((V_Abortcode == -2)||(V_Abortcode == -3))
tunegreturn=NaN
endif
endtry
caly[i]=min(tunegreturn,400)
wKK0[WC_idloc(idstr)]+=caly[i]-(KK0 + KK1 * calx[i] + KK2 * calx[i]^2) // Set the offset for the
// next point based on this point. This usually speeds up
// the next search.
endfor

CurveFit /Q /NTHR=0 poly 3, caly /X=calx /D //Fit the parabola.

wKK0[WC_idloc(idstr)]=K0 //Store the results.

wKK1[WC_idloc(idstr)]=K1
wKK2[WC_idloc(idstr)]=K2
end

B.4 Tuning hints

Here are outline instructions for tuning up a double-double device with a gate layout similar

to that shown in Figure 5.1(a).

Good tuning starts with appropriate positive bias during cooldown. If you’ve success-

fully tuned up your device (or a similar one) before, pick the same positive bias. If it’s

76
a new design, I suggest cooling with 200-300 mV on all gates. If previous tuneups with

the same device were unsuccessful, try to modify the potential landscape by adjusting the

cooldown bias on individual gates; as a general rule, more positive bias leads to a bigger

gate footprint once the dot is formed. Although ∼ 300 mV of cooldown bias seems to be

optimal for device stability, you can go at least 300 mV in either direction without much

trouble.

Once you’re cold, establish what voltage on gates 8 and 16 is needed to pinch off the

center of the device. Measuring the conductance between top and bottom sets of ohmic

contacts, and with all other gates grounded, do a two-dimensional scan of V8 and V16 . With

luck, you will get smooth pinchoff down to zero. Where this happens, you can guarantee

that the two pairs are separated. Usually I repeat the scan with the other gates all depleted

(say at -500 mV), to get an idea of how much they help to pinch off the center.

Next, try to form dots in the top and bottom pairs. Set up transport measurements

through each pair, and deplete gates 3-5 and 11-13. Over a range of V8 and V16 values

sufficient to pinch off the center, do 2-D scans (‘wall-walls’) of V2 and V6 , and of V10 and

V14 . Try to find a value of V8 and V16 where Coulomb blockade lines appear in both upper

and lower pairs.

Having found Coulomb blockade, try to establish clear double-dot behavior, manifested

by Coulomb blockade lines forming a honeycomb in the wall-walls. In a good device, the

most likely scenario is for each pair of dots to be merged initially, so you will need to decrease

V4 and V12 to separate them. Quite often, it will happen that one dot of a pair will be too

open to show Coulomb blockade or too isolated from its partner to make a honeycomb. In

these cases, you will have to adjust V3 − V5 and V11 − V13 to try to create space for your

dots in favorable locations. You will find that tuning one pair messes up the tuning in the

77
other; to make both work, switch between them every few hours until you end up with both

tuned at the same time.

Assuming you are successful, set V2 , V6 , V10 , and V14 to values in the honeycomb regions

of the wall-wall, and switch to sweeping plunger gates 3, 5, 11 and 13. Once you have verified

that you can also see honeycombs sweeping these gates, set up reflectometry measurements.

Use a network analyzer to find the resonances of your tank circuits, and set the carrier

frequencies to match them. You are now ready to hunt down the last electrons in each dot.

Do 2-D scans sweeping V3 and V5 , then V11 and V13 , going to more negative values until no

more charge sensing transitions appear. This can be a sign that the device is empty, but

more likely it means the dots are latching - becoming to pinched off for electrons to leave.

A strong indication of latching is charge sensing lines that distort or fade away in the lower

left of the gate voltage plane. the best way to prevent it is usually to pull back with the

wall voltages.

Eventually you should be able to measure a clear stability diagram similar to Figure 1.3

in each pair, showing a series of well-defined charge transitions down to the last electron.

