University of Alabama in Huntsville

LOUIS

Dissertations UAH Electronic Theses and Dissertations

2023

Exploring reliability issues in non-volatile memories using

numerical simulation
Sijay Huang

Follow this and additional works at: https://louis.uah.edu/uah-dissertations

Recommended Citation
Huang, Sijay, "Exploring reliability issues in non-volatile memories using numerical simulation" (2023).
Dissertations. 283.
https://louis.uah.edu/uah-dissertations/283

This Dissertation is brought to you for free and open access by the UAH Electronic Theses and Dissertations at
LOUIS. It has been accepted for inclusion in Dissertations by an authorized administrator of LOUIS.
 EXPLORING RELIABILITY ISSUES IN NON-VOLATILE
MEMORIES USING NUMERICAL SIMULATION

Sijay Huang

A DISSERTATION

Submitted in partial fulfillment of the requirements

for the degree of Doctor of Philosophy
in
The Department of Electrical and Computer Engineering
to
The Graduate School
of
The University of Alabama in Huntsville
May 2023

Approved by:

Dr. Biswajit Ray, Research Advisor

Dr. James K. Baird, Committee Member
Dr. Ravi Gorur, Committee Member
Dr. Sivaguru S. Ravindran, Committee Member
Dr. B. Earl Wells, Committee Member
Dr. Aleksandar Milenkovic, Department Chair
Dr. Shankar Mahalingam, College Dean
Dr. Jon Hakkila, Graduate Dean
 Abstract

EXPLORING RELIABILITY ISSUES IN NON-VOLATILE

MEMORIES USING NUMERICAL SIMULATION

Sijay Huang

A dissertation submitted in partial fulfillment of the requirements

for the degree of Doctor of Philosophy

The Department of Electrical and Computer Engineering

The University of Alabama in Huntsville

May 2023

Semiconductor non-volatile memory (NVM) devices are predominantly used for

storing the large amount of data produced by the proliferation of the Internet of Things

(IoT), mobile networks, and sensor node applications. Since data are growing at an

exponential rate, future NVM solutions need to provide even higher bit densities, superior

performance, better energy efficiency, and improved resilience. Technological scaling has

been the primary driving force for increasing the bit density of semiconductor NVM

technologies. However, the scaling leads to reliability challenges, such as endurance, data

retention, and thermal noise effects. This dissertation numerically evaluates reliability

issues associated with the two important classes of semiconductor memory technology: (1)

electron charge-based 3D NAND flash memory and (2) electron spin-based

magnetoresistive random-access memory (MRAM). A commercial device simulator

(called Taurus Medici) from Synopsys has been used for the reliability analysis of the 3D

NAND flash memory. The simulator solves Poisson's equation and the drift-diffusion

transport equations to predict the electrical properties of the cell based on its geometry and

ii
material characteristics. A number of novel device design ideas are proposed and validated

by numerical simulation, which can potentially enable further scaling of the future

generation of 3D NAND technologies. MRAM reliability is studied by focusing on the

magnetic tunnel junction (MTJ), which is the core component of the memory cell. A

simulation framework has been developed by numerically solving the Landau–Lifshitz–

Gilbert equation, which describes the dynamics of the magnetization as a function of the

external magnetic field, the spin current, and the ambient thermal fluctuations. Finally, a

Monte Carlo model is established to analyze the dielectric breakdown of insulating layers,

which are common components in both types of semiconductor memory.

iii
 ©

Sijay Huang

All Rights Reserved

iv
 Acknowledgements

My dissertation is concerned with semiconductor, dielectric, and magnetic

materials, which are the major components built in the latest designs of semiconductor non-

volatile memory. This interdisciplinary approach involved research across physics, applied

mathematics, materials science, and electrical engineering fields. The credit for the

formation of this dissertation is attributed to these exceptional individuals.

First and foremost, I would like to express my sincere appreciation to my advisor,

Dr. Biswajit Ray, for his unwavering support and guidance throughout my learning of

semiconductor device physics and electrical engineering Ph.D. journey. His vast

knowledge, dedication, and research passion have inspired and motivated me.

I would like to express my deep gratitude to my dissertation committee members,

Dr. James K. Baird, Dr. Ravi Gorur, Dr. Sivaguru S. Ravindran, and Dr. B. Earl Wells, for

their invaluable feedback, insightful suggestions, and expertise in their respective fields.

Dr. Baird's expertise in chemical physics, particularly in thermodynamics, chemical

kinetics, and electrodynamics, provided me with a strong theoretical foundation. Dr.

Gorur’s expertise in electrical insulation improved my understanding of the fundamental

principles underlying my dielectric breakdown research. Dr. Ravindran’s expertise in

numerical computation was instrumental in helping me develop robust numerical

simulations. Finally, Dr. Wells’ computer engineering expertise helped me incorporate

physics and materials science ideas into developing effective computational codes for

electrical engineering.

I would like to thank my university, UAH, for offering me the teaching assistantship

that has allowed me not only to foster my teaching abilities but also to contribute to the

v
education of others. I want to extend my thanks to the Defense Advanced Research

Projects Agency (DARPA) for the research assistantship in working with Dr. Zubin Jacob's

research group at Purdue University. The collaboration in micromagnetics and spintronics

enriched my understanding of the field and enabled me to apply my modeling and

simulation skills with brilliant experimental designers. In addition, I thank Dr. Timothy

Boykin for the dielectrics research collaboration and Dr. Shangbing Ai for the excellent

mathematics lectures with after-class discussions. Furthermore, I thank Ms. Annette

Archer and retired staff member Mrs. Jackie Siniard at the Department of Electrical and

Computer Engineering for their educational assistance. Finally, I appreciate the

department chairman Dr. Aleksandar Milenkovic’s suggestion on my dissertation revision.

I am deeply grateful to my father, Mr. Chun-Nan Huang, and my mother, Mrs. Pi-

Chu Hsieh, for their encouragement and love. Their encouragement played a crucial role

in my decision to pursue higher education in the United States. Their love and guidance

have been a constant source of strength and inspiration. I want to express my gratitude

towards my sister, Yichien, for her selfless assistance and care for our parents while I

pursued my academic aspirations. Her devotion to our family has inspired me, and I am

genuinely thankful for her assumption of responsibility. I am also thankful to my brother,

Siyao, whose independence and self-sufficiency let me focus on my studies without

worrying about his well-being. I have felt proud and inspired by his capacity to care for

himself and work towards his objectives. I am truly blessed to have such a loving family

whose unconditional love and steadfast support have helped me achieve this significant

milestone in my life.

vi
 My heartfelt thanks also go to my dear friends Steve and Kathy Owens, who

welcomed me into their home and treated me like family. Their warmth and hospitality

have offered me a sense of belonging and comfort. In addition, I am grateful to Alice and

Yanfang Hwang for leading the Ark Fellowship, which provided a nurturing community.

I benefited from the UAH InFOCUS Fellowship helpers through their kindness and

generosity. Thank you all from the bottom of my heart. I appreciate my friends, Fathi

Aldukali, Wen-Shuo Chang, Chi-Long Feng, Yuan Hsiang Lin, Milton Shen, and Bo Xu,

for backing me and sharing their life experiences. I am very grateful to my colleagues in

the Ray research group for research discussions and problem solutions.

Last but not least, I would like to reiterate my thanks to Dr. Baird for his assistance

in refining my dissertation. I also thank Ms. Simone Frazier for her timely advice during

the final stages of my dissertation, which helped me submit a well-crafted document.

To all those who have supported me in various ways throughout my academic

journey, thank you for being a part of this meaningful achievement in my academic career.

I dedicate this dissertation to my parents. I love you!

vii
 Table of Contents

Abstract.............................................................................................................................. ii

Acknowledgements ........................................................................................................... v

Table of Contents ........................................................................................................... viii

List of Figures.................................................................................................................... x

List of Tables .................................................................................................................. xiii

List of Symbols ............................................................................................................... xiv

Chapter 1. Introduction ........................................................................................... 1

1.1 3D NAND Flash Memory............................................................................. 2

1.2 Magnetoresistive Random-access Memory .................................................. 5

1.3 Scope and Outline ......................................................................................... 6

Chapter 2. 3D NAND Flash Memory Cell Current Enhancement ...................... 8

2.1 Overview ....................................................................................................... 8

2.2 Modeling and Simulation Framework ........................................................ 12

2.3 Analysis of the Cell Current Bottlenecks.................................................... 17

2.4 New Designs to Boost the Cell Current ...................................................... 21

2.5 Summary ..................................................................................................... 24

Chapter 3. Dynamics in Magnetoresistive Random-access Memory ................. 28

3.1 Background ................................................................................................. 28

3.2 Micromagnetics and Magnetic Tunnel Junction ......................................... 31

viii
 3.3 Magnetic Energy and Torques .................................................................... 34

3.4 Dynamics of Magnetization and Landau–Lifshitz–Gilbert Equation ......... 36

3.5 Spin-polarized Current and Spin-transfer Torque....................................... 39

3.6 Numerical Framework ................................................................................ 40

3.7 Simulation Setting....................................................................................... 46

3.8 Simulation of Magnetization without an External Torque ......................... 48

3.9 Simulation of Magnetization Due to External Magnetic Field ................... 49

3.10 Simulation of Magnetization Due to Spin-transfer Torque ...................... 52

3.11 Simulation of Magnetization Due to Thermal Fluctuations ..................... 56

3.12 Simulation of the Spin Current Modulation on Thermal Fluctuations ..... 69

3.13 Summary ................................................................................................... 72

Chapter 4. Electrical Treeing and Breakdown in Composite Dielectrics.......... 73

4.1 Overview ..................................................................................................... 73

4.2 Computation Method .................................................................................. 74

4.3 Electrical Treeing in Pure Polymer and Composite Films.......................... 79

4.4 Impact of Filler Concentrations on the Breakdown Time .......................... 80

4.5 Impact of the Filler Geometry on the Breakdown Time ............................. 81

4.6 Summary ..................................................................................................... 84

Chapter 5. Conclusions and Future Work ........................................................... 86

References ........................................................................................................................ 90

ix
 List of Figures

Figure 1.1 NAND flash technology development ..............................................................4

Figure 1.2 MRAM technology development ......................................................................6

Figure 2.1 3D NAND vertical layer stacking over the years ..............................................8

Figure 2.2 2D NAND cell vs. 3D NAND cell ....................................................................9

Figure 2.3 3D NAND layers and resistance ......................................................................10

Figure 2.4 3D GAA geometry and cross-sectional view of flash cell in series ...............14

Figure 2.5 Schematic circuit diagram of the simulation setting .......................................15

Figure 2.6 Simulated I-V curves for erased and programmed states (linear, semi-log). ..16

Figure 2.7 Channel resistance, fringing field, and electron distribution ...........................18

Figure 2.8 Electron distribution in the channel at radius 0, 15 nm, and 30 nm . ..............20

Figure 2.9 Charge embedding in the interlayer oxide.and ON-current at 5V ...................22

Figure 2.10 Electrode shape modulation and ON-current at 5V.......................................23

Figure 2.11 Interlayer channel doping and ON-current at 5V ..........................................24

Figure 2.12 Comparison on I-V characteristics of the three design ideas ........................25

Figure 2.13 2D contour plots of electron concentration n and potential V. ......................26

Figure 3.1 1T1MTJ MRAM cell structure, AP-state and P-state of the p-MTJ. ..............32

Figure 3.2 Energy barrier between the two stable states...................................................33

Figure 3.3 Bar magnet, magnetic field, force, and torque ................................................36

Figure 3.4 Precession locus of magnetization impacted by torques .................................38

Figure 3.5 Geometry and physical quantities on the MTJ model. ....................................41

Figure 3.6 Projection of the magnetization m affected by torque on the z-axis. ..............46

Figure 3.7 Initial condition of the magnetization for precession simulation ....................48

x
Figure 3.8 Precession and damping trajectories and initial positions in 500 ns ...............49

Figure 3.9 Initial position of the magnetization m under the effect of Bext ......................50

Figure 3.10 Magnetization reversal due to increasing of Bext ...........................................51

Figure 3.11 Magnetization precession speeded up by Bext................................................52

Figure 3.12 Initial condition for the simulation of spin-transfer torque effect ................53

Figure 3.13 Magnetization switching due to amplifying the spin current pulse ...............54

Figure 3.14 Magnetization trajectory under the impact of spin-transfer torque ...............55

Figure 3.15 Spin current vs. time to switch ......................................................................56

Figure 3.16 Magnetization m experiencing omnidirectional thermal fluctuations ...........59

Figure 3.17 The thermal fluctuation perturbs the precession............................................60

Figure 3.18 Stochasticity of thermal fluctuations cause random walks ............................61

Figure 3.19 Zoom in the process of magnetization under thermal fluctuations ...............62

Figure 3.20 Statistical analysis of relaxation time under thermal fluctuations .................64

Figure 3.21 Screen of magnetization switching with different thermal amplitudes .........65

Figure 3.22 Relaxation time versus thermal field amplitude ............................................66

Figure 3.23 Switching due to the impact of the high-energy thermal pulse. ....................67

Figure 3.24 Relaxation time versus thermal pulse amplitude. ..........................................68

Figure 3.25 Combined effects of the positive spin current and thermal fluctuations. ......70

Figure 3.26 Combined effects of the negative spin current and thermal fluctuations.. ....70

Figure 3.27 Relaxation time versus spin current changes from -30 mA to +30 mA ........71

Figure 4.1 Random process of electrical treeing.in pure polymer .. .................................76

Figure 4.2 Random process of electrical treeing.in polymer-filler composite..................78

Figure 4.3 Electrical treeing in pure polymer vs. polymer-filler composite... ..................79

xi
Figure 4.4 Insulating filler concentration and breakdown time CDF... ............................80

Figure 4.5 Insulating filler shapes and CDF of breakdown time ......................................82

Figure 4.6 Insulating filler sizes and CDF of breakdown time.. .......................................83

Figure 4.7 Insulating filler dispersed locations and CDF of breakdown time. .................84

Figure 5.1 p-MTJ vs. i-MTJ ..............................................................................................89

Figure 5.2 Simulated easy-axes of p-MTJ and i-MTJ .....................................................89

xii
 List of Tables

Table 1.1 Comparison of novel memory characteristics .................................................... 2

Table 2.1 Important Parameters and Setting .................................................................... 15

Table 3.1 Values and Units of Parameters or Constants for Simulation .......................... 47

xiii
 List of Symbols

Symbol Description

D Diameter of cylinder

LCG Length of control gate

LSD Length of source/drain

DCH Channel diameter

LCH Three cells in series channel length

VCG , VWL Control gate voltage or Word-line voltage

VD , VBL Drain voltage or Bit-line voltage

VTH Threshold voltage

VPASS Passing voltage of the top and bottom cells

Qf Fixed uniform charge density

RCG Resistance under the control gate

RINL Resistance in the interlayer

RON Channel resistance

I ON ON-current or Cell current

V , e Electric potential or voltage

σ Channel conductivity

n Electron concentration or number density

xiv
Symbol Description

nCG Electron concentration under the control gate

nINL Electron concentration in the interlayer region

p hole concentration or number density

N D+ Donor ionized impurity concentration

N A− Acceptor ionized impurity concentration

s Surface charge density

Gn Electron generation rate

Gp Hole generation rate

Rn Electron recombination rate

Rp Hole recombination rate

Jn Electron current density

Jp Hole current density

J Net current density

n Electron mobility

p Hole mobility

Dn Electron diffusivity or diffusion coefficient

Dp Hole diffusivity or diffusion coefficient

E Electric field

xv
Symbol Description

ε Absolute permittivity

d Distance between control gates

A Cross-sectional area of the channel

q Positive charge

e Elementary charge

RP Magnetoresistance of parallel state

RAP Magnetoresistance of anti-parallel state

 Thermal stability factor

Eb Energy barrier

kB Boltzmann constant

T Absolute temperature

UK Uniaxial anisotropy energy

UP Easy-plane anisotropy energy

UH Applied magnetic field energy

U Total magnetic potential energy

HK Uniaxial anisotropy field

K Uniaxial anisotropy constant

KP Easy-plane anisotropy constant

xvi
 Symbol Description

Ms Saturation magnetization

M Magnetization

m Unit magnetization

mx , my , mz Components or Projection of m on x, y, z

B Magnetic flux density or Magnetic induction

H Magnetic field

H eff Effective magnetic field

Bext , H ext Applied or external magnetic field

lm Bar magnet length

lm Lever arm

F Force

Γ Torque acting on magnetization

ΓU Magnetic energy torque

Γ1 Uniaxial anisotropy torque

Γ2 Easy-plane anisotropy torque

Γ3 External magnetic field torque

Γ4 Spin-transfer torque

Γ5 Thermal torque

xvii
Symbol Description

α Damping factor

γ Gyromagnetic ratio or Magnetogyric ratio

τ STT Spin-transfer torque

0 Permeability of vacuum

η Spin-polarization efficiency

N Number of majority eigenstate

N Number of minority eigenstate

⌀ Diameter of MTJ

tF Thickness of MTJ free layer

VF Volume of MTJ free layer

I Spin current

S Spin angular momentum

nS Unit spin angular momentum

h Dimensionless magnetic field

hP Dimensionless easy-plane anisotropy field

hS Dimensionless spin current amplitude

hT Dimensionless thermal field

hT 0 Dimensionless thermal amplitude

xviii
 Symbol Description

 Polar angle

φ Azimuthal angle

 Angle of external magnetic field

ϕ Angle of incident spin-polarized current

RNG(-1,1) Random number generator, number ∈ [-1, 1)

 ij Kronecker delta function

 (t ) Dirac delta function

N Attempt rate

NP P-state attempt rate

N AP AP-state attempt rate

tevent Time of event

tSW Time to switch

 Relaxation time

P P-state relaxation time

 AP AP-state relaxation time

 av Average relaxation time

W Width of dielectric film

T Thickness of dielectric film

xix
 Symbol Description

𝑉𝑜 Voltage of top metal contact

E Energy required to break the bonding

e (i ', j ') Electric potential of the node (i ', j ')

Probability density function of bonding

p(i, j → i ', j ')
(i, j → i ', j ')

pF Probability of breakdown path when encountering

the filler

xx
 Chapter 1. Introduction

Exponential growth in data created, captured, copied, and consumed by people and

machines has created the Zettabytes (one billion terabytes) era [1]. The International Data

Corporation (IDC) forecasted tremendous growth in digital information, from 33 zettabytes

(ZB) in 2018 to 175 ZB by 2025 [2]. To cope with exponential data growth, future storage

solutions must provide even higher bit densities, superior performance, better energy

efficiency, and improved resilience. In order to address the challenges, the semiconductor

industry is seeking newer information storage technologies. Among these are 3D NAND

flash memories involving charge storage; magnetoresistive random-access memory

(MRAM) based on switching of magnetoresistance states; resistive random-access

memory (RRAM) relying on ohmic resistance switching; ferroelectric random-access

memory (FeRAM) based on polarization switching of a ferroelectric layer; and phase-

change memory (PCM) associated with the switching between amorphous and crystalline

states. All of these memory technologies have certain advantages and disadvantages.

Table 1.1 summarizes the performance metrics of various types of non-volatile memories

[3]–[8].

This dissertation focuses on the reliability issues associated with the two important

classes of semiconductor memory technologies: (1) electron charge-based 3D NAND flash

memory and (2) electron spin-based MRAM. Numerical simulation frameworks are

developed and applied to study the reliability challenges due to device scaling.

