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.