Now you want to open the dots up to the leads until a measurable current (∼ 1 pA at

∼ 200 µV) is found at the (1,1)-(0,2) (or (0,2)-(1,1)) transitions. Usually this means pulling

back with V2 , V6 , V10 and V14 (and possibly V8 and V16 as well). Once you are close, charge

sensing can help you work out which of the three tunnel barriers in each pair of dots limits

the current: In a plot such as Figure 1.4(e), the charge sensing triangles will be dominated

by the charge state which is slowest to decay.

Now you need to optimize spin blockade. For bias voltages Vsd ∼ ±400 µV, attempt to

reproduce the data shown in Figure 1.4. If you can’t find spin blockade at the (1,1)-(0,2)

transition, try the (0,2)-(1,1) transition. For spin readout, you will need an exchange energy

78
of at least J02 ∼ 300 µV. If J02 is too small, you may be able to tune it to a higher value

by adjusting gate voltages.

I never got beyond this point with a quadruple dot, but have tuned up several double

dots, so I can make an educated guess what the next steps would be. To demonstrate spin

readout, apply the pulse schemes of [66] or [41] and look for the pulse triangle in charge

sensing, taking care not to be confused by latching. Often the clearest signal is due to the

S-T+ degeneracy [41], which gives a sharp line in gate voltage space with unmistakable field

dependence. The visibility of the pulse triangle can often be improved by tweaking gate

voltages.

The last step of tuning is to measure coherent exchange oscillations. For these to show

up, the intra-dot tunnel coupling needs to be fairly large. A good way to measure it is by

measuring the field difference of the S-T+ degeneracy; aim to have it broaden out fairly

smoothly, as in Figure 3.2(c). Experiment with various pulse parameters within the scheme

of Figure 4A in [41] until the oscillations become clear. Congratulations! You have achieved

what eluded me for nearly two years.

79
Appendix C

Synchronizing two Tektronix

AWG520s

When running an experiment involving more than two pulsed gates, synchronizing the

pulses becomes an issue. Recently Tektronix introduced the AWG5014, whose automatically

synchronized outputs solve the problem for up to four gates (at a price). I have also devised

a technique to synchronize at least four (and probably arbitrary many) AWG520 outputs.

I sketch it briefly below.

For convenient operation you will need the Igor procedure files AWGhighlevel.ipf,

AWGPTmanipW.ipf, AWGsequencesW.ipf, AWG520W.ipf, AWG520DictW.ipf, AWG710W.ipf

and AWG710DictW.ipf, available on the Marcuslab Wiki. I will explain briefly what the

key steps are.

Begin by synchronizing the clock frequencies of your AWGs by running an exter-

nal 10 MHz reference (for example from an HP3325A) into the 10MHZ REF IN ports

of your AWGs. Then choose one of your AWGs as the master, and run one of its marker

channels (arbitrarily chosen) to the TRIG IN port of the slave(s). Setting the model vari-

able to 2520, run the Igor command InitAWG() in the file AWGhighlevelW.ipf, available

on the MarcusLab Wiki. This will set the AWG clocks to use the reference signal, and set

the slave(s) to generate externally triggered output.

80
 Edit the pulse table, which specifies the output waveform, so that the marker channel

that supplies the trigger is set to 1 during the last pulse step and 0 during the rest of the

cycle. To output pulses, run setpulsesAWG(). This will upload the pulse table waveform to

the AWGs. For triggered operation, it will automatically rotate the slave waveform forward

by the trigger delay (typically 31 ns, although you may have to tweak this depending on

cable lengths.)

Your AWGs will now be synchronized within one clock cycle. The largest source of

remaining asynchronicity is that the length of the trigger cable does not correspond to an

integer number of clock cycles. To correct for this, examine the outputs of master and

slave on an oscilloscope while adjusting the marker delay of the master to bring them into

alignment. By this technique, synchronicity of ∼ 70 ps should be achievable.

81
Bibliography

[1] A. Romito and Y. Gefen, Weak values of electron spin in a double quantum dot, Phys.

Rev. Lett. 100, 4 (2008).

