dynamic modulation of material properties by solid state
TRANSCRIPT
Dynamic Modulation of Material Properties by Solid
State Proton Gating
by Aik Jun Tan
Submitted to the Department of Materials Science and Engineering in partial fulfilment of the
requirements for the degree of
Doctor of Philosophy in Materials Science and Engineering
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2019
© Massachusetts Institute of Technology 2019. All rights reserved
Author…………………………………………………………………………………………........
Department of Materials Science and Engineering
May 8th,2019
Certified by……………………………………………………………………………………........
Geoffrey S. D. Beach
Professor of Materials Science and Engineering
Thesis Supervisor
Accepted by…...………………………………………………………………………………........
Donald R. Sadoway
Chairman, Department Committee on Graduate Studies
Dynamic Modulation of Material Properties by Solid
State Proton Gating
by Aik Jun Tan
Submitted to the Department of Materials Science and Engineering on May 8th, 2019 in partial
fulfilment of the requirements for the degree of
Doctor of Philosophy in Materials Science and Engineering
Abstract
As functionalities become more abundant in solid state devices, one key capability which
remains lacking is an effective means to dynamically tune material properties. In this thesis, we
establish a pathway towards this capability by utilizing the simplest ion known to mankind: the
proton. We demonstrate for the first time dynamic control of magnetic properties in an all-solid-
state heterostructures using solid state proton gating in a metal/oxide heterostructure. We also
demonstrate dynamic modulation of magnetic anisotropy at a metal-metal interface through
hydrogen insertion in a heavy metal adjacent to a ferromagnet. Besides magnetic properties,
solid state proton gating also enables dynamic modulation of optical properties in a thin film
oxide. We observe fast gating of optical reflectivity by ~10% at timescale down to ~20ms in a
metal/oxide/metal heterostructure. Finally, we also demonstrate a room temperature reversible
solid oxide fuel cell based on hydrogen storage. The cell has a small form factor which is
suitable for energy storage in solid state microelectronics application. Our work hence provides a
platform for complete control of material properties through solid state proton gating.
Thesis Supervisor: Geoffrey S.D. Beach
Title: Professor of Materials Science and Engineering
Acknowledgements
First and foremost, I would like to thank my advisor, Prof. Geoffrey Beach. From him, I learned
how to properly conduct a research, and how to effectively present the findings of the research.
He sets a good example in terms of his hardwork and his relentless pursuit of perfection. And he
gave me a lot of autonomy when it comes to my research. He is exactly the advisor I have hoped
for and I am grateful to him.
I would also like to thank my thesis committee Prof. Harry Tuller and Prof. Eugene Fitzgerald.
Prof. Tuller has been very kind to me, and I have learned so much about solid state ionics from
him. Prof. Fitzgerald has given me very useful inputs on my thesis.
I would like to thank all my teammates in the Beach group: Max, Lucas, Mantao, Felix, Ivan,
Sara, Jason, Can, Kohei, Uwe, Satoru, Liz, Parnika, Minae, Shwoo, Chi Feng, Usama, Daniel,
and Siying. They are wonderful people to work with, and my life in MIT was made colorful by
them.
I would like to thank the staff in DMSE and MRL who have made this thesis possible: David
Bono, Charlie, Libby, Mike, Tara, Angelita, Elissa, Jessie, Dominique, John, and Jessie. They
are some of the nicest people I know, and they have given me help whenever asked.
Finally, I would like to thank my family. My parents are a source of wisdom and strength, and
they are the most important people in my life. My elder brother, who always looks after his
younger siblings, is someone I can always have a frank conversation with. And my younger
sister, who is the most intelligent among us, is someone I can always poke fun at.
Contents Chapter 1: Introduction ................................................................................................................... 1
1.1 Motivation ............................................................................................................................. 2
1.2 Thesis Outline ....................................................................................................................... 3
Chapter 2: Background ................................................................................................................... 6
2.0 Magnetic Hysteresis Loop .................................................................................................... 7
2.1 Magnetic Anisotropy ............................................................................................................ 9
2.2 Magnetization Dynamics and Spin Current ........................................................................ 20
2.3 Magneto-Electric and Magneto-Ionic Effects ..................................................................... 24
2.4 Solid Oxide Fuel and Electrolyzer cell ............................................................................... 31
2.5 Solid Oxide Proton Conductors .......................................................................................... 36
2.6 Water Electro-Catalysis ...................................................................................................... 44
2.7 Electrodes for Solid Oxide Cells......................................................................................... 47
2.8 Electrochemical Impedance Spectroscopy ......................................................................... 51
Chapter 3: Experimental Methods ................................................................................................ 54
3.1 Sputter Deposition .............................................................................................................. 55
3.2 Sample Structure and Patterning ......................................................................................... 61
3.3 Magneto-Optical Kerr Effect .............................................................................................. 67
3.4 Anomalous and Planar Hall Effect ..................................................................................... 72
3.5 Time Resolved Hall Magnetometry under different Atmospheric Conditions ................... 74
3.6 Spin-torque Ferromagnetic Resonance ............................................................................... 77
3.7 Solid Oxide Cell Characterization ...................................................................................... 79
Chapter 4: Effect of H2O on Voltage-induced Co Oxidation in a Pt/Co/GdOx Heterostructure .. 81
4.1: Experimental Methods ....................................................................................................... 84
4.2: Probing Water Uptake in GdOx ......................................................................................... 86
4.3: Voltage-induced Co Oxidation in Hydrated and Non-hydrated Pt/Co/GdOx Devices ...... 90
4.4: H2 evolution during Voltage-induced Co Oxidation in Pt/Co/GdOx ................................. 94
4.5: In-situ XAS probe of Co during Voltage-induced Co Oxidation in Pt/Co/GdOx ............. 98
Chapter 5: Magneto-ionic Control of Magnetism using a Solid-state Proton Pump .................. 100
5.1: Experimental Methods ..................................................................................................... 103
5.2: Co Redox through Water Electrolysis ............................................................................. 105
5.3: Modulation of Magnetic Anisotropy through Proton Injection ....................................... 112
5.4: Magnetic Response under Short Circuit and Open Circuit.............................................. 117
5.5: Electrical Gating of Magnetic Anisotropy at a Heavy-metal/ferromagnet Interface ....... 120
5.6 Comparison between Au and Pt Top Electrodes .............................................................. 124
Chapter 6: Voltage Gating of Magnetic Damping and Spin-Orbit Torques using Proton .......... 126
6.1 Experimental Methods ...................................................................................................... 129
6.2 Spin Torque Ferromagnetic Resonance to Probe Voltage Gating of Spin Orbit Torque and
Magnetic Damping.................................................................................................................. 131
Chapter 7: Voltage-induced Magneto-Ionic Effect in Pt/Co/MOx Heterostructure (M= Gd, Y, Zr,
and Ta) ........................................................................................................................................ 138
7.1 Experimental Methods ...................................................................................................... 141
7.2 Rate of Voltage-Induced Magnetic Modulation at Positive Bias ..................................... 142
7.3 Rate of Voltage-Induced Magnetic Modulation at Negative Bias .................................... 145
Chapter 8: Room Temperature Reversible Solid Oxide Fuel Cell ............................................. 148
8.1 Experimental Methods ...................................................................................................... 150
8.2 Proton Conductivity of GdOx............................................................................................ 152
8.3 Cell Performance and Scalability...................................................................................... 154
8.4 Gating of Magnetism using Built-in Voltage .................................................................... 158
Chapter 9: Voltage Gating of Optical Properties ........................................................................ 161
9.1 Experimental Methods ...................................................................................................... 163
9.2 Voltage Gating of Optical Reflectivity in Pt/GdOx/Au Heterostructure .......................... 164
9.3 Source of Irreversible Optical Change in Pt/GdOx/Au Heterostructure ........................... 169
9.4 Voltage Gating of GdOx Heterostructures with different Top and Bottom Electrodes .... 172
9.5 Optical Modulation outside Active Region due to Hydrogen Diffusion .......................... 175
Chapter 10: Electrical Properties of GdOx .................................................................................. 178
Chapter 11: Summary and Outlook ............................................................................................ 186
11.1 Summary ......................................................................................................................... 187
11.2 Outlook ........................................................................................................................... 189
11.2.1 Integration of Hydrogen Storage in Proton Magneto-Ionic Device ......................... 189
11.2.2 Proton Magneto-Ionics for Memory and Logic Devices ......................................... 190
11.2.3 Proton Magneto-Ionics to Quantify Proton Conductivity in Thin Film Oxides ...... 193
References ................................................................................................................................... 194
1
Chapter 1:
Introduction
2
1.1 Motivation
As solid state devices continue its path towards miniaturization, there are two important trends to
note: (1) interfaces play an increasingly important role in determining material properties1, and
(2) the electric field, which is the primary tool by which we control device functions, becomes
progressively larger. Simultaneously, the need to dynamically toggle material properties have
become more crucial due to limited functionalities and performance imposed by static devices.
For instance, in magnetic memory, it is extremely difficult to achieve both thermal stability and
low writing power simultaneously because thermal stability implies large energy barrier to
magnetic switching, while low writing power implies low energy barrier to magnetic
switching2,3. These are two conflicting requirements which are very difficult to optimize in a
static device.
In this thesis, we take advantage of functional interfaces and large electric field in nanoscale
devices in order to provide a mechanism by which we can dynamically induce large changes in
material properties. We demonstrate dynamic modulation of material properties in thin film
devices through solid state proton gating. Proton is used because it has high mobility at room
temperature which allows for fast device operation. At the same time, it can induce very large
changes in device properties because it disrupts chemical bonding at functional interfaces. In this
sense, it captures the best of both worlds in terms of speed of electronic modulation and
magnitude of ionic modulation. The term “magneto-ionic” is used to refer to magnetic-ionic
coupling where ions are used to modulate magnetic properties.
3
1.2 Thesis Outline
This thesis is written with the aim to enable a wide audience with little knowledge of magnetism,
ionics, or electrochemistry to interpret the new findings.
In chapter 1, we give a preliminary introduction to dynamic modulation of material properties
using solid state proton gating.
In chapter 2, we provide some scientific background which is necessary to understanding the
results presented in this thesis. The chapter starts off with introduction to magnetism since a
large part of the thesis is focused on modulation of magnetic properties. The chapter eventually
transitions into introduction to proton conductors and water electro-catalysis because we source
protons from water and transport them through an oxide electrolyte to modulate material
properties.
In chapter 3, we provide a detailed overview of the primary experimental techniques used to
perform the research in this thesis. The chapter starts off with description of the sample
structures and fabrication steps. This is followed by characterization techniques used to probe the
magnetic and electrochemical properties of the samples.
In chapter 4, we present the first experimental evidence that voltage-induced Co oxidation in a
Pt/Co/GdOx/Au heterostructure is dominated by water oxidation instead of oxidation by oxygen
ions. The findings represent an important breakthrough in understanding of voltage induced
redox in ferromagnet/oxide systems and show that water can play a crucial role in previously
observed magneto-ionic effect.
4
In chapter 5, we show for the first time solid state gating of magnetic anisotropy using proton.
This discovery allows for 90° toggling of the magnetization in a Pt/Co/GdOx system at
unprecedented speed and cyclability. Proton magneto-ionics also allow for gating of magnetic
anisotropy at a metal-metal interface through hydrogen insertion in a heavy metal adjacent to a
ferromagnet.
In chapter 6, we demonstrate voltage gating of magnetic damping and spin torques probed using
spin-torque ferromagnetic resonance. The results show that a wide range of fundamental
magnetic properties can be gated using ions.
In chapter 7, we compare the rates of voltage-induced magnetic changes in different Pt/Co/MOx
heterostructures, where M is Gd, Y, Zr, or Ta. The results show that speed of magnetic
modulation can change significantly depending on the choice of MOx as the proton conducting
electrolyte.
In chapter 8, we introduce a room temperature reversible solid oxide cell based on hydrogen
storage in a thin film oxide. The cell is a miniaturized version of a conventional solid oxide fuel
cell, and can be operated like a battery for microelectronics. The finding is important because it
shows the applicability of solid-oxide fuel cells for energy storage in microelectronics.
In chapter 9, we demonstrate voltage-induced modulation of optical properties in a thin film
metal/oxide/metal heterostructure using proton. Fast optical response down to 20ms was
achieved with good cyclability. This establishes the wide applicability of voltage-induced
protonics to modulate material properties in thin film heterostructures.
In chapter 10, we revisit some electrical properties of GdOx since it is used as the proton
conducting oxide in most of our gated devices.
5
In chapter 11, we summarize the findings of the thesis and discuss their impacts. This is followed
by a list of recommended future work to fill the remaining gap in our understanding proton
gating of material properties. We end by providing a brief outlook for this new field of research.
6
Chapter 2:
Background
7
2.0 Magnetic Hysteresis Loop The simplest and densest way to describe a magnetic sample is a magnetic hysteresis loop4. A
magnetic hysteresis loop is a plot of magnetization vs magnetic field as shown in Figure 2.1(a).
In magnetic hysteresis loops, the same magnetic field which is applied does not always produce
the same magnetization. The magnetization, M depends not only on the magnetic field which is
currently applied, it also depends on its previous state (This is why it is called a “hysteresis”
loop).
Figure 2.1. Magnetic hysteresis loop. (a) Easy-axis magnetic hysteresis loop. The coercive
field, HC is typically half the width of the hysteresis loop, while the saturation magnetization, MS
is half the height of the hysteresis loop. (b) A general profile of the applied magnetic field. (c)
The measured magnetization corresponding to the magnetic field in (b). (d) The direction of the
applied magnetic field and the magnetization. Both are pointing along the y-axis.
To generate a magnetic hysteresis loop, one sweeps the magnetic field and measures the
magnetization at each field according to the profile shown in figure 2.1b. For a sample which is
(1) uniformly magnetized in the direction of the magnetic field, and (2) where the magnetization
rotates coherently as a single domain (Stoner-Wohlfarth model), the resulting magnetization is
shown in figure 2.1c. When we first increase the magnetic field in the +y direction, the magnet
8
starts out pointing in a –y direction (hysteresis from its previous state). When the magnetic field
exceeds a critical field, the magnet “abruptly” flips 180º to the positive direction. This critical
field is called the coercive field, HC and it quantifies the energy barrier to switch the
magnetization 180º (assuming Stoner-Wohlfarth model). As one increases the magnetic field
further in the positive direction, the measured magnetization remains flat. The value at which this
magnetization plateaus is called the saturation magnetization, MS. When we next increase the
magnetic field in the –y direction, the magnetization again flips 180º at HC and plateaus at MS
but with a negative sign. If we plot the magnetization versus the magnetic field, we get the
magnetic hysteresis loop shown in figure 2.1a.
A magnetic hysteresis loop is measured using a variety of techniques such as magneto-optical
Kerr effect, and anomalous Hall effect. In these cases, M is measured indirectly as Kerr rotation
or Hall resistance respectively. These will be described in greater details in chapter 3. While we
can always extract HC and MS from a magnetic hysteresis loop, the magnetic information is not
limited to the two quantities. For instance, if the magnetic sample is not in a single domain state,
then the hysteresis loop can give us qualitative information on the domain configuration. It can
also give us information on the magnetic anisotropy depending on which direction the magnetic
field is applied in. The details will be discussed in subsequent sections.
In magnetism, there are two kinds of units which are frequently used: cgs (centimeter-gram-
second) and SI units. While SI units are more standardized, the magnitude is often too large for
thin film magnetism. For this thesis, we will mainly be using cgs units unless stated otherwise.
9
2.1 Magnetic Anisotropy
In an anisotropic system, the magnetization vector tends to point in certain preferred axis
because it minimizes its energy. This is called magnetic anisotropy4. The minimum energy axis
is called an easy magnetization axis whereas the maximum energy axis is called the hard
magnetization axis. In thin film magnetism, the term “magnetic anisotropy” is typically used in a
more specific sense; it represents an energy density, KU needed to rotate the magnetization from
the easy-axis to the hard-axis direction. This value is extremely important because it quantifies
the energy barrier which the magnetic moment needs to overcome in order switch180º. In other
words, the moment goes through the hard axis as it switches from one direction to the opposite
direction4,5.
10
Figure 2.2. Magnetic anisotropy. (a) Energy coordinate of M with the magnetic field applied
along the easy axis in –y direction. (b) Easy axis magnetic hysteresis loop. (c) Energy coordinate
of M with the magnetic field applied along the hard axis in +z direction. (d) Hard-axis magnetic
hysteresis loop. Adapted from reference5
In figure 2.2, we assumed the easy-axis is the y-axis while the hard axis is the z-axis and we
start out with the magnetization pointing in the +y direction. When we apply a magnetic field, H
along the easy axis in the -y direction (figure 2.2(a),), we decrease the energy of the –y state
while increasing the energy of the +y state. When the energy due to the magnetic field (Zeeman
energy) equals the energy barrier arising from the magnetic anisotropy, KU, the magnetization
switches from the +y direction to the -y direction. If one looks at the easy-axis hysteresis loop
which shows the y-projection of M (figure 2.2(b)), this happens at the coercive field, HC. If we
instead apply the magnetic field along the hard axis in the +z-direction (figure 2.2(c)), we reduce
the energy of the +z state while keeping the ±y state the same. As a result, there is no 180º
switching of the magnetization; the magnet simply rotates 90º from +y to +z. If one looks at the
hard-axis hysteresis loop (figure 2.2(d)), the z-projection of M gradually increases and saturates
at the anisotropy field, HK. Both HC and HK are equal to 2𝐾𝑈
𝑀𝑆 , hence we can see why magnetic
11
anisotropy is very important in determining the threshold magnetic field to switch or rotate the
magnetization.
Magnetic anisotropy arises from two microscopic origins: dipolar interactions and spin-orbit
coupling4–7. Dipolar interactions refers to the interaction between magnetic dipoles, and
intuitively, one can think of the dipolar energy being minimized when the one magnetic dipole is
aligned along the magnetic field generated by a second magnetic dipole. Dipolar interactions
lead to magnetic anisotropy because moments lower their energy when they are aligned along a
common axis. In bulk systems, there will be a very large number of these microscopic dipoles;
hence to find the stable magnetic configuration, one has to integrate all the individual dipoles to
find the minimum dipolar energy of the system. The second microscopic origin of magnetic
anisotropy is spin-orbit coupling (SOC). Spin-orbit coupling refers to the coupling between the
spin and orbital moments of each atom.
Figure 2.3. Spin orbit coupling. (a) Trajectory of electron from reference frame of the nucleus.
(b) Trajectory of the nucleus from reference frame of the electron
When electron spin orbits the nucleus of an atom, it experiences an effective magnetic field from
the nucleus because from the electron’s reference frame, it “sees” the positively charged nucleus
orbiting itself (a moving charge generates a magnetic field) (figure 2.3). The magnetic field
12
“generated” from the orbiting nucleus is represented by the orbital moment, L while the electron
spin is represented by the spin moment, S. As a result, the energy of spin-orbit coupling is
expressed as 𝐸𝑆𝑂 = −𝜉𝐿. 𝑆 where 𝜉 is the coefficient which represents the strength of the spin-
orbit coupling4–7. SOC leads to magnetic anisotropy because it couples the spin moment to the
orbital moment. And because the orbital moment is in turn coupled to the crystal lattice, a form
of magnetic anisotropy arises where the magnetic moments are stabilized along certain high
symmetry axes of the crystal lattice. This form of anisotropy is called magnetocrystalline
anisotropy.
In this thesis, we will be focusing on thin film 3d ferromagnet systems where the thickness of the
magnetic film is on the order of nm. While the microscopic origins (the two stated above) of
magnetic anisotropy are the same for all magnetic systems, in these thin film systems, we
classify the magnetic anisotropy into two contributions: a volumetric and a surface contribution.
The volumetric contribution, KV (energy per unit volume) is mainly due to shape anisotropy.
This anisotropy refers to the dependence of the magnetization easy axis on the shape of the
magnet and it arises due to dipolar interactions. For each shape and magnetization direction,
there is a corresponding total dipolar energy, or magnetostatic energy, 𝐸𝑚𝑎𝑔𝑛𝑒𝑡𝑜𝑠𝑡𝑎𝑡𝑖𝑐 which is
given by:
𝐸𝑚𝑎𝑔𝑛𝑒𝑡𝑜𝑠𝑡𝑎𝑡𝑖𝑐 =1
2𝐻𝐷⃗⃗ ⃗⃗ ⃗. �⃗⃗� (Equation 2.1)
𝐻𝐷⃗⃗ ⃗⃗ ⃗ = −𝑁𝐷�⃗⃗� (Equation 2.2)
Here, 𝐻𝐷⃗⃗ ⃗⃗ ⃗ is the demagnetizing field, 𝑁𝐷 is a geometry-dependent tensor, and �⃗⃗� is the
magnetization vector. The demagnetizing field 𝐻𝐷⃗⃗ ⃗⃗ ⃗, is an effective field experienced by each
magnetic dipole due to all other dipoles in the magnet. Equation 2.1 and 2.2 show that shape
13
anisotropy arises because there is certain magnetization direction relative to the magnet shape
where 𝐸𝑚𝑎𝑔𝑛𝑒𝑡𝑜𝑠𝑡𝑎𝑡𝑖𝑐 is minimum. One can further simplify the picture and think of a stable
magnetic configuration as one that results in the minimum area of “free” magnetic poles (figure
2.4).
Figure 2.4. Shape anisotropy. The magnetic configuration in (b) results in smaller area of
“free” magnetic poles (shaded in red) in the magnet compared to (a); as a result, (b) has smaller
Emagnetostatic.
In an ideal thin film which (1) stretches infinitely in the in-plane directions (x and y-axis) and
where (2) the magnetization points out-of-plane (z-axis), 𝐻𝐷 = 4πMS (MS is the saturation
magnetization) (figure 2.5) 4–7. This field points opposite to the out-of-plane magnetization,
resulting in 𝐸𝑚𝑎𝑔𝑛𝑒𝑡𝑜𝑠𝑡𝑎𝑡𝑖𝑐 of 2πMS2. If the magnetization points in any in-plane direction, the
resulting 𝐸𝑚𝑎𝑔𝑛𝑒𝑡𝑜𝑠𝑡𝑎𝑡𝑖𝑐 is 0. Hence, for a thin film magnet, assuming a negative convention for
in-plane anisotropy (magnet prefers to point in-plane), the volumetric magnetic anisotropy, KV =
-2πMS2.
14
Figure 2.5. Demagnetizing Field. Cross section of a thin film magnet showing the directions of
the magnetic moment and the demagnetizing field.
For practical applications, it is the out-of-plane magnetic configuration which is desired in order
to achieve the highest memory density2,8–11. Another commonly used term for out-of-plane
magnetic anisotropy is perpendicular magnetic anisotropy (PMA). Fortunately, in a thin film
system, the surface magnetic anisotropy, KS (energy per unit area) tends to favor PMA.
Generally speaking, in thin film ferromagnet with broken inversion symmetry, there is splitting
in degeneracy of the different orbitals (figure 2.6). Due to electrostatic interactions, these lower
energy orbitals tend to have an out-of-plane anisotropy.
Figure 2.6. Surface magnetic anisotropy. Schematic illustration of splitting of degeneracy in
thin film magnet with broken inversion symmetry. Adapted with permission from reference12.
15
Because PMA is generated by filling bands with out-of-plane orbital anisotropy, the position of
the fermi level is very important. One way to tune the fermi level is through hybridization with
heavy metals at the ferromagnetic interface. For instance, it has been known for many years that
large PMA exists in in Co/Pt and Co/Pd multilayers13,14 due to hybridization of the Co 3d orbitals
with the 5d and 4d orbitals of Pt and Pd respectively15–19. This hybridization tunes the fermi level
such that the out-of-plane orbitals are maximally filled and provides large spin-orbit coupling
from the Pt and Pd atoms. This hybridization is localized at the Co/Pt or Co/Pd interface and, for
a Co thickness of at least a few monolayers, this interface constitutes the majority of the PMA
across the entire Co film (figure 2.7)17.
Figure 2.7. Orbital moment near the Co/Pt interface. The perpendicular magnetic anisotropy
is largest closest to the Co/Pt interface. Adapted with permission from reference17.
More recently, it has been shown that hybridization between the s-p orbitals of the oxygen atom
and dz-orbitals of a 3d ferromagnetic metals such as Co and Fe can also stabilize PMA20,21. This
is remarkable given the low spin-orbit coupling strength in oxygen. It was also found that the
16
position of the oxidation front is very important in determining the strength of the PMA. By
starting out with a Pt/Co/Al trilayer structure and subjecting the top Al layer to different duration
of plasma oxidation front, Manchon et al found that the largest PMA is obtained when the
oxidation front is right at the Co/Al interface, where the structure is exactly Pt/Co/AlOx(figure
2.8)22,23. When the Al is underoxidized, the Co magnetization points in-plane whereas if it is
overoxidized (Pt/Co/CoO/AlOx), the PMA starts to decrease again.
Figure 2.8. PMA as function of interfacial oxidation. (a) Anisotropy field, Han as a function of
plasma oxidation time of the Pt/Co/Al trilayer structure. (b) Hall effect magnetic hysteresis loops
corresponding to different plasma oxidation times. Largest PMA is obtained when the oxidation
front is exactly at the Co/Al interface. Adapted with permission from reference22.
The total magnetic anisotropy, KU is given by the sum of the surface (KS) and volumetric
magnetic anisotropy (KV), according to the equation16,18:
17
𝐾𝑈 = 𝐾𝑉 +𝐾𝑆
𝑡 Equation 2.3
Where 𝑡 is the thickness of the ferromagnetic film. Due to the competition between KS and KV,
there is a critical thickness below which one gets PMA and above which one gets an in-plane
magnetic state. This is shown in figure 2.9, where the authors plotted Keff *tCo against tCo (the
symbol Keff is used in place of KU, Co is the magnetic material). Note that the intercept is 2KS
because there are two surfaces where KS is present.
Figure 2.9. Keff*t versus t for a Pd/Co multilayers. Keff*t corresponds to in-plane magnetic
anisotropy while Keff*t > 0 corresponds to perpendicular magnetic anisotropy. Adapted with
permission from reference16.
While we have so far focused on surface anisotropy arising from interfacial hybridization in a
static structure, it has also been known that gas adsorption on ultra-thin magnetic films can alter
magnetic anisotropy. One of the most studied gas is hydrogen, H2 which can either physiosorb as
molecules or chemisorb on metallic thin film after splitting into individual H atom24–31. Mankey
et al first found that hydrogen chemisorption reduces the surface magnetization of Co while
18
enhancing that of Fe24. According to the authors, this is because the majority sub-band is nearly
filled in Co whereas it is only partially filled in Fe; as a result, when electrons are added due to
hybridization with hydrogen, they are added to the minority sub-band in Co and majority sub-
band in Fe. This changes the Co and Fe magnetizations in opposite directions. Valvidares et al
have found through magneto-optical Kerr imaging that adsorption of hydrogen on a Pt(111)
substrate before the deposition of a thin Co layer (t < 1.3nm) decreases the PMA significantly25.
Using ab-initio calculations, they found a large reduction in magnet moments of the Co, which in
turn reduces the magnetic moment in Pt induced by proximity effect. As Pt atoms constitute the
largest source of spin-orbit-coupling, this results in a significant reduction in the PMA. Similarly,
Munbodh et al have found that hydrogen absorption in Co/Pd multilayers decreases both the
magnetic moment and PMA in the structure due to decreased induced moments in the Pd layer26.
Unlike Co heterostructures, Sander et al instead found that exposure of Ni/Cu(001) to hydrogen
induces PMA, and by cycling the hydrogen partial pressure they were able to cycle the
magnetization between in-plane and out-of-plane state reversibly (figure 2.10)27. In the case of
Ni, the authors attribute the change in magnetic anisotropy mainly to tetragonal distortion in the
Ni crystal structure.
19
Figure 2.10. Reversible hydrogen gas induced magnetization switching. Polar magneto-
optical Kerr signal (proportional to out of plane magnetization) of a Ni/Cu(001) thin film
structure as a function of partial pressure of hydrogen. Inset shows the direction of the
magnetization. Adapted with permission from reference27.
20
2.2 Magnetization Dynamics and Spin
Current
So far, we have pictured magnetization as moments which are aligned in a static configuration
along a magnetic field. However, magnetic moments undergo a precessional motion when
subjected to the slightest torque exerted by either a magnetic field or a spin current.
Magnetization dynamics is usually modelled by the Landau-Lifshiftz-Gilbert (LLG) equation
which is given by:
𝜕𝑚
𝜕𝑡= −𝛾𝑚 × 𝐻𝑒𝑓𝑓 + 𝛼𝑚 ×
𝜕𝑚
𝜕𝑡 Equation 2.4
Where 𝑚 is the magnetic moment vector, 𝜕𝑚
𝜕𝑡 is evolution of the magnetic moment as a function
of time, and 𝐻𝑒𝑓𝑓 is an effective field which includes contribution from the demagnetizing field,
interfacial anisotropy, and the applied field, H. 𝛾 is the gyromagnetic ratio and 𝛼 is the Gilbert
damping parameter. The first term on the right (−𝛾𝑚 × 𝐻𝑒𝑓𝑓) represents the field-like torque
which causes the magnetization to precess around 𝐻𝑒𝑓𝑓 while the second term on the right is the
damping-like torque (𝛼𝑚 ×𝜕𝑚
𝜕𝑡) which causes the magnetic moment to gradually “dampens”
towards the direction of 𝐻𝑒𝑓𝑓. Pictorially, this can be represented as:
21
Figure 2.11. Magnetization dynamics according to the LLG equation. Adapted from
reference32.
