FERROELECTRIC DOPED HAFNIUM OXIDE
AND ITS APPLICATION ON ELECTRONIC DEVICES
BY
HOJOON RYU
THESIS
Submitted in partial fulfillment of the requirements
for the degree of Master of Science in Electrical and Computer Engineering
in the Graduate College of the
University of Illinois at Urbana-Champaign, 2019
Urbana, Illinois
Adviser:
Assistant Professor Wenjuan Zhu
ii
ABSTRACT
Ferroelectricity is the material property such that we can induce spontaneous polarization,
reverse it and modulate it by varying the applied electrical field on the ferroelectric material.
Recently, doped hafnium oxide (HfO2) has garnered attention with its excellent scalability,
reliability, and compatibility with the current CMOS process. This thesis introduces two research
projects aimed at improving the electrical properties of ferroelectric-doped HfO2 for various
device applications. In the first project, we demonstrate a high-performance ferroelectric
aluminum (Al) doped HfO2 capacitor with Ti/Pd gate electrode. The remnant polarization
reaches up to 20 µC/cm2, endurance higher than 108 cycles and retention over ten years at room
temperature. In the second research, we demonstrate a ferroelectric tunneling junction (FTJ)
based on metal/aluminum oxide/zirconium doped HfO2/silicon structure. We show that this FTJ
has artificial synaptic behavior with symmetric synaptic weight change and tunable conductance.
We also show spike-timing-independent plasticity (STDP) can be obtained in this device, which
proves the possibility of using our FTJ as a neuromorphic computing chip.
iii
To my family
iv
ACKNOWLEDGMENTS
My utmost gratitude goes to my adviser, Professor Wenjuan Zhu, for her invaluable guidance
and consistent support of my academic research. I would also like to express my appreciation to
our group members, who provided insightful comments in overcoming problems I faced through
my study and research. Last but not least, I would like to thank my family—my parents and my
younger brother—for their unfailing spiritual support and continuous encouragement throughout
my research life here. This accomplishment would not have been possible without their support.
Thank you.
v
TABLE OF CONTENTS
CHAPTER 1: INTRODUCTION ........................................................................................1
CHAPTER 2: FERROELECTRICITY IN DOPED HAFNIUM OXIDE ...........................4
CHAPTER 3: NEW METAL ELECTRODE FOR FERROELECTRIC ALUMINUM-
DOPED HAFNIUM OXIDE ........................................................................7
3.1. Motivation of research ...............................................................................................7
3.2. Experimental process .................................................................................................8
3.3. Device result and discussion ......................................................................................9
CHAPTER 4: ALUMINUM OXIDE/ZIRCONIUM DOPED HAFNIUM
OXIDE FERROELECTRIC TUNNELING JUNCTION FOR
NEUROMORPHIC COMPUTING ............................................................15
4.1. Motivation of research .............................................................................................15
4.2. Experimental process ...............................................................................................16
4.3. Device result and discussion ....................................................................................16
CHAPTER 5: CONCLUSION AND FUTURE STUDY ..................................................23
REFERENCES ..................................................................................................................24
1
CHAPTER 1: INTRODUCTION
Ferroelectricity is the material property of having spontaneous polarization and reversible
switching under applied electrical field. The polarization induced by ferroelectric dipole is
retained even after external power is cut off, and the external electrical field is zero, as shown in
Fig 1.1. In the early 1950s, the concept of memory using the ferroelectric property was published
[1, 2]. Since then, ferroelectricity has been studied for application in electronics, such as the
nonvolatile semiconductor device. The operating principle and structure of ferroelectric memory
(FeRAM) are quite simple. As we can see in Fig 1.2, the FeRAM consists of only one, single
transistor (1T) structure. Here, researchers replace the conventional dielectric oxide material with
the ferroelectric material. When we apply a positive (negative) pulse to the gate, the polarized
charge in the ferroelectric layer is aligned downward (upward). This polarization results in
attracting electrons (holes) at the interface between the ferroelectric layer and substrate, which
makes the channel region. The conductance depends on the charge concentration in the channel
layer. The conductance variation leads to threshold voltage shift and drain current flowing
through the channel. This alteration affected by polarization is divided into different states
corresponding to ON state or OFF state. We distinguish the states by setting the reading voltage
within the range of two different threshold voltages. By sensing those “on” and “off” states
differently, we can assign different bit information, “0” and “1”, and use it as a binary bit
memory cell. Because ferroelectricity is non-volatile, we can save the long-term data after the
power is turned off [2].
