1-D CHALCOGENIDE NANOMATERIALS FOR ELECTRONICS: PHASE-

CHANGE MEMORY AND TOPOLOGICAL INSULATORS

A DISSERTATION

SUBMITTED TO

THE DEPARTMENT OF MATERIALS SCIENCE AND ENGINEERING

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Stefan Meister

August 2010

http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/vh072qq0446

© 2010 by Stefan Meister. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.

ii

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Yi Cui, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

David Goldhaber-Gordon

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Paul McIntyre

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

iv

Abstract

Chalcogenides are ubiquitous in academia and industry. We focus on

chalcogenides for phase-change memory and topological insulators. Phase change

memory (PCM) utilizes the resistivity contrast between the crystalline and amorphous

phases of certain chalcogenides to store information. Industry is making progress in

PCM, which could someday compete with flash memory due to its superior scalability,

fast switching and high endurance. In contrast, the field of topological insulators is

very young. Topological insulators, bulk insulators with conductive surface states,

have attracted great attention since their discovery five years ago. They have exotic

properties and are exciting for both fundamental studies and potential applications.

Our studies of chalcogenide nanomaterials for PCM and topological insulators

provided new insights into these fascinating materials.

We developed a method to switch PCM cells inside a transmission electron

microscope (TEM), allowing direct correlation of electrical behavior with structural

changes. Using this technique, we found that the electrical behavior of Ge2Sb2Te3

lateral phase-change cells depends strongly on the highly variable microstructure of

the amorphous domain. In search of new PCM materials, we developed the VLS

synthesis of GeTe and Sb2Te3 phase-change nanowires. In situ TEM observation of

these nanowires revealed a new switching mechanism via the opening and closing of

voids. We also synthesized single-crystalline topological insulator nanoribbons,

whose large surface to bulk conduction ratio increases the relative contribution of the

surface states. We successfully measured Aharonov-Bohm oscillations in the

nanoribbons, thereby observing one of the first electrical signatures of the topological

states in this material system.

v

Acknowledgements

First of all, I am grateful to my advisor Professor Yi Cui for his contagious

enthusiasm, valuable guidance and constant support throughout my graduate years.

I’m glad to have been one of his first students; the opportunity to help set up the lab

and see the group grow was a tremendous learning experience. Next, I would like to

thank the members of my reading committee, Professors David Goldhaber-Gordon

and Paul McIntyre, for their valuable input into this dissertation as well as for serving

on my defense committee along with Professor William Nix and defense chair

Professor Yoshio Nishi. Finally, I am grateful for the support and kindness of

Professors Robert Huggins, Xiao-Liang Qi, Zhi-Xun Shen, Philip Wong, and Shou-

Cheng Zhang during my time at Stanford.

I am grateful to SangBum Kim and David Schoen for the fruitful collaboration

on the phase-change memory project. I am also thankful for the exciting collaboration

on topological insulators with Hailin Peng, Desheng Kong, Keji Lai, Judy Cha and

Yulin Chen. The group dynamic and team spirit were the best I have ever experienced.

Being part of the Cui group has been both enjoyable and exciting; I am thankful to

have had the opportunity to work with these great people and I value the friendships

that we have formed.

Finally, I am grateful for the loving support of my wife, family and friends.

vi

Table of Contents

Abstract ........................................................................................................................ iv

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

1 Introduction ..........................................................................................................1