The original LLG equation was later modified to include spin-current induced torques
𝜕𝑚
𝜕𝑡= −𝛾𝑚 × 𝐻𝑒𝑓𝑓 + 𝛼𝑚 ×
𝜕𝑚
𝜕𝑡+ 𝜏𝐹𝐿
𝑠×𝑚
|(𝑠×𝑚)|+ 𝜏𝐷𝐿
𝑚×(𝑠×𝑚)
|𝑚×(𝑠×𝑚)| Equation 2.5
Where 𝜏𝐹𝐿 and 𝜏𝐷𝐿 are magnitudes of spin-current induced field-like and damping-like torques,
and s is the polarization of the spin current. Here, the third 𝜏𝐹𝐿𝑠×𝑚
|(𝑠×𝑚)| and the fourth
(𝜏𝐷𝐿𝑚×(𝑠×𝑚)
|𝑚×(𝑠×𝑚)|) terms are called the spin current induced field-like and damping-like
(Slonczewski) torques respectively because they have the same symmetry as the corresponding
field induced field-like and damping like torques. In this case, the spin current provides an
effective exchange field along the direction of its spin polarization. The third term drives the
moment to precess around this spin polarization direction while the fourth term drives the
moment to “dampen” towards the spin polarization direction. One way to generate spin current is
using the spin Hall effect where spin-orbit coupling in heavy metals drives spin current to flow in
tranverse direction to the charge curren (figure 2.12)33–35 .
22
Figure 2.12. Spin current induced by the spin Hall effect in Pt layer. The spin current
injected into the CoFe ferromagnetic layer generates a torque to drive magnetization dynamics or
switching. In this schematic, the charge current is along the x-axis, the spin current is along the
z-axis, and the spin polarization is along the y-axis. There is an exchange field due to the spin
current along the spin polarization direction. Adapted with permission from reference36.
Magnetization dynamics can be characterized using ferromagnetic resonance (FMR). In
conventional FMR, a magnetic field at high frequency is applied to a magnetic sample to provide
a torque (equation 2.4) which drives the moments to precess. Simultaneously, the absorption of
microwave radiation by the magnetic sample is measured in a microwave cavity. At resonance,
there will maximum absorption of radiation by the sample. The resonance frequency (or field)
then provides quantitative information about the 𝐻𝑒𝑓𝑓 experienced by the sample while the width
of the absorption peak is proportional to its damping parameters.
In this thesis, we instead rely on a variation of conventional FMR called spin-torque FMR (ST-
FMR) to characterize the magnetic properties of the magnetic sample37–40. A heavy metal such as
Ta or Pt is used to generate spin current using the spin Hall effect39. The generated spin current is
then used to drive magnetization precession in an adjacent ferromagnetic layer. The precession
of the magnetic layer manifests itself electrically as a change in DC resistance through
23
magnetoresistance, which can be measured as a mixing voltage(figure 2.13). A detailed
description of the method and analysis will be given in chapter 6.
Figure 2.13. Spin-torque ferromagnetic resonance (ST-FMR). (a) Schematic of the ST-FMR
measurement configuration. (b) Current induced change in magnetoresistance (ref. spin torque
diode effect). Adapted with permission from references38,39.
24
2.3 Magneto-Electric and Magneto-Ionic
Effects In magnetic memory, the ultimate goal is to have high memory density, low writing power, and
high thermal stability2. To achieve high memory density, the size of magnetic bit can be made
smaller. However smaller bits lead to degradation of the thermal stability because the energy
barrier to flip the magnetization is given by ∆𝐸 = 𝐾𝑢𝑉, where V is the volume of the magnetic
bit. ∆𝐸 has to be at least 60kT in order to have stable recording for more than 10 years (k is the
Boltzmann’s constant). One may compensate for this size reduction by increasing the magnetic
anisotropy, 𝐾𝑢 but as discussed in the previous section (chapter 2.1) the magnetic field, 𝐻𝑐
required to switch the magnetization is proportional to 𝐾𝑢. As a result, the total writing power
also increases. This challenge in achieving high memory density, low writing power, and high
thermal stability simultaneously in magnetic memory is called the trilemma.
One approach to overcoming the trilemma is to utilize a dynamic system where we reduce the
switching barrier temporarily during the writing operation. This can be done most effectively
using a gate voltage, and one of the mechanisms which allows for such modulation is the
magneto-electric effect. Magneto-electric effect refers to a wide range of phenomena which
allow for control of magnetization using an electric field. Magneto-electric phenomena can be
broadly classified into three main classes of materials: dilute magnetic semiconductors (DMS),
multiferroic materials, and ultra-thin metallic ferromagnet/oxide bilayers. The magneto-electric
coupling can be an intrinsic feature of the single phase material, or it can be coupled in
heterogeneous media through strain or interfacial interactions.
25
Dilute magnetic semiconductors (DMS) are semiconductors which exhibit magnetic properties
due to low level doping by ferromagnetic metals such as Mn41,42. The most studied of these DMS
is Mn-doped InGaAs which exhibits ferromagnetic properties due to hole-mediated interaction41–
44. With decreasing hole concentration, a super-exchange interaction which favors
antiferromagnetic configuration starts to dominate; as a result an electric field can reversibly
change the magnetic properties by changing the hole concentration through charge injection43–45.
This was first demonstrated by Chiba et al where they electrically gate the magnetic moment of
(In,Mn)As through modulation of its Curie temperature (figure 2.14)44. However, the major
problems with these systems are their extremely low Curie temperature (typically <50K) and the
complex fabrication steps to induce the ferromagnetism.
Multiferroics, on the other hand, are materials which exhibit magnetic and ferroelectric orders
simultaneously46. All of these materials have perovskite structures. In these materials, non-zero
magneto-electric coupling coefficient can arise from structural asymmetry; as a result, an electric
field can induce a change in magnetization and vice versa. Besides magneto-electric coupling in
a single phase system with non-zero magneto-electric coefficient, magneto-electric coupling can
also be engineered in heterogeneous ferroic media47–50. In this case, exchange or elastic coupling
between a ferroelectric and ferromagnetic material can allow for electric field control of
magnetization. Figure 2.15 shows an example of such system, where magnetic CoFe2O4
nanopillars are embedded with out-of-plane epitaxy in a ferroelectric BaTiO3 medium. When an
electric field is applied, the resulting strain in the ferroelectric BaTiO3 is imparted to the
CoFe2O4 nanopillars, causing a change in its magnetization48,49. While these systems have very
interesting physics, the main problem remains their complex fabrication process and the stringent
26
growth requirements (substrate epitaxy, growth temperature etc.) which are incompatible with
current CMOS processes.
Figure 2.14. Electric field control of ferromagnetism in a dilute magnetic semiconductor. (a) Schematic of device operation of a magnetic (In,Mn)As during electrical gating. (b) Hall
hysteresis loops of the (In,Mn)As at different gate voltages. Adapted with permission from
reference44.
Figure 2.15. Magneto-electric coupling in a heterogeneous multiferroic system. (a)
Schematic of magnetic CoFe2O4 (CFO) nanopillars embedded in a ferroelectric BaTiO3
(BTO)matrix. (b)-(c) Magnetic force microscope images of the CFO-BTO composite before and
after electrical poling at +12V. The region in red circle represents magnetization reversal upon
electrical poling. The green circle represents multi-domain formation. Adapted with permission
from references48,49.
27
The third class of materials where magneto-electric coupling has been observed is ultra-thin
metallic ferromagnet/ oxide bilayer systems12,51,52. In these devices, the ferromagnet needs to be
thin (<1nm) due to Coulombic screening in metals. Electric field-induced magnetic changes in
these materials has been attributed to a few factors: (1) spin-dependent screening of the electric
field can induce a net surface magnetization in an ultra-thin ferromagnet53. (2) electric field can
also lead to different occupation of the 3d orbitals at the surface layer of the ferromagnet; this in
turn changes the magnetic anisotropy54. (3) Besides changing the orbital occupation, it was also
proposed that an electric field directly changes the band structure of the ferromagnet and hence
the magnetic anisotropy55. The magneto-electric efficiency that has been predicted and
demonstrated in these ultra-thin ferromagnet is on the order of ~10 fJ/Vm. This value however is
too small for practical device applications. To put the value into context, an electric field of
~1MV/cm only induces a magnetic anisotropy change of a few percent in thermally stable
ferromagnetic devices. Figure 2.16 shows an example of an ultra-thin film ferromagnet/oxide
magnetoelectric device and its operation.
Figure 2.16. Magneto-electric effect in ultra-thin film ferromagnet/oxide device. (a)
Magneto-optical Kerr effect hysteresis loops of the Fe80Co20 magnetic film in (a) at V = ±200V.
(c) Simulated change in orbital magnetic anisotropy and magnetic anisotropy energy (MAE) as a
function of electric field for a free standing 15 monolayer thick Fe (001) film. Adapted with
permission from reference56,57.
28
Besides magneto-electric effect, another promising route to effective voltage control of
magnetism is through ionic modulation of magnetic interfaces58–65. This approach relies on
voltage-induced ionic migration and electrochemistry to modulate magnetic properties ranging
from magnetic anisotropy to spin-orbit torques. One example of magneto-ionic control of
magnetism is reversible oxidation and reduction of the Co ferromagnet layer in a Pt/Co/GdOx/Au
heterostructure (figure 2.17). The redox of the Co layer is confirmed by electron energy loss
spectroscocpy (EELS) and x-ray magnetic circular dichroism (XMCD) analyses60,61. While the
voltage-induced redox process is originally confined to the electrode edge, further optimizations
enabled uniform redox across the entire device region (figure 2.14). Oxygen-based magneto-
ionics were also demonstrated in Co/SrCoOx66,67, Co/HfOx
68, Co/ZnO69, and CoFeB/MgOx
systems70.
Figure 2.17. Magneto-ionic control of magnetism through voltage-induced redox of Co. (a)
Schematic of a Pt/Co(0.9nm)/GdOx/Au device. (b) EELS spectrum of the normalized O-K edge
count as a function gate voltage (Ubias). (c) Polar magneto-optical Kerr effect hysteresis loop of
the Pt/Co/GdOx/Au device under different gate voltages at 100ºC. Adapted from reference61.
29
Besides oxygen-ion, Group I ions like lithium ions and protons have also been used to induce
changes in magnetic properties67,71–73. For instance, Zhang et al have demonstrated large changes
in magnetic moments by reversible intercalation of lithium ions in spinel structures like
Fe2O3(figure 2.18)73. These experiments are done in a battery-like configuration with the spinel
structure acting as the cathode inside a liquid electrolyte. Similarly, proton-induced modulation
of magnetic properties have also been demonstrated by Nan et al in an acidic solution through
reversible absorption and desorption of hydrogen on an ultra-thin Co film74. More recently, Lu et
al have demonstrated that hydrogen and oxygen-induced phase transformation in SrCoOx can
lead to reversible transition between a paramagnetic, ferromagnetic and antiferromagnetic state
(figure 2.18b)67.
Magneto-ionic gating of magnetism has garnered great interest in recent years due to the
extremely large magneto-electric efficiency that can achieved, which is on the order of
~5000fJ/Vm61. This allows its implementation in devices to be more practical in terms of power
saving. However, there are a few major issues which remain to be addressed. For oxygen-ion
gating, it has been shown that irreversible structural and chemical degradation of the target
ferromagnet always accompany the magnetic property changes62. In addition, while oxidation of
the ferromagnet has indeed been observed through TEM-EELS studies, there is a lack of
understanding on the mechanism of oxidation and the source of oxidant. For lithium ion gating,
the major problem remains the incompatibility of most Group I ions with CMOS processing due
to formation of traps and defects in Si or SiO275. The exception to this is proton, where it is
relatively innocuous in its standard state. While proton gating is promising, it has only been
demonstrated using liquid electrolyte. This thesis henceforth aims to fill the gap in understanding
30
of oxygen-ion gating of magnetic properties, and to demonstrate for the first time proton gating
of magnetic properties in an all-solid-state device.
Figure 2.18: Magneto-ionic control of magnetism using lithium ion and proton. (a)
Reversible change in magnetization of Fe through intercalation of Li in Fe2O3 spinel structure.
(b) Reversible insertion of protons and oxygen ions in SrCoO2.5 to change the magnetization
between antiferromagnetic, paramagnetic, and ferromagnetic states. Adapted with permission
from references67,73.
31
2.4 Solid Oxide Fuel and Electrolyzer cell
The simplest fuel cell produces power from reaction of H2 and O2 to produce H2O, while an
electrolyzer cell produces H2 fuel and O2 from electrolysis of H2O76–78. Both cells share the same
device structure; whether it is a fuel or electrolyzer cell depends on the mode of operation. A
solid oxide cell, unlike conventional liquid electrolyte cells, uses a solid oxide electrolyte such as
Yttria-stabilized Zirconia (YSZ); as a result they are typically operated at high temperature (>
700C). A solid oxide cell can be further classified into a proton conducting oxide cell or an
oxygen-ion conducting oxide cell79,80. The main difference lies in the type of ions conducted
across the oxide; in the former case, it is a proton which gets transported, while in the latter case,
it is an oxygen ion. Figure 2.16 shows schematics of a proton-conducting and oxygen-ion
conducting solid oxide cell respectively run in electrolysis mode. The cells consist of two
electrodes sandwiching a solid oxide electrolyte layer. One electrode is the anode, where
oxidation takes place, while the other is the cathode where reduction takes place. Because the
anode and cathode assignment can change depending on operations modes, the electrodes facing
O2 and H2 are also known as the air and hydrogen electrode respectively79,80.
32
Figure 2.19. Solid oxide cell in electrolysis mode. (a) Proton conducting solid oxide cell. (b)
Oxygen-ion conducting solid oxide cell.
During operation of a proton conducting electrolyzer cell, a positive bias larger than the
thermodynamic potential of water splitting is applied to the air electrode (anode), which splits
water to produce proton and oxygen gas. The proton, 𝐻+ then gets transported across the oxide
due to the applied electric field, where it gets reduced at the hydrogen electrode (cathode) to
produce hydrogen gas. The reactions are shown below:
Proton conducting oxide electrolyzer cell
Anode: 2𝐻2𝑂 → 𝑂2 + 4𝐻+ + 4𝑒− Equation 2.6
Cathode: 4𝐻+ + 4𝑒− → 2𝐻2 Equation 2.7
33
For an oxygen-ion conducting solid oxide electrolyzer cell, the reaction starts at the hydrogen
electrode (cathode), where a negative bias splits water to produce oxygen ions and hydrogen gas.
The oxygen ions are driven by the electric field to the air electrode (anode), where it gets
oxidized to oxygen gas.
Oxygen-ion conducting oxide electrolyzer cell
Cathode: 2𝐻2𝑂 + 4𝑒− → 2𝐻2 + 2𝑂2− Equation 2.8
Anode: 2𝑂2− → 𝑂2 + 4𝑒− Equation 2.9
Equation 2.6 and 2.9 are known as the oxygen evolution reaction (OER) while equation 2.7 and
2.8 are known as the hydrogen evolution reaction (HER) (More discussion in Chapter 2.6). The
operation of a fuel cell is essentially an electrolyzer cell run in reverse mode; for instance, in
figure 2.19(a), hydrogen gas would be oxidized to H+, which gets transported to the cathode
where it recombines with O2 gas to form H2O.
Some examples of well-known proton conducting oxide electrolyte include barium cerate,
BaCeO3 (BCO) and barium zirconate, BaZrO3 (BZO) while examples of oxygen-ion conducting
oxide electrolyte include yttria-stabilized zirconia, YxZr1-xO2 (YSZ) and gadolinium-doped ceria,
Ce1-GdxO2 (GDC) 81–83. Common air electrodes include lanthanum strontium manganite, La1-
xSrxMnO3 (LSMO) while hydrogen electrodes include Ni-ceramic composites such as Ni-
YSZ84,85. A more extensive discussion of solid oxide cell electrodes will be given in Chapter 2.6
to 2.7.
One figure of merit to characterize performance in a solid oxide fuel cell is power density. It is
measured by sourcing current from a fuel cell and measuring the resulting voltage at a specific
temperature and partial pressure of H2 fuel. This is repeated for increasing current values to get
34
the power density curve. Power density depends on a wide range of factors such as overpotential
(discussed below), number of active sites on the electrode catalyst, and gas transport of the
reactants to the active sites. Figure 2.20(a) shows an example of a power density curve for a solid
oxide fuel cell with 0.3mm thick YSZ. An analogous plot can also be generated for an
electrolyzer cell (Figure 2.20(b)); in this case, low voltage and power are desired for operation of
the cell.
Figure 2.20. Performance of a solid oxide fuel cell and electrolyzer cell. (a) Exemplary power
density curve of a fuel cell. (b) Analogous performance plot for an electrolyzer cell. Adapted
with permission from reference78,86
In an electrolyzer cell, the goal is to maximize the rate of H2 production from water splitting at
minimum potential. For a fuel cell, the goal is to achieve highest power output at maximum
potential from hydrogen-oxygen recombination to form water. In both cases, the maximum (fuel
cell) and minimum (electrolyzer cell) achievable voltage is the thermodynamic potential, which
is 1.23V for H2O at standard conditions. To characterize deviations from this thermodynamic
potential, a parameter called overpotential, 𝜂 is used, which essentially tells us how many
35
additional volts above the thermodynamic potential is required to split water in an electolyzer
cell, or how many volts less than the thermodynamic potential that we can extract from water
generation in a fuel cell. Studies of electrolyte and electrode materials are intended to minimize
this overpotential, as it represents a loss. There are in general three types of overpotentials:
activation overpotential, ohmic overpotential, and concentration overpotential. Activation
overpotential arises because, the kinetics of charge transfer at thermodynamic potential is very
slow; as a result an additional voltage is required to drive this reaction in the anodic or cathodic
direction. Ohmic overpotential arises from ohmic loss and is mitigated by reducing the overall
resistance of the cell. Concentration overpotential arises from limited kinetics of the mass
transport of either the reactants to the active sites or products from the active sites. For solid
oxide cells, activation and concentration overpotential are primarily minimized through
optimization of the electrodes. For ohmic overpotential, the overall resistance is reduced through
optimization of the electrodes, electrolyte, and also the electrode/electrolyte interface. The net
voltage, 𝑉 that can be extracted from a fuel cell after accounting for all sources of overpotential
is given in equation 2.10. Similarly, the voltage that is needed in an electrolyzer cell after
accounting for all sources of overpotential is given in equation 2.11. Another term which is
commonly used in place of overpotential is polarization loss.
𝑉 = 𝐸𝑜 − 𝜂𝑜ℎ𝑚𝑖𝑐 − 𝜂𝑎𝑐𝑡𝑖𝑣 − 𝜂𝑐𝑜𝑛𝑐 Equation 2.10
𝑉 = 𝐸𝑜 + 𝜂𝑜ℎ𝑚𝑖𝑐 + 𝜂𝑎𝑐𝑡𝑖𝑣 + 𝜂𝑐𝑜𝑛𝑐 Equation 2.11
Here, 𝜂𝑜ℎ𝑚𝑖𝑐 , 𝜂𝑎𝑐𝑡𝑖𝑣, and 𝜂𝑐𝑜𝑛𝑐 are the ohmic, activation and concentration overpotentials
respectively.
36
2.5 Solid Oxide Proton Conductors
Solid oxide proton conductors have mainly been developed as electrolytes for fuel cell
applications. The main motivation for choosing an electrolyte with the largest proton
conductivity is to reduce the ohmic overpotential and to maximize the power density that can be
extracted from the fuel cell.
Proton conduction can in general be classified into two broad categories: one based on the
Grotthuss mechanism and another based on the vehicle mechanism81,82,87. In Grothuss
mechanism, proton migration is achieved by proton “hopping” between oxygen host lattice sites,
and the net transfer of proton depends crucially on factors like lattice dynamics and distance
between the oxygen host atoms. If proton migration only depends on proton “hopping” between
static host oxygen atoms, the conductivity should increase with decreasing distance between
neighboring sites. However, in many cases, this is completely opposite; structures with larger
oxygen-oxygen distance exhibits larger proton conductivity. The reason for this is the transfer of
proton involves cooperative movement of neighboring oxygen atoms which get closer/further
apart momentarily due to lattice vibrations. This allows proton to break and reform hydrogen
bonds with neighboring atoms, resulting in a net migration of proton. The lattice dynamics of
oxygen atoms is crucial for hydrogen bond breaking; and for short stiff bonds where the oxygen
atoms approximate a static lattice, the proton remains essentially “trapped” in a symmetrical
bond. In general, bulk proton conductivity (non-grain boundary) of solid-oxide proton
conductors can be attributed to the Grotthuss mechanism.
37
Figure 2.21. Coordinates of proton conduction through Grotthuss mechanism. (a)
Conduction through proton “hopping” between static oxygen lattice atoms. (b) Conduction
involving both proton “hopping” and oxygen atom lattice vibration. (c) Conduction involving
only oxygen atom lattice vibration to break hydrogen bonds. Adapted with permission from
reference88.
The second type of proton conduction mechanism is the vehicle mechanism, where proton
migration through the electrolyte is a mediated by a vehicle, such as a H2O molecule. In this
case, the rate of diffusion of the vehicle can play a crucial role in determining the overall proton
conductivity (in addition to the rate hydrogen bond breaking). The vehicle mechanism dominates
in a liquid-like environment where molecules can easily diffuse around, unlike a solid oxide
lattice where the atoms are relatively rigid.
The most studied solid oxide proton conductors are the perovskite-structure oxides such as doped
BaCeO3 and BaZrO379,89–91. Other well-known proton conductors include phosphates like LaPO4,
38
rare earth oxides such as Gd2O3 , orthoniobates and orthotantalates such as LaNbO492–94. A
compilation of their conductivity values are shown in figure 2.22.
Figure 2.22. Proton conductivities of different oxides. Adapted with permission from
reference82.
In almost all proton conducting oxides, acceptor doping is required to generate oxygen
vacancies, and these vacancies in turn absorb water to form hydroxide defect according to the
reaction:
𝐻2𝑂 + 𝑂𝑂𝑥 + 𝑉𝑂
∙∙ → 2𝑂𝐻𝑂∙ Equation 2.12
Here 𝑂𝑂𝑥 represents an oxygen ion on a normal oxygen site, 𝑉𝑂
∙∙ an oxygen vacancy with net
double positive charge relative to the normally occupied lattice site, and 𝑂𝐻𝑂∙ a singly positively
39
charged proton localized around an oxygen ion sitting on a normal oxygen site. These localized
protons form the charge carriers, and because proton conductivity (𝜎) is a function of both the
concentration of carriers (𝑛), charge (𝑞) and mobility (𝜇), according to the reaction 𝜎 = 𝑛𝑞𝜇, the
larger the amount of dissolved water, the larger is the proton conductivity. While large acceptor
doping can increase the concentration of carriers, excess doping will lead to lattice reordering
and strain effects, which in turn reduce the mobility, 𝜇. In addition, not all oxygen vacancies that
are created by doping will react with water to form hydroxide defects according to equation 2.12.
This will depend on the enthalpy of dissolution of water in the oxide. In general, the more
exothermic is the enthalpy, the more water can be dissolved in the oxide. For rare earth oxides,
higher stability of oxide leads to larger exothermic enthalpy and hence dissolution of water in the
oxide according to equation 2.12. This is counterintuitive because it is the least stable rare earth
oxides (La2O3) that is most reactive with water to form hydroxide. On the other hand, for
perovskites such as BaZrO3 and BaCeO3, it is the larger basicity that has been attributed to larger
dissolution of water. In these perovskites, their stability is inversely related to their capacity for
dissolution of water. For instance, in the case of BaCeO3, it decomposes to Ba(OH)2 and CeO2 at
high partial pressure of water95.
40
Figure 2.23. Rotation and proton transfer in a perovskite structure. Adapted with
permission from reference82
For the mobility of proton carriers, the trend is more complicated, as it depends on a series of
steps such as rotation around the oxygen host atom and proton transfer (involving hydrogen-
bond breaking). There are however some general trends: (1) the reduction in symmetry of the
lattice reduces the mobility, such as in SrCeO3, because dissimilar environment for proton
hopping reduces net permutations of ways to get from one site to another96. (2) For rare earth
oxides, mobility decreases as the lattice parameter of the lattice decreases. So far, the best known
solid oxide proton conductors are BaCeO3 and BaZrO3, with proton conductivities between 10-
1S/cm and 10-3S/cm at ~700C. However, as mentioned, BaCeO3 is not stable under high water
partial pressure. BaZrO3 on the other hand, cannot be sintered well which leads to large volume
of grain boundaries and higher resistance. A lot of research effort is hence focused on improving
the stability of BCO and sinterability of BZO through doping and special fabrication techniques.
Besides proton conductivity through grains, there has also been evidence of proton conductivity
through surface and grain boundaries in some nanocrystalline materials. In these systems, there
is water uptake in the grain boundaries and surface, and because the volumetric ratio of both
41
components to grain is very large, this form of proton conduction can dominate97–99. One
disadvantage of such proton conductors is increasing temperature will decrease the proton
conductivity as water vaporizes, as shown for nanocrystalline ceria in figure 2.24(a) and
nanocrystalline YSZ in figure 2.24(b). At higher temperature, the ionic conductivity is
dominated by another charge carrier, such as oxygen vacancies. In addition, the difference in
ionic conductivities in wet and dry atmospheres can be a few orders of magnitude at room
temperature.
Figure 2.24. Proton conductivity of nanocrystalline oxide. (a) Conductivity plot of
nanocrystalline ceria. (b) Conductivity plot of nanocrystalline YSZ. Ionic conduction in both
oxides are dominated by proton conduction through water. Adapter with permission from
references98,99.
For this thesis, there are some differences between the devices studied and majority of solid
oxide cells in literature. While there are many proton conducting oxides that have been
42
investigated at intermediate temperatures (~500C to 700C), almost none has been studied at
room temperature (25C). For voltage-gating of functional interfaces using protons, we are mainly
only interested in operation at low temperatures (<100ºC). From extrapolation of the
intermediate temperature data, we can already observe significantly different trends and
candidates from what were generally known as “good” proton conductors. Another significant
difference between conventional studies and this thesis is the dimension of the device. The
thinnest proton conducting electrolyte that has been studied for fuel and electrolyzer cells is a
few hundred nm; this thickness is constrained by mechanical strength of the electrolyte
membrane when subjected to high partial pressure of reactant gases. However, the thickness of
the proton conductor used in this thesis is down to 4nm, which is >100x smaller. At such
dimension, the electric field is extremely large (1-10MV/cm) and can have drastic, non-linear
effect on the motion of ions, as seen in many memristive devices100. Figure 2.25 shows some
driving forces which may be present under such large fields.
Figure 2.25. Driving forces for ionic migration in memristive devices under large fields. (a)
Driving force due to drift. (b) Driving force due to electromigration (c) Driving force due to
concentration gradient. (d) Driving force due to temperature gradient. Adapted with permission
from reference100.
43
Finally, for voltage gating of functional interfaces using protons, the primary figure of merit for
fast device response is proton mobility rather than proton conductivity as long as there is
sufficient concentration of proton to alter the interfaces. While oxides with large proton
conductivities generally have large proton mobilities, this is not always the case. One example of
this is BaTiO3, where the proton conductivity is very low due to low solubility of hydroxide
defects, but the proton mobility can be very high101.
44
2.6 Water Electro-Catalysis
While electrolyte conducts ions across a cell, water splitting/recombination take place at the
anode and the cathode. In water electro-catalysis, there are four types of reactions: oxygen
evolution reactions (OER), oxygen reduction reactions (ORR), hydrogen evolution reaction
(HER), and hydrogen oxidation reaction (HOR)96,102–109. The first two reactions: the OER and
HER are involved during electrolysis of water (electrolyzer mode) to produce O2 and H2
respectively, while the latter two reactions: the ORR and HOR are involved during consumption
of O2 and H2 to produce H2O (fuel cell mode).
The net ORR and HOR are just the reverse reactions of the OER and HER respectively.
Typically, the oxygen reactions (OER and ORR) constitute the largest source of overpotential
because strong oxygen bonds need to be broken in both processes110–113. This can be understood
by looking at the intermediate species which are produced during the reactions. Rossmeisl et al.
broke down the OER reaction into four distinct elementary steps in an acidic electrolyte111,112,
each involving the transfer of one electron, according to equation 2.13. The elementary steps for
HER is shown in equation 2.14. The corresponding reaction coordinates for OER are shown in
figure 2.26.
2𝐻+ + 2𝑒− → 𝐻+ + 𝐻∗ + 𝑒− Equation 2.14
→ 𝐻2
2𝐻2𝑂 → 𝐻𝑂∗ + 𝐻2𝑂 + 𝐻+ + 𝑒− Equation 2.13
→ 𝑂∗ + 𝐻2𝑂 + 2𝐻+ + 2𝑒−
→ 𝐻𝑂𝑂∗ + 3𝐻+ + 3𝑒−
→ 𝑂2 + 4𝐻+ + 4𝑒−
45
Figure 2.26. Reaction coordinates of the OER in acidic electrolyte solution. Adapted with
permission from reference112.
From the figure, one can see the Gibbs free energy of H2O and O2 species are exactly equal at an
applied bias of 1.23V (the thermodynamic potential), however the intermediate products are at
different energy levels at this bias. Because the reactions happen in series, the kinetics are
essentially limited by the largest energy barrier between the intermediate products. The reaction
kinetic is then quantified by the current, 𝑖 according to the equation:
𝑖 = 𝑖𝑘exp (−∆𝐺𝑟
𝑘𝑇) Equation 2.15
Where 𝑖𝑘 is a constant, ∆𝐺𝑟 is the Gibbs free energy barrier of the rate limiting step, k is the
Boltzmann constant, and T is the absolute temperature. An overpotential, or more specifically
the activation overpotential, hence serves to reduce the ∆𝐺𝑟 of the rate-limiting step.
For water electro-catalysis, the best catalysts usually neither have the highest nor the lowest
binding energy to the reactants, but rather have intermediate values, according to the Sabatier’s
principle. As a result, when one plots the catalytic activity against the reactant binding energy,
46
one typically gets a “volcano” plot. Some of the plots for metal and binary oxide OER catalysts
are shown in figure 2.27. Generally speaking, this “volvano” trend exists because too weak of
binding energy leads to low conversion rate of reactants to intermediates, whereas too strong of
binding energy leads to low conversion rate of intermediates to final products. The ideal catalyst
is one where the energy splitting between all the intermediates are the same, as shown in figure
2.24(c) for the case of an OER catalyst. Pt and RuOx are the best metal and oxide (non-
perovskite) OER catalysts because their binding energies to the intermediates are closest to the
ideal catalyst (equi-energy splitting). Note that because ORR is just the reverse of OER, one
should expect the best OER catalyst to also be the best ORR catalyst.