2
Figure 1.1 The polarization (P)–electric field (E) loop of ferroelectric material when applying
electric field. It shows large hysteresis loop which contains two different remnant polarization
states (+Pr and -Pr) at zero electric field.
3
n n
SiO2
+ + + + +
- - - - -
+ + + + +
p
Si
VD
Writing “0”“Off”
Ferroelectric materials
(a)
n n
SiO2
+ + + + +
- - - - -
p
Si
VD
Writing “1”“On”
(b)
- - - - -
Vread
Figure 1.2. The operation principle of ferroelectric memory based on single transistor
structure, which uses ferroelectric material as dielectric layer. The plot shows the hysteresis of
drain current (Id) as a function of gate voltage. [2]
4
CHAPTER 2: FERROELECTRICITY IN DOPED HAFNIUM OXIDE
FeRAM is considered the most promising candidate among emerging new memory. Unlike
volatile memory like DRAM, FeRAM is a non-volatile memory which means it can store
information for a long period after the power is cut off. Compared to the current mainstream
non-volatile memory, FLASH memory, FeRAM has better endurance, higher writing speed and
lower power consumption [3]. Furthermore, with the capacitor-based structure of FeRAM and
ferroelectric field-effect transistor (FeFET), other charge storing devices using ferroelectricity
are possible. One possibility is ferroelectric tunneling junction (FTJ) using electron/hole
tunneling through the oxide layer [4]. FTJ has not only the non-destructive read-out scheme but
also better scalability compared to FeRAM because the FTJ scalability is not limited by the
amount of polarized charge storage. Recently, the ferroelectric device also has been considered
as a way to fabricate an artificial synaptic device for a neuromorphic computing system, which
has the advantages of handling both analog and digital information, massive parallel calculation,
and low power consumption [5]. Compared to other candidates for analog synapses, such as
resistive switching memory (RRAM), the ferroelectric device has advantages in terms of
symmetry in conductance variation during potentiation and depression, good recognition
accuracy, and high speed [5]. However, using conventional ferroelectric material as a device has
some critical disadvantages. The conventional ferroelectric perovskite oxide materials, such as
lead zirconate titanate (PZT), have a scalability problem. When scaling down, a device based on
conventional ferroelectric material loses its polarization due to size-effect [6]. However, hafnium
oxide (HfO2) doped with specific elements has emerged as a new material to overcome the
scalability limitation. In 2011, T. Böscke at el. [7] reported silicon doped HfO2 has a specific
5
crystal phase (orthorhombic phase, o-phase), showing ferroelectricity even for film thickness
below 10 nm. In addition to scalability, doped HfO2 also shows high coercive and breakdown
voltage helping to prevent charge depolarization. As we can see in table 2.1, the compatibility
with the current CMOS process and capacity for atomic layer deposition (ALD) are other
advantages [8]. The origin of ferroelectricity in doped HfO2 film comes from its structural
transformation and atomic displacement because the dipoles inducing ferroelectricity are closely
related to the lattice structure. As shown in Fig 2.1, doped HfO2 undergoes a phase
transformation from the initial state (tetragonal, t-phase or monoclinic phase, m-phase) to non-
centrosymmetric orthorhombic phase (o-phase) which shows ferroelectricity under specific
conditions [9]. Several factors, like doping effect, mechanical stress from the top and bottom
substrate, annealing recipe, interface, and film thickness affect the formation of ferroelectric
phase [10]. Here, we mainly focus on doping element and the top electrode for material
optimization because those factors are easily tunable to improve ferroelectricity.
6
Figure 2.1 Several factors influence the ferroelectric properties in doped HfO2 (here, Zr-
doped HfO2, HZO) and the atomic structure variation depending on the polarization state.
[10]
Table 2.1: Comparison of properties of conventional ferroelectric perovskite oxide materials,
such as SBT and PZT, and the ferroelectric doped HfO2. [8]
7
CHAPTER 3: NEW METAL ELECTRODE FOR FERROELECTRIC
ALUMINUM-DOPED HAFNIUM OXIDE
3.1. Motivation of research
To make HfO2 much more desirable, researchers studied several atomic elements or doping
methods. Several dopants have been reported to be able to induce ferroelectricity in HfO2,
including silicon (Si) [11], aluminum (Al) [12], gadolinium (Gd) [13], and yttrium (Y) [14].