1.1 Background and motivation ...........................................................................1

1.2 Outline of dissertation ....................................................................................2

1.3 Phase-change materials for memory ..............................................................3

1.3.1 Introduction ............................................................................................3

1.3.2 The phase-change mechanism................................................................5

1.3.3 Threshold switching ...............................................................................7

1.3.4 Common phase-change memory cells....................................................9

1.3.5 Materials ...............................................................................................10

1.4 In situ TEM...................................................................................................14

1.4.1 Introduction and motivation .................................................................14

1.4.2 In situ TEM holder ...............................................................................15

1.4.3 TEM window manufacturing ...............................................................16

1.4.4 History and challenges of in situ TEM technique ................................18

1.5 Vapor-liquid-solid growth ............................................................................20

1.5.1 Introduction ..........................................................................................20

1.5.2 Growth parameters ...............................................................................22

1.6 Topological insulators ..................................................................................24

1.6.1 Introduction ..........................................................................................24

1.6.2 Materials ...............................................................................................26

2 In situ transmission electron microscopy of phase-change memory .............28

2.1 Introduction ..................................................................................................28

2.2 Method..........................................................................................................30

2.2.1 Device fabrication ................................................................................30

2.2.2 TEM characterization of sputtered GST films .....................................31

2.2.3 Electrical testing and TEM observation ...............................................32

vii

2.3 Switching experiments .................................................................................33

2.3.1 Demonstration of in situ switching.......................................................33

2.3.2 Two types of amorphous phases...........................................................35

2.3.3 Dependence of phase transition on voltage pulse.................................36

2.4 Threshold field measurements......................................................................40

2.4.1 Significance of threshold field..............................................................40

2.4.2 Measurements.......................................................................................41

2.4.3 Discussion.............................................................................................45

2.5 Control of quenching rate via device geometry ...........................................47

2.6 Conclusions ..................................................................................................49

3 Synthesis and characterization of GeTe phase-change nanowires ................51

3.1 Introduction ..................................................................................................51

3.1.1 Motivation and overview......................................................................51

3.1.2 Background...........................................................................................51

3.1.3 Materials ...............................................................................................52

3.2 Synthesis.......................................................................................................53

3.3 Characterization............................................................................................54

3.3.1 Scanning electron microscopy..............................................................54

3.3.2 Transmission electron microscopy .......................................................57

3.4 Sb2Te3 nanowires..........................................................................................59

3.5 Recent developments and future outlook .....................................................59

3.6 Conclusion....................................................................................................61

4 Electrical characterization and in-situ TEM of phase-change nanowires ....62

4.1 Introduction ..................................................................................................62

4.1.1 Background and motivation .................................................................62

4.1.2 Device fabrication ................................................................................64

4.1.3 Instruments and measurements.............................................................66

4.2 Conventional (black-box) switching experiment .........................................67

4.2.1 Electrical characterization ....................................................................67

4.2.2 Nanowire switching..............................................................................69

viii

4.2.3 Investigation of high resistance in off state via SPM ...........................70

4.3 Direct TEM Observation of NWs during switching.....................................71

4.3.1 TEM observation of off state................................................................71

4.3.2 Real-time observation of dynamic behavior: opening and closing of

voids 76

4.4 Conclusion....................................................................................................80

5 Topological insulator nanoribbons ...................................................................82

5.1 Introduction and motivation .........................................................................82

5.2 Synthesis of topological insulator nanoribbons............................................83

5.3 Electrical characterization of nanoribbons ...................................................87

5.4 Aharonov-Bohm interference in topological insulator nanoribbons ............92

5.4.1 Introduction to Aharonov-Bohm effect ................................................92

5.4.2 Aharonov-Bohm measurement.............................................................94

5.4.3 Discussion and conclusion ...................................................................95

5.5 Conclusion....................................................................................................96

6 Conclusion ...........................................................................................................97

ix

List of Figures

Figure 1.1: Schematic of two phases of phase-change materials. The amorphous phase

has a high resistivity with typical resistances in the MΩ range. The crystalline

phase has a low resistivity with typical resistances in the kΩ range. Switching

between the two phases is accomplished via crystallization and melt-quenching. ....4

Figure 1.2: Schematic of the temperature of the active region in a phase-change cell as

a function of time. During the RESET operation, the device must exceed the

melting temperature for a short time and then be quenched rapidly to freeze the

atoms in place. During the SET operation the temperature must be raised above the

crystallization temperature for sufficient time for crystallization to take place. The

cool-down rate is for the SET operation is not important for the phase transition to

occur. ..........................................................................................................................6

Figure 1.3: Schematic of the IV response of a typical phase-change memory cell. The

graph is for a current scan showing typical snapback behavior. ................................8

Figure 1.4: Schematic of vertical (left) and horizontal (right) phase-change memory

cells. The light red shows the approximate region where the phase-transition occurs.

..................................................................................................................................10

Figure 1.5: Ternary phase-diagram for Ge, Te, and Sb showing important

compositions.............................................................................................................13

Figure 1.6: Schematic comparison between conventional PCM measurements that

treat the memory element as a black box and in situ TEM measurements that supply

detailed information on microstructural changes during switching. ........................14

Figure 1.7: Photograph of custom electrical biasing holder and SEM images of TEM

Si3N4 window used in the holder. The light squares in the SEM image are gold pads

for making electrical contact via wire bonding. .......................................................16

Figure 1.8: Schematic of silicon nitride membrane on silicon substrate. a. Cross-

section of silicon substrate. b. Top-view of substrate with gold pads. ....................17

Figure 1.9: Design challenges for in situ TEM PCM cells...........................................19

x

Figure 1.10: Schematic of VLS growth furnace and plot of temperature distribution

inside furnace............................................................................................................21

Figure 1.11: SEM of early growth stage of GeTe NWs (left: top view; right: side view)

..................................................................................................................................22

Figure 1.12: Schematic of major growth parameters. The growth only works if the

parameters are all in a “target range” indicated by the blue ellipsoid. .....................23