[2] N. Katz, M. Neeley, M. Ansmann, R. C. Bialczak, M. Hofheinz, E. Lucero,

A. O’Connell, H. Wang, A. N. Cleland, J. M. Martinis, and A. N. Korotkov, Re-

versal of the weak measurement of a quantum state in a superconducting phase qubit,

Phys. Rev. Lett. 101, 200401 (2008).

[3] I. T. Vink, K. C. Nowack, F. H. L. Koppens, J. Danon, Y. V. Nazarov, and L. M. K.

Vandersypen, Locking electron spins into magnetic resonance by electron-nuclear feed-

back, Nat Phys 5, 764 (2009).

[4] D. J. Reilly, J. M. Taylor, E. A. Laird, J. R. Petta, C. M. Marcus, M. P. Hanson,

and A. C. Gossard, Measurement of temporal correlations of the Overhauser field in

a double quantum dot, Phys. Rev. Lett. 101, 236803 (2008).

[5] J. A. Jones, S. D. Karlen, J. Fitzsimons, A. Ardavan, S. C. Benjamin, G. A. D. Briggs,

and J. J. L. Morton, Magnetic field sensing beyond the standard quantum limit using

10-spin NOON states, Science 324, 1166 (2009).

82
 [6] M. A. Nielsen and I. A. Chuang, Quantum Computation and Quantum Information

(Cambridge University Press, 2000).

[7] N. D. Mermin, Quantum Computer Science (Cambridge University Press, 2007).

[8] R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K. Vandersypen,

Spins in few-electron quantum dots, Rev Mod Phys 79, 1217 (2007).

[9] D. Loss and D. P. DiVincenzo, Quantum computation with quantum dots, Phys. Rev.

A 57, 120 (1998).

[10] P. Shor, Algorithms for quantum computation: Discrete logarithms and factoring, in

Proceedings, 35th Annual Symposium on Foundations of Computer Science (1994).

[11] L. Grover, Quantum mechanics helps in searching for a needle in a haystack, Phys.

Rev. Lett. 79, 325 (1997).

[12] D. P. DiVincenzo, 2-bit gates are universal for quantum computation, Phys. Rev. A

51, 1015 (1995).

[13] A. Barenco, A universal 2-bit gate for quantum computation, P Roy Soc Lond A Mat

449, 679 (1995).

[14] A. Kitaev, Quantum computations: algorithms and error correction, Russ Math

Surv+ 52, 1191 (1997).

[15] A. Steane, Multiple-particle interference and quantum error correction, P Roy Soc

Lond A Mat 452, 2551 (1996).

[16] D. P. DiVincenzo and D. Loss, Quantum information is physical, Superlattice Microst

23, 419 (1998).

83
[17] F. H. L. Koppens, C. Buizert, K. J. Tielrooij, I. T. Vink, K. C. Nowack, T. Meunier,

L. P. Kouwenhoven, and L. M. K. Vandersypen, Driven coherent oscillations of a

single electron spin in a quantum dot, Nature 442, 766 (2006).

[18] J. Elzerman, R. Hanson, L. van Beveren, B. Witkamp, L. Vandersypen, and L. P.

Kouwenhoven, Single-shot read-out of an individual electron spin in a quantum dot,

Nature 430, 431 (2004).

[19] C. Barthel, D. J. Reilly, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Rapid single-

shot measurement of a singlet-triplet qubit, Phys. Rev. Lett. 103, 160503 (2009).

[20] A. C. Johnson, Charge sensing and spin dynamics in GaAs quantum dots, Ph.D.

Thesis, Harvard University (2005).

[21] A. G. A. Huibers, Electron transport and dephasing in semiconductor quantum dots,

Ph.D. Thesis, Harvard University (1999).

[22] T. Drummond, W. T. Masselink, and H. Morcok, Modulation-doped GaAs/(Al,Ga)As

heterojunction field-effect transistors - modfets, P Ieee 74, 773 (1986).

[23] B. J. van van Wees, L. P. Kouwenhoven, C. J. P. M. Harmans, J. Williamson, C. Tim-

mering, M. Broekaart, C. Foxon, and J. Harris, Observation of zero-dimensional states

in a one-dimensional electron interferometer, Phys. Rev. Lett. 62, 2523 (1989).