Figure 2.27. Volcano plots. Volcano plots for (a) metallic and (b) binary oxide OER catalysts.
(c) Energy profile of an ideal OER catalyst. Adapted with permission from references104,113.
47
2.7 Electrodes for Solid Oxide Cells
So far we have focused on the electronic structure of catalyst which primarily affects charge
transfer at the electrode/electrolyte interface. However, mass transport can also be a significant
source of overpotential; in fact it often dominates in systems like solid oxide cell where gas
phase diffusion and ionic transfer across the electrode/electrolyte interface are involved114,115.
Fundamentally, most losses in solid oxide cell electrodes boil down to the fact that a single phase
cannot conduct all three species involved in a gas phase reaction: an electron, an ion, and a gas
molecule. Only an electronically conductive material can transport electrons, only an ionically
conductive material can transport ions, and only a gas phase can transport a gas molecule. As a
result, a complete reaction can only occur at the intersection of these three phases; this
intersection is called a triple phase boundary (TPB). Figure 2.28 shows an example of a TPB at
the cathode of an oxygen-ion conducting oxide fuel cell.
Figure 2.28. Triple phase boundary. Schematic illustration of the triple phase boundary at the
cathode of an oxygen-ion conducting solid oxide fuel cell. Adapted with permission from
reference114.
48
The necessity of three phases for a complete reaction not only reduces the total number of
reaction sites, it also increases the overall complexity due to added intermediate steps and
species. For instance, it has been proposed that the oxygen reduction reaction happening at the
cathode of a proton conducting oxide fuel cell consists of eight elementary steps, each with its
own reaction order resulting in a net reaction of 𝑂2 + 4𝐻+ + 4𝑒− → 2𝐻2𝑂 (table 2.1). Because
these steps happen in series, the slowest step will be the rate-limiting step and will determine the
total resistance from the electrode115. For electrodes in solid oxide cells, they are mainly
characterized by their area specific resistance (ASR), with 𝑅𝑒𝑙𝑒𝑐𝑡𝑟𝑜𝑑𝑒 =𝐴𝑆𝑅
𝐴𝑟𝑒𝑎 . Here, 𝑅𝑒𝑙𝑒𝑐𝑡𝑟𝑜𝑑𝑒 is
the total electrode resistance. ASR is not normalized to the thickness of the electrode because in
most cases, the resistance of the electrode does not scale with its thickness; it is a function of
extrinsic properties such as microstructure and porosity.
Table 2.1. Elementary reaction steps of ORR at the cathode of a proton conducting solid oxide
fuel cell. Reproduced with permission from reference115.
In the past, to increase the area of the triple phase boundary, porous Pt was commonly used as
both the anodes and cathodes. However due to the prohibitive cost of Pt and the limited added
TPB from porosity alone, new methods have been employed. These methods have mainly
revolved around using materials with mixed ionic and electronic conductivities (MIEC). By
49
using MIEC, one only needs the intersection of two phases for reactions to take place because
one of the phases conduct both electrons and ions. As a result, the overall activity increases due
to larger area where electrons, ions, and gas molecules can coexist. MIEC can exist as a one-
phase material such as (La,Sr)(Co,Fe)O3-δ (LSCF) or two-phase material consisting of a
composite of electronic conductor and an ionic conductor, such as Ni-YSZ and LSM-
YSZ84,85,116,117. Schematics of operation of an MIEC cathode in a proton conducting fuel cell is
shown in figure 2.29.
LSCF, a one-phase MIEC, is a perovskite oxide, where the La and Sr atoms are at the A-site, and
the Co and Fe atoms are at the B-site. Sr doping induces a change in oxidation state of both the
Co and Fe from +3 to +4 resulting in p-type electronic conductivity. Simultaneously, there is also
charge compensation through formation of oxygen vacancies, resulting in increased oxygen-ion
conductivity. As a result, LSCF has both electronic and ionic conductivities at high temperature.
It is most commonly used with a ceria electrolyte. LSM ( La1-xSrxMn O3-δ ), which is typically
the electronically conductive component of a two-phase MIEC, is a perovskite where the A-site
La3+ is doped with Sr with an oxidation state of +2. This acceptor doping is mainly compensated
by change in oxidation state of the Mn from +3 to +4 which results in p-type electronic
conductivity. However, in this case, there is minimal change in the oxygen vacancy
concentration. LSM is usually mixed with YSZ, (oxygen ion conductor) to form a two-phase
composite MIEC structure. This cathode is most commonly used on a YSZ electrolyte due to the
excellent match in thermal expansion coefficients. For the anode, the most commonly used
material is Ni-YSZ cermet (cermet is short for ceramic-metal), which is a two-phase composite
MIEC. Ni is used due to three reasons. (1) it is one of the best HOR catalyst, (2) it is cheap, and
(3) it provides mechanical support to the fuel cell, especially in cases where the electrolyte is
50
very thin. Other examples of anode include Ni-SDC (Samarium-doped ceria, SmxCe1-xO2-δ) and
Ni-GDC (Gadolinium-doped ceria, GdxCe1-xO2-δ). While the primary ionic carrier in most MIEC
is oxygen-ion, there are also MIEC such as BaCe1-xFexO3 whose main ionic conducting species
is the proton. These proton-based MIEC are typically doped BCO (BaCeO2-δ) and BZO (BaZrO2-
δ)80,115.
Figure 2.29. Active sites in different types of cathodes for a proton conducting oxide fuel
cell. (a) Porous single phase electronic conductor such as Pt. (b) Two phase mixed-electronic-
ionic conductor (MIEC). (c) Single phase MIEC such as BaCe1-xFexO3.
51
2.8 Electrochemical Impedance
Spectroscopy
In solid state ionics, one of the most used techniques to characterize ionic conductivity is
electrochemical impedance spectroscopy (EIS). It is a simple technique where a small sinusoidal
AC voltage (~10mV to 100mV) is applied to the sample and the resulting impedance is
measured. To get a full impedance spectrum, the frequency of the AC voltage is swept (~0.1Hz
to 1MHz) to probe different processes which contribute to the total conductivity. The spectrum is
then fitted with one or more RC components (figure 2.30) to get the corresponding resistance and
capacitance associated with each process118–121.
Figure 2.30. EIS spectrum and its equivalent circuits. Adapted with permission from
reference121.
1
𝑍𝑎=
1
𝑅𝑎+
1
𝐶𝑎𝑗𝜔 Equation 2.16
𝑍𝑡𝑜𝑡𝑎𝑙 = 𝑍𝑎 + 𝑍𝑏 + 𝑍𝑐 Equation 2.17
52
In real systems, the capacitor is usually replaced with a constant phase element (CPE) to model
non-ideal capacitor behavior. In this case, the capacitance of CPE can be calculated according to
equation 2.19.
𝑍𝐶𝑃𝐸 =1
𝑄(𝑗𝜔)𝑛 Equation 2.18
𝐶𝐶𝑃𝐸 = (𝑅1−𝑛𝑄)𝑛 Equation 2.19
Here, Q and n are the CPE constants. In solid oxide electrolyte, a “brick layer” model is typically
used to represent a polycrystalline material with grain boundaries (figure 2.31)122. Due to the
presence of ionic conduction in both the grains and grain boundaries, the impedance spectrum
for the oxide electrolyte is fitted with two RC circuits. Figure 2.32 shows an example spectrum
for YSZ where the grain and grain boundary contributions to ionic conduction can be clearly
resolved. The third semicircle corresponds to reactions at the electrodes. In this case, the figure
also shows increasing grain boundary resistance with decreasing grain size.
Figure 2.31. “Brick layer” model for polycrystalline solid oxide. Grains have size L and grain
boundaries have width 2b. Reproduced with permission from reference123.
53
Figure 2.32. Impedance spectra of YSZ. There are 3 RC components corresponding to oxygen-
ion conduction in the bulk (grain) and grain boundaries, and the reactions at the electrode. The
data shows increasing resistance of the grain boundary with decreasing grain size, dg.
Reproduced with permission from reference123.
54
Chapter 3:
Experimental Methods
55
3.1 Sputter Deposition
Sputtering is a physical vapor deposition technique where a target material is bombarded with
energetic particles from a plasma which results in ejection of the material from the target surface
and its eventual deposition on a substrate124. Figure 3.1 shows a schematic of the sputtering
process. Sputtering is done at high or ultra-high vacuum (<10-5Torr) to sustain the sputtering
plasma and to ensure high purity of deposited films. Ar gas is usually used as the sputtering gas.
Sputtering is a powerful technique for both laboratory and industrial scale production because
the deposited thin films can be made uniform and smooth across a large area. In addition, a large
variety of materials ranging from metals to metal oxides can be deposited.
Figure 3.1. Sputtering of Au on a substrate using Ar as the sputtering gas.
In sputtering, microstructure control is one of the most important things to consider. A very
useful guide for this is the zone structure model, which gives a qualitative description of the
expected microstructure for different sputtering parameters125. Figure 3.2(a) shows a structure
zone model developed by Thornton et al to describe the thin film microstructure as a function of
the substrate temperature, TS and Ar sputtering gas pressure126. For generality, TS is usually
56
normalized by the melting temperature of the target material, TM and the ratio is called the
homologous temperature. One can divide the different microstructures into three zones: Zone I,
Zone II, and Zone III. A fourth zone, Zone T is latter added between Zone I and II. Qualitatively,
as one goes from zone I to zone III, the kinetic energy of the sputtered atoms increase. As a
result, the atoms have higher rate of diffusion and can coalesce to form larger grains. In zone I,
the thin film is fine-grained and porous, in zone II, the thin film is columnar, and in zone III, the
thin film has large equiaxed grains. An example cross section of the different zones (including
zone T) is also shown in figure 3.2(b)127.
Figure 3.2. Structure zone model. (a) Structure zone model for different argon sputtering
pressure and substrate temperature. (b) Cross sections of microstructures in different zones.
Adapted with permission from reference126,127.
Besides the substrate temperature and sputtering gas pressure, other factors that affect the
microstructure of sputtered films include background pressure and the impurity level. If one uses
oxygen for sputtering, partial oxygen pressure can also affect the overall grain size, as shown in
figure 3.3(e) and (f) for the case of ITO128,129.
57
Figure 3.3. Microstructure as a function of oxygen partial pressure. (a)-(b) SEM images of
ITO grains deposited at 4x10-5mbarr (e) and 4.75x10-5mbarr (f) of oxygen partial pressure.
Adapted with permission from reference128.
In this thesis, all the samples are fabricated using sputtering system shown in figure 3.4. The
base pressure of the chamber reach down to 10-7Torr using a Pfeiffer turbomolecular pump
backed by an Alcatel 2008A roughing pump. The system consists of four sputtering sources
which are mounted on the floor of the sputter chamber. Each sputtering source has its own
chimney and shutter for better control of the sputtering area and timing. For depositing metallic
layers, metal targets are sputtered using DC power sources. For depositing oxide layers, there are
two approaches: (1) a metal target is reactively sputtered using a DC power source under a
partial pressure of oxygen, or (2) an oxide target is directly RF sputtered (also under oxygen
partial pressure to control microstructure of the film). The argon sputtering gas pressure is
between 2-5mTorr in all cases, whereas the oxygen partial pressure can range between
0.01mTorr to 0.9mTorr depending on the materials.
In the sputtering chamber, there is a rotating substrate table with slots for four substrate holders,
and a masking table with slots for 6 masks. The masking table allows one to sputter selectively
on different substrate slots. Typically, sputtering is done while rotating the substrate and mask
58
table in order to achieve better uniformity. The rotation rate of the tables is ~36/min. All
subtrates are kept at room temperature during sputtering.
Figure 3.4. Schematic of sputtering system. (a) Substrate table. (b) Substrate holder slot. (c)
Liquid nitrogen reservoir. (d) Gears for rotating the substrate table. (e) Mask table. (f) Mask slot.
(g) Pins to align substrate and mask tables. (h) Chamber floor (i) Jack and bellows below the
chamber floor. (j) Chimney for sputter source. Adapted from reference130
59
Figure 3.5. Image of substrate holder.
The thicknesses of the sputtered layers are calibrated using x-ray reflectivity. For metallic layers,
the deposition rates are between 1 to 5nm/min at 0.4A of sputtering current. The rates are very
similar for oxide layers deposited using reactive sputtering of a metal target at low oxygen partial
pressure (<0.1mT). For reactive sputtering at high oxygen partial pressure, target poisoning takes
place where the surface of the metal target becomes oxidized before it is sputtered. As a result,
the deposition rate goes down significantly (~1 order of magnitude). Figure 3.6 shows the
deposition rate of a Gd target before and after target poisoning. For RF sputtering of oxide
layers, the deposition rates are typically <0.5nm/min at 100W of sputtering power. The rates
quoted are all for rotating sputter deposition; for stationary deposition the rate can be up to 7
times larger but the uniformity is poorer. Stationary sputtering is only done for contact pads or
the top electrodes.
60
Figure 3.6. Deposition rates. (a)Deposition rates of an S-gun Gd target at 0.4A current, 3mTorr
of Ar sputtering gas, and at different O2 partial pressure, PO2. (b) Oxygen flow rates required to
achieve PO2.
61
3.2 Sample Structure and Patterning
In this thesis, the general sample structure used for the magneto-ionic devices is Ta/Pt/Co/MOx
/Au , where M is Gd, Y, Zr, Ta or Mg. Ta is the adhesion layer, Pt is the bottom electrode, Co is
the magnetic layer, MOx is the ionic-conducting oxide, and Au is the top electrode. Another
magneto-ionic device structure is Ta/Pd/Co/Pd/MOx/Au where the Ta/Pd/Co/Pd form the bottom
electrode layers and Au is the top electrode. All magneto-ionic devices are sputter deposited on
p-doped Si substrate with 50nm of thermal oxide.
To probe dynamic modulation of magnetic properties in these structures, magneto-optical Kerr
effect (MOKE) polarimetry (Chapter 3.3) and Hall magnetometry (Chapter 3.4) were used. For
devices used in the MOKE experiments, the Ta/Pt/Co/MOx layers (or Ta/Pd/Co/Pd/MOx layers)
were deposited on Si substrate as continuous films, with the bottom metallic layers uncovered to
provide electrical connection to the ground (figure 3.7). 200µm diameter top Au electrodes were
then patterned on these continuous films. In total, three sputtering runs with two vacuum breaks
are required to make the gated MOKE devices (ie Ta/Pt, then Co/MOx, then Au). In cases where
the top Au electrodes need to be deposited in situ directly after the MOx layer, an in situ mask
aligner is used to pattern the Au electrodes.
62
Figure 3.7. Sample structure for MOKE magneto-ionic device. (a) Schematic of a magneto-
ionic device used for time-resolved MOKE measurement during voltage gating. (b) Layout of the
mask used for patterning the top Au electrodes. The diameter of the electrodes is 200µm.
For the devices used in Hall measurement, the Ta/Pt/Co/MOx layers are patterned into Hall bar
geometry as shown in figure 3.8. A thicker MOx layer is then deposited in the form of a
rectangular patch with its area larger than the active region to ensure no parasitic current leakage
path between the top and bottom electrodes. The top Au is patterned into a square, where there is
a strip protruding out at 45º to both the Hall bar arms to allow easy probe access to the top Au
gate. In total, four sputtering runs with three vacuum breaks are required to make the gated Hall
devices.
63
Figure 3.8. Sample structure for gated hall bar device. (a) Schematic of a gated hall bar used
for time-resolved Hall measurement during voltage gating. (b) Detailed layouts of the four masks
used to pattern the complete device. The active region is 500µm x 500µm.
For the reversible solid oxide cells studied in this thesis, Ta/Pt/GdOx/Au structure was used. In
this case, there is no magnetic layer and the GdOx serves as the charge storage layer. The
structure has a cross bar geometry with the bottom Ta/Pt bottom electrode on the horizontal arm
and the Au top electrode on the vertical arm. The GdOx is deposited as a rectangular patch
between the Ta/Pt and the Au layers. A simplified device schematic is shown in figure 3.9(a) and
the mask designs are shown in figure 3.9(b). In total, four sputtering runs with three vacuum
breaks are required to make the reversible solid oxide cells.
64
Figure 3.9. Sample structure for cross bars. (a) Schematic of cross bar used for reversible
solid oxide fuel cells. (b) Detailed layouts of the four masks used to pattern the complete device.
The active areas can range between ~10-3cm2 and 4x10-1cm2 depending on the width of the arms.
For the gated spin-torque ferromagnetic resonance (ST-FMR) device used in chapter 6, device
layout as shown in figure 3.10 was used. The overall device consists of a Ta/Pt/Co/GdOx FMR
planar waveguide on top of which we deposit an overlayer of GdOx which covers the entire
active region. The active region is then gated by a top Au electrode which is connected by an
extended arm to a contact pad. The waveguides used have dimensions of 10 µm x5µm, 10 µm
x10µm, 20 µm x5µm and 20 µm x10µm. The mask layout is also included in figure 3.10.
65
Figure 3.10. Sample structure for gated ST-FMR device. (a) Schematic of gated ST-FMR
device. (b) Detailed layouts of the four masks used to pattern a complete device.
For patterning most devices, shadow mask lithography is the primary technique used. For this, a
flexible 0.01” thick PEEK sheet serves as the mask and a laser cutter is used to define features
down to ~10µm (figure 3.11). During sample preparation for sputtering, the PEEK sheets are
pressed tight again the Si substrates and taped using a Cu tape to minimize any shadowing effect.
For patterning sub-micron to micron scale structures such as the gated ST-FMR devices, optical
lithography was used. Si substrates were coated with positive photoresist, Megaposit SPR700,
exposed, baked, and developed with Microposit MF-CD26 developer. This is repeated n number
of times where n is the number of mask layers.
66
Figure 3.11. Shadow mask lithography. (a) Image of a PEEK mask and top electrode patterns
defined using a laser cutter. (b) Backside of the PEEK mask where a substrate is attached tightly
using Cu tape.
67
3.3 Magneto-Optical Kerr Effect
Magneto-optical Kerr effect (MOKE) polarimetry is a powerful technique to detect
magnetization in ultra-thin films. With a magnetic field, it allows one to obtain magnetic
hysteresis loops. The basic idea behind MOKE is a linearly polarized light which is incident on a
magnetic sample is rotated by an angle, 𝜃𝑆 (Kerr rotation) and gains a slight ellipticity, 𝜖𝑆 (Kerr
ellipticity) (figure 3.11)131,132.
Figure 3.12. Magneto-optical Kerr effect (MOKE). Adapted from reference131.
A MOKE setup for detecting magnetization consists of a laser source, an incident polarizer, an
analyzer, and a photodiode detector. The analyzer is a polarizer which is set at (90+ δ)º to the
incident polarizer. The setup is shown in figure 3.13. The incident light first passes through the
incident polarizer and becomes linearly polarized. When the polarized light is reflected off a
magnetic surface, its polarization is rotated by 𝜃𝑆 . The photodiode detects this rotation as a
change in intensity, 𝐼 which can be expressed as equation 3.1.
68
𝐼 = 𝐼𝑜(1 +𝜃𝑆
𝛿) Equation 3.1
If one inserts a quarter wave plate before the analyzer, the detected intensity becomes
proportional to 𝜖𝑆 (Kerr ellipticity), which can be expressed as equation 3.2:
𝐼 = 𝐼𝑜(1 +𝜖𝑆
𝛿) Equation 3.2
Figure 3.13. MOKE configurations. (a) MOKE setup consisting of a laser source, incident
polarizer, analyzer, and a photodiode detector. (b) Polar MOKE configuration for detecting out-
of-plane magnetization. (c) Longitudinal MOKE configuration for detecting in-plane
magnetization which is parallel to the incidence plane. (d) Transverse MOKE configuration for
detecting in-plane magnetization which is perpendicular to the incidence plane. Adapted from
reference133.
There are three basic types of MOKE configurations for detecting magnetization in thin films: a
polar MOKE, a transverse MOKE, and a longitudinal MOKE. Their setups are shown in figure
3.13. A polar MOKE is used to detect magnetization which is out-of-plane, a longitudinal
MOKE is used to detect magnetization which is in-plane and parallel to the incidence plane,
69
while a transverse MOKE is used to detect magnetization which is in-plane but perpendicular to
the incidence plane.
For this thesis, we primarily use the polar MOKE configuration to measure out-of-plane
magnetic hysteresis loops. A simplified measurement setup is shown in figure 3.14. A laser
source with wavelength of ~655nm is used, and the laser spot which is incident on the sample
can be focused down to <10µm using an objective lens. A CCD aligned along the vertical axis
allows us to image the sample. A sample stage with an out-of-plane magnetic coil is set up which
can produce 1000Oe of out-of-plane field. To obtain MOKE hysteresis loops, we sweep the
magnetic field and measure the reflected intensity from a thin film sample. The amplitude of this
MOKE hysteresis loop is proportional to the out-of-plane magnetization (MZ) of the sample
(figure 3.15).
Figure 3.14. Polar MOKE setup with electrical probes and out-of-plane field for time-resolved
MOKE measurement during voltage gating.
70
For time-resolved measurements during voltage gating, a series of MOKE magnetic hysteresis
loops are acquired continually, and parameters like coercivities (HC) and MZ can be plotted as a
function of time (figure 3.15 and 3.16). For voltage gating, mechanically compliant CuBe probes
with radius of 25µm were used. One probe is landed on the top electrode to apply a gate voltage,
while a second probe is landed on the uncovered bottom electrode to make the ground
connection. The laser spot is focused on the middle of the top electrode during measurements
(figure 3.16).
Figure 3.15. Time resolved MOKE. (a) Magnetic field profile and measured Kerr signal
(intensity) as a function of time. Both values are continuously measured and MOKE hysteresis
loops are generated for each time period (1s). (b) Exemplary MOKE hysteresis loop generated
from time period 1.
71
Figure 3.16. Time series data. (a) Time series of MZ and HC extracted from MOKE hysteresis
loops during each cycle. (b) Optical micrograph of an Au top electrode with a CuBe probe
landed. Also shown is the focused laser spot.
72
3.4 Anomalous and Planar Hall Effect
In ordinary Hall effect, a potential difference (known as Hall voltage, VHall) develops in the
transverse direction when an electrical current flows along a strip of conductor under an applied
perpendicular magnetic field. The ordinary Hall voltage (𝑉𝑂𝐻𝐸)is generated due to the Lorentz
force from the magnetic field, and is hence proportional to the product 𝐼 × �⃗⃗� where 𝐼 is the
current vector and �⃗⃗� is the field vector.
Figure 3.17: Hall effects. (a) Coordinate system. (b) Ordinary Hall effect. (c) Anomalous Hall
effect. (d) Planar Hall effect. In all three cases, the current flow is in the +x direction and the
sample plane is the x-y plane. “+” and “-“ signs represent polarity of measured Hall voltage.
Adapted from reference134.
73
The ordinary Hall effect however, does not allow one to detect magnetization because the
measured Hall voltage is proportional to the applied magnetic field, not the magnetization of the
sample. To detect magnetization in a thin film sample, we instead rely on the anomalous and
planar Hall effect (figure 3.17)134,135. The anomalous Hall effect is used to probe an out-of-plane
magnetized sample because its Hall resistance (𝑅𝐴𝐻𝐸) is proportional to the out-of-plane
magnetization (𝑀𝑍). The Hall resistance is simply the ratio of transverse voltage to the current in
the longitudinal direction. The planar Hall effect is used to probe an in-plane magnetized sample
because its Hall resistance (𝑅𝑃𝐻𝐸)is proportional to the product of the magnetization projections
along the two in-plane orthogonal directions (𝑀𝑥and 𝑀𝑌) . Using the convention in figure 3.17,
𝑅𝐴𝐻𝐸 and 𝑅𝑃𝐻𝐸 are proportional to 𝑀𝑍and 𝑀𝑥𝑀𝑌 respectively, and their sums can be expressed
more generally as
𝑅𝐻𝑎𝑙𝑙 = 𝑅𝐴𝐻𝐸 + 𝑅𝑃𝐻𝐸
= 𝑅𝐴𝐻𝐸° 𝑀𝑐𝑜𝑠𝜃 + 𝑅𝑃𝐻𝐸
° 𝑀𝑠𝑖𝑛2𝜃𝑠𝑖𝑛2𝜑 Equation 3.3
Here 𝑅𝐴𝐻𝐸° and 𝑅𝑃𝐻𝐸
° are constants which depend on the dimensions of the sample and the
intrinsic material properties (Berry curvature etc). As mentioned earlier, if we have an out-of-
plane sample, the second term vanishes and 𝑅𝐻𝑎𝑙𝑙 = 𝑅𝐴𝐻𝐸 , whereas if we have an in-plane
magnetized sample, the first term vanishes and 𝑅𝐻𝑎𝑙𝑙 = 𝑅𝑃𝐻𝐸. We have ignored 𝑅𝑂𝐻𝐸 in the sum
for 𝑅𝐻𝑎𝑙𝑙. Hence, to measure the magnetizations, we just have to source a longitudinal current (𝐼)
and measure the resulting Hall voltage, 𝑉𝐻𝑎𝑙𝑙.
𝑉𝐻𝑎𝑙𝑙 = 𝐼𝑅𝐻𝑎𝑙𝑙 Equation 3.4
74
3.5 Time Resolved Hall Magnetometry
under different Atmospheric Conditions
Anomalous and planar Hall effects can be used to measure magnetic hysteresis loops of thin film
samples (figure 3.17). In this thesis, the anomalous and planar Hall measurement system is
shown in figure 3.18. A lock-in amplifier is used to both source the longitudinal current and
measure the transverse voltage. A lock-in frequency of 2kHz is typically used with an integration
time of 30ms while the source current used depends on the dimension and material of the
samples. An out-of-plane magnetic coil provides out-of-plane field for measurement of
anomalous Hall hysteresis loops while an in-plane magnetic coil set at 45º to the current injection
direction provides an in-plane field for obtaining planar Hall hysteresis loops. Similar to time
resolved MOKE, anomalous and planar Hall magnetic hysteresis loops are alternately acquired
continually during voltage gating experiment. This allows us to probe both the out-of-plane and
in-plane magnetizations in real time under a gate bias. A Keithley sourcemeter unit was used to
provide the gate voltage.
Figure 3.18. Hall hysteresis loops. (a) Anomalous Hall hysteresis loop. (b) Planar Hall
hysteresis loop.
75
Figure 3.19. Hall measurement system. (a) Top-down and (b) side views of the anomalous and
planar Hall measurement system.
The Hall measurement system is set up in a Lakeshore CPX-VF probe station which allows us to
probe magnetization in situ when applying a top gate voltage under different atmospheric
conditions. Hall devices can be subjected to vacuum condition down to 10-4mbarr using a
turbomolecular pump. For dry conditions, gases such as O2 and N2 (purity of >99%) can be
introduced directly into the chamber using a venting valve. For wet conditions, gases are bubbled
through water in a Fisher Scientific bubbler before being introduced into the chamber.
76
Figure 3.20. Schematic of the CPX-VF probe station with Hall measurement system and
atmospheric control.
77
3.6 Spin-torque Ferromagnetic
Resonance
For spin-torque ferromagnetic resonance (ST-FMR) experiments in chapter 6, Ta/Pt/Co/GdOx
structure was patterned into ST-FMR planar waveguide with a top GdOx/Au gate as shown in
figure 3.1039,136. A coaxial GS microwave probe (GGB Industries Picoprobe 40A series) with
pitch of 150µm to 250µm was used for electrical contact (one tip was landed on A, another on B
as shown in figure 3.21) while a signal generator was used to inject microwave frequency power
between 5.36GHz to 12 GHz into the planar waveguide. This microwave power acts a source of
current-induced torque which drives ferromagnetic resonance in the planar waveguide. For most
measurements, the power output of the signal generator was set to 15dbm. The output of the
signal generator was modulated by a lock-in amplifier in order to allow low frequency detection
of the mixing voltage (Vmix) which results from the change in DC resistance. Since the technique
involves signal input and detection at high and low frequencies respectively, a bias-tee was used
to separate the two components. To apply a gate voltage, a Keithley 2400 sourcemeter and CuBe
DC probe were used.
In all measurements, an in-plane magnetic field, 𝐻 is applied at 45° to the waveguide and the
measurement protocol involves sweeping 𝐻 while measuring 𝑉𝑚𝑖𝑥 at a fixed frequency, f. The
𝑉𝑚𝑖𝑥 can then be fitted according to equation below:
𝑉𝑚𝑖𝑥 = 𝑆𝑊2
(𝐻− 𝐻𝐹𝑀𝑅)2+𝑊2 + 𝐴𝑊(𝐻− 𝐻𝐹𝑀𝑅)
(𝐻− 𝐻𝐹𝑀𝑅)2+𝑊2 Equation 3.5
78
Where 𝑆 and 𝐴 are the amplitudes of the symmetric and antisymmetric Lorentzian components
of the resonance peak respectively, and 𝑊 and 𝐻𝐹𝑀𝑅 are the width and position of both these
Lorentzian components. Figure 3.22 shows an example of a ST-FMR spectrum measured at
8GHz.
Figure 3.21. Spin-torque ferromagnetic resonance system to probe voltage gating of magnetic
properties.
Figure 3.22. ST-FMR spectrum of a Pt(3nm)/Co(6nm)/GdOx(30nm) device at f=8GHz.
79
3.7 Solid Oxide Cell Characterization
For studying the performance of a reversible solid oxide fuel cell, the main metrics are the
capacity, power density and cyclability. For this, the first plot that is usually generated is the
discharge curve, which is a plot of cell voltage versus total charge. The measurement is
performed by repeatedly sourcing a constant current from the cell and measuring the resulting
voltage. When the cell voltage drops to a threshold value, the measurement is stopped, and the
capacity of the cell is given by the total integrated charge. The discharge curve provides
qualitative information regarding chemical reaction steps and the cell overpotential, besides the
capacity. The second plot which is generated is the power density curve. The measurement is
very similar to the discharge curve; however in this case, the sourced current is gradually ramped
up while measuring the voltage. And instead of plotting the cell voltage, it is the power at each
source current that is plotted; the peak power density from the cell can then be obtained from the
curve. To study cyclability, the cell is repeatedly charged and discharged, and the capacity
during each cycle is plotted as a function of cycle number. Exemplary plots of discharge curve,
power density curve, and cyclability are shown in figure 3.23.