Figure 1.13: Schematic of simplified topological insulator band diagram. .................25

Figure 1.14: Schematic showing crystal structure of Bi2Se3 .......................................27

Figure 2.1: Device schematic showing silicon substrate with 50nm silicon nitride

membrane suspended at the center and GST bridge that is fabricated on the

membrane. ................................................................................................................29

Figure 2.2: Schematic of device cross-section (left) and SEM image of complete

device (right). Cross section is taken along the dotted line. .....................................30

Figure 2.3: TEM characterization of annealed GST 225 film after sputter deposition. a,

b. Bright-field TEM images of as-deposited GST225 film after annealing on hot

plate at 180ºC for 1 min. c. SAD of GST225 film could be accurately indexed to

face-centered cubic phase of GST225. No other phase was observed. ...................31

Figure 2.4: Table summarizes measured and theoretical d-spacings of cubic phase of

GST225.....................................................................................................................32

Figure 2.5: TEM images and SAD of switching between crystalline and amorphous

phase. a, b. TEM image showing a bridge in the (a) crystalline phase and (b)

amorphous phase. Typical resistance values for crystalline and amorphous phases

are 10-100 kΩ and 1-8 MΩ respectively. The red circles indicate the approximate

size and location of the selected area diffraction aperture in (c) and (d) c, d. Selected

area diffraction confirming the poly-crystalline (c) and amorphous (d) nature of the

bridge........................................................................................................................34

Figure 2.6: TEM images of two-phase amorphous/crystalline (left) and single-phase

amorphous (right) domain. The coloring is provided to give an estimate of where

the material is amorphous.........................................................................................35

xi

Figure 2.7: Schematic of typical PCM resistance as a function of applied voltage

pulse. The crystal structure between crystalline and amorphous is unknown.........36

Figure 2.8: Study of crystalline to amorphous transition. a. Resistance as a function of

applied 400 ns voltage pulse. TEM images (b), (c), (d) correspond to labeled points

in the graph. The red line connects the average resistance values. b-d. TEM images

of phase-change memory bridge. Red ellipses show examples of crystals that

survived the amorphization process. Yellow circles point out small differences in

microstructure that accompany resistance change from 62 kΩ to 150 kΩ. Green

dashed lines surround areas that are predominantly amorphous. .............................38

Figure 2.9: TEM images and corresponding SADs of bridge in various states of

crystallization. a-c. Resistance of bridge in (a), (b), and (c) is 29 kΩ, 115 kΩ and

3.5 MΩ, respectively. The SADs (approximate size and location of SAD aperture

indicated by red circles) confirm the crystallinity of the states. In (a), the bridge is

essentially completely crystalline as indicated by spot pattern and absence of diffuse

ring. In (b), the bridge has some amorphous background, but the SAD still shows

significant crystallinity. In (c), the bridge is predominantly amorphous with several

crystals remaining embedded in the amorphous matrix. ..........................................39

Figure 2.10: Current spikes and destructive switching due to high threshold field.

Voltage response graph indicating the point at which breakage can occur (left).

TEM image of a sample broken during crystallization attempt (right). ...................41

Figure 2.11: Measurement of threshold switching in high-resistance bridges. a. IV

curve showing 120 μA current scan to crystallize the bridge in (c) and to measure

the threshold voltage. This bridge exhibits complicated switching behavior. b. IV

curve showing 100 μA current scan to crystallize the bridge in (d) and to measure

the threshold voltage. This bridge exhibits clean switching behavior. c. TEM image

of two-phase bridge before current scan (a). The approximate amorphous domain

has been colored blue to assist with visualization. d. TEM image of single-phase

amorphous bridge before current scan (b). The approximate amorphous domain has

been colored red to assist with visualization. ...........................................................42

xii

Figure 2.12: Measurement of threshold switching in high-resistance bridges. a. IV

curve showing 120 μA current scan to crystallize the bridge in (c) and to measure

the threshold voltage. This bridge exhibits complicated switching behavior, because

it is in a mixed phase consisting of nanocrystals interspersed in an amorphous

matrix resulting from repeated switching. b. IV curve showing 100 μA current scan

to crystallize the bridge in (d) and to measure the threshold voltage. This bridge

exhibits complicated switching behavior, because it consists of nanocrystals

interspersed in an amorphous matrix. c. TEM image of mixed phase bridge before

current scan (a). d. TEM image of bridge before current scan (b). The bridge

consists of nanocrystals that are due to annealing the as-deposited GST225 film

after sputtering at 180°C for 10 min. Despite the high density of nanocrystals, the

resistance of the bridge is still very high indicating that no connected current paths

exist. The nanocrystals appear as dark dots in TEM images due to the diffraction

contrast. Amorphous regions exhibit uniform contrast. ..........................................44

Figure 2.13: Dependence of threshold voltage on amorphous region parameters. a.