Among several dopants, we first chose aluminum because Al2O3 is commonly used as a high-k
dielectric. As we already mentioned, Al-doped HfO2 undergoes a ferroelectric phase
transformation under certain conditions. Ferroelectricity of film can also be tuned through
modulating the top electrode. During the crystallization process through annealing, the top
electrode serves as a capping layer which prevents the volume expansion and shearing of the
HfO2 unit cell. It prevents the formation of the non-ferroelectric phase and helps to stabilize
ferroelectricity. Typically, the TiN electrode is used. For Al-doped HfO2, however, previous
works show that remnant polarization (Pr) of Al-doped HfO2 with TiN is only in the range of 5-
15 µC/cm2 for planar capacitor structure [15, 16]. To further enhance the ferroelectricity for
electronics applications, a systematic study to investigate new metal electrodes is required. In the
first research, we fabricated ferroelectric capacitors based on Al-doped HfO2 with various metal
electrodes including W, Ti/Au, and Ti/Pd. We compared our cases to the conventional case using
TiN. All figures in chapter 3 are from the reference [17], © 2019 IEEE, and reprinted with
permission.
8
3.2. Experimental process
Planar metal–insulator (ferroelectric Al-doped HfO2)–semiconductor (MIS) capacitors were
fabricated on highly doped P-type Si substrates. We deposited 20 nm thick Al-doped HfO2 by
using an ALD technique. The Al concentration could be tuned by varying the cycle number ratio
between the Hf precursor and Al precursor. The Ti/Au and Ti/Pd electrodes were deposited by e-
beam evaporation, while TiN and W electrodes were deposited by sputtering. The encapsulated
HfO2 films were then annealed in a rapid thermal annealing (RTA) system. We varied annealing
temperature from 800 ℃ to 1000 ℃ for 1-2 seconds for the experiment.
50 nm
Al:HfO2
Ti/Pd
Si2 nm
(a) (b) Al-doped HfO2
Pt
Figure 3.1 (a) Cross-sectional TEM image of the Al-doped HfO2 capacitor with Ti/Pd
electrode on Si substrate annealed at 950 oC. (b) A close-up of the cross-sectional TEM image
clearly showing that the Al-doped HfO2 layer is crystallized. [17]
9
3.3. Device result and discussion
Figure 3.1 (a)-(b) show cross-sectional TEM images of the ferroelectric capacitor. Figure 3.2
(a) shows the remnant polarization (Pr) variation depending on four electrodes. The Pr value of
the sample was measured through pulse measurement called positive-up-negative-down (PUND)
measurement [18]. Among samples, the Ti/Pd electrode gives the highest remnant polarization
value in the ferroelectric capacitors. It reached 20 µC/cm2 at 10 V, which is the best of all
published results. We think that the internal stress caused by Ti/Pd makes it feasible to form the
ferroelectric phase of doped HfO2. As shown in Fig 3.2 (c), the leakage current density of
capacitors with Ti/Pd electrode is also lower than that of other cases. We speculate this result can
be from several factors caused by different electrodes, such as different resistivity, thermal
stability, and interface state. The endurance of the capacitors with Ti/Pd is much better than that
with other samples. In Fig 3.3, the endurance of the capacitors with Ti/Pd is higher than 108
cycles at the ±7 V program/erase applied voltage. Also, in Fig 3.4, the retention of the device
with Ti/Pd is much longer than that with W. For the MIM capacitor (Ti/Pd top and TiN bottom),
over 90% polarization remains after 10 years through linear extrapolation. In Fig 3.5, when the
annealing temperature increases, the pinched hysteresis loop disappears, and the Pr increases,
indicating that a larger portion of the film is transforming from less-aligned phase with multi-
domain to the ferroelectric phase with higher Pr and hysteresis loop. As we can see in Fig 3.5 (g),
the coercive voltage (Vc) increases monotonically with increasing annealing temperature, due to
the increasing thickness of the interfacial layer called ‘dead layer’. As the temperature increases,
the leakage current increases dramatically due to the crystallization of the HfO2 film, as shown in
Fig 3.5 (i). In Fig 3.6, we varied the Al concentration in the HfO2 by tuning the cycle number
10
ratio between the Hf and Al precursor. We found that an optimal Hf-to-Al ratio is within the
range of 21:1 to 25:1, regardless of the electrode. This range provides additional information on
how to design and fabricate for high enough ferroelectricity. In Fig 3.7, a steep increase in
switching polarization (Psw) can be observed for pulses with high amplitude, whereas for pulses
with low amplitude, only slow polarization reversal occurs. The amount of ferroelectric
polarization can be continuously tuned by external pulses with varying writing pulse amplitude
and width. This tunability is very important when we fabricate the ferroelectric artificial synaptic
device for neuromorphic computing, which requires tunable conductance. Capacitors with Ti/Pd
show higher Pr and larger tenability window than those with W electrode.