[24] L. P. Kouwenhoven, C. M. Marcus, P. L. McEuen, S. Tarucha, R. M. Westervelt, and

N. S. Wingreen, Electron transport in quantum dots, in Mesoscopic Electron Transport

(Kluwer Academic, Dordrecht, 1997).

84
[25] C. Buizert, F. H. L. Koppens, M. Pioro-Ladrière, H.-P. Tranitz, I. T. Vink, S. Tarucha,

W. Wegscheider, and L. M. K. Vandersypen, In situ reduction of charge noise in

GaAs/Alx Ga1−x As Schottky-gated devices, Phys. Rev. Lett. 101, 1 (2008).

[26] C. W. J. Beenakker, Theory of Coulomb-blockade oscillations in the conductance of a

quantum dot, Phys Rev B 44, 1646 (1991).

[27] M. Field, C. G. Smith, M. Pepper, D. Ritchie, J. Frost, G. A. C. Jones, and D. Hasko,

Measurements of Coulomb blockade with a noninvasive voltage probe, Phys. Rev. Lett.

70, 1311 (1993).

[28] J. R. Petta, A. C. Johnson, C. M. Marcus, M. Hanson, and A. C. Gossard, Ma-

nipulation of a single charge in a double quantum dot, Phys. Rev. Lett. 93, 186802

(2004).

[29] W. G. V. D. Wiel, S. D. Franceschi, J. Elzerman, T. Fujisawa, S. Tarucha, and L. P.

Kouwenhoven, Electron transport through double quantum dots, Rev Mod Phys 75, 1

(2003).

[30] J. Elzerman, R. Hanson, J. S. Greidanus, L. H. W. V. Beveren, S. D. Franceschi,

L. M. K. Vandersypen, S. Tarucha, and L. P. Kouwenhoven, Few-electron quantum

dot circuit with integrated charge read out, Phys Rev B 67, 4 (2003).

[31] J. R. Petta, A. C. Johnson, J. M. Taylor, E. A. Laird, A. Yacoby, M. D. Lukin, C. M.

Marcus, M. P. Hanson, and A. C. Gossard, Preparing, manipulating, and measuring

quantum states on a chip, Physica E 35, 251 (2006).

[32] A. C. Johnson, J. R. Petta, and C. M. Marcus, Singlet-triplet spin blockade and charge

sensing in a few-electron double quantum dot, Phys Rev B 72, 7 (2005).

85
[33] E. Lieb and D. Mattis, Ordering energy levels of interacting spin systems, J Math

Phys 3, 749 (1962).

[34] K. Ono, Current rectification by Pauli exclusion in a weakly coupled double quantum

dot system, Science 297, 1313 (2002).

[35] A. Abragam, Principles of Nuclear Magnetism (Oxford Science Publications, 1961).

[36] J. J. Sakurai, Modern Quantum Mechanics (Addison Wesley Longman, 1994).

[37] K. Il’in, M. Siegel, A. Semenov, A. Engel, and H.-W. Hübers, Critical current of Nb

and NbN thin-film structures: The cross-section dependence, phys. stat. sol. (c) 2,

1680 (2005).

[38] S. Amasha, Toward the manipulation of a single spin in anAlGaAs/GaAs single-

electron transistor, Proc. of SPIE 6244, 624419 (2006).

[39] J. Levy, Universal quantum computation with spin-1/2 pairs and Heisenberg exchange,

Phys. Rev. Lett. 89, 147902 (2002).

[40] J. M. Taylor, H.-A. Engel, W. Dür, A. Yacoby, C. M. Marcus, P. Zoller, and M. D.

Lukin, Fault-tolerant architecture for quantum computation using electrically con-

trolled semiconductor spins, Nat Phys 1, 177 (2005).

[41] J. R. Petta, A. C. Johnson, J. M. Taylor, E. A. Laird, A. Yacoby, M. D. Lukin, C. M.