Threshold voltage as a function of the size of the amorphous region. b. Threshold

voltage as a function of resistance in the amorphous state. We observe two different

behaviors depending on the microstructure of the amorphous phase.......................45

Figure 2.14: Switching of high-aspect-ratio bridge with lower cooling rate. a.

Crystalline bridge obtained after 60μA current scan; resistance is 18 kΩ. The red

circles show examples of large crystals embedded inside the bridge. b. The same

bridge after applying a 2V, 300ns pulse. The red circles show that the crystals in (a)

disappeared, indicating that the bridge melted completely as a result of the pulse.

However, the bridge did not turn completely amorphous but re-crystallized due to

the lower cooling rate. c. The same bridge after 60μA current scan; resistance is 29

kΩ. ............................................................................................................................47

Figure 2.15: Repeated switching of high-aspect-ratio bridge with lower cooling rate.

The figure shows TEM micrographs of the bridge after repeated switching. 60 μA

current scans were used for crystallization and various voltage pulses were applied

to achieve a high-resistance state. The series of TEM images with the resistance

xiii

shows that the bridge could not be turned completely amorphous by a 2V, 300ns

pulse. Increasing the pulse further resulted in the bridge breaking. ........................48

Figure 2.16: TEM micrograph of a short bridge that can achieve a single-phase

amorphous domain without crystalline residue due to its fast cooling-rate. ............49

Figure 3.1: Quasi-binary phase diagram between Au and GeTe (Sb2Te3). The shape of

the drawing represents the phase diagram between Au and GeTe. However, the

general shape for Au and Sb2Te3 is very similar and the diagram can be used for

showing the temperature values and general shape..................................................52

Figure 3.2: SEM image of as-grown GeTe nanowires, showing high density and a

wide size distribution with diameters ranging from roughly 20-300 nm. ................54

Figure 3.3: SEM image of long GeTe nanowires.........................................................55

Figure 3.4: SEM image of as-grown GeTe helical and straight NWs (top). The

nanowires can randomly change from straight to helical growth (bottom)..............56

Figure 3.5: TEM image of a straight GeTe NW. The dark spot at the tip corresponds

to a gold nanoparticles. Insets show a high resolution image of the lattice fringes

and SAD pattern. ......................................................................................................57

Figure 3.6: TEM Energy Dispersive X-ray Spectroscopy of GeTe NW.....................58

Figure 3.7: SEM image of as-grown Sb2Te3 NW.........................................................59

Figure 3.8: Schematic of phase-change memory that is based on NWs. The phase-

change NW is grown on top of a p-n junction for selective addressing...................60

Figure 4.1: Device fabrication steps outlined with SEM micrographs. .......................64

Figure 4.2: SEM micrographs of GeTe NW with Pt contacts deposited by FIB .........65

Figure 4.3: Four point probe GeTe nanowires device with corresponding measurement

results........................................................................................................................ 67

Figure 4.4: TEM image of GeTe NWs during heating experiment..............................68

Figure 4.5: Cycling behavior. Nanowires were switched to a high resistance state with

a 10 V, 200 ns pulse, and switched back to the low resistance state with a 100 mV/s

scan up to 5 V. ..........................................................................................................69

xiv

Figure 4.6: SEM, AFM, and EFM images of the same single nanowire device in a high

resistance state. There is a 2 V difference between the light and dark color in the

EFM image. ..............................................................................................................71

Figure 4.7: Low-magnification SEM image of a silicon nitride membrane (dark) with

gold contact pads (light). ..........................................................................................72

Figure 4.8: Nanowire before (lower left inset) and after pulsing (main panel); the

switching mechanism is identified as void formation at the top contact (upper right

inset). ........................................................................................................................ 72

Figure 4.9: a. TEM image of nanowire after a 3.5V, 200ns pulse shows only minor

changes including a bulge and small pinhole evaporation. b. Higher magnification

image of bulge and pin hole c. Higher magnification image of pinhole close to

contact.......................................................................................................................74

Figure 4.10: a. Low-magnification TEM image of nanowire after a long low

magnitude pulse. b. Banded planar defect structure. c. Void and amorphous region

formation. Area I is amorphous and Area II, like the rest of the wire, is crystalline

(SAED inset). ...........................................................................................................74

Figure 4.11: Pulse applied to NW shown in Figure 4.10a; blue is the applied voltage,

and red is the current through the device (left). EDS spectrum of areas I and II from

Figure 4.10c (right)...................................................................................................75