80
Figure 3.23. Performance plots of a reversible solid oxide fuel cell. (a) Discharge curves at
different discharge currents. (b) Power density curve. (c) Plot of charge versus cycle number to
study the cyclability of a device.
All the measurements are done using a Keithley 6430 source meter in a Lakeshore CPX-VF
probe station. The cells studied have a cross bar geometry where one arm is connected to the top
electrode and the other arm is connected to the bottom electrode. Active areas range from ~10-
3cm2 to 0.4cm2. For charging and discharging cells under different atmospheric conditions,
different gases can be introduced into the chamber as described in the previous section. In this
case, all the magnets are removed for measurement.
81
Chapter 4: Effect of
H2O on Voltage-
induced Co Oxidation
in a Pt/Co/GdOx
Heterostructure
82
Magneto-ionic effect in a Pt/Co/GdOx/Au system was first observed when a negative gate
voltage (VG) applied to the top Au causes a significant reduction in the magnetic anisotropy at
the electrode edge, which traps domain wall propagation58. Because the electrode edge is the
triple phase boundary where O2 gas, Au and O2- ions meet, the authors proposed that the
magnetic anisotropy is reduced due to voltage-driven O2- motion from GdOx into the Co layer
under a negative gate bias. By then optimizing the geometry of the devices, the authors were
subsequently able to gate the magnetic anisotropy across the entire electrode region. EELS
studies subsequently confirmed that a negative gate voltage does indeed oxidize the Co layer in
the Pt/Co/GdOx device61. However, voltage-driven O2- transport from GdOx to Co layer has still
not been directly observed eventhough it was generally assumed to be the operative mechanism
in voltage-induced Co oxidation.
While O2- can be the active oxidant, it is also well-known that many oxides readily absorb
water from the atmosphere, incorporated as hydroxide defects situated at oxygen ion vacancies
through the following defect reaction:137
𝐻2𝑂 + 𝑂𝑂𝑥 + 𝑉𝑂
∙∙ → 2𝑂𝐻𝑂∙ Equation 4.1
Here 𝑂𝑂𝑥 represents an oxygen ion on a normal oxygen site, 𝑉𝑂
∙∙ an oxygen vacancy with net
double positive charge relative to the normally occupied lattice site, and 𝑂𝐻𝑂∙ a singly positively
charged proton localized around an oxygen ion sitting on a normal oxygen site. For memristors,
it is already well established that humidity can alter the resistive switching behavior of oxides
such as TaOx138,139, HfOx
139, SiOx140 and SrTiOx
141,142 memristive cells due to its effect on bulk
oxide properties and interface reactions at the anode and cathode. The high basicity of rare earth
oxides makes them particularly hygroscopic143, and Gd2O3 is known to react with moisture to
83
form hydroxides144–148 such as Gd(OH)3 with consequent changes to electrical147 and ionic
properties92.
In this chapter, we show that it is H2O stored in GdOx as Gd(OH)3 that oxidizes Co under a
negative gate bias, and that oxygen migration plays an insignificant role. We further show that
hydrogen-induced CoO reduction leads to water uptake back into the GdOx matrix, allowing for
closed-system electrochemical and magnetic property switching without the need for
atmospheric exchange. These results provide a mechanistic understanding of magneto-ionic
switching in metal/oxide heterostructures and essential insights to enable magneto-ionic device
engineering.
84
4.1: Experimental Methods
Sample preparation. Ta(4 nm)/Pt(3 nm)/Co(0.9 nm)/GdOx(tGdOx nm) films were fabricated on
thermally oxidized Si (100) substrates using magnetron sputtering at room temperature and
3mTorr Ar pressure. The metal layers were grown by DC sputtering. All GdOx layers were
deposited using DC reactive sputtering with PO2 of 0.07mTorr except for the XAS samples,
where the deposition was done using RF sputtering with PO2 of 0.7mTorr O2. For MOKE
measurements, 200 μm diameter Au(3 nm) electrodes were patterned on top of the
Ta/Pt/Co/GdOx continuous film, with the Ta (4 nm)/Pt(3 nm) underlayer was uncovered by
GdOx at the sample edge to allow electrical contact to the back. For in situ XAS measurements,
the Ta/Pt/Co/GdOx/Au structure was patterned into a cross-bar geometry with 1mm arm width.
X-ray reflectivity (XRR) measurements. XRR was carried out using a Bruker D8 Discover
HRXRD instrument with Cu K-α radiation at wavelength of 1.54Å.
X-ray Photoelectron Spectroscopy (XPS) measurements. XPS was carried out using a
Physical Electronics Versaprobe II X-ray Photoelectron Spectrometer at a base pressure of 5x10-
9 Torr.
Polar magneto-optical Kerr effect (MOKE) measurements. MOKE measurements were
performed using a 1 mW laser with a wavelength of 655 nm focused to spot size of ~10µm.
Experiments were performed in polar geometry and hence sensitive to the out-of-plane
magnetization component. To apply gate voltage VG to the circular electrodes, a CuBe probe
was landed near the edge of the electrode and the Ta/Pt back electrode was grounded. The laser
spot was focused at the middle of the electrode. All experiments were performed at room
85
temperature. For experiments where different atmospheric conditions are required, VG was
applied ex-situ in a CPX-VF probe station. The MOKE hysteresis loop was measured before and
after VG was applied. Experiments under controlled gas environments were performed by
backfilling the chamber with either O2 gas (99.999% purity) or N2 gas (99.999% purity).
Humidity was introduced into the N2 gas flow by bubbling through water. Wet N2 and ambient
condition at 25C corresponds to ~20mT and 12mT of H2O partial pressure respectively.
Vacuum condition corresponds to a base pressure of 10-4mbarr. All experiments were performed
at room temperature.
X-Ray absorption spectroscopy (XAS). In situ XAS data were taken at the In-situ and Operando
Soft X-ray Spectroscopy (IOS, 23-ID-2) beamline at the National Synchrotron Light Source II,
Brookhaven National Laboratory. Partial fluorescence yield (PFY) spectra were acquired using a
Vortex EM silicon drift detector. The incident soft x-ray beam has a footprint of ~100 x 20 µm
and is directed at 30° relative to the sample normal, while the PFY detector is positioned at 40° to
the sample normal (Supplementary Information V). The sample used for the measurement has
crossbar geometry with sample structure Ta (4nm)/Mg(30 nm)/Pd(10 nm)/Co(0.9 nm)/GdOx (30
nm) and a 3 nm Au top gate. The XAS incident beam spot was located on the sample by first
scanning the scanning stage to locate the crossbars through the total electron yield of the top Au
electrode and chemical signature of the bottom electrode (Mg K-edge) (More details in
Supplementary Information V). Measurement is done with the VG applied in situ. For experiments
which require humidity, H2O vapor is introduced into the chamber through a leak valve and the
flow rate is adjusted to maintain PH2O of 10 Torr. At vacuum condition, the main chamber pressure
is ~2 x 10-7 Torr after H2O evacuation, and the sample is kept at room temperature throughout the
measurement.
86
4.2: Probing Water Uptake in GdOx
Figure 4.1(a) shows the effect of hydrating a GdOx thin film deposited on a SiO2/Si substrate.
The hydration treatment involves placing the sample at 90ºC under wet nitrogen gas at ambient
pressure with PH2O = 525 Torr, for up to 168 hours. X-ray reflectivity (XRR) spectra were
obtained periodically during the hydration process to follow the evolution of the film thickness
and density. During hydration, Gd2O3 is expected to react with H2O to form Gd(OH)3 according
to the reaction:146
𝐺𝑑2𝑂3 + 3𝐻2𝑂 → 2𝐺𝑑(𝑂𝐻)3 Equation 4.2
The XRR spectra were fitted by modeling the film as a bilayer of Gd2O3 and Gd(OH)3, with
variable thicknesses, roughnesses (structural and/or chemical), and mass densities (converted to
x-ray scattering length densities (SLDs). Figure 4.1(b) shows SLD profiles corresponding to fits
of the XRR spectra (Fig. 4.2), where two distinct layers are clearly resolved. With increasing
hydration time, we observe a gradual progression of the Gd(OH)3 layer deeper into the film.
Note that there is a diffuse gradient between a completely dry Gd2O3 and a fully hydrated
Gd(OH)3. In fact, the fitted roughness between the Gd2O3 and Gd(OH)3 layers arises from a
density gradient between the two layers, not from a structural roughness. During the hydration
process, the dry Gd2O3 first dissolves water molecules in the form of proton defects according to
Eq. 4.1. When the layer is completely hydrated, it forms Gd(OH)3 according to net reaction
depicted in Eq. 4.2. The transition region hence is expected to consist of a mixed phase of
hydrated Gd2O3 with dissolved water and Gd(OH)3.
87
Figure 4.1(c) shows the fitted thickness of the Gd(OH)3 layer as a function of hydration time
while figure 1(d) shows the XRR spectra of the non-hydrated (0 h) and hydrated (144 h) GdOx
films respectively. From the data, we see that a 22.8 nm as-prepared GdOx film takes
approximately 144 h to fully transform to Gd(OH)3. The fully transformed hydroxide shows an
increase in thickness of 50%, expanding from 22.8 nm to 34.1 nm and a decrease in density of
28%, from 8.3g/cm3 to 6.0g/cm3. This corresponds very well to the transformation of monoclinic
Gd2O3 to Gd(OH)3, with bulk densities of 8.3g/cm3 and 5.6g/cm3 respectively149,150.
In order to confirm the chemical state of the GdOx layer, we also performed x-ray
photoelectron spectroscopy (XPS) on the surface of a 3 nm GdOx thin film which was exposed to
ambient atmosphere for >2 weeks. Figure 4.1(e) shows the O1s spectrum of the thin film, where
the data is best fitted by two peaks at 531.8eV and 529.2eV, which correspond to the O-H bond
in Gd(OH)3 and the O-O bond in Gd2O3151,152 respectively. These results show that GdOx
readily uptakes water even in ambient conditions to form a hydroxide phase.
88
Figure 4.1. Probing water uptake in GdOx. (a) Schematic of a non-hydrated, partially
hydrated, and fully hydrated GdOx thin film on a SiO2/Si substrate. (b) Scattering length density
of the GdOx thin film as a function of hydration time. The fitted mass densities of Gd2O3 and
Gd(OH)3 are 8.3g/cc and 6.0 g/cc respectively. c)Fitted thickness of Gd(OH)3 as a function of
hydration time. d) X-ray reflectivity (XRR) spectra of a non-hydrated and hydrated GdOx thin
film (144h of hydration). The solid and dashed lines are the raw data and fits respectively. e) X-
ray Photoelectron Spectroscopy (XPS) data of a 3nm GdOx thin film surface which has been
exposed to ambient for > 2 weeks.
89
Figure 4.2. X-ray reflectivity (XRR) spectra. [(a)-(h)] XRR spectra of a GdOx thin film on a
SiO2/Si substrate as a function of hydration time. The solid and dashed lines are the raw data and
fits respectively. Hydration treatment involves placing the sample at 90C under 525Torr of PH2O.
90
4.3: Voltage-induced Co Oxidation in
Hydrated and Non-hydrated Pt/Co/GdOx
Devices
Figure 4.3 shows the comparison of voltage-induced Co oxidation between a non-hydrated
and hydrated Pt (3 nm)/Co(0.9 nm)/GdOx(10 nm)/Au(3 nm) device. Here, we probe the magnetic
state by measuring hysteresis loops probed locally using a polar magneto-optical Kerr effect
(MOKE) polarimeter. In its metallic state, the film exhibits a perpendicular magnetic anisotropy,
whereas in the oxidized state there is no magnetic signal, which provides a convenient means to
probe interfacial chemical state changes61.
For the non-hydrated device, the top Au electrode was deposited using an in situ shadow
mask immediately after the deposition of the Pt/Co/GdOx layers without vacuum break, so as to
serve as a capping layer to minimize water uptake upon exposure to ambient. Characterizations
of the non-hydrated devices were then done immediately after fabrication (within a day) in order
to preserve the non-hydrated state. For the hydrated device, the Pt/Co/GdOx structure was first
placed at 90ºC under PH2O = 525 Torr for 72 hours before the deposition of the top Au electrode.
All gate voltages (VG) were applied to the top Au electrode while the bottom Pt was grounded.
Figure 4.3(a) shows a MOKE hysteresis loop of a virgin non-hydrated device while figures
4.3(b)-(e) show MOKE hysteresis loops of the non-hydrated device after VG = -3V has been
applied for 600s in ambient, vacuum, wet N2 and dry O2 environments. No oxidation of Co is
observed under a negative bias even after 600s in dry O2 and wet N2 for the non-hydrated device.
Figure 4.3(f) shows the MOKE hysteresis loop of a virgin hydrated device while figure 4.3(g)-(j)
91
show MOKE hysteresis loops of the hydrated device after VG = -3V has been applied for 600s in
ambient, vacuum, wet N2 and dry O2. The results show that the Co layer is completely oxidized
even in vacuum, indicating that (1) the oxidant is present in the GdOx layer, and (2) O2 gas is not
required for the oxidation process. These combined results can be explained using the schematic
in figure 4.3(k)-(j). When VG =-3V is applied to the top Au gate of a non-hydrated device, the Co
layer remains metallic due to the absence of any oxidant in the GdOx film. When VG =-3V is
applied to the top Au electrode of a hydrated device, H2O stored in the oxide film in the form of
Gd(OH)3 oxidizes Co to CoO (Eq. 4.3). The proton, H+, produced from the reaction is then
driven by the electric field through the GdOx layer to the top Au electrode, where it is reduced by
electrons, e- (flowing through the external circuit) to form hydrogen gas (Eq. 4.4). The net
reactions, depicted in figure 4.3(l), are shown below:
The reaction described by Eq. 4.4 is also known as the hydrogen evolution reaction (HER) 153,154.
With Co and Gd(OH)3 densities of 8.9g/cc and 6.0g/cc respectively, the oxidation of 0.9 nm of
Co would require the decomposition of only ~ 3 nm of Gd(OH)3. It is likely that Gd(OH)3 does
not completely transform to dry Gd2O3 during the Co oxidation process. Rather, the Gd(OH)3
should instead transform to a semi-hydrated GdOx with dissolved water, leading to a gradient in
the water content adjacent to the Co interface. We note that the interface reaction corresponding
to the case in which the GdOx adjacent to Co is not the fully transformed hydroxide phase but
rather a hydrated oxide phase would be described by
Anode: 2𝑂𝐻𝑂+ + 𝐶𝑜 → 𝐶𝑜𝑂 + 2𝐻+ + 𝑂𝑂
𝑥 + 𝑉𝑂2+ + 2𝑒− Equation 4.5
Anode: 2𝐺𝑑(𝑂𝐻)3 + 3𝐶𝑜 → 𝐺𝑑2𝑂3 + 3𝐻2𝑂 + 3𝐶𝑜
→ 𝐺𝑑2𝑂3 + 6𝐻+ + 3𝐶𝑜𝑂 + 6𝑒− Equation 4.3
Cathode: 6𝐻+ + 6𝑒− → 3𝐻2 Equation 4.4
92
In this case, the cathode reaction will be the same as Eq. 4.3. This reaction could be controlling
under conditions where the cell is exposed to humid environments, but not as high as in this
study where formation of hydroxide is observed. In both cases, the bottom Co acts as the anode
while the top Au acts as the cathode. Note that if oxygen were available at the cathode, the
hydrogen formed at the cathode would react with the oxygen to form water as described in
Chapter 5.
93
Figure 4.3. Effect of GdOx hydration on voltage induced Co oxidation. [(a)-(e)] MOKE
hysteresis loops of non-hydrated Pt(3nm)/Co(0.9nm)/GdOx(10nm)/Au(3nm) device in virgin
state (a) and after VG = -3V has been applied for 600s in ambient (b), vacuum (c), wet N2 (d),
and dry O2 (e). [(f)-(j)] MOKE hysteresis loops of hydrated
Pt(3nm)/Co(0.9nm)/GdOx(10nm)/Au(3nm) device in virgin state (f) and after VG = -3V has been
applied for 600s in ambient (g), vacuum (h), wet N2 (i), and dry O2 (j). [(k)-(l)] Schematic of
voltage-induced reaction in a non-hydrated(k) and hydrated (l) device.
94
4.4: H2 evolution during Voltage-induced
Co Oxidation in Pt/Co/GdOx
In order to further confirm the oxidation of Co by H2O in GdOx, we also fabricated
thicker Au (15 nm) electrodes on Pt/Co/GdOx to isolate the device from its surrounding
atmosphere. Figure 4.4(a) shows the hysteresis loop of a virgin device with thicker Au electrode
while figure 4.4(b)-(c) shows the hysteresis loops after VG = -3V has been applied for 600s in
ambient and in vacuum respectively. With thicker Au, we still observe complete oxidation of Co
in roughly the same time as the thinner 3 nm Au device (fig. 4.5). This further confirms that the
oxidant is stored in the GdOx layer.
Figures 4.4(d)-(e) show optical micrographs of a hydrated
Pt(3nm)/Co(0.9nm)/GdOx(10nm) device with 15nm thick Au top electrodes, before and after
applying VG = -3V for 600s to completely oxidize the Co layer . Gas bubble formation is clearly
observed in the device. In order to verify if the gas bubbles are formed at the top GdOx/Au
interface or bottom Co/GdOx interface, we also performed a similar experiment with a Pt(3
nm)/Co(0.9 nm)/GdOx(10 nm) device with 3 nm thick Au after VG = -3V is applied for 600s
(Figure 3f). In this case, no gas bubbles are observed because the electrode is porous and the
evolved hydrogen gas escapes to the surrounding61. This indicates that the bubbles seen in the
thicker Au electrode case are formed at the GdOx/Au interface, and this gas is necessarily H2,
since the Au acts as the cathode at VG = -3V. Note that gas bubble formation is also observed
when VG = -3V is applied for 600s to Pt/Co/GdOx/Au devices in vacuum. This hydrogen
evolution reaction was confirmed in the next chapter where Pd and Mg layers were inserted in a
95
substrate/Mg/Pd/GdOx/Au stack structure and the formation of PdHx and MgHx was confirmed
using XAS upon applying a gate bias. In this case, a positive bias (VG > 0) was applied to the top
Au electrode to insert hydrogen in the bottom Pd and Mg layers.
Figure 4.4. Hydrogen gas bubble evolution. MOKE hysteresis loops of hydrated Pt(3
nm)/Co(0.9 nm)/GdOx(10 nm)/Au(15 nm) device in virgin state (a) and after VG = -3V was
applied for 600 s in ambient (b) and vacuum (c). [(d)-(e)] Optical micrographs of Pt(3
nm)/Co(0.9 nm)/GdOx(10 nm)/Au (15 nm) devices before (d) and after (e) bias voltage
application (VG = -3V for 600 s) showing generation of hydrogen bubbles under the electrode. (f)
Optical micrograph of Pt(3 nm)/Co(0.9 nm)/GdOx(10 nm)/Au (3 nm) after applying VG = -3V for
600s. The scratch marks on the side of the Au electrodes are due to the CuBe probes.
96
Figure 4.5. Voltage-induced Co oxidation in Pt/Co/GdOx/Au heterostructures probed by
MOKE polarimetry. Number is parentheses represent thickness of Au top electrode.
Interestingly, some of the hydrogen which is produced at the top Au electrode under VG < 0
can be stored in the GdOx film. In order to demonstrate this, we compared the magneto-ionic
response of two Pt/Co(CoO)/GdOx/Au devices with a hydrated GdOx layer, where the Co layer
has been oxidized. In the first device, the Co layer is metallic in its as-deposited state, and is then
oxidized completely by first applying VG = -3V for 300s to the top 3nm Au (figure 4.6a). In the
second device, the Co layer is deposited in its oxidized state by reactive sputtering with oxygen
gas (figure 4.6d). A positive bias (VG = +3V or +2V) bias is then applied to both devices in
vacuum and in ambient in order to reduce the CoO layer to metallic Co. This process has
previously been shown to occur through injection of protons to the CoO layer, where they react
with CoO to reduce it to a metallic state155. Figure 4.6(b) and (c) show the results for the first
device in vacuum and ambient respectively, while figure 4.6(e) and (f) show the corresponding
results for the second device. For the first device, we can clearly see that some of the CoO is
reduced in vacuum. This implies that some hydrogen is stored in the device after the initial VG =
-3V is applied which allows for closed-system electrochemical and magnetic property switching
without the need for atmospheric exchange. The reduction of CoO to metallic Co by the stored
97
hydrogen occurs through the reverse reactions described in Eqs. 4.3 and 4.4. In this case, at VG >
0V, the stored hydrogen donates its electrons driving the reduction of CoO to metallic Co. On the
other hand, for the second device where the Co layer is oxidized during deposition, we do not
observe any CoO reduction at VG = +3V in vacuum (figure 4.6e). The CoO layer is only reduced
when the positive gate bias is applied in ambient, where humidity is present so that a water-
splitting reaction can occur to provide a source of protons. In both devices, a positive bias in
ambient initially results in Co with perpendicular magnetic anisotropy. The Co magnetization
then rotates in-plane as more hydrogen is accumulated at the bottom interface155.
Figure 4.6. Hydrogen storage in GdOx film. (a) MOKE hysteresis loop of hydrated Pt(3
nm)/Co(0.9 nm)/GdOx(10 nm)/Au(3 nm) after VG = -3V has been applied for 300s in ambient.
[(b)-(c)] MOKE hysteresis loops after VG = +3V is applied to device in (a) in vacuum (b) and in
ambient (c). (d) MOKE hysteresis loop of hydrated Pt(3 nm)/CoO(0.9 nm)/GdOx(10 nm)/Au(3
nm) in virgin state. [(e)-(f)] MOKE hysteresis loops after VG = +2V is applied to device in (d) in
vacuum (e) and in ambient (f).
98
4.5: In-situ XAS probe of Co during
Voltage-induced Co Oxidation in
Pt/Co/GdOx
To further verify the oxidation of Co through a direct chemical probe, we also performed in-situ
x-ray absorption spectroscopy on a hydrated Pd(10nm)/CoO(0.9nm)/GdOx(30nm)/Au(3nm)
device while applying gate biases under different atmospheric conditions. In this case, the
hydration treatment at 90oC and 525Torr is performed for only 24 hours in order to retain a non-
hydrated state at the CoO/GdOx interface. The first and second columns of figure 4.7 show the
data for Co L2, L3 edge22,23 and their corresponding first derivatives, while the third column
shows schematically the chemical state of the Co and the GdOx layer near the interface . In its
virgin state, the Co layer is initially oxidized (figure 4.7a). When VG = +3V is applied to the top
Au in vacuum, the CoO layer remains oxidized (figure 5b). However, when VG = +3V is applied
in 10Torr of PH2O, the CoO layer is reduced to metallic Co by H+ sourced from H2O (figure 4.7c)
155. The H2O that is produced from this reaction is reincorporated back into GdOx in the form of
Gd(OH)3. To confirm this, we next applied VG = -3V to the metallic Co device in vacuum. The
data shows partial reoxidation of the metallic Co back to CoO, consistent with the hypothesis.
Similarly, if VG = -3V is applied to the metallic Co in 10Torr of PH2O instead of vacuum, partial
reoxidation of the Co is also observed.
99
Figure 4.7. X-ray absorption spectra (XAS) of Co L2 and L3 edge. The first, second, and third
columns are the raw data, the first derivatives, and the chemical state of the Co/GdOx interface.
[(a)-(c)] XAS spectra of Pd(10 nm)/Co(0.9 nm)/GdOx (30 nm)/Au(3 nm) device in virgin state(a)
and after VG = +3V has been applied for 600 s in vacuum(b) and 10 Torr of PH2O (c)
respectively. The experiments from (a) to (c) are done sequentially. [(d)-(e)] XAS spectra of the
device in (c) after VG = -3V is applied for 600 s in vacuum and in 10 Torr of PH2O respectively.
Two different devices in (c) are used for experiment in (d) and (e).
100
Chapter 5: Magneto-
ionic Control of
Magnetism using a
Solid-state Proton
Pump
101
Most research efforts on magneto-ionics have focused on voltage-induced oxidation and
reduction of a ferromagnetic metal to reversibly modulate its magnetic anisotropy, exchange bias,
and magnetization58–65. The problem with oxidation-based magneto-ionics is that magnetic
property changes are accompanied by chemical and structural changes in the target ferromagnet.
This often leads to irreversibility62 and would be detrimental to devices such as magnetic tunnel
junctions whose performance depends critically on structure and electronic properties of the
ferromagnet.
Alternatively, Group I ions such as Li can be inserted into a target ferromagnet to alter
magnetic properties without changing the chemical phase or structure72,73,156. Small ion size and
the possibility of super-ionic conduction makes this a promising approach to achieving fast,
reversible magnetic property switching, but most Group I ions are incompatible with CMOS,
limiting their viability for practical applications. The exception is H+, which is relatively
innocuous, and is at the same time the simplest possible ion, making it ideal for inducing rapid
electric field driven property changes in solid-state structures.
Here we show that H2O hydrolysis in ambient atmosphere catalyzed by a rare-earth
oxide/noble metal interface can serve as a solid-state proton pump that enables non-destructive
magnetic property gating with a modest voltage. We demonstrate reversible 90o magnetization
switching in a thin Co film at room temperature by either inserting H+ at its interface with an oxide
or loading hydrogen into an adjacent heavy metal layer. The mechanism permits both unipolar
toggle switching and nonvolatile state retention, with no discernible irreversibility in magnetic
properties of the ferromagnet after >2000 cycles. Moreover, since heavy metals like Pt and Pd that
exhibit strong spin-orbit coupling are also well-known hydrogen storage materials157,158 that can
be driven between a metal and metal-hydride phase, a host of spin-orbit induced phenomena at
102
heavy-metal/ferromagnetic interfaces33,159,160 becomes accessible to voltage gating despite the fact
that electric fields cannot be applied directly.
103
5.1: Experimental Methods
Sample preparation: Ta(4 nm)/Pt(3 nm)/Co(0.9 nm)/GdOx(tGdOx nm)/Au(3 nm) layers were
fabricated on thermally oxidized Si (100) substrates using magnetron sputtering at room
temperature and 3mTorr Ar pressure. The metal layers were grown by DC sputtering. The GdOx
layer was deposited either using reactive sputtering with PO2 of 0.07mTorr or RF sputtering with
PO2 of 0.7mTorr O2. For the samples described in Fig. 5.1 with Co in the initially-oxidized state,
the Co layer was reactively sputtered with PO2 of 0.07mTorr O2 with a deposition time
corresponding to the time required to deposit 0.9 nm of metallic Co. For AHE and PHE
measurements, the structure is patterned into a Hall cross geometry with 500µm arm width and
with Au(3nm) deposited over the 0.25 mm2 active region to serve as a gate electrode. For MOKE
measurements, 200 µm diameter Au(3 nm) electrodes were patterned on top of the GdOx layer of
a continuous film, with the Ta(4nm)/Pt(3 nm) underlayer uncovered by GdOx at the sample edge
to allow electrical contact to the back. All patterning was done using shadow mask lithography.
Hall Effect measurements in different atmospheres: Anomalous Hall effect (AHE) and planar
Hall effect (PHE) measurements were performed using a lock-in amplifier with an ac injected
current of amplitude 2 mA and frequency 1kHz. For the AHE measurements, the field was swept
perpendicular to the plane; for the PHE measurements, the field was oriented in the sample plane,
at 45o to the current flow axis. AHE and PHE hysteresis loops were acquired using a 2s field sweep
time. The measurements were performed in a modified CVX-PF Lakeshore Probe Station with a
base vacuum pressure of ~10-4 mbar. Experiments under controlled gas environments were
performed by backfilling the chamber with either O2 gas (99.999% purity) or N2 gas (99.999%
purity). Humidity was introduced into the N2 gas flow by bubbling through water. Wet N2 and
104
ambient condition at 25C corresponds to ~20mT and 12mT of H2O partial pressure respectively.
All experiments were performed at room temperature.
Polar magneto-optical Kerr effect (MOKE) measurements: MOKE measurements were
performed using a 1 mW laser with a wavelength of 660 nm focused to spot size of ~10µm.
Experiments were performed in polar geometry and hence sensitive to the out-of-plane
magnetization component. To apply gate voltage VG to the circular electrodes, a CuBe probe was
landed near the edge of the electrode and the Ta/Pt back electrode was grounded. The laser spot
was focused at the middle of the electrode. All experiments were performed at room temperature.
X-Ray absorption spectroscopy (XAS): XAS data was taken at Coherent Soft X-ray Scattering
(CSX) beamline at the National Synchrotron Light Source II, Brookhaven National Laboratory
using fluorescent yield. The incident soft x-ray beam has a footprint of ~200µm and the sample is
tilted 15º relative to the incident beam. The sample used for the measurement has a hall bar
geometry with sample structure Ta(4nm)/Pd(3nm)/Co(0.6nm)/Pd(4.5nm)/GdOx (30nm) and a 3nm
Au top gate. The main chamber base pressure is ~2 x 10-9 torr, and the sample is kept at 100K
throughout the measurement.
105
5.2: Co Redox through Water Electrolysis
For a Pt/Co/GdOx/Au nominal structure, we have shown in chapter 4 that it is H2O stored in GdOx
in the form of Gd(OH)3 that oxidizes Co to CoO when VG < 0 is applied to the top Au. In this
section, we will show that for a Pt/CoO/GdOx/Au structure , a positive gate voltage (VG > 0) can
electrochemically split atmospheric water and pump protons through the GdOx, to both reduce
CoO to metallic Co and to modulate the magnetic anisotropy of a metallic Co thin film.
Figure 5.1 shows the effect of applying a gate bias to Pt/Co/GdOx/Au under several atmospheric
conditions, demonstrating the critical role of ambient moisture in magneto-ionic control of Co
oxidation state and magnetic properties. We used a sample structure
Ta(4nm)/Pt(3nm)/CoO(0.9nm)/GdOx(30nm)/Au(3nm) sputter deposited on thermally oxidized Si.