1
 Table 1.1 Comparison of novel memory characteristics.
NAND
SRAM DRAM FeRAM PCM RRAM MRAM
flash
Power
off
Volatile Non-volatile
Cell size
(F2)
> 100 6–8 4–5 15–34 8–16 >5 < 10
Read time
(ns)
< 10 10–60 25000 20 48 < 10 < 15
Erase time
(ns)
8 50 106–108 80 120 <5 30
Write time
(ns)
< 10 10–60 105–106 20 40–150 ≈ 10 < 10
Write voltage
(V)
1 2.5 > 10 <3 <3 <3 < 1.5
Switch time
(ns)
10 60 > 100 ≈ 10 ≈ 50 > 50 ≈ 10
Leakage – high low low low low low
Endurance > 1015 > 1015 104 ≈ 1010 106–109 105–108 > 1015

1.1 3D NAND Flash Memory

The invention of the floating-gate transistor by Khang and Sze at Bell Labs in the

year of 1967 marks an important milestone in the technological evolution of non-volatile

flash memories [9], [10]. Subsequently, in the late 1980s, Toshiba Corporation developed

the floating-gate-based 2D NAND flash memory array, which gained popularity in the

consumer market due to its high bit density, small form factor, fast access time, and low

cost. Finally, in the mid-1990s, the first NAND-based portable flash memory card, the

SmartMedia card, was released and became a replacement selection for the 3.5-1nch 1.44

MB magnetic floppy disk [11]. Figure 1.1 summarizes the technological evolution of

NAND flash memory technology. Note that traditional 2D NAND flash technology

reached the end of scaling with sub-15 nm technology nodes, hampered by an increasing

2
cost of lithography and fundamental cell reliability issues such as random telegraph noise,

program noise, and cell-to-cell interference [12], [13].

The flash memory industry has responded by transitioning to monolithic 3D NAND

flash fabrication processes [14]. A shift to 3D NAND technology opened up a new “scaling”

dimension—the number of vertical layers—enabling further increases in the bit density for

a given die area [15], [16]. Continual advances in this technology resulted in several

generations of 3D flash memory chips, each having a larger number of stacked layers, from

early 32- to contemporary 232-layer designs [17], as shown in Figure 1.1. These advances

promise to extend an incredible growth of bit-density over the next decade. Meanwhile,

the logical scaling according to the threshold voltage of the memory cell has evolved from

the 1-bit single-level cell (SLC), 2-bit multi-level cell (MLC), 3-bit triple-level cell (TLC),

and 4-bit quad-level cell (QLC). Although logical scaling increases bit density on the same

memory array, it degrades cell reliability. Figure 1.1 illustrates the 2D NAND and 3D

NAND flash memory structures and the scaling trends in both the spatial and the logical

aspects [12], [18].

The 3D NAND technology is inherently different from the former 2D NAND

technology in terms of cell geometry (cylindrical gate-all-around cell versus planar cell),

transistor channel material (polycrystalline silicon versus crystalline silicon), and its array

structure (vertical stacking arrangement versus planar configuration) [19]. The memory

cell's ON-current (or cell current) is reduced significantly by an increase in the number of

layers due to the increased series resistance of the vertical 3D NAND string. Since memory

cells in the 3D NAND structure utilize the polycrystalline silicon (poly-Si) channel with

undoped source and drain (S/D), the ON-current of the state-of-the-art 3D NAND

3
technology is already low compared to the 2D counterpart. Lower ON-current causes

sensing errors during the read operation of the memory array. Thus, the ON-current

problem can be a significant bottleneck to future 3D NAND flash memory generation.

One of the objectives of this dissertation research is to use numerical simulation to

quantify the limiting factors that reduce the ON-current in the 3D NAND structure. In

addition, the three new design ideas are proposed and numerically evaluated to enhance

the ON-current for the future generation of 3D NAND flash memories.

Figure 1.1 2D NAND flash memory versus 3D NAND flash memory (gate-all-around and vertical
channel type) in geometric structure, scaling trend, and bit density over the years.

4
1.2 Magnetoresistive Random-access Memory

Electrons carry not only charges but also spin. The spin current is not negligible

on the atomic scale and will affect the performance of microelectronic devices. The

electron spin-based magnetoresistive random-access memory (MRAM) was developed in

the 1990s as a byproduct of the discovery of tunnel magnetoresistance (TMR). It was

subsequently applied to the design of the magnetic tunnel junction [20]. MRAM is

characterized by a high read/write speed close to SRAM [21]. In addition, it has a non-

volatility of storage similar to flash memory, and it also can be comparable to DRAM with

respect to high density and longevity [22].

Writing to the MRAM is determined by the magnetic state of the free layer. There

are three ways write operation can be performed, making three different commercial

MRAM architectures [23]. According to the magnetization switching mechanism: (1)

field-induced magnetic switching (FIMS), which is the mechanism for the Toggle MRAM,

(2) spin-polarized current induced spin transfer torque (STT) switching, which makes the

STT-MRAM and (3) thermally assisted switching (TAS) is for the TA-MRAM. In addition,

a fourth mechanism, spin-orbit torque induced switching, known as SOT-MRAM, has

recently received attention. Both Toggle MRAM and STT-MRAM are commercially

available. The Toggle MRAM, built with the in-plane MTJ as the cell element, came to

the consumer market in 2006. Later, the STT-MRAM was released in 2015, initially using

in-plane MTJ, and the production was replaced by the perpendicular MTJ after 2017 [24].

Figure 1.2 depicts MRAM architectures and their chronological development [25], [26].

The numerical model developed in this dissertation discusses the first three switching

mechanisms and focuses on TAS and STT switching with thermal effect.

5
 Figure 1.2 MRAM architectures and their chronological development along the diagonal arrow.

1.3 Scope and Outline

The chapters of this dissertation are as follows. In Chapter 2, the reliability issues

relating to the ON-current are presented. This study performs a detailed numerical device

simulation to identify the limiting factors for the lower cell current during the ON condition.

The Taurus Medici software is used to model and analyze the flash memory electrical

properties and performance of the cylindrical gate-all-around (GAA) type. The factors

affecting the ON-current, such as channel mobility, fringing E-field, and series resistance,

are investigated. In addition, the three new design ideas are proposed to improve the ON-

current.

Chapter 3 presents the numerical simulation framework of MRAM devices

focusing on the MTJ. The numerical computation is performed by solving the Landau–

6
Lifshitz–Gilbert (LLG) equation, which describes the magnetization dynamics as affected

by the intrinsic and applied torques. Effects of the external magnetic field, the spin-

polarized current, and the ambient thermal fluctuations on the magnetization switching that

occurred in the MTJ are examined in the sections. For more realistic phenomena, thermal

fluctuations are incorporated with the effect due to spin-polarized current, and the

consequent relaxation process is explored. The simulation framework proves to be very

useful in analyzing the enhancement of the MRAM device’s reliability.

Since insulating dielectrics is a critical component for all semiconductor NVM

devices, studying the reliability of the dielectric under high gate voltage or high spin-

polarized current is very important. The electronic device no longer works once dielectric

breakdown occurs in the insulation layer. Chapter 4, based on the research work published

in 2019 [27], analyzes the time-dependent electrical tree formation and propagation

resulting in dielectric breakdown. In addition, ideas for strengthening the insulation layer

to prevent further damage or to extend the lifetime are presented in this chapter.

Chapter 5 focuses on a summary of the results from the three major areas of the

dissertation research. Finally, the conclusions are stated in this chapter, and a proposed

plan for future work is presented.

7
 Chapter 2. 3D NAND Flash Memory Cell Current Enhancement

2.1 Overview

The NAND flash memory industry has switched technology development from 2D

NAND to 3D NAND architecture with the monolithic 3D fabrication process. The 3D

NAND technology promises to extend the incredible growth of the bit-density of NAND

flash memory over the next decade with the increased number of vertical layers [28], [29].

Commercially available 3D NAND technology reached 176 layers in 2021 and 232 layers

in 2022 (as illustrated in Figure 2.1); it is predicted to go beyond 1000 layers in the near

future [30], [31]. One critical issue that needs to be resolved for the increased number of

Figure 2.1 Evolution of 3D NAND technology with the increasing number of vertical layers over the
year since commercialization.

8
layers is enhancing the cell current (i.e., ON-state current) [32], [33]. The channel material

of 3D NAND cells is made of polycrystalline silicon (poly-Si), which inherently has low

charge carrier mobility [34]. Therefore, the channel resistance of an individual 3D NAND

cell is significantly higher than that of a 2D NAND cell made of crystalline silicon. Figure

2.2 shows the comparison. In addition, the undoped source and drain (S/D) region [35] of

the 3D NAND cells aggravate the current conduction [36]–[38]. Thus, the cell current of

the 3D NAND string is reported to be significantly lower than its 2D counterparts. The

cell current, however, will be further reduced with the increased number of layers due to

the accumulated series resistance of the vertical NAND string [39], as shown in Figure 2.3.

Hence, new ideas are necessary to boost the cell current for a higher number of layers in

the 3D NAND flash memory.

(a) (b)

Figure 2.2 (a) 2D NAND flash memory cell comprises a floating gate (FG) in the oxide layer (FG-
MOSFET). Both source (S) and drain (D) are highly doped. (b) 3D NAND flash memory cell has an
FG in the oxide layer or a charge-trapped (CT) layer in a cylindrical gate-all-around (GAA) structure,
but a highly-doped source and drain are missing.

9
 High cell current is a critical factor for accurately sensing memory states during the

read operation of the memory array. The typical sensing current for a memory cell is

around 60–80 nA [33], and the ON-state current of a cell needs to be much higher than the

sensing limit for an accurate read operation. A typical NAND memory's ON-state cell

current is in the range of 100–200 nA [39], which is already getting closer to the sensing

limit. However, with the increased number of memory layers, the cell current may fall

below the sensing limit, causing significant read failures [32], especially at lower

temperatures where the cell current is found to be significantly reduced due to reduced

electron mobility in the poly-Si. Thus, enhancing cell current remains a significant

challenge for the future generation of 3D NAND technologies, especially for low-

temperature applications.

(a) (b)

Figure 2.3 (a) Schematic of a 3D NAND string with cylindrical gate-all-around (GAA) memory cells
connected in series; the corresponding circuit diagram of the NAND string. (b) ON-current vs. number
of cells in series in the NAND string assuming the bit line voltage, 𝑉𝐵𝐿 = 5V, and the resistance, 𝑅𝑂𝑁 =
500 kΩ, on each cell during the turned-ON condition.

10
 In order to improve cell current, high-mobility alternative materials are proposed

for the 3D NAND. Capogreco et al. considered the epitaxially grown III–V material,

InxGa1-x, to serve as a high mobility channel for better conduction to replace the poly-Si

channel [40]. Hou et al. investigated SixGe1-x and InGaAs for the high-mobility channel

as a substitute for poly-Si in both entire and macaroni channel structures. They found the

current reduction was less than poly-Si when stacking more layers [41]. Bhatt et al.

developed the 3D NAND flash memory string model for performance enhancement by

optimizing the simulated read current through varying poly-Si channel grain boundary

density, grain size, and channel thickness [42]. To improve the cell current, Arreghini et

al. explored modifying the grain size, the channel thickness, the interface defectivity, and

source junction engineering on the poly-Si channel of a vertical string of the 3-devised

BiCS structure [43]. Experimental research on the metal-induced lateral crystallization

process, using Ni as a catalyzer to transform the amorphous Si (a-Si) into crystalline Si (c-

Si), was conducted to improve channel conduction [44]. Other laboratory studies have

been done on integrating the oxide layers with high-k dielectric materials (such as HfO2

and TiO2), which increase the gate capacitance by offering a thicker oxide layer but has the

same equivalent oxide thickness of SiO2 to improve the drain current [45]–[48]. Recently,

Qin et al. discussed the effect of total ionizing dose radiation on increasing the electron

density in the poly-Si channel [49].

The goal of the flash memory manufacturer is to downscale and stack flash memory

cells to increase the ultimate information capacity. However, the nature of the 3D NAND

string causes cell-to-cell electrostatic interference [50] and increases the memory string's

total resistance [51]. This change further reduces the ON-current conduction, resulting in

11
a lower reading speed and causing reading errors. Therefore, instead of inherently

changing to alternative channel materials, externally utilizing the common existing

materials and introducing new design ideas are proposed to modify the conventional design

to boost the ON-current. In this research, the model is used to investigate the electric

potential, electron concentration, and electron current density in order to understand the

factors limiting the ON-current. This simulation study reveals the role played by each of

these variables in determining the ON-current of the 3D NAND string. Furthermore, the

three new device design ideas to enhance the ON-current in the next generation of 3D

NAND flash memories are numerically evaluated.

2.2 Modeling and Simulation Framework

Let the electric potential be V . Then, the governing equation that describes the

electrical behavior of the semiconductor device is Poisson’s equation:

s
 2V = −
q

(p−n+ N +
D − N A− ) −

, (2.1)

−19
where q = e = 1.6 10 C ,  is the absolute permittivity, p is the hole concentration, n is

the electron concentration,  s is the surface charge density for fixed charge in insulating

materials, N D+ and N A− are the ionized impurity concentrations of donor and acceptor. In

addition, the continuity equations for charge carriers governing the electrical behavior are

n 1
=   J n + Gn − Rn , (2.2)
t q

12
 p 1
= −   J p + G p − Rp , (2.3)
t q

where J n and J p are the electron and the hole current densities. Gn and G p are the

electron and the hole generation rates. Rn and R p are the electron recombination rate and

the hole recombination rate, respectively. From the Boltzmann transport theory, the carrier

drift and diffusion equations are

J n = nqn E + qDnn (2.4)

J p = nq  p E − qD p p (2.5)

where  n and  p are the electron and the hole mobilities, Dn and D p are the electron and

the hole diffusivities (or diffusion coefficients), and E = −V is the external electric field.

When the carriers coexist, the net current density is J = J n + J p .

Figure 2.4(a) illustrates a simplified vertical channel (VC) 3D NAND string

geometry of three cells. We numerically simulate current-voltage characteristics for this

structure using Technology Computer-Aided Design (TCAD) by Taurus Medici, a device

simulation tool from Synopsys, Inc. It solves the drift-diffusion-based transport equations

(2.4) and (2.5) as well as Poisson’s equation (2.1) self-consistently [52]. Due to the

rotational symmetry (about the z-axis) in cylindrical coordinates, the 2D curvilinear grid

13
 Figure 2.4 (a) The 3D GAA geometry, and (b) a cross-sectional view of the three cells in series
structure in cylindrical coordinates with a unit in nanometer.

space is used for the 3D NAND cell simulation [53]. Assume that the poly-Si is formed in

the core of the cylindrical body. Electrons are stored in the charge-trapping (CT) region,

and the dashed lines in Figure 2.4(b) depict all three CT regions with the same volumes.

Programmed state electron density in each CT is 51018 1/cm3, equivalent to 295e in the

region. For computational simplicity, the three cells in series structure is constructed;

however, the conclusions derived from this structure apply also to longer strings. The n+

doped source and drain are located at the column's top and bottom ends as outer junctions.

The vertical cross-sectional area of the cylinder is shown in Figure 2.4(b). The charge

storage region (50 nm  5 nm) is surrounded by oxide. Table 2.1 shows the key dimensions

and material parameters used in the simulation.

14
 Table 2.1 Important Parameters and Setting.

Parameter Value
Diameter of cylinder (D) 250 nm
Length of control gate (𝐿𝐶𝐺 ) 50 nm
Length of source/drain (𝐿𝑆𝐷 ) 25 nm
blocking oxide 10 nm
Gate stack charge-trapping region 5 nm 20 nm
tunneling oxide 5 nm
Channel diameter (𝐷𝐶𝐻 ) 230 nm
Channel doping 31017 1/cm3
Source/Drain doping 51019 1/cm3
Electron mobility (𝜇𝑛 ) 10 cm/V·s
Three cells in series channel length (𝐿𝐶𝐻 ) 250 nm
Control gate voltage (𝑉𝐶𝐺 ) 5V
Drain voltage (𝑉𝐷 ) 0.5 V

The control gate (CG) or word line (WL) voltage of the middle cell is swept from

0 to 5V to construct the current-voltage (I-V) curves. The “Vpass” voltages of the top and

bottom cells are kept at 5 V. The drain voltage ( VD ) or bit line (BL) voltage is 0.5 V, as

depicted in Figure 2.5:

Figure 2.5 Schematic circuit diagram of the simulation setting for the drain current (𝐼𝐷 ) in response to
the input control gate voltage (𝑉𝐶𝐺 ) of the middle cell.

15
 Assume the programmed cell 𝑉𝑇𝐻 = 2 V and its erased state 𝑉𝑇𝐻 = 0.5 V.

Simulated I-V curves for the erased state (green) and programmed state (red) are shown in

Figure 2.6. Note that the simulation parameters are chosen from the published literature

[54] in order to obtain the cell current characteristics similar to the experimental results

[55], [56]. Since this research aims to evaluate the new device design concepts to boost

the cell current, the simulation framework is kept very simple, capturing only the essential

physics of the NAND string.

Figure 2.6 Simulated I-V curves for erased state (green) and programmed state (red). (a) Linear scale.
(b) Semi-log scale.

16
2.3 Analysis of the Cell Current Bottlenecks

Cell current in the 3D NAND flash memory depends on several material and

geometrical factors such as the electron mobility of the poly-Si [40], [41], fringing E-field

distribution [57], [58], and series resistance [51], [59]. In the following, we first discuss

the critical bottleneck for achieving high cell current in the 3D NAND architecture and

then present new design ideas for its enhancement in the future generation of 3D NAND

technology.

The carrier transport in the 3D NAND memory cells is significantly different from

its 2D counterparts. Since 3D NAND memory cells possess undoped source and drain or

interlayer regions, the carrier transport is critically dependent on the properties of the

interlayer region. The fringing E-field generated by the CG acts as the virtual source and

drain to induce the inversion charge layer for the conduction in the channel. Figure 2.7(a)

sketches the direct electric field under the CG and the role of the fringing electric field in

inducing the inversion carriers over the interlayer region [60], [61]. Figure 2.7(b) evidently

shows that the interlayer region has weaker electric potential and smaller electron

concentration according to the simulation result.

Since ON-current is predominantly drift current, it is proportional to the channel

conductivity (𝜎 = 𝑛𝑒𝜇𝑛 ), and the high mobility enhances the ON-current conduction.

However, the 3D NAND flash device is built into the polycrystalline silicon channel, where

the mobility is low due to grain boundaries [62], [63] and misaligned crystallographic

orientations [64].

17
 Figure 2.7 (a) RCG represents the resistance under the control gate. RINL represents the resistance in the
interlayer. E-field lines (including fringing field lines), and charge carrier distribution in the channel
are sketched. (b) Electric potential distribution and electron concentration profile along the channel
(Si/oxide interface).