Figure 4.12: a. Voltage scan on a single nanowire device (inset). The blue square in

the inset shows the location of the TEM observation in (b). b. TEM images taken in

situ during the voltage scan in panel, at times I, II, III, and IV. Note the correlation

of resistance with void size.......................................................................................76

Figure 4.13: TEM micrographs showing void formation via voltage pulse................77

Figure 4.14: In situ TEM observation showing closing of void and corresponding

voltage scan. .............................................................................................................78

Figure 4.15: Material movement in void induced switching. a. SEM after multiple

cycles. b. Schematic showing the movement of material inside the NW................79

Figure 5.1: Schematic of (a) bulk sample of Bi2Se3 and corresponding (b) band

diagram. ....................................................................................................................82

xv

Figure 5.2: Schematic of topological insulator nanoribbon ........................................83

Figure 5.3: SEM of as-synthesized Bi2Se3 nanoribbons. ............................................84

Figure 5.4: SEM and AFM of Bi2Se3 NR device. Graph shows height measurement of

AFM cross section....................................................................................................85

Figure 5.5: SEM and AFM of Bi2Se3 NR device. Graph shows height measurement of

AFM cross section....................................................................................................86

Figure 5.6: Schematic of NR device. Setup allows for 4-point probe (green V-) and

Hall (blue V-) measurement on the same ribbon.....................................................87

Figure 5.7: SEM of Bi2Se3 NR device at 30° rotation. Even though the NR appears

bent it is lying flat on the substrate and the contacts are on top. ..............................88

Figure 5.8: Resistance as a function of temperature for Bi2Se3 NR. ...........................89

Figure 5.9: SEM Bi2Se3 NR device. This setup allows for 4-point probe and Hall

measurements on the same device............................................................................90

Figure 5.10: Hall measurements of two different devices showing different behaviors.

..................................................................................................................................91

Figure 5.11: Schematic of double slit experiment........................................................92

Figure 5.12: Schematic of Aharonov-Bohm experiment. ............................................93

Figure 5.13: Magnetoresistance and SEM of Bi2Se3 NR .............................................94

Figure 5.14: Fourier transform of dR/dB data derived from Figure 5.X......................95

1

1 Introduction

1.1 Background and motivation

Chalcogenide materials are chemical compounds that combine at least one

chalcogen with at least one more electropositive element. The term chalcogen

generally refers to an element from group 16 (O, S, Se, Te, Po) of the periodic table.

Since oxides are usually treated separately and polonium is rare and highly radioactive,

chalcogenides commonly refer to sulfides, selenides and tellurides only. These

ubiquitous materials have been studied for many years, because they showed promise

for many applications including solar cells1, 2, wave-guides3-5, photonic crystals6,

terahertz detectors7, 8 and emitters8, thermoelectrics9, 10 and phase-change memory11, 12.

In 2005 a new class of materials called topological insulators13-17 was

discovered. These materials attract much attention because they exhibit exotic

electronic surface states that are interesting on a fundamental level and for potential

future applications. Interestingly, with the exception of BixSb1-x, the first predicted

topological insulators18 (HgTe, Bi2Se3, Bi2Te3, Sb2Te3) are all chalcogenides. Hence,

it should come as no surprise that chalcogenides are still receiving significant attention

in academic and industrial research today.

The synthesis of Silicon and Germanium nanowires19 in 1998 caused a

resurgence of interest in the vapor-liquid-solid (VLS) growth20 as a universal method

for producing a wide variety of semiconducting NWs. These nanostructures have

shown interesting properties such as quantum confinement21, 22, a high thermoelectric

2

conversion efficiency23, 24, usefulness in sensors25, 26 and batteries27, and interesting

phase transformations28. Further, nanowires are ideal candidates for fundamental

studies due to their simplicity and compatibility with techniques such as electron

microscopy and scanning probe microscopy29, 30.

In this dissertation we use the vapor-liquid-solid growth to synthesize

nanowires for technologically interesting materials, namely phase-change materials

and topological insulators. We exploit nanostructure geometries to allow in situ

transmission electron microscopy (TEM) studies on phase-change memory cells and

to measure Aharonov-Bohm interference in topological insulators, a signature of the

surface states. A short introduction to these materials, the in situ TEM technique, and

the vapor-liquid-solid growth is provided in the following sections.