11
Ti-Au Ti-Pd TiN W0
5
10
15
20
Pr
(C
/cm
2)
Top Electrode
20nm Al:HfO2
Hf:Al=23:1
Ti-Au Ti-Pd TiN W10-5
10-4
10-3
10-2
10-1
100
101
20nm Al:HfO2
Hf:Al=23:1
J @
5V
(A
/cm
2)
Top electrode
(a) (b)
-10 -5 0 5 10
-20
-10
0
10
20
Po
lari
za
tio
n (
C/c
m2)
Voltage (V)
Ti/Au
Ti/Pd
TiN
W
(c) (d)
0 1 2 3 4 5
10-6
10-5
10-4
10-3
10-2
10-1
Cu
rre
nt
de
ns
ity
J (
A/c
m2)
Voltage (V)
Ti/Au
Ti/Pd
TiN
W
Figure 3.2 (a) P-V loops of Al-doped HfO2 capacitors with various top electrodes (Ti/Pd,
Ti/Au, W and TiN). The thicknesses of Ti/Pd, Ti/Au, and W electrodes are 40-50 nm, while
the thickness of TiN is 100 nm. (b) Statistics of the remnant polarization of the Al-doped
HfO2 capacitors with various top electrodes. (c) Leakage current density as a function of gate
voltage and (d) statistics of the leakage current density at 5 V of Al-doped HfO2 capacitors
with various top electrodes. [17]
12
100 101 102 103 104-10
-5
0
5
10
15
Pr (
C/c
m2)
Time (s)
Ti/Pd
W
MIS structure
(c)
100 101 102 103 104-9
-6
-3
0
3
6
9
12
Pr (
C/c
m2)
Time (s)
MIS structure
MIM structure
Top electrode : Ti/Pd
(d)
Highly doped Si
Al-doped HfO2
Metal
VG
Highly doped Si
Al-doped HfO2
Metal
Metal
VG
(a) (b)
Figure 3.4 Retention of Al-doped HfO2. (a) and (b) are illustrations of a MIS and a MIM
capacitor with ferroelectric HfO2. (c) Retention of MIS Al-doped HfO2 capacitors with Ti/Pd
and W top electrodes. (d) Retention of MIS and MIM Al-doped HfO2 capacitors with Ti/Pd
top electrodes. The retention is tested at room temperature. [17]
100 101 102 103 104 105 106 107 108
-30
-20
-10
0
10
20
30
Pr (
C/c
m2)
Cycle
+/- 10V
+/- 7V
Ti/Pd
(a) (b)
100 101 102 103 104 105 106 107 108
-20
-10
0
10
20P
r (
C/c
m2)
Cycle
Ti/Pd
W
TiN
Pulse: 7V
Figure 3.3 (a) Endurance of Al-doped HfO2 capacitors with various top electrodes. The
amplitude of the cycling pulses is 7 V and the amplitude of the PUND read pulses is 10 V. (b)
Endurance of Al-doped HfO2 capacitors with Ti/Pd electrode measured at two different
cycling voltages: ±7 V and ±10 V. [17]
13
Figure 3.5 The impact of annealing temperature on the properties of the Al-doped HfO2 film.
(a)-(e) P-V loops of Al-doped HfO2 annealed at various temperatures from 800 oC to 1000 oC.
The top electrodes are Ti/Pd. The amplitude of the PUND read pulses is 10 V. (f) Remnant
polarization as a function of annealing temperature for Al-doped HfO2 capacitors with Ti/Pd,
Ti/Au and W electrodes. The amplitude of the PUND read pulses is 7 V. (g) 2 Vc as a
function of annealing temperature for Al-doped HfO2 capacitors with Ti/Pd electrode. The
inset illustrates the definition of 2 Vc: the difference between positive and negative coercive
voltage in the P-V loop. (h) P-V loops of Al-doped HfO2 annealed at 800 oC before and after
cycling 104 pulses. The amplitude of the cycling pulses is 10 V. (i) Leakage current density at
5 V as a function of annealing temperature of the Al-doped HfO2 capacitor with various
electrodes. [17]
14
102 103 104 105 1064
8
12
16
20
24
Ti/Pd
Ps
w (
C/c
m2)
Pulse Width (ns)
7V
6V
5V
4V
3V
20nm Al:HfO2
Hf:Al=23:1
Preset Read pulses
Write pulse
(a)
(b)
102 103 104 105 1060
4
8
12
16
20
24
Ps
w
(C
/cm
2)
Pulse Width (ns)
8V
7V
6V
5V
4V
3V
W
20nm Al:HfO2
Hf:Al=23:1
(c)
Figure 3.7 (a) Illustration of the pulse sequence for variable write polarization measurement.