A positive gate voltage (VG > 0) was applied to the top Au while the out-of-plane and in-plane
magnetization was monitored electrically through the anomalous Hall effect (AHE) and planar
Hall effect (PHE), respectively, using a Hall bar geometry (Fig. 5.1c).
Figures 5.1d-h show out-of-plane hysteresis loops probed through the AHE resistance (RAHE)
for the initially-oxidized sample. Results are shown for the virgin state (Fig. 5.1d), and after
applying VG = +3V for 1000s at room temperature in various atmospheres (Figs. 5.1e-h).
Consistent with literature60,61, under ambient atmosphere, a positive bias results in the appearance
of out-of-plane magnetization, corresponding to the reduction of nonmagnetic CoOx to metallic
Co with perpendicular magnetic anisotropy (PMA) (Fig. 5.1e). VG has no effect under vacuum
(Fig. 5.1f), even though a lower oxygen partial pressure (PO2) environment should make oxygen
extraction more favorable. Likewise, in dry O2 (Fig. 5.1g), no magnetic changes are observed, but
106
remarkably, under wet N2 (Fig. 5.1h), positive VG leads to the appearance of PMA, implying water-
assisted reduction of CoOx to Co. These results confirm the findings in chapter 4 that moisture and
proton transport through GdOx play a crucial role in voltage-induced redox of Co and that oxygen
ion migration plays an insignificant role.
Figure 5.1. In-situ probing of magneto-ionic switching in different atmospheres. a-b. Active
region of a Pt/CoO/GdOx (a) and Pt/Co/GdOx (b) device. c, Hall cross geometry used for
anomalous Hall effect (AHE) measurement to probe out-of-plane magnetization. d, AHE
hysteresis loop of virgin structure in (a). e-h, Hysteresis loops after applying a gate voltage VG =
+3V for 1000s in ambient (e), vacuum (f), dry O2 (g), and wet N2 (h).
It is well-known that water splitting reaction can be catalyzed by noble metals with oxide
support112,161,162. Water splitting and hydrogen incorporation had been shown to significantly
impact the switching behavior and electronic properties in metal/oxide/metal memristors163–165.
During reduction of CoOx under positive VG in a Pt/Co/GdOx/Au heterostructure, H2O is
hydrolyzed at the top electrode producing H+ and O2 through the oxygen evolution reaction111,112.
107
The proton is transported to the bottom electrode via a Grotthuss-type mechanism81,82,137 in which
the H+ ion hops between adjacent lattice oxygen atoms. At the bottom electrode, the proton reacts
with CoO to form Co and H2O. The net reactions, depicted in Fig. 5.2a, are
The water evolved at the bottom is then incorporated into GdOx itself as hydroxide, namely through
the reaction146: 𝐺𝑑2𝑂3 + 3𝐻2𝑂 → 2𝐺𝑑(𝑂𝐻)3 . Co oxidation under negative VG occurs by the
reverse process (Fig. 5.2b), with Pt acting as the anode for Co oxidation and Au acting as the
cathode for H2O recombination. Note that this H2O recombination reaction only occurs under the
presence of O2. If O2 is absent, a hydrogen evolution as described in chapter 4 will take place.
Figure 5.2 Electrochemical reactions in a magneto-ionic cell . a, Schematic of CoO reduction
at gate voltage VG > 0 involving H2O hydrolysis. b, Schematic of Co oxidation at gate voltage VG
< 0 involving H2O recombination.
Anode (Au) : 2𝐻2𝑂 → 4𝐻+ + 𝑂2 + 4𝑒− Equation 5.1
Cathode (Pt) : 4𝐻+ + 2𝐶𝑜2+ + 2𝑂2− + 4𝑒− → 2𝐻2𝑂 + 2𝐶𝑜. Equation 5.2
108
To confirm the role of ambient water as the source of proton in the reduction of CoO to
Co, we also looked at the effect of Au electrode thickness on the rate of voltage-induced CoO
reduction. Figure 5.3a shows the transient magneto-optical Kerr effect (MOKE) magnetization
remanence ratio, Mr/Ms at the center of a device with structure Pt(3nm)/CoO(0.9nm)/GdOx
(20nm)/Au (t =3,4,6,10 nm) as VG = +3V is applied. We observe that the voltage induced CoO
reduction becomes slower with increasing thickness of Au, confirming that hydrogen is sourced
from H2O in ambient during the reduction process. The reduction activity is highest at the
electrode edge where the thickness of Au is thinner due to shadowing effect from shadow mask
lithography. This can be seen in the optical micrograph in Figure 5.3b for a t = 4nm Au
electrode. The bottom plot in Figure 5.3b shows a snapshot of Mr/Ms across the electrode after
VG = +3V for ~600s. One can see that CoO is reduced to Co by H injection at the edges (Fig.
5.3c). With increasing bias dwell time, this region of modified properties moves inward towards
the center. Figure 5.3d shows the transient (VG = +3V) for a t = 6nm Au electrode at the edge vs
the center, where a relatively constant Mr/Ms followed by a sharp increase further indicates a
circular “front” of changed properties that is slowly moving inward from the edge towards the
center.
109
Figure 5.3 Voltage-driven injection of hydrogen at edge of electrode. a, Magneto-optical
Kerr effect (MOKE) magnetization remanence ratio, Mr/Ms versus time as VG =+3V is applied to
a Pt(3nm)/CoO(0.9nm)/GdOx (20nm)/Au (t nm) device. b, Top: Optical micrograph of a t = 4nm
electrode. Bottom: Snapshot of Mr/Ms across the electrode at ~600s. c, Hysteresis loops
corresponding to center and edge of electrode in b at ~600s. d, Magnetization transient for t =
6nm electrode at the edge vs the center.
If positive Vg pumps H+ as proposed, sustained bias application should lead to hydrogen
accumulation and evolution at the bottom electrode, and this is indeed observed. Figure 5.4a shows
scanning electron microscope (SEM) of a Pt(3nm)/GdOx(100nm)/Au(3nm) film after applying VG
= +3V for 5 hours. The inset optical micrograph shows bubble formation, which after cross-
sectioning is revealed to arise from gas evolution and GdOx delamination at the bottom electrode,
110
which could only be H2 since O2 evolution could only occur at the anode. Figure 5.4b shows the
voltage threshold for the onset of PMA in a CoO structure to be ~1.5V, which agrees well with the
proposed reactions in equation 5.1 and 5.2111–113.
Figure 5.4 Cross section SEM to image hydrogen evolution. a, Cross-sectional scanning
electron microscope image of a Pt (3nm)/GdOx (100nm)/Au (3nm) device after VG = +3V has
been applied for 5 hours. Inset shows optical micrograph of the device. Hydrogen gas is
produced between the Pt and GdOx layer at the end of the experiment. b, Out-of-plane remanent
magnetization ratio, Mr/Ms of Pt(3nm)/Co(0.9nm)/GdOx(tGdOx)/Au(3nm) structure after applying
various VG for > 5 hours, for tGdOx = 4 nm and 30 nm.
Figures 5.5a-b show cyclic voltammetry (CV) data for a Pt(3nm)/GdOx(20nm)/Au(3nm) structure
under different atmospheric conditions (Fig. 5.5a) and at different sweep rates (Fig. 5.5b). We
assume that the top Au electrode acts as the working electrode while the bottom Pt electrode acts
as the counter/reference electrode166. When cyclic voltammetry is performed under different
atmospheres (Fig. 5.5a), significant anodic and cathodic currents are observed only when H2O is
present. Both currents are governed by the voltage sweep rate (v) (Fig 5.5b) which indicates a
process that is kinetically limited by mass transport rather than charge-transfer166. From both
information, we can conclude that water hydrolysis (VG > 0) and recombination (VG < 0) are indeed
111
taking place111–113. Despite the lack of a unique reference electrode, the classical exponential
behavior observed in oxidation is consistent with oxygen evolution reaction (OER) while the peak
observed in reduction is typical of oxygen reduction reaction (ORR)107,113.
Figure 5.5. Dependence of overall reaction rate on charge transfer and mass transport. a, Cyclic Voltammetry (CV) plot under different atmospheric conditions performed at v =
10mV/s.b, CV plot at different sweep rates performed in ambient.
112
5.3: Modulation of Magnetic Anisotropy
through Proton Injection
We now show that hydrogen insertion at the Co/GdOx interface allows the anisotropy to be
toggled from out-of-plane to in-plane without requiring redox reactions in the ferromagnetic Co
(Fig. 5.6a). Surface anisotropy is known to be sensitive to adsorbed H, as shown previously for
ultrathin ferromagnetic films in ultra-high vacuum upon exposure to molecular or atomic
hydrogen24–28. Here, we show the same behavior can be gated in solid-state devices. Starting from
a virgin state with PMA (Fig. 5.6b), the magnetization rotates in-plane when VG = +3V is applied
for 800s (Fig. 5.6c), corresponding to accumulation of hydrogen at the Co/GdOx interface. When
VG is set to 0V (grounded), PMA is spontaneously recovered (Fig. 5.6d) as the accumulated
hydrogen forms H+ and diffuses away from the bottom electrode. In-plane magnetization
reorientation is confirmed by Figs. 5.6e-g, which show that the PHE signal is absent when the film
has PMA and is present when the AHE signal vanishes (Figs. 5.6e-g) under positive bias. Figures
5.6h-j show cycling results for a device with tGdOx=4 nm, in which switching is much faster. VG
was cycled >2000 times between +3V and 0V at 0.5 Hz, and out-of-plane hysteresis loops were
acquired at 25 ms intervals using a polar magneto-optical Kerr effect (MOKE) polarimeter (see
Methods). Figure 5.6h shows the ratio of the remnant (Mr) to saturation magnetization (Ms) as a
function of time, for cycles 1-10 and 2060-2070, tracking the in-plane/out-of-plane transitions
(Fig. 5.6i). The square out-of-plane loop in the virgin state (Fig. 5.6i) is indistinguishable from
the loop after toggling the magnetization in plane, both after the first cycle (Fig. 5.6i) and after
cycle 2070 (Fig. 5.6j). We find, however, that the response time degrades slightly with repeated
cycling (Fig. 5.6h), which may be associated with increased leakage currents in the oxide. The
113
switching times at the rising and falling VG edge are 100ms and 400ms respectively (Fig. 5.7). This
switching speed is faster than any room temperature results in the literature for magneto-ionic
switching60,61. The asymmetry in switching can be mitigated by applying a negative VG to
accelerate H+ removal from the interface, but this can also lead to Co oxidation, which leads to a
progressive irreversible degradation of PMA due to irreversibility of oxygen insertion into the
magnetic layer62.
Figure 5.6. Magneto-ionic switching based on hydrogen accumulation at Co/GdOx
interface: a, Schematic of magneto-ionic switching scheme. b-d, AHE hysteresis loops in virgin
state (b), after VG = +3V is applied for 800s (c), and after VG is set to 0V for 800s (d). e-g, PHE
hysteresis loops corresponding to b-d respectively. h, Magnetization remnance ratio Mr/Ms
versus time as VG is cycled between +3V and 0V at 0.5 Hz for 2070 cycles, extracted from
hysteresis loops measured by the polar magneto-optical Kerr effect (MOKE). Results are shown
for the first and last ten cycles. i, Out-of-plane hysteresis loops corresponding to the virgin state
and the first switching cycle. j, Out-of-plane hysteresis loops corresponding to cycle 2070. The
hysteresis loop of the final relaxed state is identical to that in the virgin state.
114
Figure 5.7 Magneto-ionic switching speed of a Pt/Co/GdOx structure. a-b, Polar magneto-
optical Kerr effect magnetization remanence ratio Mr/Ms at rising (a) and falling edge (b) of VG.
To quantify the change in magnetic anisotropy energy induced by hydrogen insertion, we also
probed the magnitude of the out-of-plane magnetization in the virgin state as a function of an in-
plane field. The strength of the in-plane field required to tilt the magnetization in-plane would tell
us how large the starting perpendicular magnetic anisotropy is. Since the magnetization rotates
from out-of-plane to in-plane under VG > 0, the voltage-induced change in anisotropy would be at
least the magnitude of this starting perpendicular magnetic anisotropy. Figure 5.8a shows the out-
of-plane hysteresis loop probed through the anomalous Hall effect resistance (RAHE) for a Pt(3nm)/
Co(0.9nm)/ GdOx (30nm) structure while figure 5.8b shows RAHE as a function of in-plane field,
Hx applied along the current injection line for the virgin state. The red curve shows a fit to the
single-domain Stoner-Wohlfarth model, RAHE = Rsat cos(arcsin(Hx/HK)) in order to obtain the in-
plane saturation field, HK. Rsat is the AHE resistance when the magnetization is saturated in the
out-of-plane direction. The fitted HK is ~8.2kOe, and assuming Ms of 1400emu/cc for Co, the
uniaxial magnetic anisotropy, Ku = (HkMs)/2 is 5.7x106 erg/cc. Normalizing by the thickness of
the Co layer, tCo (0.9nm), we obtain interfacial magnetic anisotropy, Ks = tCo (HkMs)/2 of
0.52erg/cm2 for the device in virgin state. This value represents the lower bound for the change in
115
Ks when the magnetization rotates from out-of-plane to in-plane under VG > 0. With an electric
field of 1MV/cm (VG = +3V), this corresponds to a magnetoelectric efficiency of > 5200fJ/ Vm.
Figure 5.8. Quantification of magnetic anisotropy energy. a, Out-of-plane hysteresis loop
probed through the AHE resistance (RAHE) for a Pt(3nm)/ Co(0.9nm)/ GdOx (30nm) structure. b,
RAHE as a function of in-plane magnetic field. The red line shows the fit for a single domain
Stoner-Wohlfarth model.
Similar to the reduction of CoO described above, the accumulation of hydrogen which causes
this magnetization rotation in-plane only occurs in the presence of humidity. Figure 5.9a-b show
the rate of change in out-of-plane and in-plane magnetization probed using the anomalous Hall
effect (AHE) and planar Hall effect (PHE) as VG = +3V is applied to a Pt/Co/GdOx Hall cross
device under different atmospheric conditions (tGdOx = 30nm). The data clearly indicates that the
magnetization rotates from out-of-plane to in-plane only in the presence of moisture, again
confirming that H2O in the environment acts as the source of hydrogen which is accumulated at
the Co/GdOx interface167. Figures 5.9c-e show the corresponding AHE hysteresis loops in the
virgin state (Fig. 5.9c) and after VG = +3V is applied for 800s in vacuum (Fig. 5.9d) and wet N2
(Fig. 5.9e). Figures 5.9f-h show the corresponding PHE hysteresis loops.
116
Figure 5.9. Effect of positive VG in different atmospheric conditions starting from a Co
virgin state. a, Anomalous Hall effect resistance (RAHE) vs time at VG = +3V. b, Planar Hall
resistance (RPHE) vs time at VG = +3V. c-e, AHE hysteresis loops of the device in virgin state (c),
and after VG = +3V is applied for 800s in vacuum (d)and in wet N2 (e) . f-h, PHE hysteresis
loops of the device corresponding to the condition in c-e respectively.
117
5.4: Magnetic Response under Short
Circuit and Open Circuit
The electrochemical reactions at VG =+3V in the two half cells (top and bottom electrodes),
given by
cannot proceed without electron flow through the external circuit. Thus, one can realize two
operating modes by either grounding the device (VG = 0V) or leaving the device at open circuit
when VG is removed. Figure 5.10a shows that after VG = +3V is applied, the magnetization
transitions from out-of-plane (Fig. 5.10b) to in-plane (Fig. 5.10c), and when VG is set to open
circuit, the in-plane state is retained (Figs. 5.10c,d). This implies that the accumulated hydrogen
remains at the interface. When the device is subsequently set to closed circuit, PMA is
spontaneously recovered (Fig. 5.10e). Hence, the magnetization state can be switched in a
nonvolatile fashion between out-of-plane and in-plane states, or toggled with a unipolar voltage,
depending on whether the VG=0 condition is at open or closed circuit.
Anode (Au) : 2𝐻2𝑂 → 4𝐻+ + 𝑂2 + 4𝑒− Equation 5.3
Cathode (Pt) : 4𝐻+ + 4𝑒− → 4𝐻 Equation 5.4
118
Figure 5.10. Magnetic response under short circuit and open circuit. a, Evolution of Mr/Ms
vs time and the corresponding VG. Dashed line indicates open circuit while solid line at VG = 0V
indicates short circuit. VG = +3V is applied between time,t = 30s to t = 100s, at which point the
probe is lifted from the top electrode. At t = 5300s, the probe is landed again and VG set to 0V
(ground). b-e, Polar MOKE hysteresis loops corresponding to t = 25 s (b), t = 110 s (c), and t =
4200 s (d), and t = 5600s (e).
Such ability to modify the magnetic state over two very distinct timescales can be particularly
attractive for neuromorphic computing where one can modify the memory168,169 (non-volatile)
during computation and once the computation is complete, all memory elements are reinitialized
to the same state during a refresh stage (volatile).
These results show that removal of H from the bottom electrode requires the reactions in Eqs.
5.3 and 5.4 to occur in reverse. To be removed, H must first split into H+ and e- at the bottom
electrode, so that the H+ can be transported back up through the GdOx to the top electrode, where
it recombines with atmospheric oxygen to form H2O. If there is no electronic conduction path, the
reaction cannot proceed and as a result the in-plane magnetized state is retained at open circuit.
Hence, a finite leakage current through the GdOx reduces the stability of the hydrogen-loaded state
as it offers an alternate path for electron transfer from the bottom electrode to the top electrode.
119
From this, we show that H cannot simply diffuse from the bottom electrode; its insertion and
removal are governed by the anodic and cathodic electrochemical reactions in Eqs. 5.3 and 5.4.
120
5.5: Electrical Gating of Magnetic
Anisotropy at a Heavy-
metal/ferromagnet Interface
Since Pd is well known for its hydrogen loading capacity157, and atmospheric hydrogen loading
in Pd/Co/Pd has previously been shown to modulate PMA26, we next exploit this feature by
inserting a Pd layer between Co and GdOx. Figure 5.11a shows the layer schematic of a
Ta(4nm)/Pd(3nm)/Co(0.6nm)/Pd(4.5nm)/GdOx(10nm)/Au(3nm) heterostructure. Because the Co
layer is protected by Pd, negative VG does not result in oxidation of Co (Fig. 5.11b)), which allows
for applying negative gate bias to accelerate H removal and recovery of PMA. In this device,
positive and negative VG can pump hydrogen into and out of the Pd layer reversibly, switching the
anisotropy from out-of-plane to in-plane and back26 (Figs 5.11c-e). Figure 5.11f shows the
switching cycles of Mr/Ms as VG is cycled between +4V and -1V at 1Hz. Robust switching was
achieved, with a switching time of ~150 ms at both the rising and falling edges (Figs 5.11g and h).
121
Figure 5.11. Voltage gating of metal-metal interface by exploiting hydrogen loading in Pd.
a, Schematic of device operation in a Pd/Co/Pd/GdOx cell. b, Out-of-plane magnetization
remanence ratio Mr/Ms versus time at gate voltage VG = -3V. Inset shows polar magneto-optical
Kerr effect (MOKE) hysteresis loops at t = 0s and t = 4500s. c-e, Polar MOKE hysteresis loops
corresponding to virgin state (c), after VG = +4V (d), and after VG is set to -1V (e). f, Mr/Ms as VG
is cycled between +4 V and -1 V at 1 Hz. Each data point corresponds to a 25 ms MOKE
hysteresis loop g, Mr/Ms at rising edge of VG. h, Mr/Ms at falling edge of VG.
In order to directly evidence insertion of H into the heavy metal layer, we performed x-ray
absorption spectroscopy (XAS) at the 23-ID-1 beamline at National Synchrotron Light Source II.
Figure 5.12a shows an XAS spectrum of a virgin
Ta(4nm)/Pd(3nm)/Co(~0.6nm)/Pd(4.5nm)/GdOx(30nm)/Au(3nm) sample in the range between
525eV and 575eV, where one can observe the Pd M3 edge170 at ~532eV and O K-edge at
~538eV. Figure 5.12b shows comparison of the Pd M3 edge between a virgin and voltage-
122
modified device. For the voltage-modified device, VG = +3V is applied for >5min ex-situ before
the XAS spectrum was measured. From the data, one can clearly observe a peak shift of ~+0.7eV
for the Pd M3 edge upon voltage application. This shift arises from two sources. First, previous
XPS studies have shown that the binding energy of core electrons shifts slightly when Pd
becomes PdHx. This shift is ~+0.17eV for the 3d electrons and its magnitude is very small
because hydrogen in Pd is well screened in the lattice171. Secondly, there is a change in energy of
the final unoccupied 4d states of Pd by ~ +0.7eV when it becomes a hydride172,173. This change
accounts for most of the peak shift in the XAS spectrum. The total peak shift of the Pd M3 edge
should hence be ~+0.87eV, which is very close to the +0.7eV peak shift we observe in our data.
Figure 5.12 X-ray absorption spectroscopy of Pd M3 edge. a, X-ray absorption spectrum of a
Pd (3nm)/Co (0.6nm)/Pd (4.5nm)/GdOx (30nm)/Au(3nm) sample in the range between 525eV
and 575eV. b, X-ray absorption spectra of a virgin and voltage-modified (VG = +3V for >5min)
sample, clearly indicating an energy shift in the Pd M3 peak for the voltage-modified sample.
We have also confirmed this with a thick Mg as the hydrogen loading layer174 in a
Ti(3nm)/Mg(40nm)/Pd(5nm)/GdOx (30nm)/Au(3nm) structure. Fig 5.13 shows X-ray absorption
spectra the device in the virgin state and after a positive bias is applied to the top electrode in
ambient (VG =+3V, for 5 minutes). Comparison of the XAS data of the biased device and the
literature data174 shows that the Mg layer becomes hydrided upon application of a positive gate
123
bias, VG = +3V. We should note that before such a thick Mg (40nm) can be loaded with
hydrogen, the entire Pd layer will have to be penetrated by hydrogen. Such a substantial amount
of hydrogen loaded into the system is far more than what would be required to explain the
magnetic property changes observed in the thin film structures in our experiments.
Figure 5.13 X-ray absorption spectra of Mg K-edge. Mg K-edge of a Ti(3nm)/Mg (40nm)/Pd
(5nm)/GdOx (30nm)/Au(3nm) structure in virgin state (black) and after VG = +3V is applied for
5min (red). Mg K-edge of the biased device shows remarkable similarity to that of MgHx174.
With this, we show for the first time that the magnetic anisotropy at a metal/metal interface can
be modulated substantially by an electric field using electrochemical gating of a metal to its
hydride phase in an appropriately designed solid state heterostructure. This effect could never be
achieved by any other known mechanism since electric field vanishes in a metal.
124
5.6 Comparison between Au and Pt Top
Electrodes For the Pt(4nm)/Co(0.9nm)/GdOx(4nm) structure described in chapter 5.3 (figure 5.6), we also
compared the difference in magneto-ionic switching speed between a 3nm Au and 3nm Pt top
electrode. Both electrodes are deposited on the same Pt/Co/GdOx film. The MOKE results for
magnetic switching in both devices at VG = +3V and 0V cycled at 0.5Hz are shown in figure
5.14. Similar to the Au devices, the Pt devices demonstrate uniform 90° switching with excellent
cyclability. Because the proton for magnetic gating is sourced from H2O in ambient, Pt devices
would be expected to toggle much faster than Au devices due to its higher activity for water
splitting. However, this is only true for the switching process from in-plane to out-of-plane at VG
= 0V. During switching from out-of-plane to in-plane state, the speed of the Pt device is
surprising slower. The reason for this is still unclear at this point.
125
Figure 5.14. MOKE switching with VG cycled between +3V and 0V at 0.5Hz. (a) Comparison
of 10 switching cycles between Au and Pt top electrodes. (b) Switching transient at falling edge
(left) and rising edge (right) of VG.
126
Chapter 6: Voltage
Gating of Magnetic
Damping and Spin-
Orbit Torques using
Proton
127
In heavy metal/ ferromagnet/oxide heterostructures, a rich set of properties such as spin-orbit
torques175, Dzyaloshinskii-Moriya interaction (DMI)159, and magnetic damping176 can affect a
wide range of behaviors in a ferromagnet. For instance, spin-orbit torques determine how
efficient the injected spin current can drive magnetization precession or switch the
magnetization. DMI affects how domain walls are configured and how fast domain wall can be
driven by a current177. Magnetic damping on the other hand represents an energy loss mechanism
which determines dynamic properties like spin wave propagation and relaxation timescale for
magnetic precession. An effective mean to gate these fundamental properties would allow a wide
range of device behaviors to be accessible to control which would significantly boost the
functionalities and performance of current magnetic devices.
A promising approach to voltage gating is through voltage-induced ionic modulation of magnetic
interfaces, dubbed the magneto-ionic effect58–65. In this approach, a gate voltage drives ionic
migration and electrochemical reactions which in turn changes the properties of the magnetic
layer. Two ions that have been studied are oxygen ions and protons. In the former case, it has
been shown that voltage-induced Co oxidation in a Pt/Co/GdOx can remove its magnetization,
upon which a voltage of the opposite polarity can reverse the effect. For the latter case, voltage-
induced hydrogen accumulation near the metallic Co layer in a Pt/Co/GdOx structure changes its
interfacial magnetic anisotropy which results in the magnetization rotation from out-of-plane
state to in-plane (chapter 5). Thus far, the main interest has been to modulate magnetic
anisotropy in order to reduce barrier for switching. However, one should expect ions to change
other interfacial magnetic properties too.
In this chapter, we demonstrate voltage-induced ionic gating of magnetic damping and spin-orbit
torques in addition to magnetic anisotropy, probed by spin-torque ferromagnetic resonance39,40.
128
Spin torque ferromagnetic resonance is a very useful technique where a current-induced torque
drives magnetic resonance which manifests itself electrically through a change in resistance. This
can then be measured as a DC mixing voltage, 𝑉𝑚𝑖𝑥 which can be expressed as39,136
𝑉𝑚𝑖𝑥 = 𝑆𝑊2
(𝐻− 𝐻𝐹𝑀𝑅)2+𝑊2+ 𝐴
𝑊(𝐻− 𝐻𝐹𝑀𝑅)
(𝐻− 𝐻𝐹𝑀𝑅)2+𝑊2 Equation 6.1
Here, 𝑆 and 𝐴 are the amplitudes of the symmetric and antisymmetric Lorentzian components of
the resonance peak respectively, 𝑊 and 𝐻𝐹𝑀𝑅 are the width and position of both these
Lorentzian components.
129
6.1 Experimental Methods
Experiments focus on Ta(4nm)/Pt(3nm)/Co(tCo)/GdOx (36nm)/Au(3nm) structures which were
fabricated on thermally oxidized Si (100) substrates using magnetron sputtering at room
temperature and 3mTorr of Ar pressure. The metal layers were grown by DC sputtering. The
GdOx layer was deposited either using reactive sputtering with PO2 of 0.07mTorr or RF
sputtering with PO2 of 0.7mTorr O2. The base pressure during sputtering is 5x10-7 Torr. Figure
6.1(a) shows the device and measurement schematic for the gated ST-FMR device while figure
6.1(b) shows an optical micrograph from a top-down view. The waveguide has dimension of
10µm x 5µm, 20µm x 10µm, 10µm x 10µm, and 20µm x 5µm. The in-plane field, H is applied
45º to the current line. 15dbm of power at frequencies, f ranging from 5.5GHz to 12GHz is
supplied by a signal generator to the waveguide while 𝑉𝑚𝑖𝑥 is measured by a lock-in amplifier.
The modulation amplitude and frequency from the lock-in amplifier (to the signal generator) are
1V and 2kHz respectively. Figure 6.1(c) shows a set of FMR spectra measured at different f for
a tCo =4.8nm device. These spectra are fitted with two Lorentzian (symmetric and
antisymmetric) components according to equation 6.1. HDemag is then obtained from a fit of 𝐻𝐹𝑀𝑅
to the Kittel formula given by equation 6.2.
2𝜋𝑓 = 𝛾 √𝐻𝐹𝑀𝑅(𝐻𝐹𝑀𝑅 + 𝐻𝐷𝑒𝑚𝑎𝑔) Equation 6.2
A sample fit is shown in figure 6.1(d). Similarly, the widths, W of the FMR spectra at different
frequency can also be fitted to equation 6.3 to obtain the damping constant, 𝛼. The fit is shown in
figure 6.1(e).
𝑊 = 2𝜋𝑓
𝛾𝛼 Equation 6.3
130
For in situ probing of the magnetic properties during voltage gating, 𝑉𝑚𝑖𝑥 is continuously
measured at 8GHz(15dbm) of injected power while a constant gate bias VG is applied. The FMR
spectrum for each acquisition cycle is then fitted with the two Lorentzian components (equation
6.1) so that the change in magnetic parameters can be plotted as a function of time. For fittings
which require multiple frequencies such as the damping constant, the spectra for different
frequencies were acquired during specific times by setting the gate bias to a holding voltage.
Figure 6.1. Voltage gating of magnetic properties probed using spin-torque ferromagnetic
resonance (ST-FMR). (a) Device schematic of a 10µm x5µm gated device. 𝑉𝑚𝑖𝑥 is measured
using a lock-in amplifier at modulation frequency of 2kHz. (b) Optical micrograph of the gated
device for ST-FMR measurement. The magnetic field is applied at 45° to the current line. The
contact pad at the top left corner is connected to the top gate. (c) ST-FMR spectra of a
Pt(3nm)/Co(4.8nm)/GdOx (36nm)/Au(3nm) device at different frequencies. All the spectra are
fitted to equation 6.1 to obtain S,A, 𝑊, and 𝐻𝐹𝑀𝑅 (d) Fit of 𝐻𝐹𝑀𝑅 to the Kittel equation
(equation 6.2) to obtain HDemag. (e) Fit of W to equation 6.3 to obtain α.