1.2 Outline of dissertation

In this dissertation, we first give an introduction to phase-change memory, the

in situ TEM technique employed in chapter 2 and 3, the vapor-liquid-solid growth, and

topological insulators. In chapter 2, we describe an in situ TEM study of a

conventional lateral phase-change memory cell31. The TEM observation allows for a

direct correlation between microstructure and switching behavior. We find that PCM

devices with similar resistances can exhibit distinct threshold switching behaviors due

to the different initial distribution of nanocrystalline and amorphous domains,

explaining the variability of switching behaviors of PCM cells in the literature. In

chapter 3, we discuss the VLS synthesis and characterization of GeTe and Sb2Te3

phase-change nanowires as new materials for phase-change memory. Chapter 4

3

covers our TEM observations of switching in phase-change nanowires. Instead of a

crystalline-amorphous transformation, we observed that the dominant switching

mechanism during multiple cycling appears to be the opening and closing of voids in

the nanowires due to material migration. In Chapter 5 we discuss the VLS synthesis

of Bi2Se3 topological insulator nanostructures, and in chapter 6 we show transport

evidence of topological surface states through Aharonov-Bohm oscillations32 in VLS-

synthesized Bi2Se3 nanoribbons.

1.3 Phase-change materials for memory

1.3.1 Introduction

Phase-change materials are materials that can exist in at least two different

structural states: amorphous and crystalline (with possibly multiple crystalline states).

While many materials can exist in these two states, only a few satisfy the stringent

requirements for use in memory applications33, 34 (discussed in section 1.3.5). In order

for a memory element to be useful, a read operation must be able to unambiguously

differentiate between the two states. In the case of PCM, there is a large contrast in

conductivity and reflectivity between the crystalline and amorphous state. While the

difference in reflectivity has been exploited for optical storage media35, such as

rewritable DVDs, phase-change memory relies entirely on the difference in

conductivity. Figure 1.1 shows a schematic of the two phases, with typical resistance

values of actual cells and basic nomenclature. Switching from the amorphous phase to

the crystalline phase, or crystallizing the amorphous phase, is referred to as the SET

operation. Switching from the crystalline to the amorphous phase, called the RESET

4

operation, is accomplished by melt-quenching the material. Resistances in typical

devices are in the MΩ range and kΩ range for amorphous and crystalline states,

respectively.

Figure 1.1: Schematic of two phases of phase-change materials. The amorphous phase has a high resistivity with typical resistances in the MΩ range. The crystalline phase has a low resistivity with typical resistances in the kΩ range. Switching between the two phases is accomplished via crystallization and melt-quenching.

The first electrical switching between two states in chalcogenide materials was

performed in 1968 by Stanford Ovshinsky36. However, phase change memory has

only recently generated great interest as a non-volatile memory solution, since flash

memory has dominated the non-volatile memory market over the past 20 years.

Continued scaling resulting in better performance and favorable economies of scale in

flash memory made it difficult for competing memory technologies to enter the market.

Only in recent years has the scaling of flash encountered fundamental problems that

rekindled research into several other non-volatile memory technologies37, 38. These

technologies include magnetoresistive RAM39 (MRAM), ferroelectric RAM40, 41

(FeRAM), other resistive RAM42 (RRAM) types, and phase-change memory.

5

Applications for non-volatile memory are ubiquitous. The wide use of

portable electronics such as mobile phones, MP3 players, personal multimedia players,

e-book readers, navigational devices and advanced laptops contribute to the growing

market for non-volatile memory and push developers to make memory cheaper, faster

and denser while consuming less power at the same time. Indeed, with the recent

advances there is hope for a universal memory that has all the attributes for an ideal

memory and can replace current non-volatile and volatile memory technologies.

Phase-change memory is believed to be a candidate for such universal memory43.

Today, phase-change materials are widely used in optical storage technologies

such as rewritable CDs and DVDs. Though, it is not clear which memory technology

will prevail for future non-volatile memory applications. However, on April 28th 2010,

Samsung announced shipment of the industry’s “First Multi-chip Package with a

PRAM Chip for Handsets” claiming that they use an alloy of germanium, antimony

and tellurium, which “provides three-times faster data storage performance per word

than NOR chips.” The device has a capacity of 512 megabit. This entrance of PCM

into the market place is an important step towards future memory technologies that are

not based on flash. It also demonstrates the maturity that PCM has reached at this

point and its potential for the future given that industry is already heavily invested in

its development.

1.3.2 The phase-change mechanism

Switching between the two phases is accomplished via Joule heating by a

voltage pulse44. In order to switch from the stable crystalline phase to the metastable

6

amorphous phase, the material is melted and rapidly quenched, freezing the atoms in

place before they have time to re-crystallize. For the reverse transformation, from

amorphous to crystalline, the material is heated above its crystallization temperature

for a sufficient time. Figure 1.2 depicts a schematic of typical temperature vs. time

behavior that induces switching in a PCM cell. While the RESET operation generally

requires higher power, the SET operation is the time-limiting step (and hence data rate

limiting), because it requires significantly more time to switch. Although the SET

operation tends to be the slower step, crystallization times faster than 1ns have been

reported45 providing a positive future outlook for very fast memory switching.