For the 3D NAND flash memory cell, the resistance of the ON-current, RON,

comprises not only the resistance under the cell CG coverage, 𝑅𝐶𝐺 , but also the resistance

in the interlayer region (outside of CG coverage), 𝑅𝐼𝑁𝐿 . In our model, the 𝑅𝑂𝑁 of the

reference (middle) cell is:

𝑅𝑂𝑁 = 𝑅𝐶𝐺 + (1/2)𝑅𝐼𝑁𝐿 + (1/2)𝑅𝐼𝑁𝐿 . (2.6)

The existence of 𝑅𝐼𝑁𝐿 is the feature of 3D NAND flash memory that distinguishes the

architecture. When a constant electric field is applied, the current flow in the channel,

which depends on the conductivity, is proportional to the carrier’s concentration and

18
inversely proportional to the resistivity. Hence, 𝑅CG and 𝑅INL can be expressed as:

1 𝐿𝐶𝐺
𝑅𝐶𝐺 = , (2.7)
𝑛𝐶𝐺 𝑒𝜇𝑛 𝐴

1 𝐿𝐼𝑁𝐿
𝑅𝐼𝑁𝐿 = , (2.8)
𝑛𝐼𝑁𝐿 𝑒𝜇𝑛 𝐴

where A is the cross-sectional area of the string channel, 𝐿𝐶𝐺 is the length of CG, and 𝐿𝐼𝑁𝐿

is the length of the interlayer region. The two types of resistance, 𝑅𝐶𝐺 (under the control

gate) and 𝑅𝐼𝑁𝐿 (in the interlayer region), can be determined locally by monitoring the

concentrations of charge carriers 𝑛𝐶𝐺 (under the control gate) and 𝑛𝐼𝑁𝐿 (in the interlayer

region) throughout the channel. The carrier concentration in the channel is position

dependent, leading to a resistance that is also position dependent. The greater carrier

concentration represents the smaller resistance in the channel. In order to know the carrier

distribution in the three-cell string, the electron concentration profile according to the

different cross-sectional planes of the 3D NAND column is investigated. Figure 2.8

demonstrates the simulated electron concentration profiles along the channel for three

different radial positions: (i) at the oxide-channel interface (𝑟 = 30 𝑛𝑚, green), (ii) at the

center of the cylindrical body (𝑟 = 0, red), and (iii) at an intermediate radial position (𝑟 =

15 𝑛𝑚, blue) for the ON condition (all the CGs are at 5 V).

The plot suggests that the ON-current mostly flows along the oxide-channel

interface (where the 𝑛𝐶𝐺 is much greater than 𝑛𝐼𝑁𝐿 ; thus, 𝑅𝐶𝐺 is much smaller than 𝑅𝐼𝑁𝐿 )

and the interlayer region offers a significant bottleneck to the ON-current due to the lower

19
electron concentration. Based on simulation results, it is estimated that the interlayer

resistance will be about six times higher compared to the under CG channel resistance of

each cell (viz., 𝑅𝐼𝑁𝐿 ≈ 6𝑅𝐶𝐺 ). When the attention departs from the interface and shifts to

the center of the column, both 𝑛𝐶𝐺 and 𝑛𝐼𝑁𝐿 decrease, resulting in low current conduction.

Moreover, the accumulation of interlayer resistance becomes the major contributor to the

series resistance; especially when the stacking number is enormous.

As channel mobility is an intrinsic material property, changing or selecting the

appropriate material for a more significant current flow is outside the scope of this

Figure 2.8 Electron distribution and electron number density profiles in the axial direction along the
channel at (red), nm (blue), and nm on the oxide-channel interface (green), where
𝑛𝐶𝐺 ≈ 6𝑛𝐼𝑁𝐿 leading to the ratio 𝑅𝐼𝑁𝐿 /𝑅𝐶𝐺 ≈ 6.

20
dissertation. Instead, it focuses on strengthening the fringing field and reducing the series

resistance by altering the carrier concentrations with new designs to enhance the cell

current.

2.4 New Designs to Boost the Cell Current

In the following sub-sections, this research proposes and numerically evaluates

three new device design concepts capable of boosting the ON-current of the 3D NAND

string made of typical poly-Si material. Essentially, the proposed ideas are based on

strengthening the fringing field and reducing the series resistance of the interlayer region

by altering the carrier concentrations. All these ideas are orthogonal to one another and

can be applied in tandem to improve the ON-current of the future 3D NAND array design.

2.4.1 Charge Embedding in the Interlayer Oxide Region

Figure 2.9(a) illustrates the first idea, proposing the embedding of a fixed positive

charge layer in the interlayer oxide region. Since embedded positive charges help to invert

the interlayer region, the embedding improves the ON-current. Suppose the embedded

charges are placed 2 nm above the poly-Si channel. The uniform charge density (𝑄𝑓 ) in

the charged region is assumed, keeping all other dimensions and material parameters the

same as the reference cell. This idea is numerically evaluated in Figure 2.9(b) and

discovers that the ON-current can be increased by 35% with an embedded positive charge

density of 51018 1/cm3. Moreover, there are charging methods to embed positive charges

into the oxide near the Si/oxide interface. For instance, building positively charged ions

during plasma etching [65] and embedding Au nanocrystals in the oxide to store holes may

attract electrons to increase the density in the poly-Si channel [66].

21
 Figure 2.9 (a) Design of positive charge embedding in the interlayer oxide. (b) Transfer characteristics
of the design and corresponding ON-current values at 5 V in comparison to the reference model.

2.4.2 Electrode Shape Modulation

Figure 2.10(a) shows our second idea, proposing an altered CG electrode geometry

with its edges inserted into the oxide. The inserted electrode will intensify the fringing

field around the interlayer region and attract charge carriers resulting in a higher ON-

current, as confirmed by simulated I-V curves in Figure 2.10(b). The shape modulation

concentrates the fringing field under the control gates. Lee et al. introduced the extended

metal gate design to enhance the electric field and the ON-current [61]. However, the

three-side enclosure may affect the distribution of trapped charges and vary the threshold

voltage. Therefore, the inserted parts must be kept short and separated from the charge-

trapping region, especially in the programmed state, to minimize cell-to-cell interference

22
and mitigate the trapped charge uneven distribution or screening that causes data retention

issues. The simulation indicates that the ON-current at 5 V increases by 46% of the original

non-modified (reference) model when both ends of the electrode are embedded 10 nm into

the oxide.

Figure 2.10 (a) Design of electrode shape modulation. (b) Transfer characteristics of the design
corresponding ON-current values at 5 V comparing to the reference model.

2.4.3 Interlayer Channel Doping

The GAA vertical channel NAND structure only has a source and a drain at both

ends of the string. The internal undoped interlayers cause high resistance (due to low

carrier concentration) and lower the ON-current. In order to retain the high number density

of the electrons in the interlayer regions, extra S/D junctions are created, as illustrated in

Figure 2.11(a). The interlayer S/D junctions are highly doped with electrons to reach 1019

23
1/cm3. Since the junctions offer sufficient charge carriers to flow in the interlayer channel

region, the string’s series resistance decreases, and the ON-current increases. Our

simulation shows the idea of selective doping of the interlayer region. Simulated I-V

curves reveal that a 42% increase in ON-current is achievable with S/D doping about

51018 1/cm3 (see Figure 2.11(b)). The ON-current value then becomes the function of the

interlayer doping level.

Figure 2.11 (a) Design of Interlayer channel doping. (b)Transfer characteristics of the design and
corresponding ON-current values at 5 V comparing to the reference model.

2.5 Summary

This simulation investigates the impact of series resistance on the ON-current in

the string of 3D NAND flash memory and presents new designs to boost the ON-current.

Instead of searching for high-mobility candidates in a comprehensive variety of materials,

this research emphasizes the common constituent of the 3D NAND flash memory to obtain

24
optimal current results. It demonstrates that higher ON-current can be achieved through

CG shape modulation and the positive charge embedding in the oxide in the interlayer

regions. Both ideas are to enhance the ability of the CG electric field and fringing field to

attract more conducting charge carriers in the channel. The third idea is to utilize the

interlayer channel doping for creating additional source and drain junctions, which provide

considerable charge carriers into the channel and mitigate the series resistance to further

boost the ON-current. Finally, this research makes a comparison of all three designs with

appropriate measures that achieve higher ON-current values. The designs and associated

values are: (1) embedding positive charges (51018 1/cm3) into oxide in the interlayer

region near the channel, (2) inserting re-shaped CG electrodes at both ends (10 nm) into

the oxide layer, and (3) allocating negative dopants (51018 1/cm3) into the interlayer to

form the source and drain (see Figure 2.12). These ideas can be the cornerstone for future

design and manufacturing upgrades.

Figure 2.12 Comparison of transfer characteristics of the three designs and corresponding ON-current
values at 5 V compared to the reference model.

25
 Next, we visualize the phenomena from different angles by providing filled 2D

contour plots of electric potential (V) and electron concentration (n) for a better

understanding of their variations, as shown in Figure 2.13. Let the reference and all new

design cells be in the erased state. The reference electric potential profile V0 highlights the

three E-field effective regions; the reference electron concentration profile n 0 shows three

distinct high electron concentration “island” structures in the channel under CG coverage

areas. The high concentration (red color) part is located in each island's center (directly

enveloped by CGs).

For the design idea of charge embedding (labeled 1), the electric profile V1 plot

shows that the charge distribution in the two interlayer oxide regions serves to increase the

Figure 2.13 Filled 2D contour plots of electric potential (V) and the electron concentration (n) for the
reference model and three new designs. The color bar ranges the electric potential 0–5 V and the
electron concentration 0–61019 1/cm3.

26
fringing E-field strength, which results in more inversion electrons in the channel interlayer

regions (see profile n 1 plot). As a result, the concentration rises and links the three electron

concentration islands, which then form a continuing bright stream that leads to ON-current

enhancement.

The design of electrode shape modulation (labeled 2) intensifies the E-field mainly

due to the high fields on the electrode tips, which are inserted into the center of the oxide

layer and provide close attraction and confinement to the channel electrons. In the electron

concentration plot, each lengthened island shows two high-concentration areas (red color)

resulting from the separate tips of each electrode. The lengthened and double high-point

islands structure serves to amplify the current conduction.

The interlayer channel doping design with highly doped S/D (labeled 3) supplies

carriers for conduction in the channel, which promotes the linkage of electron islands

shown in the electron concentration plot. However, the appearance of the E-field pattern

due to the associate electrode is similar to the reference plots. This fact reveals the ON-

current enhancement is mainly attributed to interlayer channel doping.

27
 Chapter 3. Dynamics in Magnetoresistive Random-access Memory

3.1 Background

Early in the nineteenth century, magnetoresistance was discovered in the form of

anisotropic magnetoresistance (AMR). This type of resistance is high when the current

flows in the same direction as the magnetic force and low when both directions are

perpendicular, as is the case in iron and nickel [67]. In 1988, Baibich et al. discovered a

large resistance in iron superlattices with a thin chromium layer (Fe–Cr–Fe structure) and

named it giant magnetoresistance (GMR) to distinguish the AMR [68]. Almost

simultaneously, Binasch et al. reported a similar result. They showed that the resistance

was much stronger than the usual AMR by exploring the coupling between different

magnetic films and non-magnetic (or antiferromagnetic) interlayers [69]. By sandwiching

a non-magnetic metal spacer between two magnetic layers of the electrode on the

nanometer scale, the GMR is a quantum mechanical effect in which a minor change in the

magnetic field creates a significant difference in electrical resistance. The electrical

resistance depends on the magnetization alignment of the two ferromagnetic layers. The

parallel alignment allows more significant current conduction than the antiparallel

alignment. Perkin et al. reported that the Co/Cu multilayer structure magnetoresistance

ratio was more than 65% at 295 K and 115% at 4.2 K compared to 13% and 80% in the

counterpart Fe/Cr [70]. In 2007, the French physicist Albert Fert, and the German physicist,

Peter Grünberg, shared the Nobel Prize for the discovery of GMR, which led to the novel

28
method of materials characterization and contributed to the scaling of electronic and

magnetic devices [71]. The discovery of GMR labels the beginning of spin-based

electronics [72]. The GMR concept was applied to the invention of the spin valve [73],

which contains magnetic layers and an interconnected non-magnetic conductive layer. The

spin valve is used for hard disk drive (HDD) read heads and magnetic sensors [74], [75].

In 1997, the first spin valve-based HDD was commercialized by IBM [76].

Research on tunnel magnetoresistance (TMR) was initiated by Jullière in 1975

using oxidized germanium instead of the non-magnetic metal as the spacer between iron

and cobalt. At 4.2 K, the resistance increased by 14% [77]. The sandwiched spacer was

changed to a thin insulating film (in a few nanometers), so the spin-polarized electrons

could quantum mechanically tunnel through and flow into another ferromagnetic layer.

Electrons are more likely to tunnel through the insulating film if the magnetizations are in

the same rather than an opposite orientation. TMR experiments did not exhibit significant

resistance changes until 1994. Miyazaki and Tezuka employed a thin amorphous Al2O3 as

a barrier material to make a Fe–Al2O3–Fe structure and achieve relative resistance

differences by 30% at 4.2 K and 18% at room temperature [78]. By stacking CoFe and Co

separated by an amorphous Al2O3 insulator at room temperature, Moodera et al.

demonstrated the resistance ratio R / R = 11.8% [79]. Wei et al. reported an 80% of

tunneling magnetoresistance ratio with CoFeB–Al2O3–CoFeB structure at room

temperature [80]. After replacing the oxide tunneling barrier with crystalline MgO, the

magnetoresistance ratio increased drastically at room temperature. In 2004, Yuasa et al.

reported a ratio of 180% from their Fe–MgO–Fe junction [81]. Later, the stacking based

on the CoFeB–MgO–CoFeB structure showed an increase of 230% [82], and design

29
improvements by Ikeda et al. increased this to 604% [83] at room temperature.

Additionally, thanks to thin-film deposition improvement and photolithography progress,

the advances in magnetic tunnel junction (MTJ) structure enhanced the higher TMR ratio

[84]. MRAM performance relies on the MTJ device, which can provide a high TMR.

Conventionally, the writing process of the Toggle MRAM needed a current-

induced magnetic field to change the magnetization orientation. However, generating a

magnetic field from the current was not energy-efficient, and the structure was difficult to

scale down to achieve higher densities. Early in 1996, two independent research groups,

Slonczewski [85] from IBM and Berger [86] from Carnegie Mellon University,

theoretically predicted the spin-transfer torque (STT) effect as a new magnetic state

switching mechanism. Near the beginning of 2000, experimentally performed STT-

induced switching MRAM, which no longer needed an external magnetic field, has been

reported [87]–[89]. In 2004, SONY fabricated the STT-MRAM, of which MTJ comprised

CoFeB layers and a crystalized MgO tunnel barrier, capable of 4 kb and 1/30 write current

reduction to switch compared to the early generation MRAM [90]. On the critical trend of

scaling for high density storage [91], the attention transition has occurred from the in-plane

to the perpendicular anisotropy type of the MTJ [92], [93]. Recently, the 1 Gb (Gigabit)

STT-MRAM adopting p-MTJ technology has been launched to the consumer market [94].

The perpendicular anisotropy STT-MRAM has become mainstream due to its advantages

in scalability, thermal stability, TMR ratio, and energy efficiency [95], [96].

High AMR, GMR, or TMR at room temperature are essential for advanced

electronic device development, such as magnetoresistive random-access memories [97]

and HDD read heads [98]. This dissertation applies the TMR behavior to the MRAM

30
simulation model. The simplest cell of the MRAM is structured as a junction of two

ferromagnetic layers separated by an oxide layer. For the STT-MRAM, spin polarization

happens when the incoming spin current enters the fixed ferromagnetic layer. Then, the

spin-polarized electrons tunnel through the oxide layer and enter the free ferromagnetic

layer. Eventually, the electrons’ change of spin angular momentum over time exerts spin-

transfer torque to interact with the magnetization of the magnetic body. This research

analyzes the spin current interacting with the free ferromagnetic layer. Discussion of the

physical phenomena that occur before entering the free layer is outside the scope of this

dissertation.

3.2 Micromagnetics and Magnetic Tunnel Junction

Micromagnetics is concerned with understanding the magnetic behavior of

magnetized bodies on a sub-micrometer scale. The scale is larger than the material’s

atomic structure, and the atomic property can be ignored; however, it is small enough to

describe the structure of magnetic domains. Micromagnetics is a continuum theory of

ferromagnetism that handles: (i) the static equilibrium of the system in steady state with

minimum magnetic energy and (ii) the dynamic evolution of magnetization under the

effective magnetic field as described by the solution of the time-dependent dynamical

equation [99]. Consequently, the theory of micromagnetics may be applied to the study of

the magnetic nanostructure in MRAM.

MRAM is a ferromagnetic thin-film nanodevice, which is non-volatile and

attractive because of its endurance, low energy consumption, radiation hardened, and high

speed. The magnetic tunnel junction, based on the TMR for the binary switching

mechanism, is the fundamental information storage element of MRAM. An MTJ cell is a

31
two-terminal device composed of two ferromagnetic layers on either side of a thin

insulating layer (in a few nanometers), which is thin enough for electron tunneling [99] and

sandwiched to ensure current conduction between the two electrodes. One ferromagnetic

layer is called the fixed layer or pinned layer, and the other is called the free layer or storage

layer. The current flows from the bottom fixed layer via tunneling in the dielectric layer

to reach the top of the free layer, as shown in Figure 3.1. When the magnetization vectors

of the two magnetic layers are both in the same direction (0°), it is called parallel (P),

whereas when the two magnetization vectors are in opposite directions (180°), it is called

antiparallel (AP). During the AP-state, the current encounters the highest resistance (logic

“1”); in the P-state, it has the lowest resistance (logic “0”). Manipulation of the

magnetization status on the free layer distinguishes between the two distinct resistance

states, which can be used for the read sensing of the data stored. Besides the intrinsic

(a) (b)

Figure 3.1 (a) One-transistor-one-MTJ (1T1MTJ) cell structure. The MTJ acts as a variable resistor
and connects to the bit line (BL) and the transistor, which controls the current flow in BL by the word
line (WL) voltage. (b) An MTJ unit. “P” represents the low resistance state (left); “AP” represents the
high resistance state (right).

32
magnetic anisotropy and damping mechanism, the switching of the magnetization can be

performed by the external magnetic field, the spin current, and the thermal energy.

Two factors that characterize the magnetization switching of the MTJ are: (1) the

tunneling magnetoresistance (TMR) ratio, which is defined by TMR% = ( RAP − RP ) / RP ,

where RAP and RP are resistances of antiparallel and parallel states, respectively. High

TMR% ensures read accuracy between 0 and 1. (2) the thermal stability factor  = Eb / k BT ,

where Eb is energy barrier, which is an intrinsic property of the ferromagnet, k B is the

Boltzmann constant, and T is the absolute temperature. The thermal stability factor 

specifies the relative populations in the two states and determines the thermal ability as

well as data retention.

A high energy barrier is preferable for memory devices [100], [101], while a low

energy barrier is suitable for the design of MTJ sensors [102]. The two stable states are

Figure 3.2 Energy barrier (green) between the two stable states for magnetization (red). P-state
(magnetization vectors are in the same direction with an angle equal to 0) and AP-state (magnetization
vectors are in opposite directions with an angle equal to π).

33
separated by the energy barrier Eb , which depends on the materials and determines the

functionality of the MTJ device. Figure 3.2 depicts how the anisotropy energy E plotted

versus the rotation angle  compares with respect to the energy barrier separating the two

stable states. The barrier energy E = U K , which is the uniaxial anisotropy energy, appears

in the first relation in (3.1) below.

In the following sections, the simulation framework will be established to

demonstrate the characteristics of the three magnetization switching mechanisms [103]:

field-induced magnetic switching (FIMS), spin-transfer torque (STT) switching, and

thermally assisted switching (TAS). In addition, the STT switching, as it is affected by

thermal fluctuations, will be demonstrated.