131
6.2 Spin Torque Ferromagnetic
Resonance to Probe Voltage Gating of
Spin Orbit Torque and Magnetic Damping
Figure 6.2(a) shows the ST-FMR spectra at 8GHz for different Co thicknesses. The spectra show
three trends with increasing Co thickness (1) 𝑊 decreases, (2) HDemag shifts to smaller field
values, and (3) there is an increasing contribution from an antisymmetric Lorentzian component.
To quantify these trends, the Kittel formula (Equation 6.2) is first fitted for different frequencies
at each Co thickness to obtain 𝐻𝐷𝑒𝑚𝑎𝑔. 𝐻𝐷𝑒𝑚𝑎𝑔 as a function of Co thickness is shown in figure
6.2(b). At small thicknesses, 𝐻𝐷𝑒𝑚𝑎𝑔 is small at ~3kOe due to the large interfacial perpendicular
anisotropy at the Pt/Co and Co/GdOx interfaces. At larger thicknesses (>6nm), the demagnetizing
field approaches ~14kOe which is equivalent to 4πMCo. This indicates shape anisotropy is much
larger than the interfacial anisotropy and dominates at such thicknesses. Figure 6.2(c) shows the
trend in magnetic damping constants as a function of Co thickness. The data shows decreasing
damping constant with increasing Co thickness, consistent with previous literature. To get the
spin orbit torque quantities, the values obtained from the Lorentzian fits are substituted in
equation 6.4 to obtain 𝜉𝐹𝑀𝑅.
𝜉𝐹𝑀𝑅 = 𝑆
𝐴
𝑒𝜇0𝑀𝑆𝑡𝐶𝑜𝑡𝑃𝑡
ℏ√1 +
4𝜋𝑀𝑆
𝐻𝐹𝑀𝑅 Equation 6.4
Here 𝜉𝐹𝑀𝑅 is the spin hall angle in the absence of a field-like torque from the spin current, 𝑀𝑆 is
the saturation magnetization, 𝑡𝐶𝑜 and 𝑡𝑃𝑡 are the Co and Pt thicknesses respectively, and 𝑒 and ℏ
are the electron charge and Planck’s constant. By plotting the 1/𝜉𝐹𝑀𝑅 versus 1/tCo (figure
132
6.2(d)),we can then get damping-like (𝜉𝐷𝐿) and field like (𝜉𝐹𝐿) torque coefficients from equation
6.5.
1
𝜉𝐹𝑀𝑅=
1
𝜉𝐷𝐿 (1 +
ℏ
𝑒)
𝜉𝐹𝐿
𝜇0𝑀𝑆𝑡𝐶𝑜𝑡𝑃𝑡 Equation 6.5
At Co thickness of less than ~2.5nm, 1/𝜉𝐹𝑀𝑅 is negative because the field-like torque is
dominated by the torque from the spin-current whereas at >3nm, 1
𝜉𝐹𝑀𝑅 is positive because it is
dominated by the Oersted field. 𝜉𝐷𝐿 obtained from the intercept of the linear fit is ~0.03,
consistent with literature178. A plot of 𝑆
𝐴 versus 𝑡𝐶𝑜 is also shown in figure 6.1(e) for reference.
In this case, 𝑆
𝐴 shows the ratio of the total damping-like torque to the total field-like torque
regardless of whether the torques are due to spin current or the Oersted field.
Figure 6.2. Thickness dependence of HDemag, α, and 1/𝝃𝑭𝑴𝑹. (a) ST-FMR spectra of
Pt(3nm)/Co(tCo)/GdOx (36nm)/Au(3nm) device at 8GHz. (b) HDemag as a function of tCo. (c) α as a
function of tCo. (d) 1/𝜉𝐹𝑀𝑅 as a function of 1/tCo (e)S/A ratio as a function of tCo.
133
Figure 6.3(a) shows the ST-FMR spectra of a Pt/Co/GdOx/Au device with 1.2nm thick Co at
different frequencies. At this thickness, the spectra are dominated by a symmetric Lorentzian
component, indicating a large damping-like torque relative to the field-like torque. Figure 6.3(b)
shows the ST-FMR spectra after VG = +4V has been applied for 8200s while figure 6.3(c) shows
the ST-FMR spectra after VG is set to 0V for >400000s. Quick inspection shows a clear change
in resonance peak positions, widths and shapes. Figure 6.3(d)-(f) show the fits of the extracted
Lorentzian parameters to equation 6.2, 6.3, and 6.4 to obtain to obtain HDemag, α, and 𝜉𝐹𝑀𝑅
respectively. At VG of +4V a large increase in demagnetizing field from 3.5kOe to 8kOe is
observed due to large increase in in-plane magnetic anisotropy. Similarly, the magnitude of 𝜉𝐹𝑀𝑅
also increases from -0.040 to -0.084. On the other hand, the magnetic damping constant, α
decreases from 0.090 to 0.037. These changes are largely reversible when VG is set to 0V, where
HDemag, α, and 𝜉𝐹𝑀𝑅 revert back to values of 3.7kOe, 0.088, and -0.042 respectively. Figure 6.2
(b) –(d) shows the corresponding time series data of three magnetic parameters extracted from
the ST-FMR spectra under the gate biases. The black data points correspond to values extracted
by directly inserting the resonance peak field at 8GHz into equations 6.2 and 6.4 while the red
data points correspond to values obtained from fits to 4 different frequencies (5.4GHz, 6GHz,
7GHz, and 8GHz). During the measurement of the spectra for the 4 frequencies, a holding
voltage of +3V was used. Regarding the mechanism which leads to the modulation of the
magnetic anisotropy, damping, and spin torques, we propose that there are two electrochemical
processes which are taking place. The first process is voltage-induced hydrogen generation and
removal. In the previous chapter, we proposed that hydrogen is primarily accumulated at the
Co/GdOx interface. However, since magnetic damping in Co/Pt heterostructures primarily
originate from the Co/Pt interface, it is very likely that some hydrogen also penetrates the Co and
134
disrupts this interface. The second process is voltage-induced oxidation and reduction of the
magnetic Co layer. In this case, the device starts out with some oxidized dead layer and a
positive bias reduces it to magnetic Co. When the VG is next set to 0V, the reduced Co gradually
reoxidizes again. The redox of Co modulates the magnetic parameters primarily by changing the
effective thickness of the Co layer.
Figure 6.3. Voltage gating of HDemag, α, and 𝝃𝑭𝑴𝑹. (a)-(c) ST-FMR spectra of a
Pt(3nm)/Co(1.2nm)/GdOx (36nm)/Au(3nm) device in virgin state (a), after VG = +4V for 8200s
(b) and after setting VG back to +0V for 410000s. (d)-(f) Fit to the Kittel equation 6.1(d),
equation 6.2(e) and equation 6.3(f) to obtain HDemag, α, and 𝜉𝐹𝑀𝑅 respectively.
135
Figure 6.4. Time series data for tCo = 1.2nm device. (a) HDemag ,(b) α, and (c) 𝜉𝐹𝑀𝑅 at VG =
+4V and 0V for device shown in figure 6.3
To verify this conclusion, we also performed voltage gating on a Pt/Co/GdOx/Au structure with
𝑡𝐶𝑜=4.2nm. In this case, under a positive bias, there is minimal change in the demagnetizing field
(magnetic anisotropy) and damping constant, while there is 30% reduction in the torque ratio. If
we look at figure 6.1, while the increase in the effective thickness of the Co layer due to
reduction of any oxidized layer only lead to small changes in damping constant at 4nm, it should
lead to relatively large change in demagnetizing field. The fact that this is not observed strongly
suggests that the large changes observed in the 1.2nm Co device is primarily due to hydrogen-
136
induced modulation of the magnetic interfaces. In the 4.2nm Co device, the Co thickness is too
large for hydrogen to penetrate it, as a result there is minimal change in the magnetic anisotropy
and damping. As for the spin orbit torques, it has been shown that the strength of the field-like
torque depends crucially on the ferromagnet/oxide interface; as a result hydrogen accumulation
at the Co/GdOx can alter the ratio of the different torques even if the Pt/Co interface is not
affected.
Figure 6.5. Time series data for tCo = 4.2nm device. (a) HDemag ,(b) α, and (c) 𝜉𝐹𝑀𝑅 at VG =
+4V.
137
In conclusion, we have shown that magnetic properties which manifest themselves at interfaces
can be significantly modulated by ions driven using a gate voltage. These properties include
magnetic anisotropy, magnetic damping, and spin-orbit torques. Dynamic modulation of these
properties represent an important advancement in the field of voltage control of magnetism, and
can enable important new functionalities and device design.
138
Chapter 7: Voltage-
induced Magneto-
Ionic Effect in
Pt/Co/MOx
Heterostructure (M=
Gd, Y, Zr, and Ta)
139
In this chapter, we demonstrate the generality of voltage-induced magneto-ionic effect in
Pt/Co/MOx structure (M = Gd,Y,Zr, Ta) probed magnetically by magneto-optical Kerr effect. For
both rates, ROxide of voltage-induced Co oxidation and reduction in the Pt/Co/MOx structure,
there is a trend which follows the order: RGdOx > RYOx > RZrOx > RTaOx. RGdOx is very similar to
RYOx while RZrOx is ~ 2 orders of magnitude smaller. For the TaOx device, no redox of Co is
observed under an applied bias throughout the experiment for 24hours. The difference in
magneto-ionic rates for the four oxides is attributed to the difference in their proton
conductivities. This shows that significant improvement in speed of magneto-ionic switching can
be achieved through careful selection of material as the solid state electrolyte.
Electrical gating of magnetism has shown tremendous versatility for dynamic control of
various magnetic properties. Pioneering work has demonstrated reversible control of magnetic
anisotropy using an electric field at a metal/metal oxide interface12,51,52. However, such electronic
effect is small due to Coulombic screening in a metal. An alternative approach called magneto-
ionic switching60–63,66 exploits mobile ionic species in a solid state heterostructure to modulate
magnetic anisotropy with an efficiency of up to 5000fJ/Vm60,61 compared to ~10fJ/Vm in electric
field gating12. However, its speed is slower because it relies on ionic motion instead of electron
flow. To improve this speed, it is necessary to find an optimal metal oxide dielectric which is
simultaneously an excellent electrolyte to allow for fast ionic transport of the active species.
Thus far, studies on magneto-ionic switching has mostly focused on thin film GdOx60–62as
the electrolyte in a heavy metal/ferromagnet/dielectric (electrolyte) structure. The reliance on
only one material makes it difficult to identify trends or desirable characteristics of the
electrolyte. In this chapter, we introduce three new materials, namely YOx, ZrOx, and TaOx and
compare their rates of voltage-induced magneto-ionic effect to that of GdOx in a Pt/Co/MOx (M
140
= Gd, Y, Zr, and Ta) structures. We chose these materials for three reasons. 1) They are binary
oxides which allow for simple comparison to GdOx144, 2) they cover a wide spectrum in terms of
the similarity of their crystal structures179–183 to GdOx , and 3) they are high-k dielectric179 with
good breakdown characteristics.
141
7.1 Experimental Methods
Experiments focus on Ta(4nm)/Pt(3nm)/Co(0.9nm)/MOx (10nm) (M = Gd,Y,Zr, and Ta)
structures which are fabricated by DC magnetron sputtering on thermally oxidized Si substrates.
All layers are sputtered at argon pressure,PAr of 3mT and base pressure of 5 x 10-7 torr. For the
Co layer, the initial state is either metallic or oxidized. The oxidized state is sputtered with an
additional oxygen pressure PO2 of 0.07mT. For the MOx layer, reactive sputtering was performed
at PO2 of 0.06mT, followed by PO2 of 0.7mT during the last 2 minutes of sputtering. This last step
acts like an additional plasma oxidation step. To allow for electrical contact, a portion of the
Ta/Pt layers (bottom electrode) is uncovered by the MOx layer while round Au (3nm) electrodes
of 200µm diameter (top electrode) is sputtered on top of the MOx layer using shadow mask
lithography (Figure 7.1). The bottom Pt is grounded while the gate voltage, VG is applied to the
top Au by contacting the edge of the electrode with a CuBe probe. All magnetic measurements
are done using magneto-optical Kerr effect (MOKE) polarimeter with a 1mW laser at
wavelength of 655nm. The laser is focused at the middle of the Au electrode.
Figure 7.1. Measurement schematic for in situ probe of Co magnetization using magneto-
optical Kerr polarimetry.
142
7.2 Rate of Voltage-Induced Magnetic
Modulation at Positive Bias
Figure 7.2(a) shows the change in ratio of the remnant (Mr) to saturation magnetization
(Ms) obtained from hysteresis loops as a gate voltage, VG = +3V is applied to Pt/CoO/MOx
structures. In its virgin state, Mr /Ms ~0 because the Co layer is oxidized. As VG = +3V is applied,
CoO layer is reduced to Co leading to an increase and eventual plateau in Mr /Ms of the GdOx ,
YOx and ZrOx structures, which is consistent with literature60,61. As VG is applied for longer,
interestingly, the Mr /Ms values decreases. For the TaOx structure, no discernable reduction of the
Co layer is observed over a period of 24 hours. The data shows a trend in the rates of Co
reduction, Roxide which proceeds in the following order: RGdOx > RYOx > RZrOx > RTaOx .
To identify the magnetization states during application of VG = +3V, we plotted the
hysteresis loops at different times of the experiment in figure 7.2(a). Figure 7.2(b)-(e) show the
hysteresis loops for GdOx, YOx , ZrOx and TaOx devices respectively. At t =0s, in all cases, there
is no magnetic signal. As VG = +3V is applied, we observe a similar trend in the GdOx, YOx, and
ZrOx devices, where upon complete reduction of CoO to Co, there is large perpendicular
magnetic anisotropy (PMA) in the layer. As VG = +3V is applied for longer, the hysteresis loops
evolve into a hard-axis loop, which corresponds to the rotation of the magnetization in-plane.
The observed reduction in Mr /Ms in figure 7.2(a) originates from this rotation and not from
vanished magnetization. This rotation is due to hydrogen accumulation at the Co/MOx interface
(chapter 4) as a positive VG is applied beyond reduction of CoO to Co. For the TaOx device,
figure 7.2(e) confirms that the Co layer remains oxidized throughout the experiment.
143
Figure 7.2. Voltage-induced magneto-ionics at positive bias. (a) Rates of change in Mr / Ms at
VG = +3V for different Pt/Co/MOx /Au structures, where M = Gd, Y, Zr, and Ta. (b)-(d) MOKE
hysteresis loops at different times in (a), corresponding to M = (b) Gd, (c)Y, (d) Zr, and (e) Ta.
Figure 7.3(a) shows the change in Mr /Ms for the GdOx, YOx, and ZrOx devices at VG = 0V after
VG = +3V has rotated the magnetization in plane. The data shows relaxation of the magnetization
back to an out-of-plane state, with the rate of relaxation for the four oxides following the same
144
trend as the Co reduction rate. Figure 7.3(b) shows a representative set of hysteresis loops for the
GdOx, YOx and ZrOx devices before and after the relaxation, clearly indicating a rotation of the
magnetization from in-plane to out-of-plane. This rotation is due to removal of hydrogen from
the bottom Co/MOx interface as the hydrogen forms protons and is transported to the top Au
electrode where it recombines with O2 to form water through the oxygen reduction reaction.
Figure 7.3. Voltage induced magneto-ionics at 0V after positive bias gating. (a) Rate of
change in Mr / Ms for the Pt/Co/MOx structures at VG = 0V after VG = +3V has been applied. (b)
Hysteresis loops before and after VG is set to 0V.
145
7.3 Rate of Voltage-Induced Magnetic
Modulation at Negative Bias
Figure 7.4(a) shows the change in Mr /Ms obtained from hysteresis loops as a negative
bias, VG = -2V is applied to the Pt/Co/MOx structures. All four devices exhibit large PMA in their
initial state. As VG = -2V is applied, we observe a gradual decrease in Mr /Ms of the GdOx, YOx
and ZrOx devices to ~0. Unlike the situation at VG = +3V where the reduction in Mr /Ms
corresponds to rotation of the magnetization in-plane, in this case Mr /Ms decays due to the
oxidation of Co to CoO60,61. This is confirmed by hysteresis loops in figure 7.4(b) to (d)
corresponding to GdOx, YOx and ZrOx devices respectively where instead of a hard-axis loop,
we observe no magnetic signal even at a maximum field of ~500Oe after a prolonged negative
VG. For the TaOx device, we do not observe any oxidation of Co after 24hours. The rates of Co
oxidation for the different oxides again follow the same trend as CoO reduction in figure 7.2(a).
146
Figure 7.4. Voltage induced magneto-ionics at negative bias. (a) Rates of change in Mr / Ms at
VG = -2V for different Pt/Co/MOx /Au structures. (b)-(e) MOKE hysteresis loops for
Pt/Co/MOx/Au structure at different times in (a), where M = (b) Gd, (c)Y, (d) Zr, and (e) Ta.
The metal oxide layer plays a very important role in the overall magneto-ionic process. For
voltage-induced Co oxidation, the difference in rates can be attributed to the basicity of the
different oxides. GdOx and YOx have very high basicity compared to ZrOx and TaOx, and form
hydroxides143,144 when they are in contact with H2O. During voltage-induced Co oxidation, it is
the H2O stored in the form of hydroxides that oxidizes the Co layer, as shown in chapter 4. As a
147
result, the Co oxidation rate is much higher in GdOx and YOx compared to ZrOx and TaOx.
Besides acting as a store for H2O, the metal oxide layer also acts as the electrolyte for proton
transport. In fact, the trend in the rates of all voltage-induced magneto-ionic effects tracks the
proton conductivities of the different oxides studied81,82,137. This is consistent since proton
transport has to take place for all electrochemical reactions to happen. GdOx and YOx have high
proton conductivities due to large concentration of carriers and high carrier mobility. On the
other hand, ZrOx and TaOx have low proton conductivities; as a result their rates of voltage-
induced magneto-ionic effect are very slow.
In conclusion, we have demonstrated that the choice of metal oxide adjacent to the
ferromagnet affect substantially the rate of voltage- induced magneto-ionic effect. Through
judicious choice of materials, we can potentially reduce the timescale of voltage gating by a few
orders of magnitude, making it viable for practical low power memory applications.
148
Chapter 8: Room
Temperature
Reversible Solid Oxide
Fuel Cell
149
The scientific community has approached miniaturization of power sources for
electronics by focusing mostly on solid state Li-ion microbatteries184–188. Another power storage
and generation system: the solid oxide fuel cells, produce power through recombination of
hydrogen fuel with oxygen to form water. However, the operation of SOFCs have largely been
limited to very high temperature (>700C) due to sluggish kinetics of charge transfer and mass
transport. In addition, SOFCs were mainly studied in large bulky form factors with gaseous
hydrogen as fuel; these are specifically customized for large scale power generation system, not
microelectronics. In fact, most of the gaseous hydrogen fuel is generated by a separate methane
reforming cell, and the fuel is stored in a gas tank. In this chapter, we demonstrate a miniaturized
room temperature reversible solid oxide fuel cell for small-scale systems such as
microelectronics. The cell can be “charged” by splitting water in ambient111,112. It can then
produce power when needed through the water recombination reaction113. Hydrogen is stored in
a thin GdOx film where good scalability in terms of area and thickness of the GdOx storage layer
is observed. This is a significant because we demonstrate for the first time the miniaturization of
a solid oxide fuel cell for room temperature operation. This opens up a new range of applications
such as microelectronics where solid oxide cells can be employed. The use of a rechargeable
hydrogen cell also puts forth a new alternative to lithium-ion battery.
150
8.1 Experimental Methods
Experiments focus on Ta(4nm)/Pt(10nm)/GdOx (tGdOx nm) /Au cross bar structures (figure
8.1) which are fabricated on thermally oxidized Si substrates. The Ta, Pt and Au layers are
sputtered using DC magnetron sputtering at argon pressure,PAr between 3mTorr and 3.5mTorr.
The GdOx layer is RF sputtered at PAr of 3mTorr and oxygen partial pressure, PO2 of 0.7mTorr.
The base pressure in all cases is ~5 x 10-7 torr. CuBe probes are used for electrical contact. For
electrical characterization, a Keithley 6430 source meter unit is used for charging and to measure
the discharge and power density curves. All measurements are performed in a CPX-VF probe
station using mechanically compliant CuBe probes. Experiments under different gas environments
were performed by backfilling the chamber with either O2 gas (99.999% purity) or N2 gas
(99.999% purity). Humidity was introduced into the N2 gas flow by bubbling through water.
Ambient condition at 25°C corresponds to 12mT of H2O partial pressure respectively. All
experiments were performed at room temperature.
Figure 8.1. Device schematic of cross bar solid oxide cell.
As mentioned in chapter 4, the Pt/Co/GdOx/Au cell stores hydrogen when VG > 0 is applied to
the top Au gate. Figure 8.2 summarizes the charging, storage, and discharging processes for a
Pt/GdOx/Au cell. In this chapter, we are primarily interested in the capacity and power output of
151
the cell for energy storage applications. During charging, H2O is split into O2 and H+ through the
oxygen evolution reaction. The H+ which is produced is driven to the bottom Pt electrode,
reduced to neutral hydrogen, and stored in GdOx layer. During discharge, the stored hydrogen
provides the electrons which drives the load in the external circuit.
Figure 8.2. Reaction schematic of reversible solid oxide fuel cell.
152
8.2 Proton Conductivity of GdOx
To characterize the lower bound for proton conductivity of the GdOx layer, we measured
its dc conductivity. Figure 8.3(a) and (b) show the total current flow at VG = +2.5V for a
hydrated GdOx in ambient and in vacuum respectively while figure 8.3(c) and (d) show the same
measurements for a non-hydrated GdOx. We observe a large difference in dc current flow
between the hydrated GdOx cell in ambient and in vacuum indicating that a majority of current
flow is due to ionic current alone. We also observe that the current is significantly larger in a
hydrated GdOx compared to a non-hydrated GdOx in ambient, indicating a much larger
conductivity in the hydrated GdOx cell. Specifically, it is the ionic conductivity which is much
larger in the hydrated vs the non-hydrated GdOx, as will be shown later.
Figure 8.3. Charging current of solid oxide cell at VG = +2.5V at different temperatures (a)
Hydrated GdOx in ambient. (b) Hydrated GdOx in vacuum. (c) Non-hydrated GdOx in ambient.
(d) Non-hydrated GdOx in vacuum.
153
By subtracting the current that flows in a hydrated GdOx sample in vacuum from the
current flow in ambient, we obtain the current flow due to reactions and proton transport alone.
This is because in vacuum, the water splitting reactions depicted in equation 2.6 and 2.7 cannot
take place, whereas in ambient, they do. Figure 8.4 shows the conductivity calculated from this
current flow as a function of temperature. At room temperature, the total conductivity is between
10-12 and 10-11 S/cm. This value fits well with the extrapolated proton conductivity of doped-
GdOx in literature82,92 of between 10-11and 10-10 S/cm. Note however that the ionic conductivity
calculated from dc current include resistance contributions from reactions at both electrodes; ie
the oxygen evolution and hydrogen evolution reactions (equation 2.6 and 2.7). Hence, the
conductivity value given here only represents a lower bound to actual proton conductivity of the
oxide. By linear fitting the conductivity values, we obtained an activation energy of 0.23eV.
While this activation energy is not a true representation of the proton mobility in the oxides
(since it includes several other contributions), it is still interesting to note that the value is very
low compared to the activation energy for proton transport in other oxides81,82,92,137.
Figure 8.4. Conductivity of solid oxide cell.
154
8.3 Cell Performance and Scalability
Figure 8.5 summarizes the performance of the cell. At VG = +2.5V, the hydrated GdOx
cell can be charged to 100% capacity of 1.4µAh/cm2 in ~60min while the non-hydrated cell does
not show any stored charge up to 80min of charging time. The amount of charge stored in the
hydrated cell is more than 3 orders of magnitude larger than what is calculated from the expected
capacitance of the GdOx layer. The absence of any stored charge in the non-hydrated cell shows
that the small current flow in figure 8.2 is primarily due to low proton conductivity of the non-
hydrated oxide. This is consistent with the findings in chapter 4,5, and 6 which show voltage-
induced magnetic change in a Pt/Co/GdOx heterostructures are only observed in hydrated GdOx
devices. Figure 8.5(b) shows the discharge curve of the GdO at ~4µA/cm2 for the different
charging times. We can observe a plateau at ~1V which corresponds well to the thermodynamic
potential for the water recombination at room temperature. Figure 8.5(c) shows power density
curves for the Pt/GdOx(40nm)/Au cells at different charging times where a maximum power
density of ~50µW/cm2 was achieved at current density of 80µA/cm2. Figure 8.5(d) shows the
cyclability of a 40nm thick cell where charging was done at VG = +3V for 10min. A constant
capacity of ~0.5µAh/cm2 was obtained with no significant decrease in capacity for 50cycles.
Figure 8.5(e) shows the corresponding discharge curves for 5 out of the 50 cycles, with
reproducible behavior throughout the 50 cycles. Figure 8.5(f) shows the efficiency of the
charging process at VG = +2.5V. The cell exhibits efficiency of ~10% at room temperature. The
efficiency values are calculated as the ratio of the stored energy to the input energy. The stored
energy is obtained by integrating the area under the discharge curves, while the input energy is
simply the product of the charging voltage, charging current, and charging time. In the GdOx
155
cell, possible losses include large overpotentials from oxygen reduction reaction113, ohmic losses
and electronic leakage across the oxide.
Figure 8.5. Performance of solid oxide cell. (a) Capacity as function of charging time. (b)
Discharge curves. (c) Power density curves (d) Cyclabililty. (e) Discharge curves at different
cycles in (d). (f) Cell efficiency.
In figure 8.6 (a)-(c), we show that the capacity and power density of the cell scale with
the thickness of the GdOx. This thickness scalability confirms that the dominant charge storage
in the device is electrochemical in nature and not capacitive. It also shows that the charge in the
form of hydrogen is stored across the thickness of the GdOx instead of at the interfaces. Note
however that maximum cell voltage for a 80nm thick cell is only ~ 0.78V compared to ~1V in
thinner cells. This is likely due to larger Ohmic overpotential for proton transport during
discharge of hydrogen stored further away from the Pt interface. In figure 8.6(d)-(e), we also
demonstrate that the capacity scales with cell area. The largest capacity of 1.8x10-3µAh is
156
achieved with an area of 4mm2. In these GdOx thin film however, the exact charge storage
mechanism is still not well understood; one possibility is hydrogen is stored in the crystal
structure of GdOx as interstitials.
Figure 8.6. Scalability of solid oxide cell. (a) Capacity and peak power density as function of
GdOx thickness. (b) Discharge curves of cell with different GdOx thicknesses. (c) Power density
curves of cell with different GdOx thicknesses. (d) Capacity as function of cell area for a 40nm
thick GdOx. (e) Discharge curves of cell for different cell areas
In chapter 4, we showed that voltage-induced reduction of CoO in a Pt/CoO/GdOx/Au
heterostructure at VG > 0 (applied to Au) is due to hydrogen. This was done by probing the
magnetic signal of the CoO layer using Hall magnetometry; when charged in vacuum, the
magnetic signal remain absent due to . On the other hand, when the cell is charged in ambient
(50% humidity), there is large magnetic signal indicating the reduction of CoO to Co by
hydrogen generated at the bottom Pt electrode. Here, we again show that the stored charge is
hydrogen by comparing the capacity of the cell charged in ambient vs in vacuum. While the
157
maximum capacity is 1.4µAh/cm2 when charged in ambient (figure 8.3), no measurable charge is
detected after charging in vacuum. This is expected since the calculated capacitance for the
GdOx cell dimensions (1.23x10-3cm2) is ~0.5nF assuming no hydrogen storage. When charged
at +2.5V in vacuum, the total stored charge is only ~1nC, which is too small to be measured.
Figure 8.7 also shows the power density curves for a 1mm2, 20nm thick GdOx cell charged at
+3V for 30s and discharged under different atmospheric conditions. The results are again
consistent with the hypothesis that H2O acts as the source of hydrogen for energy storage in the
cell.
Figure 8.7. Power density of GdOx cell in different atmospheric conditions. The charging
and discharging conditions are listed in order.
158
8.4 Gating of Magnetism using Built-in
Voltage
As discussed in chapter 5, in a Pt/Co/GdOx/Au heterostructure, a positive gate voltage,
VG applied to the top Au electrode drives the Co magnetization from out-of-plane to in-plane due
to the generated hydrogen. When VG is next set to 0V (short circuit), the magnetization reverts
back to an out-of-plane state due to removal of hydrogen from the Co interface. If the VG is
instead set to open circuit after +3V, the magnetization remains in the in-plane state as the
hydrogen remain “trapped” at the Co interface due to the absence of an external circuit through
which it can donate its electron.
In this section, we further show that the Co can be switched between a high and a low
perpendicular anisotropy state by toggling between short circuit (SC) and open circuit (OC). In
the experiment, we first charge the Pt(4nm)/Co(0.9nm)/GdOx(10nm)/Au(3nm) device at VG =
+3V for 600s before alternating between short and open circuit(figure 8.8). Figure 8.9 shows the
switching cycles of Mr/Ms and coercivity between the two conditions. From the data, we can
clearly observe an increase in Mr/Ms and coercivity at short circuit which corresponds to an
increased perpendicular anisotropy while the opposite is observed for open circuit. The
corresponding polar MOKE hysteresis loops in the first cycle are also shown. The magnitude of
change is largest during the first cycle, and slowly decreases over cycles. This is because during
the first cycle immediately after VG = +3V, the amount of hydrogen stored in GdOx is maximum.
With increasing short circuit duration, the amount of charge stored is gradually decreased due to
the removal of hydrogen through water recombination reaction. The toggling between high and
159
low perpendicular anisotropy states happen in the absence of an external power source; this
means it is the internal voltage in the GdOx which is driving this magnetic change. At short
circuit, the voltage difference between the top and bottom electrode is zero while at open circuit,
the voltage difference is given by the potential for the water recombination reaction. The detailed
mechanism which drives the magnetic modulation is however not well understood at this point.
Figure 8.8. Experimental sequence for toggling Co magnetization between short circuit and
open circuit
Figure 8.9. Modulation of magnetic anisotropy at open circuit (OC) and short circuit (SC)
conditions. (a) Switching cycles of coercivity at OC and SC. (b) Switching cycles of Mr/Ms at
OC and SC. (c) Polar MOKE hysteresis loops corresponding to the first cycle.