Figure 1.2: Schematic of the temperature of the active region in a phase-change cell as a function of time. During the RESET operation, the device must exceed the melting temperature for a short time and then be quenched rapidly to freeze the atoms in place. During the SET operation the temperature must be raised above the crystallization temperature for sufficient time for crystallization to take place. The cool-down rate is for the SET operation is not important for the phase transition to occur.

The impressive speed at which phase-change materials can crystallize has been a topic

of great interest and will be briefly discussed in section 1.3.5 on phase-change

7

materials. Section 1.3.5 will also cover the atomic arrangements of the crystalline and

amorphous phase for typical phase-change materials.

1.3.3 Threshold switching

Depending on device geometry and size, a typical phase-change cell needs

hundreds to thousands of μA of current to melt or switch from crystalline to

amorphous46-50. Since the typical resistance range of the amorphous phase is in the

MΩs, one may wonder how it is possible to supply sufficient current for the RESET

operation. If the resistance behavior was strictly linear, hundreds to thousands of volts

would be required to achieve the necessary current. It turns out that despite the high

resistivity of the amorphous phase, crystallization can be achieved by a relatively

small voltage pulse because PCM materials exhibit threshold switching51-53.

Threshold switching occurs when a critical electric field is applied across the

amorphous region resulting in conductive filaments that can facilitate sufficient

heating to induce crystallization.

8

Figure 1.3: Schematic of the IV response of a typical phase-change memory cell. The graph is for a current scan showing typical snapback behavior.

Figure 1.3 shows a schematic of an IV curve for a typical PCM cell for a current scan

for a cell that is initially in the amorphous state. Even though the voltage is the

dependent variable, it is customary to plot it on the horizontal axis. As the current is

increased, the required voltage scales nearly linearly until a certain threshold voltage

(Vth) is reached. At this point, the IV curve shows a typical snapback behavior

indicative of a sudden decrease in resistance. The resistance drops on the order of

nanoseconds, much faster than the typical crystallization process, which takes 10s to

100s of nanoseconds. At this point the cell is in a dynamic ‘on state’ in which the

resistance is very similar to the resistance in the crystalline state. However, it is a

volatile phenomena and the cell is still amorphous. As the current is further increased,

the cell temperature increases above the crystallization temperature due to joule

heating and permanent crystallization (switching) occurs. For clarity, Figure 1.3

9

shows the current scan extending into the temperature region where melting occurs.

However, to SET the cell, the current must only be large enough to cause heating

above the crystallization temperature (Figure 1.2). To read the state of the cell, a

voltage (Vread) that is smaller than the threshold voltage is necessary to obtain an

unambiguous response and to avoid inadvertent switching. Applying a voltage that is

larger than Vth typically results in sufficient current for crystallization.

The mechanism of threshold switching is not yet completely understood. It is

thought that threshold switching is due to carrier multiplication through inelastic

scattering as a result of the electric field53. There have been several theoretical

studies52, 54-56 of threshold switching, but they do not agree on the exact details of this

mechanism.

1.3.4 Common phase-change memory cells

There are several different designs for phase-change memory cells, but they

roughly fall into two categories: horizontal57, 58 and vertical31, 46-48 cells. Figure 1.4

shows a schematic for a basic vertical and horizontal cell. In one kind of horizontal

cell, often called a “T-cell” (left in Figure 1.4), a thin film of phase-change material is

contacted by a bottom electrode, the heater. During the phase-change, current is sent

through the heater causing localized heating and phase-change in the active material.

Similarly, the vertical cell forces the current through a narrow neck to localize the

phase transition.

10

Figure 1.4: Schematic of vertical (left) and horizontal (right) phase-change memory cells. The light red shows the approximate region where the phase-transition occurs.

Vertical cells are standard in industry, because they require significantly less

area on the wafer. However, horizontal cells have received significant attention due to

their ease of manufacture and low power consumption, because heating is confined to

the phase-change material only. In this dissertation, we focus on horizontal cells,

because they allow for a direct observation with the TEM and scanning probe

microscopy without destroying the cell. In fact, one of the main accomplishments of

this dissertation is the development of a horizontal line cell that can be switched in situ

the TEM repeatedly to study the details of the microstructural changes during

switching (Chapter 2).

1.3.5 Materials

Phase-change materials are one of the most important aspects of PCM, because

they dictate switching speed, power consumption, scalability and endurance. Thus,

finding the best possible phase-change materials will be a decisive factor in the

11

success of PCM. In the past, the discovery of new phase-change materials was mostly

driven by optical storage applications. However, the requirements that define a good

phase-change material are similar for optical and electronic applications. It turns out

that the most popular materials for optical applications also show exceptional

electronic performance. However, it is not clear that these materials will have

optimum performance for PCM.