Marcus, M. P. Hanson, and A. C. Gossard, Coherent manipulation of coupled electron

spins in semiconductor quantum dots, Science 309, 2180 (2005).

[42] S. Foletti, H. Bluhm, D. Mahalu, V. Umansky, and A. Yacoby, Universal quantum

control of two-electron spin quantum bits using dynamic nuclear polarization, Nat

Phys 5, 1 (2009).

86
[43] E. A. Laird, C. Barthel, E. I. Rashba, C. M. Marcus, M. P. Hanson, and A. C. Gossard,

Hyperfine-mediated gate-driven electron spin resonance, Phys. Rev. Lett. 99, 246601

(2007).

[44] E. A. Laird, C. Barthel, E. I. Rashba, C. M. Marcus, M. P. Hanson, and A. C.

Gossard, A new mechanism of electric dipole spin resonance: hyperfine coupling in

quantum dots, Semicond. Sci. Technol. 24, 064004 (2009).

[45] E. I. Rashba, Properties of semiconductors with an extremum loop: 1. Cyclotron and

combinational resonance in a magnetic field perpendicular to the plane of the loop,

Sov. Phys. Sol. State. 2, 1109 (1960).

[46] R. Bell, Electric dipole spin transitions in InSb, Phys. Rev. Lett. 9, 52 (1962).

[47] B. McCombe, S. Bishop, and R. Kaplan, Combined resonance and electron g-values

in InSb, Phys. Rev. Lett. 18, 748 (1967).

[48] E. I. Rashba and V. I. Sheka, Landau Level Spectroscopy (Elsevier Science Publishers,

1991).

[49] F. Jelezko, T. Gaebel, I. Popa, A. Gruber, and J. Wrachtrup, Observation of coherent

oscillations in a single electron spin, Phys. Rev. Lett. 92, 076401 (2004).

[50] D. Rugar, R. Budakian, H. J. Mamin, and B. W. Chui, Single spin detection by

magnetic resonance force microscopy, Nature 430, 329 (2004).

[51] M. Xiao, I. Martin, E. Yablonovitch, and H. Jiang, Electrical detection of the spin

resonance of a single electron in a silicon field-effect transistor, Nature 430, 435

(2004).

87
[52] S. I. Pekar and E. I. Rashba, Combined resonance in crystals in inhomogeneous mag-

netic fields, Sov. Phys. - JETP 20, 1295 (1965).

[53] Y. Kato, R. Myers, D. C. Driscoll, A. C. Gossard, J. Levy, and D. Awschalom, Giga-

hertz electron spin manipulation using voltage-controlled g-tensor modulation, Science

299, 1201 (2003).

[54] Y. Tokura, W. G. V. D. Wiel, T. Obata, and S. Tarucha, Coherent single electron

spin control in a slanting Zeeman field, Phys. Rev. Lett. 96, 4 (2006).

[55] V. N. Golovach, M. Borhani, and D. Loss, Electric-dipole-induced spin resonance in

quantum dots, Phys Rev B 74, 10 (2006).

[56] K. C. Nowack, F. H. L. Koppens, Y. V. Nazarov, and L. M. K. Vandersypen, Coherent

control of a single electron spin with electric fields, Science 318, 1430 (2007).

[57] M. Pioro-Ladrière, T. Obata, Y. Tokura, Y. S. Shin, T. Kubo, K. Yoshida,

T. Taniyama, and S. Tarucha, Electrically driven single-electron spin resonance in

a slanting Zeeman field, Nat Phys 4, 776 (2008).

[58] H.-A. Engel and D. Loss, Detection of single spin decoherence in a quantum dot via

charge currents, Phys. Rev. Lett. 86, 4648 (2001).

[59] M. Gueron and C. Ryter, Overhauser effect in metallic lithium, Phys. Rev. Lett. 3,

338 (1959).

[60] M. Dobers, K. V. Klitzing, J. Schneider, G. Weimann, and K. H. Ploog, Electrical

detection of nuclear magnetic-resonance in GaAs-Alx Ga1−x As heterostructures, Phys.