160
In conclusion, we demonstrated a room temperature miniaturized reversible solid oxide fuel cell
based on hydrogen storage in a thin film GdOx. Hydrogen is sourced from ambient through the
water splitting reaction during charging while power is produced during discharge through the
water recombination reaction. For a self-contained energy storage unit, we propose the
integration of a hydrogen storage unit on top of the GdOx which will enable implementation in
practical devices. We also showed that by inserting a thin magnetic layer beneath the GdOx, the
voltage of the cell can be used to modulate interfacial magnetic anisotropy by toggling between
short and open circuit. This work demonstrates the viability of oxide fuel cell miniaturization for
solid state device applications.
161
Chapter 9: Voltage
Gating of Optical
Properties
162
Besides gating of magnetic properties, we also observe reversible color change in a
Pt/GdOx/Au heterostructure induced by a gate voltage. Electrochromic films present an attractive
way to build energy efficient smart windows, coatings and mirrors189. These films exhibit a
change in their optical transmission and absorption under an applied voltage. Typically,
electrochromic films consists of three layers: an ion storage layer, an electrolyte and an active
layer such as WO3 or NiOx189–191. The three layers are then sandwiched by transparent
conductors. When voltage is applied across the device, ions are pumped into and out of the active
layer through the electrolyte, modulating the optical characteristics. Due to the trilayer design,
the complexity and cost of such films have limited their broader application, and for certain
applications, the optical switching speed of common electrochromic materials is still
prohibitively low.
In this chapter, we describe a thin-film reflective electrochromic device consisting of a
single ultra-thin layer of GdOx that acts simultaneously as electrolyte and active layer and operates
without a separate ion storage layer. We achieve gate-voltage-induced reversible modulations in
reflectivity of ~10% and switching time down to 20 ms at room temperature. Measurements of
the change in reflectivity as a function of inserted charge192 show a figure of merit of ~37.6 cm2/C
for a 20nm thick film, comparable to conventional electrochromic materials, making this simple
structure a viable alternative to more complex multilayer stacks. Systematic experiments
demonstrate a strong dependence of device behavior on the top electrode thickness, which
indicates that the rate-limiting step is the water splitting reaction at the top electrode112 (chapter
5). Because the GdOx layer serves both as electrolyte and active layer, and the atmosphere serves
as a hydrogen reservoir, this eliminates the need for an ion storage layer and greatly simplifies the
device architecture.
163
9.1 Experimental Methods
Our work focuses on Ta(4 nm)/Pt(10 nm)/GdOx(tox)/Au(tAu) sputter-deposited on a Si
substrate with 50nm of native oxide. Here tox and tAu represent the thickness of the corresponding
layers. Figure 9.1(a) shows a schematic of the device structure. The metal layers were deposited
by dc magnetron sputtering at room temperature with an Ar pressure of 3 mTorr. The GdOx layer
was RF sputtered from a stoichiometric Gd2O3 target at room temperature at a rate of ~ 0.2nm/min
in an oxygen partial pressure of 0.7mT. The Ta layer acts as an adhesion layer while the Pt layer
acts as the back electrode and serves as a reflecting mirror. The GdOx is the electrolyte and active
layer whose optical properties change depending on the applied voltage. On top of these layers,
Au electrodes with a diameter of ~200µm and with tAu=3 nm except where noted, were patterned
using shadow mask lithography to serve as the top gate. At this thickness, the Au layer is semi-
transparent and porous61. Experiments were performed by contacting the top electrode with a CuBe
microprobe and grounding the back electrode (figure 9.1). The optical reflectivity was monitored
in situ with a wide-field microscope and CCD camera, as well as with a with a focused laser spot
(655 nm wavelength) whose reflected intensity was recorded by a photodiode for time-resolved
measurements.
164
9.2 Voltage Gating of Optical Reflectivity
in Pt/GdOx/Au Heterostructure
Figure 9.1(b) shows optical micrographs of the electrode in its virgin state, after applying
+3V for 10 s, and after applying -2 V for 10 s, respectively, to the top gate. The reflected color
changes from yellow to brown under positive bias, and this color change can be reversed by
inverting the gate voltage polarity.
Figure 9.1. Voltage gated optics. (a) Schematic illustration of the Ta(4nm)/Pt
(10nm)/GdOx(10nm) /Au(3nm) optical device. (b) Optical micrographs of the top electrode in its
virgin state, at V = +3V and at V= -2V.
165
Figure 9.2(a) shows the normalized reflectivity measured using the laser probe during
several cycles of a square-wave voltage waveform applied to the gate. Here the voltage amplitude
was 3V and the frequency was 0.5 Hz. We find that in the initial cycle (not shown here) the
reflectivity change proceeds rather slowly and is accompanied by some irreversible darkening,
whereas in subsequent cycles, the device response is quite fast and largely reversible. As can be
seen in figure 6.2(a), the reflectivity can be reversibly switched by ~10% over many cycles. Figure
9.2(b) shows the reflectivity near the rising and falling edge of the voltage waveform. For the
rising edge, the reflectivity abruptly decreases by 5% within ~ 40ms, followed by a slower
continued decrease. At the falling edge, the response is faster; reflectivity increases by 8% within
20 ms. Robust reflectivity switching is observed over >200 cycles, with an overall decrease in
reflectivity indicating some degree of irreversibility. We then correlate the change in reflectivity
with the amount of inserted charge in order to determine the figure of merit for the device193–195.
We performed the experiment by applying a positive gate voltage, which we anticipate leads to
the injection of hydrogen into the system (chapter 5), and measuring the current and reflectivity
simultaneously. Figure 9.2(c) shows data for the device reflectivity, normalized to the virgin state,
versus integrated charge, for a typical device after initial cycling. A linear fit to the data yields a
slope of 37.6 cm2/C. We note that while this figure of merit was measured in reflectivity, similar
coloration efficiency is expected for transmission195.
166
Figure 9.2. Switching transient. (a) Switching cycles of reflectivity and the corresponding
voltage profile. (b) Reflectivity at rising edge and falling edge of the gate voltage. The voltage
was applied at relative time = 0ms. (c) Reflectivity as a function of inserted charge density. The
slope yields a figure of merit of 37.6 cm2/C.
We first examine the variation of optical properties on tAu, which provides evidence that
water splitting and proton transport are key to achieving the observed optical changes. As has
been shown in chapter 5, the rate of water splitting depends on the thickness and morphologies of
the anodes. In the GdOx optical device, the same trend is observed. In figure 9.3(b)-(e) we show
optical micrographs for four different electrodes with different tAu deposited on the same GdOx
film, after applying a gate bias of + 3V for 600 s. For tAu=3 nm, darkening of the film happens
rapidly and relatively uniformly across the electrode area, which we attribute to the porous Au
microstructure (see figure 9.3(a), inset). As tAu is increased, it is evident that the change in optical
167
properties begins at the electrode edge and proceeds inward. This corresponds to the triple phase
boundary, where electrons, protons and H2O are all present and electrochemical reactions114,115 are
expected to proceed most efficiently. Figure 9.3(f)-(i) show a sequence of images during voltage
application for tAu=4 nm, where it is evident that darkening progressively moves from the edge to
the center.
168
Figure 9.3. In situ probe of optical gating. (a)-(d) Optical micrographs of device with tAu=
3nm (a), 4nm(b), 6nm(c), and 10nm(d) after +3V for 600s. Scale bar is 500nm for inset of (a).
(e)-(h) Optical micrographs tAu= 4nm device at +3V after 0s (e), 250s (f), 450s (g) and 1200s
(h). The red dot indicates the laser spot.
169
9.3 Source of Irreversible Optical Change
in Pt/GdOx/Au Heterostructure
Finally, we consider the origin of the long-term irreversible decrease in reflectivity in the
device, which contributes to degradation during repeated cycling. The optical micrographs in
Figure 9.5(a)-(b) shows the virgin state and the state after applying a long-term bias of +3V for
10000s. The dramatic decrease in reflectivity (by 40% compared to the virgin state), is largely
irreversible. In order to understand the chemical change causing the irreversibility, we performed
X-ray photoelectron spectroscopy (XPS) on the sample in its virgin state and at the end of the
applied bias. Figures 9.5(c) shows the XPS spectra for Au4f where we observe a 0.5eV shift
towards a higher binding energy for the Au4f peaks, which likely indicates the oxidation of Au at
large anodic potential96. This oxidation in turn causes the irreversible browning of the device.
Figure 9.4: XPS analysis of irreversible optical changes. (a)-(b) Optical micrographs of the
top electrode in its virgin state(a) and after +3V for 10000s(b).(c) XPS spectrum of Au4f (c)
energy region for the device in virgin state and after bias application. Au4f has two peaks
corresponding to the doublet f 5/2 and f 7/2.
170
Figure 9.5 shows that the irreversibility in optical reflectivity is more pronounced when a larger
VG is applied. In the case of VG = +2V, although optical gating takes longer, the changes are largely
reversible. On the other hand, the optical changes at VG = +4V while large, are mostly irreversible.
In these Pt/GdOx/Au structures, the magnitude of reversible reflectivity change is ~0.1. One can
also notice the appearance of a bright region surrounding the electrode in figure 9.5(b). The details
will be provided in chapter 9.5.
Figure 9.5. Dependence of reversibility on VG. (a)-(b) Modulation of optical reflectivity at VG =
+2V (a) and +4V(b)
Figure 9.6 shows the change in optical reflectivity at different VG. The reflectivity is
calculated by taking the ratio of the current reflected intensity to its initial intensity. Note that VG’s
are applied to virgin devices in these experiments; as a result, the optical changes are slower.
Subsequent cycles yield much faster optical changes. At this point, it is not clear why VG = +1V
171
produces a change in optical reflectivity even though the thermodynamic water splitting potential
is 1.23V at room condition.
Figure 9.6. Rate of modulation of optical reflectivity at different VG’s.
172
9.4 Voltage Gating of GdOx
Heterostructures with different Top and
Bottom Electrodes
Figure 9.7 shows modulation of optical reflectivity in Pt(10nm)/GdOx(20nm)
heterostructures with top Pt(3nm) and Cu(3nm) top electrodes respectively while figure 9.8
shows optical modulation in Au(10nm)/GdOx(20nm)/Au(3nm) and
Pt(10nm)/YOx(10nm)/Au(3nm) heterostructures (different bottom electrode and oxides). The
presence of optical modulation in all devices show that all three layers might contribute to
changes in the optical properties.
173
Figure 9.7. Voltage gating of optical reflectivity in Pt/GdOx structures with Pt(a) and
Cu(top) electrodes
174
Figure 9.8. Voltage gating of optical reflectivity in Au/GdOx/Au (a) and Pt/YOx/Au (b)
heterostructures.
175
9.5 Optical Modulation outside Active
Region due to Hydrogen Diffusion
Figure 9.9 shows the appearance of a bright region around the circular Au electrode in a
Pt(10nm)/GdOx (20nm)/Au(3nm) heterostructure. This region grows at a rate of up to 100nm/s at
VG = +2.3V. In figure 9.10, we observe that upon reaching another device which was previously
gated, the curvature of the region changes (there are signs of necking). The propagation of the
bright region is however not affected by electric field applied between two adjacent devices; in
this case the bright region retains its circular profile (results not shown). This indicates a
diffusion instead of a drift process. We also inserted a thin CoO (2nm) layer between the Pt and
GdOx layers and observed reduction of CoO to metallic Co within the bright region. These
results show that neutral hydrogen is responsible for this optical change. The hydrogen is
injected into the film through the water splitting reaction at VG > 0.
Figure 9.9. Bright region surrounding active device.
176
Figure 9.10. Propagation of bright region.
Figure 9.11. Reduction of CoO within the bright region in a Pt/CoO/GdOx/Au
heterostructure.
In conclusion, we have demonstrated voltage gating of optical properties in different thin film
oxide heterostructures. We introduced a simple electrochromic device where a single layer of
oxide acts as both as an electrolyte and as the active layer. The ion storage layer is absent because
the atmosphere can act as a reservoir of hydrogen. We also discovered the presence of neutral
177
hydrogen in the region surrounding the active device when VG > 0 is applied. This leads to
increased reflectivity in the region.
178
Chapter 10: Electrical
Properties of GdOx
179
Because a majority of this thesis is focused on heterostructures with GdOx as the proton
conducting layer, we will also provide some characterizations on the electrical properties of
GdOx. We deposited 30nm of GdOx on Ta(4nm)/Pt(3nm) by reactive sputtering at different
partial pressure of oxygen (PO2). 200µm of circular Pt(3nm) electrodes were then patterned on
the continuous GdOx film with a short vacuum break between the fabrication of the GdOx and
the Pt top electrodes. The gate voltage (VG) is applied to the top circular electrodes while the
bottom electrode is grounded. The device schematic is shown in figure 10.1a. Figure 10.1b
shows the total current which flows at VG between ±1V for GdOx deposited at different oxygen
flow rates. VG is swept according to the sequence: 0V→1V→-1V→0V. The PO2 which
corresponds to the oxygen flow rates are shown in figure 10.1c. We observe that GdOx deposited
at higher PO2 shows a higher total resistance compared to lower PO2 samples, with the difference
in resistance between the 2.3sccm and 5sccm samples being more than 3 orders of magnitude. In
all three samples, we observe rectifying behavior in electrical conduction.
Figure 10.2 shows the voltage breakdown characteristics of the different GdOx layers
deposited at different PO2. The VG sequence follows 0V→11V→-11V→0V. Consistent with
earlier results, deposition at higher PO2 leads to samples with larger breakdown voltage. For the
samples in figure 10.2(e), the GdOx layer is first deposited at oxygen flow rate of 2.5sccm
followed by a plasma oxidation step for 2minutes. This step is performed by sputtering the
metallic Gd target at 5.5sccm which causes target poisoning. In the case of GdOx sputtered at
5sccm, no breakdown is observed up to more than ±11V.
180
Figure 10.1. Electrical properties of GdOx. (a) Schematic of a
Ta(4nm)/Pt(3nm)/GdOx(30nm)/Pt(3nm) device. (b) IV curves of GdOx deposited at different
PO2. (c) PO2 as a function of oxygen flow rate for reactive sputtering of GdOx. The shutter covers
the chimney around the sputter source. (d) Data in (b) in linear scale.
181
Figure 10.2. Voltage breakdown characteristics of GdOx reactively sputtered under different
oxygen flow.
To correlate the performance of magneto-ionic device to the electrical properties of
GdOx, we also fabricated Ta(4nm)/Pt(3nm)/Co(0.9nm)/GdOx(4nm or 10nm) structures with
either Au(5nm) or Ta(1.4nm)/Au(5nm) circular electrodes patterned on top of the GdOx. The
182
four commonly observed IV curves are shown in figure 10.4, where the VG sequence follows
0V→11V→-11V→0V. The electrical characteristic in figure 10.4(a) and 10.4(b) are very similar
where they both start out with a rectifying behavior; however there is a change to Ohmic
behavior (breakdown) at large voltage for the case in figure 10.4(b). The electrical behavior in
figure 10.4(c) and 10.4(d) are also very similar where they start out with an Ohmic behavior;
however in the case of 10.4(d), there is a change to rectifying behavior at large voltage. The most
likely explanation for these observations is the presence of parallel electronic conduction paths
in the GdOx. Ohmic behavior is caused by an electronically conductive filament formed through
the GdOx, while rectifying behavior can either mean the absence of any filaments or the presence
of a filament with Schottky junction at the interfaces.
Typically, devices which exhibit IV curve shown in figure 10.4(c) show minimal
modulation in magnetic properties with a gate voltage. In this case, the magnetization remains
out-of-plane throughout the voltage sweep sequence (figure 10.5(a)). The “best” devices have IV
curves shown in figure 10.4(a) or (b) where they exhibit large modulation in magnetic properties
with a small gate voltage. In this case, the magnetization rotates in-plane (figure 10.5(b)) at VG >
0V, which is reversible upon setting VG = 0V (figure 10.5(c)). At VG < 0V, the magnetization
vanishes as the Co layer oxidizes (figure 10.5(d)). Device in figure 10.4(d) also exhibits
modulation in magnetic properties; however a large gate voltage has to be applied during the first
cycle to “break” the conductive filament. In this case, the magnetization remains out-of-plane
with minimal change (figure 10.5(a)) from VG = 0V to 6V. At +7V, upon “breaking” the
filament, the magnetization rotates in-plane (figure 10.5(b)). The device now operates with IV
curve as shown in figure 10.4(a) or (b).
183
Note that while we mention that the “best” devices have rectifying behavior, the opposite is not
always true. In another word, devices with rectifying behavior does not necessarily show
magnetic modulation under a gate voltage. The rectifying behavior only implies low electronic
current but it does not tell us much about the ionic current that flows. The total current consists
of both electronic and ionic current superimposed. In order to have large ionic current from
proton transport, the GdOx has to be hydrated. Essentially, magneto-ionic devices with no
electronic current but maximum ionic (proton) current is desired.
To give a sense of the ionic current required to induce magneto-ionic switching in a
0.9nm thick Co layer, there is 5x10-12mol of Co atoms in a 200µm diameter circular device.
Assuming 1H is needed for every Co atom, ~500nC of charge will need to flow. If the timescale
of switching is ~100ms, there will be 5µA of ionic current.
Figure 10.3. Schematic of a magneto-ionic device.
184
Figure 10.4. Typical IV curves for magneto-ionic device. (a) Rectifying behavior. (b)
Rectifying behavior with breakdown. (c) Ohmic behavior. (d) Ohmic behavior with abrupt
change to rectifying behavior at large voltage.
185
Figure 10.5. Polar MOKE hysteresis loops at varying VG. (a)-(d) Hysteresis loop in initial
state (a), at VG > 0V (b), at VG back to 0V (c), and at VG < 0V (d).
186
Chapter 11: Summary
and Outlook
187
11.1 Summary
In summary, we have significantly advanced the mechanistic understanding of voltage-
induced ferromagnet oxidation in a ferromagnet/oxide heterostructures. We debunked the
conventional belief that Co oxidation in a Pt/Co/GdOx heterostructure is dominated by oxygen-
ion migration from the GdOx layer into the Co. Instead, we showed that Co oxidation is primarily
caused by water stored in the GdOx matrix under a gate voltage. The conclusion was supported
by magnetic data showing voltage-induced Co oxidation only in Pt/Co/GdOx device which has
been hydrated under high partial pressure of water and chemical spectroscopic data showing Co
oxidation.
We also demonstrated for the first time voltage gating of magnetism through solid state
proton gating. This significant breakthrough allowed for magneto-ionic gating of magnetism in
solid state devices without oxidation of the ferromagnetic layer. We achieve 90° magnetization
switching by proton injection for >2000 cycles at timescale down to 100ms and observed new
magnetic responses which could be employed for neuromorphic applications. We also modulated
magnetic damping and spin torques in a planar waveguide by solid state proton pumping which
allows us to tune a range of dynamical properties in a magnetic device. Finally, by insertion of
hydrogen in a heavy metal layer like Pd, we also showed for the first time gating of magnetic
anisotropy at a metal-metal interface. Because a wide range of magnetic interactions such as
Dzyaloshinskii-Moriya interaction and RKKY coupling manifest themselves at such interfaces,
this work paves the way to a new field of voltage-controlled spin orbitronics.
Besides voltage-induced modulation of magnetic properties, we also demonstrate that solid state
proton gating can be similarly employed to tune the optical properties of a thin film
188
metal/oxide/metal device. We showed gating of the optical reflectivity by ~10% at timescale
down to 20ms. The observation of optical gating is general across a wide range of thin film
structures with different electrodes and oxides, indicating the versatility of proton gating for
optical modulation.
Finally, we demonstrate a miniaturized room temperature reversible solid oxide fuel cell
for small scale system which is based on hydrogen storage. The cell can be “charged” by
splitting water in ambient. It can then produce power when needed through the water
recombination reaction. Hydrogen is stored in a thin GdOx film where good scalability in terms
of area and thickness of the GdOx storage layer is observed. This is a significant development
because we demonstrate for the first time miniaturization of a solid oxide fuel cell for room
temperature operation. This opens up a new range of applications such as microelectronics where
solid oxide cells can be employed. The use of a rechargeable hydrogen cell also puts forth a new
alternative to lithium-ion battery.
189
11.2 Outlook
In this thesis, we bridge the concept of proton-based solid oxide cell which has largely been
confined to high temperature energy storage applications to dynamic modulation of material
properties in nanoscale devices. Below, we provide a brief outlook into potential developments
and applications in the future.
11.2.1 Integration of Hydrogen Storage in
Proton Magneto-Ionic Device
So far, we have primarily relied on proton sourced from humidity in ambient (through
water splitting) to modulate magnetic properties or for energy storage. This represents a
significant barrier towards practical implementation in electronics because operation which
varies according to season and places is simply too unreliable for use. Besides, the water splitting
reaction is a complicated process which involves multiple charge transfer and mass transport
steps. As a result, the reaction is very energy inefficient and can be rate limiting. For practical
devices, a hydrogen storage layer on top of the electrolyte will be needed to enable fast sourcing
of proton and to keep the device entirely self-contained. Some of these of hydrogen storage layer
includes oxides like WO3196 or alloys like Ni-Mg158.
190
11.2.2 Proton Magneto-Ionics for Memory
and Logic Devices
Because solid state proton gating can be used to modulate a wide range of magnetic
properties, it can be used to generate exotic spin textures such as skyrmions197 or control their
motions. One of the most promising architectures for magnetic memory device is the magnetic
racetrack, where magnetic bits can be dynamically created and moved for reading and writing
(figure 10.2)198. In a magnetic racetrack memory, a gate voltage can be used to generate
skyrmions as magnetic bits, which can then be moved by a spin current. By placing a series of
gate pads, a gate voltage can then used to synchronize the motion of these skyrmions or tune the
velocity of the skyrmionic bits (figure 11.2b)199.
Figure 11.1. Proton magneto-
ionic device with an integrated
hydrogen storage
191
Figure 11.2. Magnetic racetrack memory. (a) Schematic of a skyrmionic magnetic racetrack
memory where skymion is used to represent data bits. (b) Periodic gate pads to synchronize
current driven skyrmionic motion. Adapted from reference199,200.
An example application for proton magneto-ionics in logic operation is a logic gate as shown in
figure 11.3201. In this device, a gate voltage, VG can be used to control whether a cell is switched
by a current (IIN) or not. The cell output is then the Hall voltage readout (Rxy) from the cell.
Through a combination of a few cells, one can get a complementary AND or an OR gate.
192
Figure 11.3. Logic gate based on voltage controlled spin-orbit torques. (a) Schematic of the
logic gate. (b) Rxy output as a function of VG and IIN. (c) Table for the logic operations. Adapted
from reference201.
193
11.2.3 Proton Magneto-Ionics to Quantify
Proton Conductivity in Thin Film Oxides
While there is a rich literature of proton conductivity in bulk oxides at high temperature,
relatively little is understood about proton conductivity of oxides at room temperature and in
ultra-thin dimensions. By fabricating Pt/Co/MOx heterostructures using different metal oxides as
the electrolyte, proton magneto-ionics can be utilized to provide quantitative information about
proton conductivity and even proton mobility in these oxides.
In these experiments, MOKE or Hall magnetometry can be used for time resolved probe
of the Co magnetization while we keep as constant both the electrodes, and the ambient
humidity. The thickness of the oxide can be varied to study the effect of electric field on proton
conductivity. The effect of microstructure and phase on proton conductivity of the different
oxides can also be studied by varying the fabrication parameters of the oxide.
194
References
195
1. Hwang, H. Y. et al. Emergent phenomena at oxide interfaces. Nat. Mater. 11, 103–113
(2012).
2. Chappert, C., Fert, A. & Van Dau, F. N. The emergence of spin electronics in data
storage. Nat. Mater. 6, 813–823 (2007).
3. Reiss, G., Schmalhorst, J., Thomas, A., Hütten, A. & Yuasa, S. Magnetic tunnel junctions.
Springer Tracts Mod. Phys. 227, 291–333 (2007).
4. O’Handley, R. C. Modern Magnetic Materials: Principles and Applications. John Wiley
Sons New York Chichester Weinheim Brisbane Singapore Toronto (2000).
doi:10.1109/MEI.2005.1490004
5. Beach, G. S. D. Hysteresis in Ferromagnetic Materials. Lecture
6. Bayreuther, G. Magnetic Anisotropy. Lecture
7. Emori, S. Magnetic Domain Walls Driven by Interfacial Phenomena. PhD Thesis (2013).
8. Ross, C. A. Patterned Magnetic Recording Media. Annu. Rev. Mater. Res. 203–35 (2001).
9. Iwasaki, S. ichi & Ouchi, K. Co-Cr recording films with perpendicular magnetic
anisotropy. IEEE Trans. Magn. 14, 849–851 (1978).
10. Iwasaki, S. Perpendicular Magnetic Recording. IEEE Trans. Magn. 16, 71–76 (1980).
11. Tudu, B. & Tiwari, A. Recent Developments in Perpendicular Magnetic Anisotropy Thin
Films for Data Storage Applications. Vacuum 146, 329–341 (2017).
12. Maruyama, T. et al. Large voltage-induced magnetic anisotropy change in a few atomic
196
layers of iron. Nat. Nanotechnol. 4, 158–161 (2009).
13. Lin, C. J. et al. Magnetic and structural properties of Co/Pt multilayers. J. Magn. Magn.
Mater. 194–206, (1991).
14. Carcia, P. F., Meinhaldt, A. D. & Suna, A. Perpendicular magnetic anisotropy in Pd/Co
thin film layered structures. Appl. Phys. Lett. 47, 178–180 (1985).
15. G. H. O. Daalderop, P. J. K. and M. F. H. S. Magnetic anisotropy of a free-standing Co
monolayer and of multilayers which contain Co monolayers. Phys. Rev. B 50, 9989
(1994).
16. Johnson, M. T., Bloement, P. J. H., Broeder, F. J. A. den & Vries, J. . de. Magnetic
Anisotropy in Metallic Multilayers. Rep. Prog. Phys. 59, 1409–1458 (1996).
17. Nakajima, N. et al. Perpendicular Magnetic Anisotropy Caused by Interfacial
Hybridization via Enhanced Orbital Moment in Co/Pt Multilayers: Magnetic Circular X-
Ray Dichroism Study. Phys. Rev. Lett. 81, 5229–5232 (1998).
18. Heinrich, B. & Cochran, J. F. Ultrathin metallic magnetic films: magnetic anisotropies and
exchange interactions. Adv. Phys. 42, 523–639 (1993).
19. D. Weller, Y. Wu, J. Stohr, and M. G. Samant, B. D. Hermsmeier, C. C. Orbital magnetic
moments of Co in multilayers with perpendicular magnetic anisotropy. Phys. Rev. B 49,
888–896 (1994).
20. Yang, H. X. et al. First-principles investigation of the very large perpendicular magnetic
anisotropy at Fe|MgO and Co|MgO interfaces. Phys. Rev. B 84, 054401 (2011).
21. Dieny, B. & Chshiev, M. Perpendicular magnetic anisotropy at transition metal/oxide
197
interfaces and applications. Rev. Mod. Phys. 89, (2017).
22. Manchon, A. et al. Analysis of oxygen induced anisotropy crossover in Pt / Co / MOx
trilayers. J. Appl. Phys. 104, 043914 (2008).
23. Manchon, A. et al. X-ray analysis of the magnetic influence of oxygen in Pt/Co/AlOx
trilayers. J. Appl. Phys. 103, (2008).
24. Mankey, G. J., Kief, M. T., Huang, F. & Willis, R. F. Hydrogen chemisorption on
ferromagnetic thin film surfaces. J. Vac. Sci. Technol. A Vacuum, Surfaces, Film. 11,
2034–2039 (2002).
25. Valvidares, S. M., Dorantes-Dávila, J., Isern, H., Ferrer, S. & Pastor, G. M. Interface-
driven manipulation of the magnetic anisotropy of ultrathin Co films on Pt(111): Substrate
deposition of hydrogen and model calculations. Phys. Rev. B 81, 1–11 (2010).
26. Munbodh, K., Perez, F. A., Keenan, C. & Lederman, D. Effects of hydrogen / deuterium
absorption on the magnetic properties of Co / Pd multilayers. Phys. Rev. B 83, 094432
(2011).
27. Sander, D. et al. Reversible H-induced switching of the magnetic easy axis in Ni/Cu(001)
thin films. Phys. Rev. Lett. 93, 247203 (2004).
28. Santos, B. et al. Hydrogen-induced reversible spin-reorientation transition and magnetic
stripe domain phase in bilayer Co on Ru(0001). Phys. Rev. B 85, 3–8 (2012).
29. Hsu, P., Rózsa, L., Finco, A., Schmidt, L. & Palotás, K. Inducing skyrmions in ultrathin
Fe films by hydrogen exposure. arXiv 1711.06784v1
30. Dijken, S. Van, Vollmer, R., Poelsema, B. & Kirschner, J. The influence of CO and H
198
adsorption on the spin reorientation transition in Ni / Cu ( 0 0 1 ). J. Magn. Magn. Mater.
210, 316–328 (2000).
31. Gallego, S., Sanchez, N., Martin, S., Muñoz, M. C. & Szunyogh, L. Reversible
enhancement of the magnetism of ultrathin Co films by H adsorption. Phys. Rev. B -
Condens. Matter Mater. Phys. 82, 1–6 (2010).
32. Barman, A. & Haldar, A. Time-Domain study of magnetization dynamics in magnetic thin
films and micro- and nanostructures. Solid State Physics - Advances in Research and
Applications 65, (Elsevier Inc., 2014).
33. Hoffmann, A. Spin Hall Effects in Metals. Ieee Trans. Magn. 49, 5172–5193 (2013).
34. Liu, L. et al. Spin-Torque Switching with the Giant Spin Hall Effect of Tantalum. Science
(80-. ). 336, 555–559 (2012).
35. Liu, L., Lee, O. J., Gudmundsen, T. J., Ralph, D. C. & Buhrman, R. A. Current-Induced
Switching of Perpendicularly Magnetized Magnetic Layers Using Spin Torque from the
Spin Hall Effect. Phys. Rev. Lett. 109, 096602 (2012).