(b) Switching polarization (Psw) as a function of pulse width at various pulse voltages for 20
nm Al-doped HfO2 with Ti/Pd electrode on silicon substrate. (c) Switching polarization (Psw)
as a function of pulse width at various pulse voltages for 20 nm Al-doped HfO2 with W
electrode on silicon substrate. [17]
16 20 24 28 320.0
2.5
5.0
7.5
10.0
12.5
Pr (
C/c
m2)
Hf : Al cycle ratio
20nm Al:HfO2
TiN
10V
16 18 20 22 24 26 281
2
3
4
5
6
7
8P
r (
C/c
m2)
Hf:Al cycle ratio
920oC anneal
880oC anneal
20nm Al:HfO
Ti/Pd
7V
(a) (b)
Figure 3.6 Remnant polarization as a function of Hf-to-Al cycle ratio for 20 nm Al-doped
HfO2 capacitors with (a) Ti/Pd electrodes and (b) TiN electrodes. The amplitude of the PUND
read pulse is 7 V for the Ti/Pd capacitors, while it is 10 V for the TiN capacitors. [17]
15
CHAPTER 4: ALUMINUM OXIDE/ZIRCONIUM DOPED HAFNIUM OXIDE
FERROELECTRIC TUNNELING JUNCTION
FOR NEUROMORPHIC COMPUTING
4.1. Motivation of research
Ferroelectric tunneling junction (FTJ), with tunable tunneling electroresistance (TER), is
promising for many emerging applications, including non-volatile memories and neurosynaptic
computing. Nowadays, doped hafnium oxide (HfO2) has emerged as a new ferroelectric material
[19]. Among various doped HfO2, Zr-doped HfO2 (HZO) has garnered attention because of its
low annealing temperature (500-600 ℃) with acceptable Pr value [20]. In FTJs based on
metal/HZO/metal (MFM) structure, to obtain enough TER ON/OFF ratio to use, the thickness of
HZO needs to be scaled down to sub-5 nm. However, getting proper polarization in the ultra-thin
layer is challenging [21]. In this research, we fabricate a new type of FTJ based on
metal/aluminum oxide (Al2O3)/HZO/Si structure. The interfacial Al2O3 layer and
semiconducting substrate enable acceptable TER ratio even though the thickness of HZO is
above 10 nm. We additionally demonstrate an artificial synaptic device based on this FTJ with
symmetric potentiation and depression characteristics and widely tunable conductance. We also
show that spike-timing-dependent plasticity (STDP), the biological property of human neurons
for memory and learning, can be harnessed from HZO based FTJs. All figures in this chapter are
from the reference [22], © 2019 IEEE, and reprinted with permission.
16
4.2. Experimental process
Planar ferroelectric metal-insulator-ferroelectric-semiconductor (MFIS) capacitors were
fabricated on highly doped P-type Si substrates, illustrated in Fig 4.1 (a). 12 nm thick zirconium
(Zr) doped HfO2 was deposited by using an ALD. We alternately stacked HfO2 and ZrO2 layers
by using the Hf precursor and Zr precursor, fixed with a 1:1 cycle ratio. Then, we deposit Al2O3
by using ALD again. The encapsulated HfO2 films were then annealed in a rapid thermal
annealing (RTA) system. The 10/40 nm thick Ti/Au electrodes were deposited by e-beam
evaporator for the top electrode.