3.3 Magnetic Energy and Torques

In the free layer, M is the magnetization vector, which is the volume density of

permanent or induced dipole moment of the magnetic material. H is the applied (external)

magnetic field which makes an angle  with respect to the z-axis, S is the spin angular

momentum making an angle  to the z-axis, and I is the spin current. Both angles  and

 lie on the yz-plane (see Figure 3.5). Assume that the ferromagnet has perpendicular

magnetic anisotropy. The z-axis is defined as the easy axis, which is the stable direction

for the magnetization vector to align. The magnetic potential energy U stored in the

magnetic field is composed of the energies including uniaxial anisotropy energy U K , easy-

plane anisotropy energy U P , and an applied magnetic field energy U H . These potential

energies make the following forms [104], [105]:

34
 U K = K sin 2  and K = (1 / 2) 0 M s H K ,

U P = K P ( sin 2  sin 2  ) , (3.1)

U H = − M  H = − MH ( sin  sin  sin + cos cos ) ,

where H K is the uniaxial anisotropy field, and both K and K P are uniaxial anisotropy and

easy-plane anisotropy constants, respectively. The total magnetic potential energy

becomes

U = UK +U P +UH . (3.2)

As a uniform magnetic field exerts a net torque on magnetic dipoles, the various

magnetic torques can be derived from magnetic energy equations. Consider a bar magnet

of length lm , placed in a uniform magnetic field (see Figure 3.3). The external magnetic

force F acts on the magnet, producing the torque ΓU to re-align the magnetization vector

parallel to the field lines and store the potential energy U in the magnetic field. The torque

ΓU is the vector product of the lever arm l m and the force F = −U ; namely,

ΓU = l m  F = l m m  ( − U ) = −lmm U , where m = M / M = M / M s is the unit

magnetization vector pointing from the south pole to the north pole of the magnet. The

relation of the potential energy and mechanical torque to the unit magnetization vector is

ΓU
= −m U . (3.3)
lm

35
 Figure 3.3 A fictitious bar magnet represents the vector of magnetization. The magnet is placed in the
magnetic field and experiences a force couple and torque due to the magnetic field.

Applying the principle of moments, we let the resultant torque be the algebraic sum

of the three torques ΓU = Γ1 + Γ2 + Γ3 due to the magnetic energy of (1) the uniaxial

anisotropy U K , (2) the easy-plane anisotropy U P , and (3) the external magnetic field U H ,

respectively. Using (3.3) for the total torque exerted by the magnetic field, we write

ΓU Γ1 + Γ2 + Γ3
= = −m (U K + U P + U H ) . (3.4)
lm lm

3.4 Dynamics of Magnetization and Landau–Lifshitz–Gilbert Equation

The dynamics of the magnetization M (in SI units: A/m) on the ferromagnetic free

layer can be described by the Landau–Lifshitz–Gilbert (LLG) equation [106]:

36
 dM   dM 
+ M  =  MB, (3.5)
dt M s  dt 

where t is the time variable and B is magnetic flux density or magnetic induction (in SI

units: T). When the temperature is below the Curie point, the magnetization M = M s m ,

where M s is the saturation magnetization value of the ferromagnet, α is the Gilbert

damping factor, and  = gB /  2 B / is the gyromagnetic ratio, where  B is the Bohr

magneton. Equation (3.5) expresses the time evolution of the magnetization vector. The

 M  B term describes the Larmor precession in the magnetic field, while

( / M s )M  dM / dt is the phenomenological damping term representing the energy

dissipation. Eventually, it causes the axis of revolution to be aligned with the applied

magnetic field. The movement of the magnetization vector in response to the torques is

illustrated in Figure 3.4. Factor out the saturation magnetization M s on LHS of (3.5) and

write

dm  dm  
+ m = MB. (3.6)
dt  dt  M s

The magnetic flux density B = 0 H , where 0 is the permeability of free space,

and H is the magnetic field strength. Equation (3.6) can be expressed by using the effective

magnetic field H eff (in SI units A/m ), which includes the torques associated with the

anisotropy and the applied field as [103]

37
 dm  dm  
+ m = 0 M  H eff . (3.7)
dt  dt  M s

Г is the torque acting on the magnetization vector of the magnetic body where

Γ = M  B = 0 M  H eff . Substituting this relation into (3.7) leads to

dm  dm  
+ m = Γ. (3.8)
dt  dt  M s

Introducing the uniaxial anisotropy energy K = (1 / 2) 0 M s H K to the RHS of (3.8),

we obtain

Figure 3.4 Precession locus of magnetization impacted by the damping torque, the effective magnetic
field torque, and the spin-polarized current caused spin transfer torque.

38
 dm  dm  1 Γ
+ m  = 0 H K , (3.9)
dt  dt  2 K

where H K is the anisotropy magnetic field. If considering three energy-associated torques

acting on the magnetic body in (3.4), where Γ1 is from the uniaxial anisotropy, Γ 2 is the

easy-plane anisotropy, Γ3 is the applied magnetic field, then the following form results

dm  dm  1 3
Γi
dt
+ m

 =
dt  2
0 H K 
i =1 K
. (3.10)

In the next section, additional torques due to the spin-polarized current ( Γ 4 ) and

the thermal fluctuations ( Γ5 ) in combination with the dynamic equation of magnetization

will be discussed.

3.5 Spin-polarized Current and Spin-transfer Torque

An electron bears charge and spin. The current flow carries not only electron

charge but also electron spin angular momentum. On the microscopic scale, electron spin

plays an important role as it interacts with the magnetic object. Because the rate of change

of spin angular momentum is the torque, this serves as the mechanism by which the spin

angular momentum is transferred through the torque, which affects the magnetization

status. For the dynamics of the spin-transfer torque (STT) interacting with the

magnetization of the MTJ free layer, the τ STT term is introduced into the LLG equation as

39
 dm  dm   
+ m = 0 M  H eff + τ STT . (3.11)
dt  dt  M s Ms

The spin-transfer torque τ STT = sm  (n S  m) , where n S is the unit vector of the

initial spin orientation. The spin current density s and charge current density J have the

relation, s = / (2e) J . The spin-polarization efficiency  = ( N  − N  ) / ( N  + N  ) ,

which N  means the number of the majority spin eigenstate and N  is the number of the

minority spin eigenstate [107]. As the torque may alter the magnetization, the STT-driven

switching from high resistance AP-state to low resistance P-state and vice versa becomes

possible. Our simulation demonstrates this effect.

3.6 Numerical Framework

The computation method follows Sun’s work [108] with our modification. The

simulation of the free-layer magnetization dynamics for the MTJ element is based on the

Stoner–Wohlfarth model, which describes the magnetization behavior of a single

ferromagnetic domain body. Assume that the cylindrical free layer of the MTJ is situated

on the xy-plane, and the z-axis is the easy axis corresponding to the stable direction of

spontaneous magnetization [109] of the free-layer ferromagnet. Assume that the device

has perpendicular magnetic anisotropy and the magnetization vector is out-of-plane or

perpendicular to the cross-sectional plane of the free layer. It is convenient to describe the

rotational motion of the magnetization vector using spherical polar coordinates ( r, , ) .

Let the unit vector of the magnetization m = M / M s = e r be located at the time-varying

40
polar angle  =  (t ) and azimuthal angle  =  (t ) (in physics convention), as illustrated

in Figure 3.5.

Figure 3.5 Geometry and physical quantities on the model of the MTJ. “M” is the spontaneous
magnetization vector of the free layer. “Hext” stands for the external magnetic field. “S” represents the
spin angular momentum and the initial spin direction.

In the spherical polar coordinates, the LHS of (3.10) can be expressed in terms of

 and  components as

dm  dm   d d   d d 
+ m = −  sin   e +  sin  +  e . (3.12)
dt  dt   dt dt   dt dt 

41
 Next, calculate the Γ’s. For the unit magnetization m = e r , using (3.1) and (3.4)

for Γ1 lm , Γ 2 lm , and Γ3 lm , we obtain the following relations:

−m  U K = −2 K sin  cos  e ,

−m  U P = 2 K P ( sin 2  sin  cos  ) e − ( sin 2  sin  cos  ) e  , (3.13)

−m  U H = MH  − ( cos  sin ) e + ( cos sin  sin − sin  cos ) e  .

It follows that

Γ1
= −2sin  cos e . (3.14)
lm K

Γ2
= 2hP ( sin 2  sin  cos  ) e − (sin 2  sin  cos  ) e  , (3.15)
lm K

where the dimensionless easy-plane anisotropy field hP = K P / K .

Γ3
= −2h ( cos  sin ) e − ( cos sin  sin − sin  cos ) e  , (3.16)
lm K

where the dimensionless magnetic field h = H / H K = H / (2 K / M s ) .

42
 For the spin current, assume that the majority electron spin orientation is the angle

 measured with respect to the z-axis. The transport of the spin current is along the z-axis.

Suppose the spin direction varies only in the yz-plane, so the spin unit vector is defined as

nS = (sin  )e y + (cos  )e z . In the spherical polar coordinates, n S becomes:

nS = ( sin  sin  sin  + cos cos  ) er + ( cos sin  sin  − sin  cos  ) e + cos e .

Recall the directional unit vector of magnetization m = e r . Expressing the vector triple

product of  STT in (3.11) using Γ 4 , we calculate the vector triple product:

Γ4 = sm  (n S  m) = s ( cos sin  sin  − sin  cos  ) e + cos  e  .

It follows that

Γ4
= 2hS ( cos sin  sin  − sin  cos  ) e + cos  e  , (3.17)
lm K

/ (2e) J / (2e) I
where the dimensionless spin current hS = = .
2lm K 2 ( / 2)2 lm K

Applying (3.12) with (3.14), (3.15), and (3.16), we can construct a system of scalar

differential equations of  and  for vector quantities in the LLG equation, which

considers torques of the magnetic potential energy and spin-polarized current.

43
 1  d d   d d  1 4
Γi
 −  sin   e +  sin  +  e = 0 H K 
lm  dt dt   dt dt  2 i =1 lm K

or

 d d   d d  1 4
Γi
 −  sin  +
  
e sin  +   
e =  0 K
H . (3.18)
 dt dt   dt dt  2 i =1 K

Comparing components of e and e on LHS and RHS of (3.18), we can construct the

simultaneous differential equations for each torque and their combined influences.

To consider only the effect of Γ1 with damping, we can decompose the LLG

equation into spherical polar coordinates:

 d 0 H K
 dt = 1 +  2 ( − sin  cos ) ,

 (3.19)
 d  H
 = 0 K ( − cos ) .
 dt 1 +  2

Regarding the effect of Γ 2 with damping, the LLG equation can be decomposed into:

 d 0 H K 
 dt = 1 +  2  hP sin  ( sin  cos + sin  sin  cos  )  ,
2


 (3.20)
 d  H
 = 0 2K  hP ( sin 2  cos − sin  sin  cos  )  .
 dt 1+

44
For the purpose of exhibiting the effect of Γ3 with damping, the LLG equation can be

decomposed into:

 d 0 H K
 
 dt = 1 +  2 − h  ( sin  cos − cos  sin  sin ) + cos  sin  ,

 (3.21)
 d  H
 = 0 K  − h ( sin  cos − cos  sin  sin −  cos  sin )  (1 / sin  ) .
 dt 1 +  2 

For the purpose of exhibiting the effect of Γ 4 with damping, the LLG equation can be

decomposed into:

 d 0 H K
 dt = 1 +  2 hS ( cos  + cos sin  sin  − sin  cos  ) ,

 (3.22)
 d  H
 = 0 K hS cos  −  ( cos sin  sin  − sin  cos  )  (1 / sin  ) .
 dt 1 +  2 

Considering all torques affecting the magnetization of the ferromagnet, we may write the

following equations:

 d 
 dt  0 H K
2 ( 1
 = Γ + Γ 2 + Γ3 + Γ 4 ) . (3.23)
 d  1 + 
 dt 

45
With the initial condition ( (0), (0) ) , the trajectory of magnetization can calculated and

obtained.

3.7 Simulation Setting

Assume that the free layer ferromagnet has perpendicular magnetic anisotropy with

the z-axis as the easy axis. The magnetization M is described by the unit vector

m = ( mx , my , mz ) = (sin  cos  ,sin  sin  ,cos ) , whose components (directional cosines)

satisfy the relation mx2 + m2y + mz2 = 1 . As the MTJ is characterized by the perpendicular

magnetic anisotropy (p-MTJ), the equations indicate that the projection of the

magnetization on the z-axis, mz (i.e., the z-component of the magnetization), accounts for

the switching process. Thus, for the switching behavior, we will concentrate on the

variation of mz = cos , which varies between mz = −1 , corresponding to the AP-state, and

mz = 1 , corresponding to the P-state. Figure 3.6 shows the relation:

Figure 3.6 Projection of the magnetization m on the z-axis. The rotation of 𝑚𝑧 in the unit circle
represents the switching between the two states.

46
 In addition, assume the angle of the external or applied magnetic field H ext with

the z-axis is  = 0 , and the angle of the incident spin-polarized current S with the z-axis is

 = 0 (i.e., sin = 0 , cos = 1 , sin  = 0 , and cos  = 1 ). The parameters and constants

for the numerical calculation of the free-layer magnetization dynamics are listed in Table

3.1. For the computation, the time step is smaller than the duration of any applied physical

quantity. The vector form LLG equation will be decomposed into a system of scalar LLG

equations in spherical polar coordinates. Finally, the explicit Runge–Kutta method of order

four will be performed for the numerical integration of the resulting non-linear ordinary

differential equations.

Table 3.1 Values and Units of Parameters or Constants for Simulation.

Parameters / Constants Value Unit

Boltzmann constant (𝑘𝐵 ) 1.38110-23 J/K

Reduced Planck constant (ℏ) 1.05510-34 J·s
Permeability of vacuum (𝜇0 ) 4π10-7 T·m/A
Elementary charge (e) 1.60210-19 C
Diameter of MTJ (⌀) 20010-9 m
Thickness of free layer (𝑡𝐹 ) 1.0810-9 m
Volume of free layer (𝑉𝐹 ) 3.3910-23 m3
Gyromagnetic ratio (γ) 1.761011 (free electron) (rad)s-1T-1
Damping factor (α) 0.02
Anisotropic constant (K) 1.13105 T·A/m or J/m3
Saturation magnetization (𝑀𝑠 ) 1.4/𝜇0 A/m
Uniaxial anisotropy field (𝐻𝐾 ) 7.96103 A/m
Spin-polarization efficiency (η) 0.95

47
3.8 Simulation of Magnetization without an External Torque

When external torques are ignored, the magnetic body will only experience the

uniaxial anisotropy torque Γ1 , which is determined by the crystalline structure of the

magnetic body. The torque causes the Larmor precession along with the damping torque

due to the nature of energy dissipation. The latter serves to align the magnetization vector

with the easy axis. In order to observe the spontaneous movement, the magnetization m

has been assumed to lie initially on the surface of the thin film at the polar coordinates

( − / 2,0) ; namely, z = 0 (xy-plane) of the Cartesian coordinates, as shown in Figure 3.7:

Figure 3.7 Initial positioning of the magnetization vector m in the unit sphere for the simulation of
precession and alignment.

By setting the starting position slightly above the surface, it is allowed to evaluate

(3.19) by simulation and observe the trajectory (in red color) of the precession of the

magnetization, m, moving clockwise and eventually aligning with the easy axis in the + z

48
direction. In contrast, if the magnetization vector starts the precession slightly below the

xy-plane, the magnetization, m, precesses counterclockwise and eventually aligns with the

easy axis towards − z direction, as shown in Figure 3.8:

(a) (b)

Figure 3.8 Precession and damping of the magnetization vector, m, during 500 ns in the unit sphere. (a)
the initial position of m is slightly above the xy-plane, resulting in a clockwise precession; (b) the initial
position of m is slightly below the xy-plane, resulting in a counterclockwise precession.

3.9 Simulation of Magnetization Due to External Magnetic Field

This section considers the combination of a uniaxial anisotropy field and an

external magnetic field, which is parallel to the easy axis (z-axis). The magnetization, m,

initially corresponds to the AP-state, which is perpendicular to the film surface located by

the initial spherical polar coordinates ( r, , ) = (1, − ,0) or Cartesian coordinates

49
( x, y, z ) = (0,0, −1) . An external magnetic field, Bext , pointing in the + z direction is

applied to the system. Figure 3.9 indicates the arrangement.

Figure 3.9 Initial position of the magnetization vector m in the unit sphere for evaluating the effect of
the external magnetic field.

The net influence of torques is Γ1 + Γ3 . With the application of (3.19) and (3.21),

the system of equations for the dependence of the angles is

 d 
 dt  0 H K  H  − sin  cos − h sin  
2 ( 1
 = Γ + Γ3 ) = 0 2K  . (3.24)
d
   1 +  1 +   − cos  − h 
 dt 

50
 Figure 3.10 Magnetization reversal due to increasing of an external magnetic field from (1) 16 mT to
(8) 20 mT in a 250 ns time interval.

Figure 3.10 shows that the external magnetic field increased evenly from 16 mT (level 1)

to 20 mT (level 8), accelerating the precession and inducing the magnetization switching.

Figure 3.11 demonstrates the corresponding trajectory in the unit sphere of the different

magnitudes of the applied magnetic field. According to the National Oceanic and

Atmospheric Administration (NOAA), the magnitude of the Earth's magnetic field strength

at its surface is roughly between 25–65 μT (0.25–0.65 G). Hence, the geomagnetic field

does not affect the magnetization dynamics of the layers on the MTJ. A typical refrigerator

magnet with a magnetic field strength of 1–5 mT (10–50 G) may also not trigger the

switching and affect the functionality of the MTJ. MRAM devices are also capable of

magnetic shielding by a layer of mu-metal, which is a very high permeability ferromagnetic

alloy, the magnetization can be adjusted to attenuate the external magnetic field [110],

[111].

51
 Figure 3.11 Magnetization switching due to a uniaxial anisotropy field and an external magnetic field,
which varied the magnitude from (1) 16 mT to (8) 20 mT in a 250 ns time interval.

3.10 Simulation of Magnetization Due to Spin-transfer Torque

In this section, spin-transfer torque switching is demonstrated by simulation.

Figure 3.5 shows the incoming spin current flowing in the +z direction to the MTJ with the

assumption of the spin angle  = 0 and the spin-polarization efficiency  = 0.95 . A

square current pulse is applied to the system during the 500 ns monitoring window. The

current pulse is turned on at 114 ns and ended at 284 ns after a duration of 170 ns. In order

to observe the magnetization reversal in the simulation, the square current pulse magnitude

is varied from 0.0526 mA to 0.0695 mA (or 1.67 105 A/cm2 to 2.21 105 A/cm2 ).

52
 In the physics aspect, the spin-transferred torque Γ 4 superimposes to the uniaxial

anisotropy torque Γ1 , creating the net influence of torques on the free layer of MTJ, which

is expressed by Γ1 + Γ 4 . Mathematically, using (3.19) in conjunction with (3.22), we

establish the system of differential equations for the time variation of the angles:

 d   − sin  cos + hs cos  

 dt  0 H K  H  .
2 ( 1
 = Γ + Γ4 ) = 0 2K
 − cos + hs  cos  +   
(3.25)
 d  1 +  1+
  sin   
 dt 

Let the magnetization, m, rest in the AP-state located by the initial spherical polar

coordinates ( r, , ) = (1, − ,0) , as follows in Figure 3.12.