Rev. Lett. 61, 1650 (1988).

88
[61] J. Baugh, Y. Kitamura, K. Ono, and S. Tarucha, Large nuclear overhauser fields

detected in vertically coupled double quantum dots, Phys. Rev. Lett. 99, 4 (2007).

[62] M. S. Rudner and L. S. Levitov, Electrically driven reverse Overhauser pumping of

nuclear spins in quantum dots, Phys. Rev. Lett. 99, 4 (2007).

[63] D. Goldhaber-Gordon, H. Shtrikman, D. Mahalu, D. Abusch-Magder, U. Meirav, and

M. Kastner, Kondo effect in a single-electron transistor, Nature 391, 156 (1998).

[64] R. Hanson, B. Witkamp, L. M. K. Vandersypen, L. H. W. V. Beveren, J. Elzerman,

and L. P. Kouwenhoven, Zeeman energy and spin relaxation in a one-electron quantum

dot, Phys. Rev. Lett. 91, 196802 (2003).

[65] F. H. L. Koppens, J. A. Folk, J. Elzerman, R. Hanson, L. H. W. V. Beveren, I. T.

Vink, H.-P. Tranitz, W. Wegscheider, L. P. Kouwenhoven, and L. M. K. Vandersypen,

Control and detection of singlet-triplet mixing in a random nuclear field, Science 309,

1346 (2005).

[66] A. C. Johnson, J. R. Petta, J. M. Taylor, A. Yacoby, M. D. Lukin, C. M. Marcus,

M. P. Hanson, and A. C. Gossard, Triplet–singlet spin relaxation via nuclei in a double

quantum dot, Nature 435, 925 (2005).

[67] K. Ono and S. Tarucha, Nuclear-spin-induced oscillatory current in spin-blockaded

quantum dots, Phys. Rev. Lett. 92, 256803 (2004).

[68] L. Khazan, Y. Rubo, and V. Sheka, Exchange-induced optical spin transitions in

semimagnetic semiconductors, Phys Rev B 47, 13180 (1993).

89
[69] E. A. Laird, J. R. Petta, A. C. Johnson, C. M. Marcus, A. Yacoby, M. P. Hanson, and

A. C. Gossard, Effect of exchange interaction on spin dephasing in a double quantum

dot, Phys. Rev. Lett. 97, 4 (2006).

[70] E. I. Rashba, Theory of electric dipole spin resonance in quantum dots: Mean field

theory with Gaussian fluctuations and beyond, Phys Rev B 78, 195302 (2008).

[71] I. A. Merkulov, Electron spin relaxation by nuclei in semiconductor quantum dots,

Phys Rev B 65, 8 (2002).

[72] L. S. Levitov and E. I. Rashba, Dynamical spin-electric coupling in a quantum dot,

Phys Rev B 67, 5 (2003).

[73] I. Rabi, Space quantization in a gyrating magnetic field, Phys Rev 51, 0652 (1937).

[74] F. H. L. Koppens, D. Klauser, W. A. Coish, K. C. Nowack, L. P. Kouwenhoven,

D. Loss, and L. M. K. Vandersypen, Universal phase shift and nonexponential decay

of driven single-spin oscillations, Phys. Rev. Lett. 99, 106803 (2007).

[75] D. Paget, G. Lampel, B. Sapoval, and V. Safarov, Low field electron-nuclear spin

coupling in gallium-arsenide under optical-pumping conditions, Phys Rev B 15, 5780

(1977).

[76] L. Dicarlo, H. Lynch, A. C. Johnson, L. I. Childress, K. Crockett, C. M. Marcus,

M. Hanson, and A. C. Gossard, Differential charge sensing and charge delocalization

in a tunable double quantum dot, Phys. Rev. Lett. 92, 226801 (2004).

[77] A. Khaetskii, D. Loss, and L. Glazman, Electron spin decoherence in quantum dots

due to interaction with nuclei, Phys. Rev. Lett. 88, 186802 (2002).