4.3. Device result and discussion
FTJ based on HZO: The device structure is illustrated in Fig 4.1 (a) and the polarization-
voltage (P-V) loops are shown in Fig 4.1 (b). The MIFS capacitor shows Pr around 20 µC/cm2 at
10 V. We found that the Al2O3/HZO structure shows current switching with the non-volatile,
hysteretic loop as shown in Fig 4.2 (a). When we apply positive (negative) pulses, it induces a
low (high) current which means OFF (ON) state. The corresponding I-V curve to both states is
shown as the inset of Fig 4.2 (b). We can see that the TER ratios of FTJs based on
metal/Al2O3/HZO/Si are higher than those on metal/HZO/Cr and metal/Al2O3/HZO/Cr
structures. The tunneling phenomena can be explained by the following analysis. Figure 4.3
shows the plot of versus , where I is the tunneling current and V is the voltage we
applied. At high bias, decreases linearly with , indicating that Fowler-Nordheim
(F-N) tunneling dominates the charge transport. At low bias, increases logarithmically
with , which is consistent with direct tunneling [23]. For the TER hysteresis loop shown in
17
-10 -5 0 5 10
-20
-10
0
10
20
Po
lari
za
tio
n (
C/c
m2)
Voltage (V)
7V
7.5V
8V
8.5V
9V
9.5V
10V
P type Si
Zr doped HfO2
Ti/AuAl2O3
(a) (b)
Figure 4.1 (a) Illustration of FTJs based on Al2O3/HZO/Si. (b) Polarization-voltage (P-V)
loops of a FTJ with 3nm Al2O3/12nm HZO. [22]
0 1 20
1
2
3
Cu
rre
nt
(nA
)
Read voltage (V)
After -10V pulse
After +10V pulse
Ti/Al2O3/HZO/Si
-10 -5 0 5 100.0
0.5
1.0
1.5
Co
nd
uc
tan
ce
(n
S)
Write Pulse Voltage (V)
(a) (b)
0
1
2
3
4
5
Ti/HZO
/Cr
Ti/Al2O3
/HZO/Cr
TE
R O
n/O
ff r
ati
o
Ti/Al2O3
/HZO/Si Figure 4.2 (a) Hysteresis loops of the tunneling conductance. The pulse trains are shown
schematically in the inset. (b) TER ratio of FTJs based on Ti/HZO/Cr, Ti/Al2O3/HZO/Cr, and
Ti/Al2O3/HZO/Si. The inset shows the IV curves measured after -10 V and +10 V program
pulses. [22]
18
Fig 4.2 (a), the read voltage is 2 V, which is in the F-N tunneling dominant zone. The energy
diagrams of the FTJs after positive (OFF state) and negative (ON state) pulses are shown in Fig
4.4 (a)-(b). When a negative pulse is applied to the top electrode, the polarization in HZO points
to the top electrode. The screening charge drives p-type silicon into accumulation, which will
reduce tunneling width and increase the tunneling current. A positive read voltage will further
enhance the band tilt in HZO and increase the F-N tunneling current. These two factors will lead
to high conductance in FTJ. In our experiment, the TER ratio shows its peak around 3 nm thick
Al2O3 and 10-12 nm HZO, as shown in Fig 4.5. This is because too-thin HZO with Al2O3 makes
it hard to get proper Pr value. On the other hand, if the layers are too thick, it is difficult for the
charge to tunnel through due to the barrier thickness. Figure 4.6 shows the area and reading
voltage effect on the TER ratio of our FTJ for better design and performance.
For the artificial synaptic device that mimics the human brain, the tunneling conductance
needs to be continuously tunable to emulate biological synapses. Figure 4.7 (a)-(c) show the
effect of multiple pulse schemes on the conductance of the device. All figures show the device
-5 0 5 10 15-26
-24
-22
-20
-18
-16
ln(I
/V2)
1/V(V-1)
10nm
12nm
15nm
F-N Tunneling Direct Tunneling
Figure 4.3 versus for FTJs with various HZO thicknesses. [22]
19
current increases and decreases with a pulse, which means artificial synaptic weight can be tuned
by the designed pulse scheme. Among them, pulse amplitude modulation shows the best data in
terms of discrete multilevel states, linearity, and symmetry. For biologic synapses, STDP is
considered as an important mechanism for unsupervised learning. Figure 4.8 shows the STDP
characteristics of our device. We emulate the spikes from pre- and post-neurons by using the
waveforms shown in Fig 4.8 (b). Depending on pulse arrival time difference between pre- and
post-neuron, the synaptic connection changes.