Figure 3.12 Initial position of the magnetization vector m and spin-polarized current S in the unit
sphere for evaluating the impact from spin-transfer torque.

53
 Figure 3.13(a) shows the input spin current pulse for eight different levels of

magnitude denoted by (1) to (8). The current starts as low as 0.0526 mA (level 1) and is

increased unevenly to 0.0695 mA (level 8). The corresponding Figure 3.13(b) shows the

resultant magnetization. The switching event happens while the input current pulse rises

up to 0.0562 mA (level 4). Current magnitudes at 0.0532 mA (level 2) and 0.0544 mA

(level 3) were unable to support the magnetization reversal. The simulation also tells us

that the magnetization will stay at P-state but not fall back to AP-state once the switching

from AP-state to P-state occurs. Thus, it allows the ability to control the state of

magnetization.

(a) (b)

Figure 3.13 (a) Spin current pulse input during 114 ns to 284 ns. (b) Magnetization switching due to
increasing of the spin-polarized current induced spin transfer torque from (1) 0.0526 mA to (8) 0.0695
mA in the time interval of 500 ns.

54
 Figure 3.14 Magnetization switching due to a uniaxial anisotropy field and spin-polarized current
inducing spin-transfer torque. The pulse of the spin current intervenes with the precession. It becomes
the dominant factor causing the magnetization reversal, while the magnitude of the pulse increases from
(1) to (8) from 114 ns to 284 ns.

Figure 3.14 is the trajectory of the magnetization in Figure 3.13. During the input

of levels 1, 2, and 3 of the current pulse, the precession dominates the motion; however,

the monitoring time frame of 500 ns is not enough to foster the switching event. When the

input spin current is higher in levels 4, 5, 6, 7, and 8, the spin-transfer torque takes effect,

and the switching takes place. In addition, the figure indicates that with the higher input

spin current pulse, the switching is faster, and the trajectory is shorter. Although the spin

current is large enough to promote switching, the precession and damping are ongoing and

cause the lag to switch. That is, the input of spin current pulse and magnetization switching

55
cannot coincide, as shown in Figure 3.15. The spin current that initiates the spin-transfer

torque compensates for the damping [112].

Figure 3.15 Magnetization switching due to strong spin-polarized current induced spin-transfer torque
during 114 ns to 284 ns. The larger the spin current, the shorter the time to switch 𝑡𝑆𝑊 .

3.11 Simulation of Magnetization Due to Thermal Fluctuations

In MRAM, changing magnetization between resistance states needs to apply energy.

Thermally agitated magnetization fluctuations may cause switching when the thermal

energy is comparable to the magnetic storage energy of the MTJ device. Once the magnetic

device is subject to heavy usage, self-heating or heating having its origin in the

environment, may cause the storage of the information to be vulnerable to the temperature.

The thermal stability factor  = (1 / 2) M s H K kBT , where kB is the Boltzmann constant

and T is temperature, is the physical quantity that determines the resistance to thermal

56
fluctuations. The fluctuation changes the state of magnetization, resulting in thermally

assisted switching. To formulate the effect of thermal fluctuations in addition to the LLG

equation, we introduce the time-dependent random thermal field vector hT (t ) , and the

components of hT (t ) in spherical coordinate system are hT (t ) (in the  direction) and

hT (t ) (in the  direction). The statistical averages of the thermal field components are

expressed as hT (t ) and hT (t ) . In order to ensure stochasticity, the random thermal

field hT (t ) must be stationary, with each component distributed normally (Gaussian) and

both components having mean values, hT (t ) = 0 and hT (t ) = 0 . In addition, if the

process is purely random, the correlation of the thermal field components

hT (t ), hT (t + t ) =   ( t ) , where  is constant,   is Kronecker delta and  ( t )

is Dirac delta function [113]. In order to satisfy these statistical properties, let

Γ5 = [ hT hT ]T represent the thermal fluctuation effect, where hT (t ) = hT 0 RNG(−1,1)

and hT (t ) = hT 0 RNG(−1,1) . Both hT and hT are independent components of the time-

dependent random thermal fluctuation field hT = hT (t ) , and hT 0 is a number that stands

for the constant thermal field amplitude depending upon the environment where the

proportionality hT 0  k BT holds. The amplitude is kept the same for the two components.

The RNG( −1,1) represents a pseudo-random number generator, generating random

floating-point numbers fluctuating between -1 and 1. Considering the thermal field in

addition to the uniaxial anisotropy field, we know that

57
  d 
 dt  0 H K  H  − sin  cos + hT 
2 ( 1
 = Γ + Γ5 ) = 0 2K  . (3.26)
 d  1 +  1 +   − cos + hT 
 dt 

Equation (3.26) will be used to describe the magnetization dynamics when only

thermal fluctuations affect the magnetization externally. Figure 3.16(a) depicts a

magnetization vector receiving ambient thermal energy impact in all directions,

representing a stochastic manner. Each directional thermal torque alters the movement of

the magnetization. To observe how the thermal noise affects the precessional movement

of magnetization, we let the initial condition be (r, , ) = (1, − / 2,0) . The switching

behavior is seen through the magnetization vector projection on the z-axis. The thermal

fluctuation is random; however, if the first hT  0 , the magnetization will initially be led

to the northern hemisphere, whereas if the first hT  0 , the magnetization will initially be

led to the southern hemisphere, as shown in Figure 3.16(b). The final location of the

magnetization depends on the magnitude of the thermal fluctuation Γ5 in comparison to

the strength of the uniaxial anisotropy Γ1 and the strength of the damping.

58
 (a) (b)

Figure 3.16 (a) Magnetization vector m experiencing omnidirectional thermal fluctuations, which
affect the movement. (b) The thermal stochasticity influences the direction of precession, so the
projection 𝑚𝑧 is oscillating, depending on the sign of the random component ℎ 𝑇𝜃 of the thermal
fluctuation torque encountered at the time.

Particularly, if the thermal noise is slight, the switching will not occur, but a minor

trajectory disturbance can be observed. The end of a movement is determined by the sign

of the initial hT . Comparing Figure 3.17 with Figure 3.8, we see the perturbation in the

precession trajectory caused by the slight thermal fluctuation. The slight thermal

fluctuation only perturbs the precession orbit but does not affect the switching. Therefore,

the precession and the damping are dominant in the magnetization dynamics.

59
 (a) (b)

Figure 3.17 Magnetization vector m experiencing thermal fluctuation. The trajectory is perturbed in the
unit sphere over 500 ns. (a) The thermal fluctuation caused clockwise precession as initial ℎ 𝑇𝜃 > 0;
(b) The thermal fluctuation caused counterclockwise precession as initial ℎ 𝑇𝜃 < 0.

However, if the magnetization is initially placed at a stable AP-state or P-state, the

switching or the reversal of magnetization will not occur solely due to the precession and

the damping. When the magnitude of the thermal fluctuation is sufficiently large (e.g.,

hT 0 = 2.0 ), the thermal effect may lead to the movement of magnetization, and the

thermally assisted switching will happen. The time and the path before a switching event

cannot be predicted due to the stochasticity of thermal fluctuations. Figure 3.18 shows the

randomness associated with the trajectory before the magnetization vector first reaches the

P-state after starting from the initial condition, ( (0), (0)) = ( − / 2,0) , which is preset at

the AP-state. The case in Figure 3.18(a) shows the path to the P-state, and that in Figure

3.18(b) indicates more time spent and a longer distance of the random walk traveled in

order to arrive at the P-state.

60
 (a) (b)

Figure 3.18 Stochasticity of the thermal fluctuation assisted switching by the same condition (thermal
amplitude ℎ 𝑇0 = 2.0) resulting in different trips. (a) 6 ns with 53 random steps; (b) 30 ns with 262
random steps to reach the P-state.

61
 The instability caused by thermal fluctuations may result in magnetization reversal,

thus changing the state and further reducing data retention. Figure 3.19(a) shows the entire

sequence of switching events in the temporal evolution of the magnetization subject to a

thermal amplitude hT 0 = 2.0 . When zooming into the first 100 ns, as shown in Figure

3.19(b), the switching events between the AP- and P-states can be clearly seen. The event

time of each state is measured by the duration between two consecutive switching events.

In Figure 3.19(b), the elapsed time between 10 ns and 26 ns is an AP-state event, whereas

the elapsed time from 28 ns to 43 ns is a P-state event. The same events from the digital

signal point of view are the OFF event and the ON event, respectively, as displayed in

Figure 3.19(c).

Figure 3.19 Temporal evolution of magnetization (a) Magnetization switching due to thermal
fluctuation observed during 0 to 5000 ns. (b) Magnetization state varying by zooming into the
beginning 100 ns. (c) Digitized magnetization of the ON-state and the OFF-state switching in the first
100 ns.

62
 The characteristic time for determining the MTJ stability; namely, the duration of

staying in each stable state, is based on the Néel–Arrhenius law, where the relaxation time

 is given by

 Eb 
 = 0 exp  . (3.27)
 k BT 

In (3.27),  0 is the attempt time period needed to overcome the energy barrier depending

on the uniaxial anisotropy field [113], and typically  0  1 ns. The switching event is a

Poisson process [105]. Practically, the relaxation time  can be determined from the curve

fitting of the exponential distribution. The attempt rate, N, between two switching events

[114] separated by the time, tevent , is

1  t 
N = exp  − event  . (3.28)
   

The simulation ran up to several thousand switching trials in a 5000 ns interval in

order to acquire enough data for the statistical analysis. Figure 3.20 demonstrates the

normalized histograms of the event time with the curve fitting of the exponential

distribution function for both AP-state and P-state [114]. By fitting the distribution curve

described in (3.28) to the normalized histogram, we obtain  AP = 5.144 ns for the relaxation

time in AP-state and  P = 5.146 ns for the relaxation time in P-state. Note that  AP   P

results are under conditions where the thermal field amplitude hT 0 = 2.5 and the anisotropy

63
field 0 H k = 10 mT when observed over a 5000 ns interval. The averaged relaxation time

 av is defined by the geometric mean of  A and  AP , which is  av =  A AP for the entire

process. The average relaxation time  av = 5.145 ns is calculated for this simulation.

(a) (b)

Figure 3.20 The attempt rate fitted with the exponential distribution and the relaxation time. (a) AP-
state relaxation time 𝜏𝐴𝑃 = 5.144 ns (b) P-state relaxation time 𝜏𝑃 = 5.146 ns, and the average
relaxation time 𝜏𝑎𝑣 = 5.145 ns from the simulation under thermal amplitude ℎ 𝑇0 = 2.5 over a 5000 ns
interval.

The relaxation time is the function of the ambient thermal field amplitude hT 0 ,

which varies with temperature. By increasing the value of hT 0 from 1.5 to 3.25, we can

see the switching frequency rising, which results in the color of the magnetization plot

being denser over the elapsed time of 5000 ns, as shown in Figure 3.21.

64
 Figure 3.21 Magnetization switching screen when the thermal field amplitude from ℎ 𝑇0 = 1.5, via
ℎ 𝑇0 = 2.5 to ℎ 𝑇0 = 3.25.

The following Figure 3.22 demonstrates the variation of  AP and  P with the

thermal field. As the thermal field amplitude hT 0 increases, the relaxation time decreases,

which indicates that in the high temperature environment, the magnetization has a higher

tendency to switch and thus affects data retention.

65
 Figure 3.22 Relaxation time versus thermal field amplitude. (a) Linear plot (b) Semi-log plot (c)
Increasing thermal field amplitude in the entire time range.

In the regular use of MRAM, the magnetization of MTJ is designed to be stable and

not to switch in order to retain the stored information. However, when there are abrupt

extreme conditions to the device, such as optical heating due to illumination by a laser

beam or bombardment by energetical particles, a time-dependent thermal field amplitude

hT 0 = hT 0 (t ) must be considered. To achieve this, assume that the background thermal field

amplitude stays low at hT 0 = 0.25 , and the incoming energetic thermal heating (in the form

of a square pulse) has the magnitude hT 0 = 5 to impact the device during the period

between 1500 ns to 2000 ns. Figure 3.23 demonstrates the scenario by which the thermal

pulse causes the switching and the resulting changes in magnetization states over the 5000

66
ns period. In addition, a random switching event due to accumulated background thermal

noise (the same sign of hT ) is observed near 3400 ns.

P-state (ON)
AP-state AP-state

(OFF) (OFF)

background background

Figure 3.23 An unexpected magnetization switching due to the impact of the high-energy thermal pulse.
The magnitude of thermal heating induces magnetization reversals from 1500 ns to 2000 ns. Although
the background thermal fluctuation is low, the magnetization may be reversed due to accumulated
thermal noise (3400 ns–3450 ns).

When thermal fluctuations occur, the relaxation time can serve as a measure of

stability. The same concept can be applied to determine the effect of an incoming thermal

pulse, which can be treated as the time-dependent thermal field amplitude. Figure 3.24

shows that while the background thermal field amplitude remains the same at hT 0 = 1.5 ,

the larger the thermal pulses, the shorter the relaxation time. The trend indicates the

incoming energy pulse serves to shorten the relaxation time due to causing more switching

events, thus increasing instability.

67
 Figure 3.24 Relaxation time versus thermal pulse amplitude. (a) Linear plot (b) Semi-log plot (c) Even
increment of the thermal pulse magnitude and acting over the 1500 ns to 2000 ns time interval.

By comparing Figure 3.22 and Figure 3.24, we understand that a very high energy

thermal pulse, one even lasting over a short period, may have a similar impact as the

application of a small but uniform background heat.

68
3.12 Simulation of the Spin Current Modulation on Thermal Fluctuations

In MRAM devices, particularly the spin-transfer type, STT-MRAM, the input spin

current is controllable and used to determine the magnetization states for storing

information, as described in Section 3.10. This section explores magnetization with the

combined influence of STT and thermal fluctuations. The spin-polarized current is applied

to suppress thermal fluctuations and further modulate the relaxation time.

To establish the simulation model, let the intrinsic uniaxial anisotropy Γ1 be

superimposed on the external spin-transfer torque Γ 4 plus the thermal torque Γ5 , which

makes the combination Γ1 + Γ4 + Γ5 in (3.23). This combination leads to

 d   − sin  cos + hs cos  + hT 

 dt   H K  HK   . (3.29)
2 ( 1
 = Γ + Γ 4 + Γ5 ) =
 − cos + hs  cos  +   + hT 
 d  1 +  1+2
 dt    sin   

With the initial condition ( (0) = − , (0) = 0) indicating that the magnetization vector

initially rests in AP-state ( mz = −1) , we integrate simultaneously in both  and 

directions to acquire the magnetization trajectory. Figure 3.25 visually shows the

combined effect of the spin current (positive) and the thermal fluctuation. As the spin

current increases from 0.1 mA to 0.4 mA, the magnetization tends to be “frozen” in the P-

state ( mz = 1) , whereas by heating the device, the switching frequency increases, and a

larger current is required to maintain stability.

69
 Figure 3.25 Combined effects of the positive spin current and thermal fluctuations. The thermal energy
increases horizontally from left to right [(a) to (c), (e) to (f)], and the spin current increases vertically
from bottom to top [(d) to (a), (e) to (b), (c) to (f)].

Figure 3.26 visually presents the combined effect of the spin current (negative) and

thermal fluctuation. The increase of the negative spin current overwhelmingly freezes the

magnetization in AP-state when the thermal fluctuation is low. Conversely, the stability

Figure 3.26 Combined effects of the negative spin current and thermal fluctuations. The thermal energy
increases horizontally from left to right [(a) to (c), (e) to (f)], and the negative spin current increases
vertically from top to bottom [(a) to (d), (b) to (e), (c) to (f)].

70
weakens when the heating is intense, and a larger spin current is required to suppress the

thermal noise.

Quantitatively, the stability by introducing the relaxation time analysis can be

determined when thermal fluctuations exist. Figure 3.27 demonstrates that the negative

spin current hinders switching from AP-state to P-state. Upon increasing the magnitude of

the negative current, the relaxation time of the AP-state  AP (green “x” symbol) increases.

Whereas the positive spin current hinders switching from P-state to AP-state, the relaxation

time of the P-state  P (blue “+” symbol) increases with the increasing magnitude of the

current. At the zero-current point, we see  AP   P   av .

Figure 3.27 Relaxation time verses spin current changes from -30 mA to +30 mA. The spin current (in
form of square pulse) modulates the thermal fluctuation. The negative current hinders the switching
events from AP-state to P-state. The positive current hinders switching events from P-state to AP-state.

71
 The simulation reveals that with the control of the input spin current, it is allowed

to change the state of magnetization, which in practice, means that we are able to modify

the information stored in the MRAM. Conclusively, an energetical thermal fluctuation

may destroy the data. Still, the spin current induced STT can be used to prevent this

alternation, leading to a retention of the stored information.

3.13 Summary

This chapter explored the free-layer dynamics of the p-MTJ by decomposing the

vector form of the LLG equation into the system of scalar equations in spherical polar

coordinates. In the absence of external torques, the simulation shows that as the

magnetization precesses, it damps and aligns to the easy axis during the entire process. The

magnetization switching between AP-state and P-state will be observed when an MTJ is

impacted by the external magnetic field, spin-polarized current, or thermal fluctuations.

Such individual or combined influences may exert torques to affect the precessional motion.

The two states, which differ in tunneling resistance, underlie the binary information storage

capacity. In addition, an increase in the spin current not only induces switching but also

overcomes the thermal noise is demonstrated. The net effect is to enhance stability and

device reliability. The thermally assisted switching becomes more frequent as the thermal

fluctuations increase. This connection between the statistical distribution of thermal

relaxation time and magnetization switching is demonstrated. The length of the relaxation

time determines the strength of the thermal instability. Evaluating the relaxation time

enabled us to monitor the stability and determine thermal energy absorption from an

external heat source.

72
 Chapter 4. Electrical Treeing and Breakdown in Composite Dielectrics

4.1 Overview

Polymer-based dielectrics are commonly used in many electronic devices as they

offer good electrical insulation. However, polymers degrade over a period of time while

operating under a sub-critical electric field [115]. The common approach for enhancing

the reliability of polymeric materials is mixing with non-conducting (ceramic) filler

materials [116]. It is experimentally reported that the breakdown strength significantly

improves with the optimal filler loading. The improvement depends on the size and the

shape of the fillers as well as their spatial distribution in the composite film [117]–[119].

In addition, filler-loaded polymer composites often show large variability in breakdown

strength, which limits their large-scale industrial adoption. In short, experimental studies

show that incorporating insulating fillers improves the breakdown strength, but there is no

theoretical guideline to optimize the mixing process. Hence a theoretical framework is

important to reap the benefits from polymer-filler composites.

Electrical treeing has been identified as one of the critical factors affecting the long-

term reliability of polymer dielectrics [120]. The most commonly used model for electrical

treeing in polymer is the dielectric breakdown model (DBM), first introduced by Niemeyer

et al. [121] and describes the electric breakdown in homogeneous dielectrics as a stochastic

process producing fractal structures that are called electrical trees.