[78] A. Overhauser, Polarization of nuclei in metals, Phys Rev 91, 476 (1953).

90
[79] M. Pioro-Ladrière, Y. Tokura, T. Obata, T. Kubo, and S. Tarucha, Micromagnets for

coherent control of spin-charge qubit in lateral quantum dots, Appl. Phys. Lett. 90,

024105 (2007).

[80] P. Franken, G. Weinreich, C. Peters, and A. Hill, Generation of optical harmonics,

Phys. Rev. Lett. 7, 118 (1961).

[81] A. De, C. E. Pryor, and M. E. Flatte, Electric-field control of a hydrogenic donor’s

spin in a semiconductor, Phys. Rev. Lett. 102, 017603 (2009).

[82] T. Fujisawa, D. Austing, Y. Tokura, Y. Hirayama, and S. Tarucha, Allowed and

forbidden transitions in artificial hydrogen and helium atoms, Nature 419, 278 (2002).

[83] A. S. Bracker, E. A. Stinaff, D. Gammon, M. E. Ware, J. G. Tischler, A. Shabaev,

A. L. Efros, D. Park, D. Gershoni, V. L. Korenev, and I. A. Merkulov, Optical pumping

of the electronic and nuclear spin of single charge-tunable quantum dots, Phys. Rev.

Lett. 94, 4 (2005).

[84] P.-F. Braun, X. Marie, L. Lombez, B. Urbaszek, T. Amand, P. Renucci, V. K. Kale-

vich, K. V. Kavokin, O. Krebs, P. Voisin, and Y. Masumoto, Direct observation of

the electron spin relaxation induced by nuclei in quantum dots, Phys. Rev. Lett. 94,

4 (2005).

[85] R. Hanson, L. H. W. V. Beveren, I. T. Vink, J. Elzerman, W. J. M. Naber, F. H. L.

Koppens, L. P. Kouwenhoven, and L. M. K. Vandersypen, Single-shot readout of

electron spin states in a quantum dot using spin-dependent tunnel rates, Phys. Rev.

Lett. 94, 4 (2005).

91
[86] S. Erlingsson, Y. Nazarov, and V. Fal’ko, Nucleus-mediated spin-flip transitions in

gaas quantum dots, Phys Rev B 64, art. no. (2001).

[87] N. Turro, Influence of nuclear-spin on chemical-reactions - magnetic isotope and

magnetic-field effects (a review), P Natl Acad Sci Usa 80, 609 (1983).

[88] A. Buchachenko, Magnetic isotope effect: Nuclear spin control of chemical reactions,

J Phys Chem A 105, 9995 (2001).

[89] H. Staerk, W. Kuhnle, R. Treichel, and A. Weller, Magnetic-field dependence of in-

tramolecular exciplex formation in polymethylene-linked A-D systems, Chem Phys

Lett 118, 19 (1985).

[90] V. Tarasov, N. Ghatlia, A. Buchachenko, and N. Turro, Probing the exchange inter-

action through micelle size: 1. Probability of recombination of triplet geminate radical

pairs, J Am Chem Soc 114, 9517 (1992).

[91] W. A. Coish and D. Loss, Singlet-triplet decoherence due to nuclear spins in a double

quantum dot, Phys Rev B 72, 12 (2005).

[92] J. R. Petta, A. C. Johnson, A. Yacoby, C. M. Marcus, M. Hanson, and A. C. Gossard,

Pulsed-gate measurements of the singlet-triplet relaxation time in a two-electron double

quantum dot, Phys Rev B 72, 161301 (2005).

[93] D. M. Zumbuhl, C. M. Marcus, M. Hanson, and A. C. Gossard, Cotunneling spec-

troscopy in few-electron quantum dots, Phys. Rev. Lett. 93, 256801 (2004).

[94] D. P. DiVincenzo, D. Bacon, J. Kempe, G. Burkard, and K. Whaley, Universal quan-

tum computation with the exchange interaction, Nature 408, 339 (2000).