36. Emori, S., Bauer, U., Ahn, S. M., Martinez, E. & Beach, G. S. D. Current-driven
dynamics of chiral ferromagnetic domain walls. Nat. Mater. 12, 611–616 (2013).
37. Sankey, J. C. et al. Spin-transfer-driven ferromagnetic resonance of individual
nanomagnets. Phys. Rev. Lett. 96, 50–53 (2006).
38. Tulapurkar, A. A. et al. Spin-torque diode effect in magnetic tunnel junctions. Nature 438,
339–342 (2005).
39. Liu, L., Moriyama, T., Ralph, D. C. & Buhrman, R. A. Spin-torque ferromagnetic
199
resonance induced by the spin Hall effect. Phys. Rev. Lett. 106, 1–4 (2011).
40. Fuchs, G. D. et al. Spin-torque ferromagnetic resonance measurements of damping in
nanomagnets. Appl. Phys. Lett. 91, 20–23 (2007).
41. Dietl, T. A ten-year perspective on dilute magnetic semiconductors and oxides. Nat.
Mater. 9, 965–974 (2010).
42. Ohno, H. Making Nonmagnetic Semiconductors Ferromagnetic. Science (80-. ). 281, 951–
957 (1998).
43. Matsukura, F., Tokura, Y. & Ohno, H. Control of magnetism by electric fields. Nat.
Nanotechnol. 10, 209–220 (2015).
44. Ohno, H. et al. Electric- field control of ferromagnetism. Nature 408, 944–946 (2000).
45. Chiba, D., Yamanouchi, M., Matsukura, F. & Ohno, H. Electrical Manipulation of
Magnetization Reversal in a Ferromagnetic Semiconductor. Science (80-. ). 301, 943–946
(2003).
46. Eerenstein, W., Mathur, N. D. & Scott, J. F. Multiferroic and magnetoelectric materials.
Nature 442, 759–765 (2006).
47. Kim, S., Lee, J., Shin, S. & Wook, H. Voltage control of a magnetization easy axis in
piezoelectric / ferromagnetic hybrid films. J. Magn. Magn. Mater. 267, 127–132 (2003).
48. Zheng, H. et al. Multiferroic BaTiO3-CoFe2O4 Nanostructures. Science (80-. ). 303, 661–
664 (2004).
49. Zavaliche, F., Zheng, H., Yang, S. Y. & Zhan, Q. Electric Field-Induced Magnetization
Switching in Epitaxial Columnar Nanostructures. Nano Lett. 4, 1793–1796 (2005).
200
50. Ma, J., Hu, J., Li, Z. & Nan, C. Recent Progress in Multiferroic Magnetoelectric
Composites : from Bulk to Thin Films. Adv. Mater. 1062–1087 (2011).
51. Weisheit, M. et al. Electric field-induced modification of magnetism in thin-film
ferromagnets. Science (80-. ). 315, 349–351 (2007).
52. Shiota, Y., Maruyama, T., Nozaki, T. & Shinjo, T. Voltage-assisted Magnetization
Switching in Ultrathin Fe80Co20 Alloy Layers. arXiv 0905.0185, (2009).
53. Duan, C. et al. Surface Magnetoelectric Effect in Ferromagnetic Metal Films. Phys. Rev.
Lett. 101, 137201 (2008).
54. Tsujikawa, M. & Oda, T. Finite electric field effects in the large perpendicular magnetic
anisotropy surface Pt/Fe/Pt(001): A first-principles study. Phys. Rev. Lett. 102, 247203
(2009).
55. Nakamura, K., Shimabukuro, R., Fujiwara, Y., Akiyama, T. & Ito, T. Giant Modification
of the Magnetocrystalline Anisotropy in Transition-Metal Monolayers by an External
Electric Field. Phys. Rev. Lett. 102, 187201 (2009).
56. Duan, C. G. et al. Surface magnetoelectric effect in ferromagnetic metal films. Phys. Rev.
Lett. 101, 137201 (2008).
57. Shiota, Y. et al. Voltage-assisted magnetization switching in ultrathin Fe 80Co20 alloy
layers. Appl. Phys. Express 2, (2009).
58. Bauer, U., Emori, S. & Beach, G. S. D. Voltage-controlled domain wall traps in
ferromagnetic nanowires. Nat. Nanotechnol. 8, 411–416 (2013).
59. Emori, S., Bauer, U., Woo, S. & Beach, G. S. D. Large voltage-induced modification of
201
spin-orbit torques in Pt/Co/GdOx. Appl. Phys. Lett. 105, (2014).
60. Bi, C. et al. Reversible control of Co magnetism by voltage-induced oxidation. Phys. Rev.
Lett. 113, 267202 (2014).
61. Bauer, U. et al. Magneto-ionic control of interfacial magnetism. Nat. Mater. 14, 174–181
(2015).
62. Gilbert, D. A. et al. Structural and magnetic depth profiles of magneto-ionic
heterostructures beyond the interface limit. Nat. Commun. 7, 12264 (2016).
63. Gilbert, D. A. et al. Controllable positive exchange bias via redox-driven oxygen
migration. Nat. Commun. 7, 11050 (2016).
64. Sakamaki, M. & Amemiya, K. Observation of an electric field-induced interface redox
reaction and magnetic modification in GdOx/Co thin film by means of depth-resolved X-
ray absorption spectroscopy. Phys. Chem. Chem. Phys. 20, 20004–20009 (2018).
65. Mishra, R. et al. Electric-field control of spin accumulation direction for spin-orbit
torques. Nat. Commun. 10, 248 (2019).
66. Li, H. B. et al. Electric-field control of ferromagnetism through oxygen ion gating. Nat.
Commun. 8, 2156 (2017).
67. Lu, N. et al. Electric-field control of tri-state phase transformation with a selective dual-
ion switch. Nature 546, 124–128 (2017).
68. Zhou, X. et al. Role of Oxygen Ion Migration in the Electrical Control of Magnetism in
Pt/Co/Ni/HfO2 Films. J. Phys. Chem. C 120, 1633–1639 (2016).
69. Li, Q. et al. Electrical control of exchange bias via oxygen migration across CoO-ZnO
202
nanocomposite barrier. Appl. Phys. Lett. 109, 1–5 (2016).
70. Hirai, T. et al. Control of magnetic anisotropy in Pt/Co system using ionic liquid gating.
Appl. Phys. Express 9, (2016).
71. Dasgupta, S. et al. Intercalation-driven reversible control of magnetism in bulk
ferromagnets. Adv. Mater. 26, 4639–4644 (2014).
72. Dasgupta, S. et al. Toward On-and-Off Magnetism: Reversible Electrochemistry to
Control Magnetic Phase Transitions in Spinel Ferrites. Adv. Funct. Mater. 26, 7507–7515
(2016).
73. Zhang, Q. et al. Lithium-Ion Battery Cycling for Magnetism Control. Nano Lett. 16, 583–
587 (2016).
74. Di, N. Electric and chemical manipulation of magnetic anisotropy of ultrathin
ferromagnetic films. PhD Thesis. (2015).
75. Jani, M. G., Halliburton, L. E. & Halperin, A. Observation of simple lithium-associated
electron trap in crystalline SiO2. Phys. Rev. Lett. 56, 1392–1395 (1986).
76. Minh, N. Recent advances in solid oxide fuel cell technology. Am. Ceram. Soc. Bull. 82,
305–313 (2003).
77. Haile, S. M. Fuel cell materials and components. Acta Mater. 51, 5981–6000 (2003).
78. Laguna-Bercero, M. A. Recent advances in high temperature electrolysis using solid oxide
fuel cells: A review. J. Power Sources 203, 4–16 (2012).
79. Malavasi, L., Fisher, C. A. J. & Islam, M. S. Oxide-ion and proton conducting electrolyte
materials for clean energy applications: Structural and mechanistic features. Chem. Soc.
203
Rev. 39, 4370–4387 (2010).
80. Bi, L., Boulfrad, S. & Traversa, E. Steam electrolysis by solid oxide electrolysis cells
(SOECs) with proton-conducting oxides. Chem. Soc. Rev. 43, 8255–8270 (2014).
81. Kreuer, K.-D. Proton Conductivity: Materials and Applications. Chem. Mater. 8, 610–641
(1996).
82. Kreuer, K. D. Proton-Conducting Oxides. Annu. Rev. Mater. Res. 33, 333–359 (2003).
83. Skinner, S. J. & Kilner, J. A. Oxygen ion conductors. Mater. Today 6, 30–37 (2003).
84. Sun, C., Hui, R. & Roller, J. Cathode materials for solid oxide fuel cells: A review. J.
Solid State Electrochem. 14, 1125–1144 (2010).
85. Cowin, P. I., Petit, C. T. G., Lan, R., Irvine, J. T. S. & Tao, S. Recent progress in the
development of anode materials for solid oxide fuel cells. Adv. Energy Mater. 1, 314–332
(2011).
86. Dusastre, V., Tao, S. & Irvine, J. T. . A redox-stable efficient anode for solid-oxide fuel
cells. Mater. Sustain. Energy 259–262 (2012).
87. Kreuer, K.-D., Paddison, S. J., Spohr, E. & Schuster, M. Transport in Proton Conductors
for Fuel-Cell Applications: Simulations, Elementary Reactions, and Phenomenology.
Chem. Rev. 104, 4637–4678 (2004).
88. Kreuer, K. Proton Conductivity : Materials and Applications. Chem.Mater 8, 610–641
(1996).
89. Medvedev, D. et al. BaCeO3: Materials development, properties and application. Prog.
Mater. Sci. 60, 72–129 (2014).
204
90. Pergolesi, D. et al. High proton conduction in grain-boundary-free yttrium-doped barium
zirconate films grown by pulsed laser deposition. Nat. Mater. 9, 846–852 (2010).
91. Fabbri, E., Pergolesi, D. & Traversa, E. Materials challenges toward proton-conducting
oxide fuel cells: A critical review. Chem. Soc. Rev. 39, 4355–4369 (2010).
92. Larring, Y. & Norby, T. Protons in rare earth oxides. Solid State Ionics 77, 147–151
(1995).
93. Truls, N. Solid-state protonic conductors: principles, properties, progress and prospects.
Solid State Ionics 125, 1–11 (1999).
94. Haugsrud, R. & Norby, T. Proton conduction in rare-earth ortho-niobates and ortho-
tantalates. Nat. Mater. 5, 193–196 (2006).
95. Tanner, C. W. & Virkar, A. V. InstabiIiiy of BaCeO3 in H20-Containing Atmospheres. J.
Electrochem. Soc. 143, 1386–1389 (1996).
96. Fabbri, E. et al. Developments and perspectives of oxide-based catalysts for the oxygen
evolution reaction. Catal. Sci. Technol. 4, 3800–3821 (2014).
97. Gregori, G., Shirpour, M. & Maier, J. Proton conduction in dense and porous
nanocrystalline ceria thin films. Adv. Funct. Mater. 23, 5861–5867 (2013).
98. Scherrer, B. et al. On proton conductivity in porous and dense yttria stabilized zirconia at
low temperature. Adv. Funct. Mater. 23, 1957–1964 (2013).
99. Shirpour, M., Gregori, G., Merkle, R. & Maier, J. On the proton conductivity in pure and
gadolinium doped nanocrystalline cerium oxide. Phys. Chem. Chem. Phys. 13, 937–940
(2011).
205
100. Yang, J. J., Strukov, D. B. & Stewart, D. R. Memristive devices for computing. Nat.
Nanotechnol. 8, 13–24 (2013).
101. Waser, R. Solubility of Hydrogen Defects in Doped and Undoped BaTiO3. J. Am. Ceram.
Soc. 71, 58–63 (1988).
102. Hong, W. T. et al. Toward the rational design of non-precious transition metal oxides for
oxygen electrocatalysis. Energy Environ. Sci. 8, 1404–1427 (2015).
103. Suen, N. T. et al. Electrocatalysis for the oxygen evolution reaction: Recent development
and future perspectives. Chem. Soc. Rev. 46, 337–365 (2017).
104. Man, I. C. et al. Universality in Oxygen Evolution Electrocatalysis on Oxide Surfaces.
ChemCatChem 3, 1159–1165 (2011).
105. Kim, J. S., Kim, B., Kim, H. & Kang, K. Recent Progress on Multimetal Oxide Catalysts
for the Oxygen Evolution Reaction. Adv. Energy Mater. 8, 1–26 (2018).
106. Zeng, M. & Li, Y. Recent advances in heterogeneous electrocatalysts for the hydrogen
evolution reaction. J. Mater. Chem. A 3, 14942–14962 (2015).
107. Stamenkovic, V. R., Strmcnik, D., Lopes, P. P. & Markovic, N. M. Energy and Fuels from
Electrochemical Interfaces. Nat. Mater. 16, 57–69 (2017).
108. Sheng, W., Gasteiger, H. A. & Shao-Horn, Y. Hydrogen Oxidation and Evolution
Reaction Kinetics on Platinum: Acid vs Alkaline Electrolytes. J. Electrochem. Soc. 157,
B1529 (2010).
109. Strmcnik, D. et al. Improving the hydrogen oxidation reaction rate by promotion of
hydroxyl adsorption. Nat. Chem. 5, 300–306 (2013).
206
110. Nielsen, J. & Hjelm, J. Impedance of SOFC electrodes: A review and a comprehensive
case study on the impedance of LSM:YSZ cathodes. Electrochim. Acta 115, 31–45
(2014).
111. Rossmeisl, J., Qu, Z., Zhu, H., Kroes, G. & Norskov, J. K. Electrolysis of Water on Oxide
Surfaces. J. Electrochem. Chem. 607, 83–89 (2007).
112. Rossmeisl, J., Logadottir, A. & Nørskov, J. K. Electrolysis of water on ( oxidized ) metal
surfaces. Chem. Phys. 319, 178–184 (2005).
113. Nørskov, J. K. et al. Origin of the Overpotential for Oxygen Reduction at a Fuel-Cell
Cathode. 17886–17892 (2004).
114. Adler, S. B. Factors Governing Oxygen Reduction in Solid Oxide Fuel Cell Cathodes.
Chem. Rev. 104, 4791–4844 (2004).
115. Peng, R., Wu, T., Liu, W., Liu, X. & Meng, G. Cathode processes and materials for solid
oxide fuel cells with proton conductors as electrolytes. J. Mater. Chem. 20, 6218–6225
(2010).
116. Tsai, T. & Barn, S. A. Effect of LSM-YSZ cathode on thin electrolyte solid oxide fuel cell
performance. Solid State Ionics 93, 207–217 (1997).
117. Primdahl, S. Oxidation of Hydrogen on Ni/Yttria-Stabilized Zirconia Cermet Anodes. J.
Electrochem. Soc. 144, 3409 (2006).
118. Barsoukov, E. & Macdonald, J. R. Theory , Experiment , and. America 177, (2005).
119. Lai, W. & Haile, S. M. Impedance spectroscopy as a tool for chemical and
electrochemical analysis of mixed conductors: A case study of ceria. J. Am. Ceram. Soc.
207
88, 2979–2997 (2005).
120. Jaminik, J. Generalised Equivalent Circuits for Mass and Charge Transport.Pdf. Phys.
Chem. Chem. Phys. 3, 1668–1678 (2001).
121. Harrington, D. A. & Van Den Driessche, P. Mechanism and equivalent circuits in
electrochemical impedance spectroscopy. Electrochim. Acta 56, 8005–8013 (2011).
122. Tuller, H. L. Ionic conduction in nanocrystalline materials. Solid State Ionics 131, 143–
157 (2000).
123. Tuller, H. L. Ionic conduction in nanocrystalline materials. Solid State Ionics 131, 143–
157 (2000).
124. Greene, J. E. Review Article: Tracing the recorded history of thin-film sputter deposition:
From the 1800s to 2017. J. Vac. Sci. Technol. A Vacuum, Surfaces, Film. 35, 05C204
(2017).
125. Kaiser, N. Review of the fundamentals of thin-film growth. Appl. Opt. 41, 3053 (2007).
126. Thornton, J. A. Influence of apparatus geometry and deposition conditions on the structure
and topography of thick sputtered coatings. J. Vac. Sci. Technol. 11, 666–670 (2002).
127. Hentzell, H. T. G., Smith, D. A. & Grovernor, C. The development of grain structure
during growth of metallic thin films. Acta Metall. 32, 773–781 (1984).
128. Kerkache, L., Layadi, A. & Mosser, A. Effect of oxygen partial pressure on the structural
and optical properties of dc sputtered ITO thin films. J. Alloys Compd. 485, 46–50 (2009).
129. Choi, C. G. et al. Effects of oxygen partial pressure on the microstructure and electrical
properties of indium tin oxide film prepared by d.c. magnetron sputtering. Thin Solid
208
Films 258, 274–278 (1995).
130. Price, E. P. Characterization of transport processes in magnetic tunnel junctions. (2001).
131. Hamrle, J. Magneto-optical Kerr effect ( MOKE ) Photon-photon spectroscopies (
absorption ) I : (2014).
132. Qiu, Z. Q. & Bader, S. D. Surface magneto-optic Kerr effect (SMOKE). J. Magn. Magn.
Mater. 200, 664–678 (1999).
133. Bauer, U. Voltage Programmable Materials. PhD Thesis. (2015).
134. Garello, K. et al. Symmetry and magnitude of spin–orbit torques in ferromagnetic
heterostructures. Nat. Nanotechnol. 8, 587–593 (2013).
135. Nagaosa, N., Sinova, J., Onoda, S., MacDonald, A. H. & Ong, N. P. Anomalous Hall
effect. Rev. Mod. Phys. 82, 1539–1592 (2010).
136. An, H., Kageyama, Y., Kanno, Y., Enishi, N. & Ando, K. Spin-torque generator
engineered by natural oxidation of Cu. Nat. Commun. 7, 1–8 (2016).
137. Norby, T., Widerøe, M., Glöckner, R. & Larring, Y. Hydrogen in oxides. Dalt. Trans. 0,
3012–3018 (2004).
138. Tsuruoka, T. et al. Humidity effects on the redox reactions and ionic transport in a
Cu/Ta2O5/Pt atomic switch structure process. Jpn. J. Appl. Phys. 55, 06091–09095
(2016).
139. Lübben, M., Wiefels, S., Waser, R. & Valov, I. Processes and Effects of Oxygen and
Moisture in Resistively Switching TaOxand HfOx. Adv. Electron. Mater. 4, 1–11 (2018).
209
140. Tsuruoka, T. et al. Effects of moisture on the switching characteristics of oxide-based,
gapless-type atomic switches. Adv. Funct. Mater. 22, 70–77 (2012).
141. Messerschmitt, F., Kubicek, M. & Rupp, J. L. M. How Does Moisture Affect the Physical
Property of Memristance for Anionic-Electronic Resistive Switching Memories? Adv.
Funct. Mater. 25, 5117–5125 (2015).
142. Sediva, E., Bowman, W. J., Gonzalez-Rosillo, J. C. & Rupp, J. L. M. Investigation of the
Eightwise Switching Mechanism and Its Suppression in SrTiO3 Modulated by Humidity
and Interchanged Top and Bottom Platinum and LaNiO3 Electrode Contacts. Adv.
Electron. Mater. 1800566, 1–11 (2018).
143. Moeller, T. & Kremers, H. E. The basicity characteristics of scandium, yttrium, and the
rare earth elements. Chem. Rev. 37, 97–159 (1945).
144. Adachi, G. & Imanaka, N. The Binary Rare Earth Oxides. Chem. Rev. 98, 1479–1514
(1998).
145. Fanciulli, M. & Scarel, G. Rare earth oxide thin films: Growth, Characterization and
Applications. Springer (2007). doi:10.1007/11499893_20
146. Cohen, S. et al. The interaction of H2O with the surface of polycrystalline gadolinium at
the temperature range of 300-570 K. Surf. Sci. 617, 29–35 (2013).
147. Jeon, S. & Hwang, H. Effect of hygroscopic nature on the electrical characteristics of
lanthanide oxides (Pr2O3, Sm2O3, Gd2O3, and Dy2O3). J. Appl. Phys. 93, 6393–6395
(2003).
148. Külah, E. et al. Surface chemistry of rare-earth oxide surfaces at ambient conditions:
210
Reactions with water and hydrocarbons. Sci. Rep. 7, 1–10 (2017).
149. Villars, P. Gd2O3 C-phase Crystal Structure. Inorganic Solid Phases, SpringerMaterials
(online database). (Springer-Verlag Berlin Heidelberg & NIMS, Japan, 2016).
150. Villars, P. Gd(OH)3 Crystal Structure. Inorganic Solid Phases, SpringerMaterials (online
database). (Springer-Verlag Berlin Heidelberg & NIMS, Japan, 2016).
151. Aruna, I., Mehta, B. R., Malhotra, L. K. & Shivaprasad, S. M. Stability and hydrogenation
of ‘bare’ gadolinium nanoparticles. Adv. Funct. Mater. 15, 131–137 (2005).
152. Brundle, C. R. & Wandelt, K. The Interactions of oxygen with Gadolinium: UPS ans XPS
Studies. Surf. Sci. 157, 162–182 (1985).
153. Lasia, A. in Handbook of Fuel Cells- Fundamentals, Technology and Applications 417–
434 (2003). doi:10.1002/9780470974001.f204033
154. Tymoczko, J., Calle-Vallejo, F., Schuhmann, W. & Bandarenka, A. S. Making the
hydrogen evolution reaction in polymer electrolyte membrane electrolysers even faster.
Nat. Commun. 7, 10990 (2016).
155. Tan, A. J. et al. Magneto-ionic control of magnetism using a solid-state proton pump. Nat.
Mater. 18, 35–42 (2019).
156. Zhu, X. et al. In Situ Nanoscale Electric Field Control of Magnetism by Nanoionics. Adv.
Mater. 28, 7658–7665 (2016).
157. Adams, B. D. & Chen, A. The role of palladium in a hydrogen economy. Mater. Today
14, 282–289 (2011).
158. Schlapbach, L. & Züttel, A. Hydrogen-storage materials for mobile applications. Nature
211
414, 353–358 (2002).
159. Yang, H., Thiaville, A., Rohart, S., Fert, A. & Chshiev, M. Anatomy of Dzyaloshinskii-
Moriya Interaction at Co / Pt Interfaces. Phys. Rev. Lett. 115, 267210 (2015).
160. S.S.P.Parkin, R.Bhadra & Roche, K. P. Oscillatory magnetic exchange coupling through
thin copper layers. Phys. Rev. Lett. 66, 2152–2155 (1991).
161. Campbell, C. T. Catalyst–support interactions: Electronic perturbations. Nat. Chem. 4,
597–598 (2012).
162. Rodriguez, J. A., Liu, P., Hrbek, J., Evans, J. & Pérez, M. Water gas shift reaction on Cu
and Au nanoparticles supported on CeO 2(111) and ZnO(0001): Intrinsic activity and
importance of support interactions. Angew. Chemie - Int. Ed. 46, 1329–1332 (2007).
163. Valov, I. & Lu, W. D. Nanoscale electrochemistry using dielectric thin films as solid
electrolytes. Nanoscale 8, 13828–13837 (2016).
164. Tsuruoka, T. et al. Effects of moisture on the switching characteristics of oxide-based,
gapless-type atomic switches. Adv. Funct. Mater. 22, 70–77 (2012).
165. Yin, Q. et al. Cathode bubbles induced by moisture electrolysis in TiO2-x-based resistive
switching cells. J. Phys. D. Appl. Phys. 49, 09LT01 (2016).
166. Bard, A. J., Faulkner, L. R., Swain, E. & Robey, C. Electrochemical Methods:
Fundamentals and Applications. (John Wiley and Sons).
167. Yoon, H. et al. Reversible phase modulation and hydrogen storage in multivalent VO2
epitaxial thin films. Nat. Mater. 15, 1113–1119 (2016).
168. Burr, G. W. et al. Neuromorphic computing using non-volatile memory. Adv. Phys. X 2,
212
89–124 (2017).
169. Pershin, Y. V. & Di Ventra, M. Experimental demonstration of associative memory with
memristive neural networks. Neural Networks 23, 881–886 (2010).
170. Thompson, A. C. & D.Vaughn. X-RAY DATA BOOKLET. (2001).
171. Bennett, P. A. & Fuggle, J. C. Electronic structure and surface kinetics of palladium
hydride studied with x-ray photoelectron spectroscopy and electron-energy-loss
spectroscopy. Phys. Rev. B 26, 6030–6039 (1982).
172. Davoli, I. et al. Palladium L3 absorption edge of PdH0.6 films: Evidence for hydrogen
induced unoccupied states. Solid State Commun. 71, 383–390 (1989).
173. Tew, M. W., Miller, J. T. & van Bokhoven, J. A. Particle Size Effect of Hydride
Formation and Surface Hydrogen Adsorption of Nanosized Palladium Catalysts: L 3 Edge
vs K Edge X-ray Absorption Spectroscopy. J. Phys. Chem. C 113, 15140–15147 (2009).
174. Richardson, T. J. et al. X-ray absorption spectroscopy of transition metal-magnesium
hydride thin films. J. Alloys Compd. 356–357, 204–207 (2003).
175. Gambardella, P. & Miron, I. M. Current-induced spin–orbit torques. Philos. Trans. R. Soc.
A 369, 3175–3197 (2011).
176. Hellman, F. et al. Interface-Induced Phenomena in Magnetism. Rev. Mod. Phys. 89,
025006 (2017).
177. Emori, S., Bauer, U., Ahn, S., Martinez, E. & Beach, G. S. D. Current-driven dynamics of
chiral ferromagnetic domain walls. Nat. Mater. 12, 611–616 (2013).
178. Pai, C. F., Ou, Y., Vilela-Leão, L. H., Ralph, D. C. & Buhrman, R. A. Dependence of the
213
efficiency of spin Hall torque on the transparency of Pt/ferromagnetic layer interfaces.
Phys. Rev. B - Condens. Matter Mater. Phys. 92, (2015).
179. Robertson, J. High dielectric constant oxides. Eur. Phys. J. Appl. Phys. 28, 265–291
(2004).
180. Paton, M. G. & Maslen, E. N. A refinement of the crystal structure of yttria. Acta
Crystallogr. 19, 307–310 (1965).
181. French, R., Glass, S., Ohuchi, F., Xu, Y. & Ching, W. Experimental and theoretical
determination of the electronic structure and optical properties of three phases of ZrO2.
Phys. Rev. B 49, 5133–5142 (1994).
182. Scott, H. G. Phase relationships in the zirconia-yttria system. J. Mater. Sci. 10, 1527–1535
(1975).
183. Stephenson, N. C. & Roth, R. S. Structural systematics in the binary system Ta2O5–WO3.
V. The structure of the low-temperature form of tantalum oxide L-Ta2O5. Acta
Crystallogr. 27, 1037–1044 (1971).
184. Wang, Y. et al. Lithium and lithium ion batteries for applications in microelectronic
devices: A review. J. Power Sources 286, 330–345 (2015).
185. Pikul, J. H., Gang Zhang, H., Cho, J., Braun, P. V & King, W. P. High-power lithium ion
microbatteries from interdigitated three-dimensional bicontinuous nanoporous electrodes.
Nat. Commun. 4, 1732 (2013).
186. Lai, W. et al. Ultrahigh-energy-density microbatteries enabled by new electrode
architecture and micropackaging design. Adv. Mater. 22, 139–144 (2010).
214
187. Bates, J. B. B. et al. Thin-film rechargeable lithium batteries. J. Power Sources 54, 58–62
(1995).
188. Dudney, N. J. Solid-state thin-film rechargeable batteries. Mater. Sci. Eng. B 116, 245–
249 (2005).
189. Gillaspie, D. T., Tenent, R. C. & Dillon, A. C. Metal-oxide films for electrochromic
applications: present technology and future directions. J. Mater. Chem. 20, 9585 (2010).
190. Niklasson, G. a. & Granqvist, C. G. Electrochromics for smart windows: thin films of
tungsten oxide and nickel oxide, and devices based on these. J. Mater. Chem. 17, 127
(2007).
191. Granqvist, C. G. Recent progress in thermochromics and electrochromics: A brief survey.
Thin Solid Films 1, 1–7 (2015).
192. Triana, C. a, Granqvist, C. G. & Niklasson, G. a. Electrochromic Properties of Li + -
Intercalated Amorphous Tungsten (aWO3- x ) and Titanium (aTiO2- x ) Oxide Thin
Films. J. Phys. Conf. Ser. 559, 012004 (2014).
193. Cai, G. F. et al. Ultra fast electrochromic switching of nanostructured NiO films
electrodeposited from choline chloride-based ionic liquid. Electrochim. Acta 87, 341–347
(2013).
194. Cai, G. et al. Hierarchical structure Ti-doped WO3 film with improved electrochromism
in visible-infrared region. RSC Adv. 3, 6896–6905 (2013).
195. Lu, Y., Liu, L., Mandler, D. & Lee, P. S. High switching speed and coloration efficiency
of titanium-doped vanadium oxide thin film electrochromic devices. J. Mater. Chem. C 1,
215
7380 (2013).
196. Granqvist, C. G. Electrochromic tungsten oxide: Review of progress 1993-1998. Sol.
Energy Mater. Sol. Cells 60, 201–262 (2000).
197. Woo, S. et al. Observation of room-temperature magnetic skyrmions and their current-
driven dynamics in ultrathin metallic ferromagnets. Nat. Mater. 15, 501–6 (2016).
198. Parkin, S. S. P., Hayashi, M. & Thomas, L. Magnetic Racetrack Memory. 320, 190–195
(2008).
199. Zhao, W. et al. Complementary Skyrmion Racetrack Memory With Voltage
Manipulation. IEEE Electron Device Lett. 37, 924–927 (2016).
200. Zhang, S., Baker, A. A., Komineas, S. & Hesjedal, T. Topological computation based on
direct magnetic logic communication. Sci. Rep. 5, 1–11 (2015).
201. Baek, S. H. C. et al. Complementary logic operation based on electric-field controlled
spin-orbit torques. Nat. Electron. 1, 398–403 (2018).