73
 The key feature in the DBM is the local field-dependent stochastic breakdown

channel formation, which in turn enhances the local field and eventually leads to complete

breakdown. Irurzun et al. compared the electrical treeing on the two stochastic models of

the diffusion-limited aggregation model and the dielectric breakdown model (with η = 1)

using fractal analysis [122]. Peruani et al. extended the dielectric breakdown model with

conductor-loaded filler particles as inhomogeneous distribution and simulated the fractal

structure of the electrical tree [123]. Bergero et al. modified the dielectric breakdown

model with insulator-loaded filler conditions for composite materials. They explored

dielectric breakdown fractal dimension concerning fractions of conducting and insulating

particles [124]. However, none of the previous analyses report the effects of filler size,

shape, and position on the breakdown strength of the dielectrics. In this research, a

numerical simulation framework is developed in order to understand the role of insulating

filler on the long-term reliability of polymer-filler composites. The framework is based on

extending the DBM by accounting for the properties of filler materials. Using the

simulation framework, we systematically analyze the effects of filler distribution, size,

shape or orientation on the electrical tree formation and hence the long-term reliability.

Finally, different engineering strategies to further improve the reliability of polymer-filler

composites are proposed.

4.2 Computation Method

In this research, we have extended the stochastic model for dielectric breakdown

developed by Niemeyer et al. in [121] for polymer-composites. The point-plane electrode

system, where it is assumed the tip of the point electrode provides a sufficiently large local

electric field to initiate breakdown [125]. The essence of the stochastic model is the

74
discharge pattern growth depending on the local electric field distribution. The field

distribution is acquired by solving 2D Laplace’s equation:

∇2 𝜙𝑒 (𝑥, 𝑦) = 0. (4.1)

For computational simplicity, we restrict our study to two-dimensional Cartesian

coordinates, where 0 < 𝑥 ≤ W, 0 < 𝑦 ≤ T, T is the thickness, and W is the width of the

dielectric film. The boundary conditions at the top and bottom metal contacts are set to

𝜙𝑒 (𝑥, 0) = 0, and 𝜙𝑒 (𝑥, T) = 𝑉o . At the open sides of the dielectric film, the following

Neumann boundary conditions (assuming no field fringing) are considered:

𝜕𝜙𝑒 (𝑥, 𝑦) 𝜕𝜙𝑒 (𝑥, 𝑦)

| = 0, | = 0.
𝜕𝑥 𝑥=0
𝜕𝑥 𝑥=W

We assume that the dielectric film has a thickness of 400 𝜇m and employ a uniform

grid of size 10040 (in Figure 4.1, W=100, T=40). Thus, the separation between nodes is

10 𝜇 m, representing the branch extension length of the electrical tree growth in the

simulation. Assume that the tree tubule diameter is much smaller than the 10 𝜇m length

scale. The point electrode placed at the node (W/2, 0) has potential 𝜙𝑒 = 0, and the plane

electrode placed at 𝑦 = T has potential 𝜙𝑒 = 1. In addition, we assume that the dielectric

breakdown initiates at the center of the bottom panel. The tree itself is considered

conductive, taking the potential of the electrode inside the dielectric. In the simulation, we

have treated the electrical tree as the distributed boundary with the same potential as the

bottom electrode.

75
 Figure 4.1 Numerical treatment and random process of electrical tree formation in pure polymer.

The time-dependent electrical treeing process is modeled as follows: (i) In each

time step, a random reference probability is generated, assuming uniform probability

distribution. (ii) The Laplace equation is solved to determine the node potentials. (iii) All

possible growth sites connected to the existing electrical tree are identified. (iv) For each

of these growth sites, the probability that the bonding (𝑖, 𝑗 → 𝑖 ′ , 𝑗′) will break in this step

is determined by the following probability distribution function:

𝜙𝑖 ′ ,𝑗 ′
𝑝(𝑖, 𝑗 → 𝑖 ′ , 𝑗 ′ ) = . (4.2)
∑ 𝜙𝑖 ′,𝑗 ′

The sum in the denominator refers to all the possible tree growth processes (grey arrows in

Figure 4.1), and (𝑖′, 𝑗′) is the terminating node of the tree branch. (v) The probabilities in

76
(4.2) are compared to the reference probability from step (i), and the site for which the

probability exceeds the reference by the smallest amount is broken. Once the tree branch

is formed, the electric potential 𝜙𝑒 propagates to the associated grid point. The clock is

then incremented, and the process repeats.

The accumulation of broken sites results in electrical treeing. When a tree branch

reaches the opposite end of the film, we assume it will form a short circuit between the

electrodes, triggering a complete breakdown. The breakdown time is calculated by

assuming a uniform electrical tree growth rate and counting the number of broken sites at

the short circuit [126], [127]. According to our model, any node (grid point) connecting to

the electrical tree may develop a tree branch at any time during the process. The selection

of the next broken site depends on two factors, the local electric field and the Monte Carlo

probability. In Figure 4.1, red arrows represent tree branches, and a grey arrow means the

possibility of the next broken site. The next site to be broken among the grey candidates

is determined by the procedure in steps (i)–(v).

Based on the pure polymer breakdown model (similar to the DBM), we further

consider incorporating insulating fillers on the pure polymer matrix. For this consideration,

we first assume the random filler position distribution on the polymer matrix. The electric

field in the polymer-filler composite is calculated by solving Laplace’s equation (4.1),

ignoring the mismatch in permittivity values between the host polymer and fillers since

many insulating fillers (e.g., silica and mica) have similar permittivity values as the host

polymer [128]. The breakdown path formation when encountering the filler is determined

by the probability, 𝑝𝐹 ,

77
 ∆𝐸
−𝑘 𝑇
𝑝𝐹 ∝ 𝑒 𝐵 , (4.3)

where the material dependent ΔE is the additional energy required to break the bonding on

the ceramic filler compared to the pure polymer matrix. We assume that ΔE is sufficiently

large and 𝑝𝐹 ≈ 0. In other words, the bonding within the filler is less likely to break, and

hence the breakdown path (red arrow) is prohibited from passing through the filler site.

Instead, the red arrow is set to bypass or propagate along the edge of the insulating fillers,

which agrees with the literature results [129]. Figure 4.2 presents the concept of

incorporating insulating fillers in our model. The time to breakdown of the dielectric is

now determined by the number of tree-growth attempts, which comprise time steps for

successful breaking through the polymer and unsuccessful breaking attempts on the fillers

to form a shorted path between the electrodes.

Figure 4.2 Numerical treatment and random process of electrical treeing on composite materials. Bond
breakings are set not to occur on incorporated insulating fillers.

78
4.3 Electrical Treeing in Pure Polymer and Composite Films

In Figure 4.3, we demonstrate the evolution of the electrical treeing process in the

pure polymer and polymer-filler composite matrix by our numerical simulation. The

termination condition for the tree growth simulation is the point when the treeing reaches

the opposite boundary. From the figure, it is evident that for a given thickness of both pure

polymer and polymer-filler composites, more branches are produced in the composite film

at the point of breakdown. In other words, the insulating fillers hinder the electrical tree

growth pattern and spread the random tree branches. In this way, insulating fillers slow

down the breakdown process. However, the exact time to break down has a statistical

distribution as the filler position is considered entirely random in the film. In the next

section, the statistical nature of the time to breakdown as a function of different filler

concentrations in the polymer film will be described in detail.

Figure 4.3 Evolution of the electrical treeing in a pure polymer dielectrics and a polymer-filler
composite. The electric breakdown occurs when the treeing propagates and reaches the counter
electrode.

79
4.4 Impact of Filler Concentrations on the Breakdown Time

The concentration of the insulating filler in the polymer matrix is an important

design parameter that significantly impacts the breakdown time [130], [131]. In this

section, the effect of the filler concentration on the electrical treeing process is explored.

For clear illustration, the pure polymer (0 vol%) with 10, 20, 30, and 40 vol% of randomly

distributed insulating fillers are compared. For simplicity, we consider uniform square-

shaped filler particles of size 44 for our simulation. Figure 4.4(a) illustrates that higher

filler concentrations tend to diversify the treeing in a broader region. In addition, the fractal

analysis is performed using the box-counting method [132] on the simulated structure with

different filler concentrations. The calculation shows that the fractal dimension falls in the

range of 1.62 to 1.70. The fractal dimension gradually reduces as the filler volume fraction

increases.

Figure 4.4 (a) Electrical treeing due to different concentrations of insulating fillers in percentage by
volume. (b) CDF plots for breakdown times, comparing 0, 10, 20, 30, and 40 % fillers by volume.

80
 In fact, there will be more failed attempts for bond breaking in the presence of high

filler concentrations, which will slow down the breakdown process. In Figure 4.4(b), the

cumulative distribution function (CDF) of the time to breakdown is compared, which

indicates that the higher the insulating filler content, the longer the median breakdown time.

In addition, we find that the variability associated with breakdown time also increases with

the filler loading. This increased variability with the filler loading was experimentally

demonstrated by Imperiale et al. [133]. Our simulation framework thus provides a

theoretical justification for the observed variability in the breakdown strength based on

additional randomness introduced by the filler position in the electrical treeing process.

However, the benefit from high filler concentration saturates after a certain

threshold concentration, as experimentally reported [117], [118], [133]. Our simulation

framework does not reflect the effect of very high filler loading as the framework is

developed based on electrical treeing in the pure polymer film. Further improvement of

the framework considering the physicochemical properties of polymer–filler interfaces is

required to capture higher filler loading effects, and it remains a topic of our future research.

4.5 Impact of the Filler Geometry on the Breakdown Time

It has been experimentally reported that the geometric structures, shape, and size of

the filler also affect the breakdown process. Figure 4.5(a) illustrates the impact of filler

shape on the tree growth by choosing two representative shapes; one is 22 square-shaped,

and the other is with a long flat rectangular-shaped (14) fillers. Both concentrations are

fixed at 20 vol%. We find that the rectangular fillers spread the tree growth more

effectively than the square ones for the same volume fraction. Hence, composites

incorporating long flat rectangular shaped (14) fillers require a longer time to break down

81
compared to more symmetric square-shaped fillers, as shown in Figure 4.5(b). The

possible reason is that long flat-shaped fillers tend to impede the bond breaking vertically

and guide it horizontally to adjacent sites, which widens the tree branches and extends the

breakdown time. This result concurs with the experimental work [134]. With the

improvement of the fabrication process, the flat-shaped fillers can be designed to improve

dielectric breakdown strength, which will also increase the energy storage density of

polymer-filler composites [135], [136].

Figure 4.5 (a) Comparison of tree growth with the filler shapes of 22 (top) and 14 (bottom) on 20
vol% of concentration. (b) CDF plots for breakdown times and comparison of different shapes at 20%
fillers by volume.

Next, we investigate the effect of filler size on the electrical tree growth, keeping

their shapes the same. In order to compare, we retain the same filler concentration at 20

vol% and set the square shape of the fillers to sizes in 22 and 44. The electrical treeing

phenomena on the polymer-filler composite with both small and large insulating fillers are

compared in Figure 4.6(a). For the statistical comparison of the time to breakdown with

82
different filler sizes, we plot the CDF in Figure 4.6(b), which shows that the time to

breakdown increases with the smaller size of fillers. Furthermore, comparing the filler

sizes by experiments under DC [137] and AC [138] voltages illustrate that smaller fillers

reinforce the dielectric strength. Thus, our numerical framework explains the effect of

filler size on the dielectric breakdown based on the retarded tree-growth process in the

presence of small-sized fillers—that is, at the same concentration, smaller fillers offer a

larger specific surface area to barricade the bond-breaking.

Overall, the results of this simulation study demonstrate that the size and the shape

of insulating fillers strongly affect the electric breakdown of dielectrics. We have

considered homogeneously random dispersion of fillers in all the previous analyses. Here,

we consider the effects of the position of the fillers in the film. We first selectively place

insulating fillers at the top and bottom quarters of the film to perform the breakdown

simulation. Then, we consider placing the fillers only in the middle half of the film. The

Figure 4.6 (a) Comparison of tree growth with the square fillers between sizes of 22 (top) and 44
(bottom) at 20 % by volume. (b) CDF plots for breakdown times in comparing different square sizes of
fillers at 20 % by volume.

83
arrangements are displayed in Figure 4.7(a). The statistically analyzed CDF of time to

breakdown in Figure 4.7(b) reveals that dispersing the insulating fillers in the middle of

the dielectric causes more branches and a broader electrical tree structure for the same filler

concentration. Hence, the selective localization of the fillers in the film will provide better

reliability of the dielectrics. Our results thus provide a theoretical justification of the

experimentally demonstrated feature of highly enhanced dielectric strength on the

sandwiched three-layer film [139], which is composed of nanofibers in the central layer

and nanoparticles in the two outer layers of the polymer matrices.

4.6 Summary

In this chapter, we have developed a comprehensive numerical framework to

understand the role of insulating fillers on the electrical treeing process in the polymer

Figure 4.7 (a) Electrical treeing branches in 10 vol% 14 asymmetric filler composites by homogeneous
dispersion to the bulk (top), local dispersion to outer layers (middle), and local dispersion to the central
layer (bottom). (b) CDF comparison of time to breakdown on the 10 vol% homogeneously dispersed
asymmetric filler composite, and the two locally dispersed asymmetric filler composites.

84
dielectrics. First, we explained the experimentally observed variability in the breakdown

strength of the polymer-filler composites in terms of randomness in the electrical treeing

process. Then, using our simulation framework, we explored the optimum filler design for

improving breakdown strength. Our results indicate that adequately arranging the filler

shape and position will significantly extend the bond-breaking process to elongate the life

of dielectrics.

85
 Chapter 5. Conclusions and Future Work

This dissertation research has employed computational methods to examine

reliability concerns in semiconductor memory devices. First, it explored the mechanism

of 3D NAND flash memory cell current degradation when the third dimension is stacked

to increase storage capacity. The degradation originated in the reduced electron

concentration in the poly-Si channel and undoped source/drain in the interlayer region,

which is not presented in the 2D planar device. The research proposed three new designs

to overcome the degradation problem by either strengthening the fringing electric field

between CG coverage areas or supplying more electrons in the channel. The designs are

(1) electrode shape modulation, (2) positive charge embedment in the interlayer oxide

region, and (3) electron doping in the interlayer channel region. The first design involves

modifying the shape of CG electrodes to confine more electrons in the channel and

strengthen the fringing field significantly in the interlayer region. Still, the CG voltages

remain unchanged from the original design. The second design involves embedding

positive charges to attract more inversion electrons in the interlayer channel region. Finally,

the third design introduces more electrons in the channel. These ideas to enhance the cell

current in the 3D NAND flash memory were numerically verified using Taurus Medici, a

commercial workstation software exploiting the finite element method to solve the

boundary value problem (BVP) of the 2D drift-diffusion and Poisson’s equations.

86
 Second, the MRAM study incorporated the external magnetic field, the STT, and

thermal fluctuation torque terms in the LLG equation for diverse situations. Originally the

equation was in vector form, and computation required a reduction of the vector differential

equation to a system of scalar ordinary differential equations. The resulting initial value

problem (IVP) of the system of scalar LLG equations was solved in 3D spherical polar

coordinates. In addition, the control of MRAM by STT in the presence of background

thermal fluctuation was computationally verified.

Today, the state-of-the-art STT-MRAM is in the product stream; hence, it is vital

to understand the impact of the phenomena on the device's reliability. The simulation

describes the switching dynamics directly in terms of the magnetization rather than by

evaluating the magnetic behavior indirectly using voltage variation in the experiment. On

the basis of the simulation results, it is concluded that device reliability can be maintained

by controlling the spin current.

Ultimately, the dielectric layer is the essential component for both the 3D NAND

flash memory cell (where the dielectric separates the control gate and the channel) and the

MRAM cell (where the dielectric separates the free layer from the fixed layer of MTJ).

The functionality of the dielectric layer is closely related to the reliability and the

prevention of dielectric breakdown. In the case of dielectric breakdown, the governing

equation for the electrical treeing behavior in the dielectric layer is described by Laplace’s

equation. As the dielectric breakdown exhibits randomness, the simulation incorporates

stochasticity into the electrical treeing path and the insulating filler distribution. With the

Neumann boundary conditions, 2D Laplace’s equation was solved using the finite

difference method. Correspondingly, the electric potential obtained at every grid point of

87
the thin layer indicates possible paths leading to a breakdown. The simulation study,

supplemented by statistical analysis, enables us to evaluate the influences on the device

and propose inserting insulating fillers for the composite film to consolidate the dielectric

strength against electrical breakdown.

As for future work, the techniques presented in Chapter 2 for the 3D NAND flash

memory may be applied to the analyses of the lateral charge migration and the cross-

temperature effect on the program and read operations. These two effects are closely

related to the NAND flash memory reliability. Most 3D NAND flash memories use

charge-trapped nitride layers to capture electrons (for storing information). Due to any

uneven population of the electrons, the electrons will diffuse out and migrate into the

interlayer nitride regions or even neighboring CG coverage areas. This phenomenon

affects reading and causes serious reliability problems in the long run. In addition, using

flash memory in the presence of a temperature gradient may cause reading errors. For

instance, use the device when writing information in a low-temperature area and reading

the information in a high-temperature area or vice versa. These reliability issues may be

analyzed through modeling and simulation.

Regarding MRAM in Chapter 3, a comprehensive simulation study was carried out

to model the magnetization dynamics of the free layer in the p-MTJ. Since PMA and IMA

are the two types of magnetic anisotropy, the method of this research presented can also be

applied to the i-MTJ, which has the in-plane type of magnetic anisotropy. Figure 5.1

demonstrates the magnetization orientations in both p-MTJ and i-MTJ. Figure 5.2 shows

the simulated magnetization trajectories on p-MTJ, which has an easy axis along the z

direction, and i-MTJ, which has an easy axis along the y direction.

88
 (a) (b)

Figure 5.1 (a) Perpendicular MTJ (p-MTJ) with perpendicular magnetic anisotropy. (b) In-plane MTJ
(i-MTJ) with in-plane magnetic anisotropy.

Figure 5.2 Simulated magnetization trajectory from the initial 𝐦𝑖 (blue) to the final 𝐦𝑓 (green), which
aligned to the easy-axis in (a) perpendicular magnetic anisotropy; (b) in-plane magnetic anisotropy.

In the future, we may want to explore the difference between the STT and thermal

fluctuation effects to discover new and subtle methods for electronic device design.

Furthermore, the simulation techniques can be modified to include the spin-orbit torque,

which is exploited in the SOT-MRAM, as this device is likely to constitute another state-

of-the-art type of MRAM.

89
 References

[1] V. Janev, “Semantic Intelligence in Big Data Applications,” in Smart Connected

World: Technologies and Applications Shaping the Future, S. Jain and S.
Murugesan, Eds. Cham: Springer International Publishing, 2021, pp. 71–89. doi:
10.1007/978-3-030-76387-9_4.

[2] David Reinsel, John Gantz, and John Rydning, Data Age 2025: The Digitization of
the World–From Edge to Core. Framingham, MA, USA: International Data
Corporation, 2018.

[3] D. Kumar, R. Aluguri, U. Chand, and T. Y. Tseng, “Metal oxide resistive switching
memory: Materials, properties and switching mechanisms,” Ceramics International,
vol. 43, pp. S547–S556, Aug. 2017, doi: 10.1016/j.ceramint.2017.05.289.

[4] W. Banerjee, “Challenges and Applications of Emerging Nonvolatile Memory

Devices,” Electronics, vol. 9, no. 6, Art. no. 6, Jun. 2020, doi:
10.3390/electronics9061029.