+++
---
+++
--
DepletionEc
Ev
EF
Ti Al2O3 HfZrO P-Si
(b) After positive write pulse
Barrier width
P
+++
---
++
---
AccumulationEc
Ev
EF
Ti Al2O3 HfZrO P-Si
Barrier width
(a) After negative write pulse
P
Figure 4.4 Energy band diagrams of the FTJ after positive and negative write pulses,
respectively. [22]
20
0 5 10 15 20 25
1
2
3
4
5
TE
R r
ati
o
HZO thickness (nm)
0 1 2 3 4 5
1
2
3
4
5
TE
R r
ati
o
Al2O3 thickness (nm)
Figure 4.5 TER ratio as a function of Al2O3 thickness and HZO thickness. [22]
0.5 1.0 1.5 2.01.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
TE
R r
ati
o
Reading Voltage (V)
44 x 44 166 x 166 304 x 304 398 x 2000
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
TE
R r
ati
o
FTJ area (m2)
Figure 4.6 TER ratio as a function of FTJ area and reading voltage. [22]
21
0 10 20 30 40 50 600.0
0.5
1.0
1.5
Depression
Co
nd
ucta
nce (
nS
)
Number of Pulses
Potentiation
0 10 20 30 40 50 600.00
0.25
0.50
0.75
Depression
Co
nd
uc
tan
ce
(n
S)
Number Of Pulses
Potentiation
0 10 20 30 40 50 60
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Depression
Co
nd
uc
tan
ce
(n
S)
Number of Pulses
Potentiation
(a) (c)(b)
10µS
Figure 4.7 Tunneling conductance of FTJs as a function of pulse numbers. (a) Scheme 1:
constant pulse amplitude and width. (b) Scheme 2: modulation of pulse width. (c) Scheme 3:
modulation of pulse amplitude. [22]
Figure 4.8 (a) Sketch of pre- and post-neurons connected by a synapse. (b) Illustration of the
pre- and post-synaptic pulses applied on FTJ. (c) Measurement of STDP in a FTJ based on
Al2O3/HZO stack. [22]
22
When the post-neuron spike arrives after the pre-neuron spike, the conductance of the
FTJ increases, which means the synaptic connection is strengthened. However, when the pre-
neuron spikes come after the post-neuron, the synaptic connection is weakened. Close-timed
spikes produce a large conductance change, whereas long delays between two pulses leave the
FTJ unchanged, which also can be observed in biologic synapses. The retention of the device is
shown in Fig 4.9. Through linear extrapolation, the device still shows a residual on/off window
after 10 years. The degradation of retention may come from the depolarization field due to
unscreened charges at the oxide layer interface. As further work, through optimized interfacial
engineering, we will try to improve the reliability of our FTJ.
100 101 102 103 104 105 106 107 108 109
-10
-8
-6
-4
-2
0
2
4
6
8
10
Pr (
C/c
m2)
Time (sec)
+8V
-8V 10 years
(a) (b)
100 101 102 103 104 105 106 107 108 1090.01
0.1
1
10
Co
nd
uc
tan
ce
(n
S)
Time (sec)
ON state
OFF state
Vwrite : +/- 8V
Vread : 2V
10 years
Figure 4.9 Retention characteristics. (a) Remnant polarization of ON and OFF state as a
function of retention time in our FTJs. (b) ON and OFF state resistance as a function of
retention time in FTJs based on Al2O3/HZO stack. [22]
23
CHAPTER 5: CONCLUSION AND FUTURE STUDY
In summary, we explored three new metal materials as top electrodes for Al-doped HfO2:
Ti/Pd, Ti/Au, and W. First, we found that the capacitors with Ti/Pd electrodes have much higher
remnant polarization, and better endurance and data retention as compared to those with TiN, W,
and Ti/Au electrodes. These results indicate that Ti/Pd is a very promising electrode candidate
for ferroelectric Al-doped HfO2. Based on the optimized process conditions, we demonstrated
high-performance ferroelectric Al-doped HfO2 with remnant polarization up to 20 µC/cm2 at 10
V, endurance higher than 108 cycles and data retention persisting after 10 years. The results
demonstrated the feasibility of high-performance ferroelectric Al-doped HfO2. To explore the
application of ferroelectric doped HfO2, we fabricated ferroelectric tunneling junctions based on
Al2O3/HZO stack and tested their suitability as artificial synaptic devices. The FTJ with
optimized HZO and Al oxide thickness shows non-volatile hysteresis loop with high ON/OFF
ratio. Pre-and post-neuron pulses with precisely designed waveforms and timing induce synaptic
weight change and STDP behavior, which indicates the possibility of using an FTJ for
neuromorphic computing. This study broadens our perspective for using this ferroelectric doped
HfO2 in various applications including ferroelectric memories, tunneling junctions, and artificial
synaptic devices. For future work, we will improve device performance through fabrication
optimization and extend the applications by combining the ferroelectric hafnium oxide with two-
dimensional (2D) materials.
24
REFERENCES
[1] D. A. Buck, "Ferroelectrics for Digital Information Storage and Switching," master's
thesis, Massachusetts Institute of Technology, 1952.
[2] R. Lous, "Ferroelectric memory devices," How to Store Information of the Future, 2011.
[3] R. Jones Jr et al., "Ferroelectric non-volatile memories for low-voltage, low-power
applications," Thin Solid Films, vol. 270, no. 1-2, pp. 584-588, 1995.
[4] V. Garcia and M. Bibes, "Ferroelectric tunnel junctions for information storage and
processing," Nature Communications, vol. 5, p. 4289, 2014.