92
 [95] E. Step, A. Buchachenko, and N. Turro, Paramagnetic interactions of triplet radical

pairs with nitroxide radicals - an antiscavenging effect, J Am Chem Soc 116, 5462

(1994).

[96] A. Buchachenko and V. Berdinsky, Electron spin catalysis, Chem Rev 102, 603 (2002).

[97] L. Gaudreau, S. A. Studenikin, A. S. Sachrajda, P. Zawadzki, A. Kam, J. Lapointe,

M. Korkusinski, and P. Hawrylak, Stability diagram of a few-electron triple dot, Phys.

Rev. Lett. 97, 4 (2006).

[98] D. Schröer, A. D. Greentree, L. Gaudreau, K. Eberl, L. C. L. Hollenberg, J. P. Kot-

thaus, and S. Ludwig, Electrostatically defined serial triple quantum dot charged with

few electrons, Phys Rev B 76, 11 (2007).

[99] L. Gaudreau, A. Kam, G. Granger, S. A. Studenikin, P. Zawadzki, and A. S. Sachra-

jda, A tuneable few electron triple quantum dot, arXiv:0907.1322v1 (2009).

[100] G. Toth and C. Lent, Quantum computing with quantum-dot cellular automata, Phys.

Rev. A 63, 052315 (2001).

[101] L. Gaudreau, A. S. Sachrajda, S. A. Studenikin, P. Zawadzki, and A. Kam, Spin

blockade of quantum cellular automata effects in a few electron triple quantum dot,

Physica E 40, 978 (2008).

[102] T. Stevenson, F. Pellerano, C. Stahle, K. Aidala, and R. J. Schoelkopf, Multiplexing

of radio-frequency single-electron transistors, Appl. Phys. Lett. 80, 3012 (2002).

[103] D. J. Reilly, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Fast single-charge

sensing with a rf quantum point contact, Appl. Phys. Lett. 91, 162101 (2007).

93
[104] J. M. Taylor, J. R. Petta, A. C. Johnson, A. Yacoby, C. M. Marcus, and M. D. Lukin,

Relaxation, dephasing, and quantum control of electron spins in double quantum dots,

Phys Rev B 76, 1 (2007).

[105] G. Shinkai, T. Hayashi, Y. Hirayama, and T. Fujisawa, Controlled resonant tunneling

in a coupled double-quantum-dot system, Appl. Phys. Lett. 90, 103116 (2007).

[106] F. Perez-Martinez, I. Farrer, D. Anderson, G. A. C. Jones, D. A. Ritchie, S. J.

Chorley, and C. G. Smith, Demonstration of a quantum cellular automata cell in

a GaAs/AlGaAs heterostructure, Appl. Phys. Lett. 91, 032102 (2007).

[107] K. D. Petersson, C. G. Smith, D. Anderson, P. Atkinson, G. A. C. Jones, and D. A.

Ritchie, Microwave-driven transitions in two coupled semiconductor charge qubits,

Phys. Rev. Lett. 103, 016805 (2009).

[108] D. Austing, S. Sasaki, S. Tarucha, S. Reimann, M. Koskinen, and M. Manninen,

Ellipsoidal deformation of vertical quantum dots, Phys Rev B 60, 11514 (1999).

[109] M. Switkes, Decoherence and adiabatic transport in semiconductor quantum dots,

Ph.D. Thesis, Harvard University (1999).

[110] S. R. Patel, Electron transport and dephasing in semiconductor quantum dots, Ph.D.

Thesis, Harvard University (2002).

[111] S. M. Cronenwett, Coherence, charging and spin effects in quantum dots and point

contacts, Ph.D. Thesis, Harvard University (2001).

[112] L. DiCarlo, Mesoscopic electronics beyond dc transport, Ph.D. Thesis, Harvard Uni-

versity (2007).

[113] F. Pobell, Matter and Methods at Low Temperatures (Springer, 1996).

94
[114] D. M. Pozar, Microwave Engineering (John Wiley, 1988).

95