[5] V. Milo, G. Malavena, C. Monzio Compagnoni, and D. Ielmini, “Memristive and

CMOS Devices for Neuromorphic Computing,” Materials, vol. 13, no. 1, Art. no. 1,
Jan. 2020, doi: 10.3390/ma13010166.

[6] W. Zhao et al., “High Performance SoC Design Using Magnetic Logic and Memory,”
in VLSI-SoC: Advanced Research for Systems on Chip, Berlin, Heidelberg, 2012, pp.
10–33. doi: 10.1007/978-3-642-32770-4_2.

[7] S. Kargar and F. Nawab, “Challenges and future directions for energy, latency, and
lifetime improvements in NVMs,” Distrib Parallel Databases, Sep. 2022, doi:
10.1007/s10619-022-07421-x.

[8] K. Ishimaru, “Future of Non-Volatile Memory -From Storage to Computing-,” in

2019 IEEE International Electron Devices Meeting (IEDM), Dec. 2019, p. 1.3.1-
1.3.6. doi: 10.1109/IEDM19573.2019.8993609.

[9] D. Kahng and S. M. Sze, “A floating gate and its application to memory devices,”
The Bell System Technical Journal, vol. 46, no. 6, pp. 1288–1295, Jul. 1967, doi:
10.1002/j.1538-7305.1967.tb01738.x.

90
[10] S. M. Sze, “The floating-gate non-volatile semiconductor memory--from invention
to the digital age,” J Nanosci Nanotechnol, vol. 12, no. 10, pp. 7587–7596, Oct.
2012, doi: 10.1166/jnn.2012.6648.

[11] T. Coughlin, “A Timeline for Flash Memory History [The Art of Storage],” IEEE
Consumer Electronics Magazine, vol. 6, no. 1, pp. 126–133, Jan. 2017, doi:
10.1109/MCE.2016.2614739.

[12] K. Parat and A. Goda, “Scaling Trends in NAND Flash,” in 2018 IEEE International
Electron Devices Meeting (IEDM), Dec. 2018, p. 2.1.1-2.1.4. doi:
10.1109/IEDM.2018.8614694.

[13] D. James, “3D ICs in the real world,” in 25th Annual SEMI Advanced Semiconductor
Manufacturing Conference (ASMC 2014), May 2014, pp. 113–119. doi:
10.1109/ASMC.2014.6846988.

[14] K.-T. Park, D.-S. Byeon, and D.-H. Kim, “A world’s first product of three-
dimensional vertical NAND Flash memory and beyond,” in 2014 14th Annual Non-
Volatile Memory Technology Symposium (NVMTS), Oct. 2014, pp. 1–5. doi:
10.1109/NVMTS.2014.7060840.

[15] R. Micheloni, S. Aritome, and L. Crippa, “Array Architectures for 3-D NAND Flash
Memories,” Proceedings of the IEEE, vol. 105, no. 9, pp. 1634–1649, Sep. 2017,
doi: 10.1109/JPROC.2017.2697000.

[16] C. Monzio Compagnoni, A. Goda, A. S. Spinelli, P. Feeley, A. L. Lacaita, and A.

Visconti, “Reviewing the Evolution of the NAND Flash Technology,” Proceedings
of the IEEE, vol. 105, no. 9, pp. 1609–1633, Sep. 2017, doi: 10.1109/
JPROC.2017.2665781.

[17] “Micron Is First to Deliver 3D Flash Chips With More Than 200 Layers - IEEE
Spectrum.” https://spectrum.ieee.org/micron-is-first-to-deliver-3d-flash-chips-with-
more-than-200-layers (accessed Mar. 25, 2023).

[18] M. Raquibuzzaman, A. Milenkovic, and B. Ray, “Intrablock Wear Leveling to

Counter Layer-to-Layer Endurance Variation of 3-D NAND Flash Memory,” IEEE
Transactions on Electron Devices, vol. 70, no. 1, pp. 70–75, Jan. 2023, doi:
10.1109/TED.2022.3224420.

91
[19] H. Kim, S.-J. Ahn, Y. G. Shin, K. Lee, and E. Jung, “Evolution of NAND Flash
Memory: From 2D to 3D as a Storage Market Leader,” in 2017 IEEE International
Memory Workshop (IMW), May 2017, pp. 1–4. doi: 10.1109/IMW.2017.7939081.

[20] H. J. M. Swagten and P. V. Paluskar, “Magnetic Tunnel Junctions,” in Encyclopedia

of Materials: Science and Technology, K. H. J. Buschow, R. W. Cahn, M. C.
Flemings, B. Ilschner, E. J. Kramer, S. Mahajan, and P. Veyssière, Eds. Oxford:
Elsevier, 2010, pp. 1–7. doi: 10.1016/B978-008043152-9.02206-5.

[21] B. Lacoste et al., “Control of Sub-Nanosecond Precessional Magnetic Switching in

STT-MRAM Cells for SRAM Applications,” in 2016 IEEE 8th International
Memory Workshop (IMW), May 2016, pp. 1–4. doi: 10.1109/IMW.2016.7495262.

[22] G. Prenat et al., “Magnetic memories: From DRAM replacement to ultra low power
logic chips,” in 2014 Design, Automation & Test in Europe Conference & Exhibition
(DATE), Mar. 2014, pp. 1–1. doi: 10.7873/DATE.2014.281.

[23] H. Lim, S. Lee, and H. Shin, “A Survey on the Modeling of Magnetic Tunnel
Junctions for Circuit Simulation,” Active and Passive Electronic Components, vol.
2016, p. e3858621, May 2016, doi: 10.1155/2016/3858621.

[24] S. Ikegawa, F. B. Mancoff, and S. Aggarwal, “Commercialization of MRAM –

Historical and Future Perspective,” in 2021 IEEE International Interconnect
Technology Conference (IITC), Jul. 2021, pp. 1–3. doi: 10.1109/IITC51362.
2021.9537434.

[25] T. Coughlin, “MRAM In Perspective,” Forbes.

https://www.forbes.com/sites/tomcoughlin/2019/12/16/mram-in-perspective/
(accessed Nov. 24, 2022).

[26] B. Dieny and M. Chshiev, “Perpendicular magnetic anisotropy at transition

metal/oxide interfaces and applications,” Rev. Mod. Phys., vol. 89, no. 2, p. 025008,
Jun. 2017, doi: 10.1103/RevModPhys.89.025008.

[27] S. Huang, T. B. Boykin, R. S. Gorur, and B. Ray, “Electrical tree formation in

polymer-filler composites,” IEEE Transactions on Dielectrics and Electrical
Insulation, vol. 26, no. 6, pp. 1853–1858, Dec. 2019, doi: 10.1109/TDEI.2019.
008186.

92
[28] A. Silvagni, “3D NAND Flash Based on Planar Cells,” Computers, vol. 6, no. 4, Art.
no. 4, Dec. 2017, doi: 10.3390/computers6040028.

[29] A. Goda, “3-D NAND Technology Achievements and Future Scaling Perspectives,”
IEEE Transactions on Electron Devices, vol. 67, no. 4, pp. 1373–1381, Apr. 2020,
doi: 10.1109/TED.2020.2968079.

[30] A. Goda, “Recent Progress on 3D NAND Flash Technologies,” Electronics, vol. 10,
no. 24, Art. no. 24, Jan. 2021, doi: 10.3390/electronics10243156.

[31] J.-W. Park et al., “30.1 A 176-Stacked 512Gb 3b/Cell 3D-NAND Flash with
10.8Gb/mm2 Density with a Peripheral Circuit Under Cell Array Architecture,” in
2021 IEEE International Solid-State Circuits Conference (ISSCC), Feb. 2021, vol.
64, pp. 422–423. doi: 10.1109/ISSCC42613.2021.9365809.

[32] H. Kim, S.-J. Ahn, Y. G. Shin, K. Lee, and E. Jung, “Evolution of NAND Flash
Memory: From 2D to 3D as a Storage Market Leader,” in 2017 IEEE International
Memory Workshop (IMW), May 2017, pp. 1–4. doi: 10.1109/IMW.2017.7939081.

[33] Aritome, Seiichi, Nand Flash Memory Technologies. John Wiley & Sons, Ltd, 2015.
doi: 10.1002/9781119132639.

[34] A. S. Spinelli, C. M. Compagnoni, and A. L. Lacaita, “Reliability of NAND Flash

Memories: Planar Cells and Emerging Issues in 3D Devices,” Computers, vol. 6, no.
2, p. 16, Apr. 2017, doi: 10.3390/computers6020016.

[35] Y.-H. Hsiao, H.-T. Lue, T.-H. Hsu, K.-Y. Hsieh, and C.-Y. Lu, “A critical
examination of 3D stackable NAND Flash memory architectures by simulation
study of the scaling capability,” in 2010 IEEE International Memory Workshop, May
2010, pp. 1–4. doi: 10.1109/IMW.2010.5488390.

[36] Y.-T. Oh et al., “Impact of etch angles on cell characteristics in 3D NAND flash
memory,” Microelectronics Journal, vol. 79, pp. 1–6, Sep. 2018, doi:
10.1016/j.mejo.2018.06.009.

[37] E.-S. Choi and S.-K. Park, “Device considerations for high density and highly
reliable 3D NAND flash cell in near future,” in 2012 International Electron Devices
Meeting, Dec. 2012, p. 9.4.1-9.4.4. doi: 10.1109/IEDM.2012.6479011.

93
[38] S. Venkatesan and M. Aoulaiche, “Overview of 3D NAND Technologies and
Outlook Invited Paper,” in 2018 Non-Volatile Memory Technology Symposium
(NVMTS), Oct. 2018, pp. 1–5. doi: 10.1109/NVMTS.2018.8603104.

[39] R. Micheloni, L. Crippa, and A. Marelli, Inside NAND Flash Memories. Dordrecht:
Springer Netherlands, 2010. doi: 10.1007/978-90-481-9431-5.

[40] E. Capogreco et al., “Feasibility of InxGa1–xAs High Mobility Channel for 3-D
NAND Memory,” IEEE Transactions on Electron Devices, vol. 64, no. 1, pp. 130–
136, Jan. 2017, doi: 10.1109/TED.2016.2633388.

[41] Z. Hou, J. Yao, Z. Wu, and H. Yin, “Investigation for the Feasibility of High-
Mobility Channel in 3D NAND Memory,” ECS J. Solid State Sci. Technol., vol. 7,
no. 5, p. Q75, May 2018, doi: 10.1149/2.0011805jss.

[42] U. M. Bhatt, A. Kumar, and S. K. Manhas, “Performance Enhancement by

Optimization of Poly Grain Size and Channel Thickness in a Vertical Channel 3-D
NAND Flash Memory,” IEEE Transactions on Electron Devices, vol. 65, no. 5, pp.
1781–1786, May 2018, doi: 10.1109/TED.2018.2817920.

[43] A. Arreghini et al., “Improvement of conduction in 3-D NAND memory devices by

channel and junction optimization,” in 2019 IEEE 11th International Memory
Workshop (IMW), May 2019, pp. 1–4. doi: 10.1109/IMW.2019.8739661.

[44] S. Ramesh et al., “Understanding the kinetics of Metal Induced Lateral

Crystallization process to enhance the poly-Si channel quality and current
conduction in 3-D NAND memory,” in 2021 IEEE International Electron Devices
Meeting (IEDM), Dec. 2021, p. 10.2.1-10.2.4. doi: 10.1109/IEDM19574.2021.
9720571.

[45] J. A. Kittl et al., “High-k dielectrics for future generation memory devices (Invited
Paper),” Microelectronic Engineering, vol. 86, no. 7, pp. 1789–1795, Jul. 2009, doi:
10.1016/j.mee.2009.03.045.

[46] W. Lin et al., “Grain Boundary Trap-Induced Current Transient in a 3-D NAND
Flash Cell String,” IEEE Transactions on Electron Devices, vol. 66, no. 4, pp. 1734–
1740, Apr. 2019, doi: 10.1109/TED.2019.2900736.

94
[47] B. Padmanaban, R. Ramesh, D. Nirmal, and S. Sathiyamoorthy, “Numerical
modeling of triple material gate stack gate all-around (TMGSGAA) MOSFET
considering quantum mechanical effects,” Superlattices and Microstructures, vol.
82, pp. 40–54, Jun. 2015, doi: 10.1016/j.spmi.2015.01.021.

[48] M. Karbalaei, D. Dideban, and H. Heidari, “Impact of high-k gate dielectric with
different angles of coverage on the electrical characteristics of gate-all-around field
effect transistor: A simulation study,” Results in Physics, vol. 16, p. 102823, Mar.
2020, doi: 10.1016/j.rinp.2019.102823.

[49] Q. Qin, F. Wang, X. Zhan, Y. Li, and J. Chen, “TID Radiation Impacts on Charge-
trappingping Macaroni 3D NAND Flash Memory,” in 2020 IEEE International
Symposium on the Physical and Failure Analysis of Integrated Circuits (IPFA), Jul.
2020, pp. 1–4. doi: 10.1109/IPFA49335.2020.9260906.

[50] J.-M. Sim, M. Kang, and Y.-H. Song, “A New Read Scheme for Alleviating Cell-
to-Cell Interference in Scaled-Down 3D NAND Flash Memory,” Electronics, vol.
9, no. 11, Art. no. 11, Nov. 2020, doi: 10.3390/electronics9111775.

[51] J. Lee, J. Jang, J. Lim, Y. G. Shin, K. Lee, and E. Jung, “A new ruler on the storage
market: 3D-NAND flash for high-density memory and its technology evolutions
and challenges on the future,” in 2016 IEEE International Electron Devices
Meeting (IEDM), Dec. 2016, p. 11.2.1-11.2.4. doi: 10.1109/IEDM.2016.7838394.

[52] Taurus Medici User Guide, Q-2019.12. Mountain View, CA, USA: Synopsys, Inc.,
2019.

[53] H.-J. Cho and J. D. Plummer, “Modeling of Surrounding Gate MOSFETs with Bulk
Trap States,” IEEE Transactions on Electron Devices, vol. 54, no. 1, pp. 166–169,
Jan. 2007, doi: 10.1109/TED.2006.887521.

[54] S. Sahay and D. Strukov, “A Behavioral Compact Model for Static Characteristics
of 3D NAND Flash Memory,” IEEE Electron Device Letters, vol. 40, no. 4, pp. 558–
561, Apr. 2019, doi: 10.1109/LED.2019.2901211.

[55] H. Tanaka et al., “Bit Cost Scalable Technology with Punch and Plug Process for
Ultra High Density Flash Memory,” in 2007 IEEE Symposium on VLSI Technology,
Jun. 2007, pp. 14–15. doi: 10.1109/VLSIT.2007.4339708.

95
[56] J. Jang et al., “Vertical cell array using TCAT(Terabit Cell Array Transistor)
technology for ultra high density NAND flash memory,” in 2009 Symposium on
VLSI Technology, Jun. 2009, pp. 192–193.

[57] J.-H. Lee and S.-G. Jung, “NAND flash memory technology utilizing fringing
electric field,” Microelectronics Reliability, vol. 52, no. 4, pp. 662–669, Apr. 2012,
doi: 10.1016/j.microrel.2011.09.029.

[58] M.-S. Kim, S.-J. Choi, D.-I. Moon, J. P. Duarte, S. Kim, and Y.-K. Choi,
“Investigation of gate length and fringing field effects for program and erase
efficiency in gate-all-around SONOS memory cells,” Solid-State Electronics, vol.
79, pp. 7–10, Jan. 2013, doi: 10.1016/j.sse.2012.03.008.

[59] Y. Yanagihara, K. Miyaji, and K. Takeuchi, “Control Gate Length, Spacing and
Stacked Layer Number Design for 3D-Stackable NAND Flash Memory,” in 2012
4th IEEE International Memory Workshop, May 2012, pp. 1–4. doi:
10.1109/IMW.2012.6213656.

[60] Y. Hsiao et al., “Impact of Vpass Interference on Charge-trapping NAND Flash

Memory Devices,” IEEE Transactions on Device and Materials Reliability, vol. 15,
no. 2, pp. 136–141, Jun. 2015, doi: 10.1109/TDMR.2015.2398193.

[61] Y. G. Lee, H. S. Jung, and T. W. Kim, “Enhancement of the Electrical Characteristics

for 3D NAND Flash Memory Devices Due to a Modified Cell Structure in the Gate
Region,” J Nanosci Nanotechnol, vol. 19, no. 10, pp. 6148–6151, Oct. 2019, doi:
10.1166/jnn.2019.17017.

[62] J. Levinson, F. R. Shepherd, P. J. Scanlon, W. D. Westwood, G. Este, and M. Rider,

“Conductivity behavior in polycrystalline semiconductor thin film transistors,”
Journal of Applied Physics, vol. 53, no. 2, pp. 1193–1202, Feb. 1982, doi:
10.1063/1.330583.

[63] R. Degraeve et al., “Statistical poly-Si grain boundary model with discrete charging
defects and its 2D and 3D implementation for vertical 3D NAND channels,” in 2015
IEEE International Electron Devices Meeting (IEDM), Dec. 2015, p. 5.6.1-5.6.4.
doi: 10.1109/IEDM.2015.7409636.

96
[64] H. Jo, J. Kim, M. Kim, and H. Shin, “Extraction of Mobility in 3-D NAND Flash
Memory with Poly-Si Based Macaroni Structure,” in 2020 4th IEEE Electron
Devices Technology Manufacturing Conference (EDTM), Apr. 2020, pp. 1–4. doi:
10.1109/EDTM47692.2020.9117961.

[65] V. Ishchuk, B. E. Volland, M. Hauguth, M. Cooke, and I. W. Rangelow, “Charging

effect simulation model used in simulations of plasma etching of silicon,” Journal
of Applied Physics, vol. 112, no. 8, p. 084308, Oct. 2012, doi: 10.1063/1.4759005.

[66] D. M. Schaadt, E. T. Yu, S. Sankar, and A. E. Berkowitz, “Charge storage in Co

nanoclusters embedded in SiO2 by scanning force microscopy,” Appl. Phys. Lett.,
vol. 74, no. 3, pp. 472–474, Jan. 1999, doi: 10.1063/1.123039.

[67] W. Thomson, “XIX. On the electro-dynamic qualities of metals:—Effects of

magnetization on the electric conductivity of nickel and of iron,” Proceedings of the
Royal Society of London, vol. 8, pp. 546–550, Jan. 1997, doi:
10.1098/rspl.1856.0144.

[68] M. N. Baibich et al., “Giant Magnetoresistance of (001)Fe/(001)Cr Magnetic

Superlattices,” Phys. Rev. Lett., vol. 61, no. 21, pp. 2472–2475, Nov. 1988, doi:
10.1103/PhysRevLett.61.2472.

[69] G. Binasch, P. Grünberg, F. Saurenbach, and W. Zinn, “Enhanced magnetoresistance

in layered magnetic structures with antiferromagnetic interlayer exchange,” Phys.
Rev. B, vol. 39, no. 7, pp. 4828–4830, Mar. 1989, doi: 10.1103/PhysRevB.39.4828.

[70] S. S. P. Parkin, Z. G. Li, and D. J. Smith, “Giant magnetoresistance in

antiferromagnetic Co/Cu multilayers,” Appl. Phys. Lett., vol. 58, no. 23, pp. 2710–
2712, Jun. 1991, doi: 10.1063/1.104765.

[71] S. M. Thompson, “The discovery, development and future of GMR: The Nobel Prize
2007,” J. Phys. D: Appl. Phys., vol. 41, no. 9, p. 093001, Mar. 2008, doi:
10.1088/0022-3727/41/9/093001.