[5] M. Jerry et al., "Ferroelectric FET analog synapse for acceleration of deep neural
network training," in 2017 IEEE International Electron Devices Meeting (IEDM), 2017,
pp. 6.2.1-6.2.4.
[6] N. Gong and T.-P. Ma, "Why is FE–HfO2 more suitable than PZT or SBT for scaled
nonvolatile 1-T memory cell? A retention perspective," IEEE Electron Device Letters,
vol. 37, no. 9, pp. 1123-1126, 2016.
[7] T. Böscke, J. Müller, D. Bräuhaus, U. Schröder, and U. Böttger, "Ferroelectricity in
hafnium oxide thin films," Applied Physics Letters, vol. 99, no. 10, p. 102903, 2011.
[8] Z. Fan, J. Chen, and J. Wang, "Ferroelectric HfO2-based materials for next-generation
ferroelectric memories," Journal of Advanced Dielectrics, vol. 6, no. 02, p. 1630003,
2016.
[9] M. H. Park et al., "A comprehensive study on the structural evolution of HfO2 thin films
doped with various dopants," Journal of Materials Chemistry C, vol. 5, no. 19, pp. 4677-
4690, 2017.
[10] S. J. Kim, J. Mohan, S. R. Summerfelt, and J. Kim, "Ferroelectric Hf0.5Zr0.5O2 thin films:
A review of recent advances," JOM, vol. 71, no. 1, pp. 246-255, 2019.
[11] T. Böscke et al., "Phase transitions in ferroelectric silicon doped hafnium oxide," Applied
Physics Letters, vol. 99, no. 11, p. 112904, 2011.
[12] S. Mueller et al., "Incipient ferroelectricity in Al‐doped HfO2 thin films," Advanced
Functional Materials, vol. 22, no. 11, pp. 2412-2417, 2012.
[13] S. Mueller, C. Adelmann, A. Singh, S. Van Elshocht, U. Schroeder, and T. Mikolajick,
"Ferroelectricity in Gd-doped HfO2 thin films," ECS Journal of Solid State Science and
Technology, vol. 1, no. 6, pp. N123-N126, 2012.
[14] J. Müller et al., "Ferroelectricity in yttrium-doped hafnium oxide," Journal of Applied
Physics, vol. 110, no. 11, p. 114113, 2011.
[15] P. Polakowski et al., "Ferroelectric deep trench capacitors based on Al:HfO2 for 3D
nonvolatile memory Applications," 2014 IEEE 6th International Memory Workshop
(IMW), 2014.
[16] U. Schroeder et al., "Hafnium oxide based CMOS compatible ferroelectric materials,"
ECS Journal of Solid State Science and Technology, vol. 2, no. 4, pp. N69-N72, 2013.
[17] H. Ryu, K. Xu, J. Kim, S. Kang, J. Guo, and W. Zhu, "Exploring new metal electrodes
for ferroelectric aluminum-doped hafnium oxide," IEEE Transactions on Electron
Devices, vol. 66, no. 5, pp. 2359-2364, 2019.
25
[18] J. F. Scott, C. Araujo, H. B. Meadows, L. McMillan, and A. Shawabkeh, "Radiation
effects on ferroelectric thin‐film memories: Retention failure mechanisms," Journal of
Applied Physics, vol. 66, no. 3, pp. 1444-1453, 1989.
[19] J. Müller, P. Polakowski, S. Mueller, and T. Mikolajick, "Ferroelectric hafnium oxide
based materials and devices: Assessment of current status and future prospects," ECS
Journal of Solid State Science and Technology, vol. 4, no. 5, pp. N30-N35, 2015.
[20] J. Muller et al., "Ferroelectricity in simple binary ZrO2 and HfO2," Nano Letters, vol. 12,
no. 8, pp. 4318-4323, 2012.
[21] A. Chouprik et al., "Electron transport across ultrathin ferroelectric Hf0.5Zr0.5O2 films on
Si," Microelectronic Engineering, vol. 178, pp. 250-253, 2017.
[22] H. Ryu, H. Wu, F. Rao, and W. Zhu, "Ferroelectric tunneling junctions for neurosynaptic
computing," in 2019 77th Device Research Conference (DRC), 2019, pp. 179-180.
[23] J. M. Beebe, B. Kim, J. W. Gadzuk, C. D. Frisbie, and J. G. Kushmerick, "Transition
from direct tunneling to field emission in metal-molecule-metal junctions," Physical
Review Letters, vol. 97, no. 2, p. 026801, 2006.