[72] S. A. Wolf et al., “Spintronics: A Spin-Based Electronics Vision for the Future,”
Science, vol. 294, no. 5546, pp. 1488–1495, Nov. 2001, doi:
10.1126/science.1065389.

97
[73] B. Dieny, V. S. Speriosu, S. S. P. Parkin, B. A. Gurney, D. R. Wilhoit, and D. Mauri,
“Giant magnetoresistive in soft ferromagnetic multilayers,” Phys. Rev. B, vol. 43,
no. 1, pp. 1297–1300, Jan. 1991, doi: 10.1103/PhysRevB.43.1297.

[74] A. Hirohata et al., “Review on spintronics: Principles and device applications,”

Journal of Magnetism and Magnetic Materials, vol. 509, p. 166711, Sep. 2020, doi:
10.1016/j.jmmm.2020.166711.

[75] C. Reig, M.-D. Cubells-Beltran, and D. R. Muñoz, “Magnetic Field Sensors Based
on Giant Magnetoresistance (GMR) Technology: Applications in Electrical Current
Sensing,” Sensors (Basel), vol. 9, no. 10, pp. 7919–7942, Oct. 2009, doi:
10.3390/s91007919.

[76] B. A. Gurney et al., “Spin Valve Giant Magnetoresistive Sensor Materials for Hard
Disk Drives,” in Ultrathin Magnetic Structures IV: Applications of Nanomagnetism,
B. Heinrich and J. A. C. Bland, Eds., Berlin, Heidelberg: Springer, 2005, pp. 149–
175. doi: 10.1007/3-540-27164-3_6.

[77] M. Julliere, “Tunneling between ferromagnetic films,” Physics Letters A, vol. 54, no.
3, pp. 225–226, Sep. 1975, doi: 10.1016/0375-9601(75)90174-7.

[78] T. Miyazaki and N. Tezuka, “Giant magnetic tunneling effect in Fe/Al2O3/Fe

junction,” Journal of Magnetism and Magnetic Materials, vol. 139, no. 3, pp. L231–
L234, Jan. 1995, doi: 10.1016/0304-8853(95)90001-2.

[79] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, “Large

Magnetoresistance at Room Temperature in Ferromagnetic Thin Film Tunnel
Junctions,” Phys. Rev. Lett., vol. 74, no. 16, pp. 3273–3276, Apr. 1995, doi:
10.1103/PhysRevLett.74.3273.

[80] H. X. Wei, Q. H. Qin, M. Ma, R. Sharif, and X. F. Han, “80% tunneling

magnetoresistance at room temperature for thin Al–O barrier magnetic tunnel
junction with CoFeB as free and reference layers,” Journal of Applied Physics, vol.
101, no. 9, p. 09B501, May 2007, doi: 10.1063/1.2696590.

98
[81] S. Yuasa, T. Nagahama, A. Fukushima, Y. Suzuki, and K. Ando, “Giant room-
temperature magnetoresistance in single-crystal Fe/MgO/Fe magnetic tunnel
junctions,” Nature Mater, vol. 3, no. 12, Art. no. 12, Dec. 2004, doi:
10.1038/nmat1257.

[82] D. D. Djayaprawira et al., “230% room-temperature magnetoresistance in

CoFeB∕MgO∕CoFeB magnetic tunnel junctions,” Appl. Phys. Lett., vol. 86, no. 9, p.
092502, Feb. 2005, doi: 10.1063/1.1871344.

[83] S. Ikeda et al., “Tunnel magnetoresistance of 604% at 300K by suppression of Ta

diffusion in CoFeB∕MgO∕CoFeB pseudo-spin-valves annealed at high temperature,”
Appl. Phys. Lett., vol. 93, no. 8, p. 082508, Aug. 2008, doi: 10.1063/1.2976435.

[84] W. J. Gallagher and S. S. P. Parkin, “Development of the magnetic tunnel junction

MRAM at IBM: From first junctions to a 16-Mb MRAM demonstrator chip,” IBM
Journal of Research and Development, vol. 50, no. 1, pp. 5–23, Jan. 2006, doi:
10.1147/rd.501.0005.

[85] J. C. Slonczewski, “Current-driven excitation of magnetic multilayers,” Journal of

Magnetism and Magnetic Materials, vol. 159, no. 1, pp. L1–L7, Jun. 1996, doi:
10.1016/0304-8853(96)00062-5.

[86] L. Berger, “Emission of spin waves by a magnetic multilayer traversed by a current,”

Phys. Rev. B, vol. 54, no. 13, pp. 9353–9358, Oct. 1996, doi:
10.1103/PhysRevB.54.9353.

[87] M. Tsoi et al., “Excitation of a Magnetic Multilayer by an Electric Current,” Phys.

Rev. Lett., vol. 80, no. 19, pp. 4281–4284, May 1998, doi:
10.1103/PhysRevLett.80.4281.

[88] E. B. Myers, D. C. Ralph, J. A. Katine, R. N. Louie, and R. A. Buhrman, “Current-

Induced Switching of Domains in Magnetic Multilayer Devices,” Science, vol. 285,
no. 5429, pp. 867–870, Aug. 1999, doi: 10.1126/science.285.5429.867.

[89] J. A. Katine, F. J. Albert, R. A. Buhrman, E. B. Myers, and D. C. Ralph, “Current-

Driven Magnetization Reversal and Spin-Wave Excitations in Co/Cu/Co Pillars,”
Phys. Rev. Lett., vol. 84, no. 14, pp. 3149–3152, Apr. 2000, doi:
10.1103/PhysRevLett.84.3149.

99
[90] M. Hosomi et al., “A novel nonvolatile memory with spin torque transfer
magnetization switching: spin-ram,” in IEEE International Electron Devices
Meeting, 2005. IEDM Technical Digest., Dec. 2005, pp. 459–462. doi:
10.1109/IEDM.2005.1609379.

[91] W. Kim et al., “Extended scalability of perpendicular STT-MRAM towards sub-

20nm MTJ node,” in 2011 International Electron Devices Meeting, Dec. 2011, p.
24.1.1-24.1.4. doi: 10.1109/IEDM.2011.6131602.

[92] S. Ikeda et al., “A perpendicular-anisotropy CoFeB–MgO magnetic tunnel junction,”

Nature Mater, vol. 9, no. 9, Art. no. 9, Sep. 2010, doi: 10.1038/nmat2804.

[93] D. C. Worledge et al., “Spin torque switching of perpendicular Ta∣CoFeB∣MgO-

based magnetic tunnel junctions,” Appl. Phys. Lett., vol. 98, no. 2, p. 022501, Jan.
2011, doi: 10.1063/1.3536482.

[94] J. J. Sun et al., “Commercialization of 1Gb Standalone Spin-Transfer Torque

MRAM,” in 2021 IEEE International Memory Workshop (IMW), May 2021, pp. 1–
4. doi: 10.1109/IMW51353.2021.9439616.

[95] S. Ikeda et al., “Recent progress of perpendicular anisotropy magnetic tunnel

junctions for nonvolatile VLSI,” SPIN, vol. 2, no. 3, Sep. 2012, doi:
10.1142/S2010324712400036.

[96] B. Tudu and A. Tiwari, “Recent Developments in Perpendicular Magnetic

Anisotropy Thin Films for Data Storage Applications,” Vacuum, vol. 146, pp. 329–
341, Dec. 2017, doi: 10.1016/j.vacuum.2017.01.031.

[97] J. M. Daughton, “Magnetic tunneling applied to memory (invited),” Journal of

Applied Physics, vol. 81, no. 8, pp. 3758–3763, Apr. 1997, doi: 10.1063/1.365499.

[98] S. Mao et al., “Commercial TMR heads for hard disk drives: characterization and
extendibility at 300 gbit2,” IEEE Transactions on Magnetics, vol. 42, no. 2, pp. 97–
102, Feb. 2006, doi: 10.1109/TMAG.2005.861788.

[99] W. F. Brown, Micromagnetics, Reprint 1978 with corrections. Huntington, N.Y: R.

E. Krieger Pub. Co., 1978.

100
[100] S. S. P. Parkin et al., “Giant tunnelling magnetoresistance at room temperature with
MgO (100) tunnel barriers,” Nature Mater, vol. 3, no. 12, Art. no. 12, Dec. 2004,
doi: 10.1038/nmat1256.

[101] S.-E. Lee, J.-U. Baek, and J.-G. Park, “Highly Enhanced TMR Ratio and Δ for
Double MgO-based p-MTJ Spin-Valves with Top Co2Fe6B2 Free Layer by
Nanoscale-thick Iron Diffusion-barrier,” Sci Rep, vol. 7, no. 1, Art. no. 1, Sep. 2017,
doi: 10.1038/s41598-017-10967-x.

[102] Z. Lei et al., “High TMR for both in-plane and perpendicular magnetic field justified
by CoFeB free layer thickness for 3-D MTJ sensors,” AIP Advances, vol. 9, no. 8, p.
085127, Aug. 2019, doi: 10.1063/1.5117320.

[103] A. G. Qoutb and E. G. Friedman, “MTJ Magnetization Switching Mechanisms for

IoT Applications,” in Proceedings of the 2018 on Great Lakes Symposium on VLSI,
in GLSVLSI ’18. New York, NY, USA: Association for Computing Machinery, May
2018, pp. 347–352. doi: 10.1145/3194554.3194624.

[104] L. D. Landau, L. P. Pitaevskii, and E. M. Lifshitz, Electrodynamics of Continuous

Media: Volume 8, 2nd edition. Amsterdam Heidelberg: Butterworth-Heinemann,
1984.

[105] S. Kanai, K. Hayakawa, H. Ohno, and S. Fukami, “Theory of relaxation time of

stochastic nanomagnets,” Phys. Rev. B, vol. 103, no. 9, p. 094423, Mar. 2021, doi:
10.1103/PhysRevB.103.094423.

[106] T. L. Gilbert, “A phenomenological theory of damping in ferromagnetic materials,”

IEEE Transactions on Magnetics, vol. 40, no. 6, pp. 3443–3449, Nov. 2004, doi:
10.1109/TMAG.2004.836740.

[107] J. Z. Sun, “Current-driven magnetic switching in manganite trilayer junctions,”

Journal of Magnetism and Magnetic Materials, vol. 202, no. 1, pp. 157–162, Jun.
1999, doi: 10.1016/S0304-8853(99)00289-9.

[108] J. Z. Sun, “Spin-current interaction with a monodomain magnetic body: A model

study,” Phys. Rev. B, vol. 62, no. 1, pp. 570–578, Jul. 2000, doi:
10.1103/PhysRevB.62.570.

101
[109] Soshin Chikazumi, Physics of Ferromagnetism, Second Edition. in International
Series of Monographs on Physics. Oxford, New York: Oxford University Press,
2009.

[110] B. Bhushan et al., “Enhancing Magnetic Immunity of STT-MRAM with Magnetic

Shielding,” in 2018 IEEE International Memory Workshop (IMW), May 2018, pp.
1–4. doi: 10.1109/IMW.2018.8388828.

[111] Y.-D. Chih et al., “13.3 A 22nm 32Mb Embedded STT-MRAM with 10ns Read
Speed, 1M Cycle Write Endurance, 10 Years Retention at 150°C and High Immunity
to Magnetic Field Interference,” in 2020 IEEE International Solid-State Circuits
Conference-(ISSCC), Feb. 2020, pp. 222–224. doi: 10.1109/ISSCC19947.
2020.9062955.

[112] J. Z. Sun, “Spin-transfer torque switched magnetic tunnel junction for memory
technologies,” Journal of Magnetism and Magnetic Materials, vol. 559, p. 169479,
Oct. 2022, doi: 10.1016/j.jmmm.2022.169479.

[113] W. F. Brown, “Thermal Fluctuations of a Single-Domain Particle,” Phys. Rev., vol.

130, no. 5, pp. 1677–1686, Jun. 1963, doi: 10.1103/PhysRev.130.1677.

[114] K. Hayakawa et al., “Nanosecond Random Telegraph Noise in In-Plane Magnetic

Tunnel Junctions,” Phys. Rev. Lett., vol. 126, no. 11, p. 117202, Mar. 2021, doi:
10.1103/PhysRevLett.126.117202.

[115] Z. Zuo, L. A. Dissado, N. M. Chalashkanov, S. J. Dodd, and C. Yao, “Dielectric

breakdown at sub-critical fields,” Appl. Phys. Lett., vol. 113, no. 11, p. 112901, Sep.
2018, doi: 10.1063/1.5044718
.

[116] R. Vogelsang, T. Farr, and K. Frohlich, “The effect of barriers on electrical tree
propagation in composite insulation materials,” IEEE Transactions on Dielectrics
and Electrical Insulation, vol. 13, no. 2, pp. 373–382, Apr. 2006, doi:
10.1109/TDEI.2006.1624282.

[117] S. Siddabattuni, T. P. Schuman, and F. Dogan, “Dielectric Properties of Polymer–

Particle Nanocomposites Influenced by Electronic Nature of Filler Surfaces,” ACS
Appl. Mater. Interfaces, vol. 5, no. 6, pp. 1917–1927, Mar. 2013, doi:
10.1021/am3030239.

102
[118] C.-W. Nan, Y. Shen, and J. Ma, “Physical Properties of Composites Near
Percolation,” Annual Review of Materials Research, vol. 40, no. 1, pp. 131–151,
2010, doi: 10.1146/annurev-matsci-070909-104529.

[119] L. Zhang and Z.-Y. Cheng, “Development of polymer-based 0–3 composites with
high dielectric constant,” J. Adv. Dielect., vol. 01, no. 04, pp. 389–406, Oct. 2011,
doi: 10.1142/S2010135X11000574.

[120] L. A. Dissado and J. C. Fothergill, Electrical degradation and breakdown in

polymers, London: P. Peregrinus, 1992.

[121] L. Niemeyer, L. Pietronero, and H. J. Wiesmann, “Fractal Dimension of Dielectric

Breakdown,” Phys. Rev. Lett., vol. 52, no. 12, pp. 1033–1036, Mar. 1984, doi:
10.1103/PhysRevLett.52.1033.

[122] I. M. Irurzun, P. Bergero, V. Mola, M. C. Cordero, J. L. Vicente, and E. E. Mola,

“Dielectric breakdown in solids modeled by DBM and DLA,” Chaos, Solitons &
Fractals, vol. 13, no. 6, pp. 1333–1343, May 2002, doi: 10.1016/S0960-
0779(01)00142-4.

[123] F. Peruani, G. Solovey, I. M. Irurzun, E. E. Mola, A. Marzocca, and J. L. Vicente,

“Dielectric breakdown model for composite materials,” Phys. Rev. E, vol. 67, no. 6,
p. 066121, Jun. 2003, doi: 10.1103/PhysRevE.67.066121.

[124] P. Bergero, F. Peruani, G. Solovey, I. M. Irurzun, J. L. Vicente, and E. E. Mola,

“Dielectric breakdown model for conductor-loaded and insulator-loaded composite
materials,” Phys. Rev. E, vol. 69, no. 1, p. 016123, Jan. 2004, doi:
10.1103/PhysRevE.69.016123.

[125] E. Kuffel, W. S. Zaengl, and J. Kuffel, High voltage engineering: fundamentals, 2nd
ed. Oxford; Boston: Butterworth-Heinemann, 2000.

[126] Y. Liu and X. Cao, “Electrical tree growth characteristics in XLPE cable insulation
under DC voltage conditions,” IEEE Trans. Dielectr. Electr. Insul., vol. 22, no. 6,
pp. 3676–3684, Dec. 2015, doi: 10.1109/TDEI.2015.005222.

103
[127] T. Han, B. Du, and J. Su, “Electrical tree initiation and growth in silicone rubber
under combined DC-pulse voltage,” Energies, vol. 11, no. 4, p. 764, Apr. 2018.

[128] I. Pleşa, P. V. Noţingher, S. Schlögl, C. Sumereder, and M. Muhr, “Properties of

polymer composites used in high-voltage applications,” Polymers, vol. 8, no. 5, p.
173, May 2016, doi: 10.3390/polym8050173.

[129] M. G. Danikas and T. Tanaka, “Nanocomposites-a review of electrical treeing and

breakdown,” IEEE Electrical Insulation Magazine, vol. 25, no. 4, pp. 19–25, Jul.
2009, doi: 10.1109/MEI.2009.5191413.

[130] H. -Z. Ding and B. R. Varlow, “Filler volume fraction effects on the breakdown
resistance of an epoxy microcomposite dielectric,” in Proceedings of the 2004 IEEE
International Conference on Solid Dielectrics, 2004. ICSD 2004., Jul. 2004, vol. 2,
pp. 816-820 Vol.2. doi: 10.1109/ICSD.2004.1350557.

[131] S. F. Mendes et al., “Influence of filler size and concentration on the low and high
temperature dielectric response of poly(vinylidene fluoride)/Pb(Zr0.53Ti0.47)O3
composites,” J Polym Res, vol. 19, no. 9, p. 9967, Sep. 2012, doi: 10.1007/s10965-
012-9967-5.

[132] K. Kudo, “Fractal analysis of electrical trees,” IEEE Trans. Dielectr. Electr. Insul.,
vol. 5, no. 5, pp. 713–727, Oct. 1998.

[133] I. Imperiale et al., “Role of the Insulating Fillers in the Encapsulation Material on
the Lateral Charge Spreading in HV-ICs,” IEEE Transactions on Electron Devices,
vol. 64, no. 3, pp. 1209–1216, Mar. 2017, doi: 10.1109/TED.2016.2645080.

[134] T. Imai, F. Sawa, T. Yoshimitsu, T. Ozaki, and T. Shimizu, “Preparation and

insulation properties of epoxy-layered silicate nanocomposite [insulating material
applications],” in The 17th Annual Meeting of the IEEE Lasers and Electro-Optics
Society (LEOS), 2004, pp. 402–405. doi: 10.1109/CEIDP.2004.1364272.

[135] S. P. Fillery et al., “Nanolaminates: Increasing Dielectric Breakdown Strength of

Composites,” ACS Appl. Mater. Interfaces, vol. 4, no. 3, pp. 1388–1396, Mar. 2012,
doi: 10.1021/am201650g.

104
[136] M. Raquibuzzaman, B. Ray, T. B. Boykin, and R. S. Gorur, “Polymer-metal layered
structures for improved energy storage density,” IEEE Trans. Dielectr. Electr. Insul.,
vol. 25, no. 6, pp. 2375–2379, Dec. 2018.

[137] J. K. Nelson, Y. Hu, and J. Thiticharoenpong, “Electrical properties of TiO2

nanocomposites,” Annu. Rep. Conf. Electr. Insul. Dielectr. Phenom. (CEIDP), 2003,
pp. 719–722, doi: 10.1109/CEIDP.2003.1254955.

[138] H. -Z. Ding and B. R. Varlow, “Effect of nano-fillers on electrical treeing in epoxy
resin subjected to AC voltage,” in The 17th Annual Meeting of the IEEE Lasers and
Electro-Optics Society (LEOS), 2004, pp. 332–335.

[139] P. Hu et al., “Topological-structure modulated polymer nanocomposites exhibiting

highly enhanced dielectric strength and energy density,” Adv. Funct. Mater., vol. 24,
no. 21, pp. 3172–3178, Jun. 2014, doi: 10.1002/adfm.201303684.

105