So what makes an excellent phase-change material for PCM? Speed, long data

retention, and endurance are critical in any memory technology. Since the SET

operation is the time limiting step as discussed in section 1.3.2, a fast crystallization

speed is paramount for overall performance. For PCM to be competitive in the

marketplace, the crystallization time needs to be on the order of nanoseconds to 10s of

ns. It should be noted that most materials that crystallize quickly do so because of a

large driving force that is due to the difference in their Gibb’s free energy.

Unfortunately, this also means that the amorphous phase is metastable, which brings

us to the second requirement: the metastable amorphous phase must show sufficient

data retention. Specifically, it must stay in the amorphous phase for 10 years at 85ºC.

The conflict between fast crystallization and data retention helps explain why so few

materials are well suited for PCM. While the latter requires a relatively stable

amorphous phase, the former means that phase-change materials are inherently bad

glass formers: in order to switch to the amorphous phase a cooling-rate on the order of

1010 K/s has to be achieved. Additional requirements for PCM materials relate to

device endurance. For one, PCM materials must be chemically inert with the

surrounding materials. The RESET operation, which requires melting, is particularly

12

prone to induce chemical reactions and mixing with surrounding materials (i.e. the

contacts or encapsulation) severely reducing the endurance of the device. Since

several phase-change materials are binary or ternary materials, another requirement for

long endurance is that the material must not phase-segregate during re-solidification.

Phase-change materials have shown endurance over 109 cycles59, but 106 cycles is

more common in many cells. Few materials meet these stringent requirements.

The exceptional properties of phase-change materials typically arise from

structural properties that are shared between several of the most popular materials.

Unlike most other semiconductors, which tend to share the same short range order

present in their crystalline states with their amorphous phases, phase change materials

usually have widely different local bonding orders in each of their phases60. Typically,

the crystalline state is an octahedral arrangement characterized by a high vacancy

concentration and resonance bonding61, while the amorphous state is a more

covalently bonded, tetrahedrally coordinated arrangement. This difference in local

bonding order is responsible for the large optical contrast between the crystalline and

amorphous phases12.

13

Figure 1.5: Ternary phase-diagram for Ge, Te, and Sb showing important compositions.

The most widely used phase-change materials come from the Germanium-

Antimony-Tellurium (GST) family; Ge2Sb2Te5 (GST225) is most popular. Figure 1.5

shows a ternary phase-diagram between Ge-Sb-Te to illustrate some of the most

important compositions. The green oval indicates a range of material composition that

is used for rewritable optical storage disks in industry, and the red line shows

compositions used for PCM. While most stable compositions in the GST family are

phase-change materials, the pseudo-binary line between GeTe and Sb2Te3 has shown

fast switching and little phase-segregation upon repeated switching. Hence, for our

study we focus on GST225 for its superior properties and on GeTe and Sb2Te3 for

their simplicity and significance as the building blocks of GST225.

14

1.4 In situ TEM

1.4.1 Introduction and motivation

Common electronic measurements on phase-change memory treat PCM cells

as black boxes; microstructural changes inside the devices remain hidden (Figure 1.6).

In situ TEM has been shown to be a powerful tool to understand the microstructure

evolution during dynamic processes such as electroplating62, nanowire63-65 and

nanocrystal66 growth. Indeed, with its high resolution, selected area diffraction and

other attachments such as energy dispersive X-ray spectroscopy, the TEM seems to be

an ideal research platform to study the details of the phase-change.

Figure 1.6: Schematic comparison between conventional PCM measurements that treat the memory element as a black box and in situ TEM measurements that supply detailed information on microstructural changes during switching.

15

We have developed an in situ TEM technique enabling direct correlation

between microstructural changes and electrical behavior. To make these in situ

measurements possible, we obtained a custom made in situ electrical biasing holder

(section 1.4.2), manufactured TEM windows (section 1.4.3), and designed and

manufactured PCM cells compatible with in situ TEM (discussed in sections 1.4.4 and

2.2).

1.4.2 In situ TEM holder

The in situ TEM experiments presented in this dissertation were performed at

two locations with two different customized electrical biasing in situ holders: The

study on GeTe NWs (Chapter 3) was done at the National Center for Electron

Microscopy at Lawrence Berkeley National Laboratory using their equipment, and the

study on conventional GST225 cells was done at the Stanford Nanocharacterization

Laboratory (SNL) with a custom e-biasing holder purchased from Hummingbird

Scientific.

16

Figure 1.7: Photograph of custom electrical biasing holder and SEM images of TEM Si3N4 window used in the holder. The light squares in the SEM image are gold pads for making electrical contact via wire bonding.

Figure 1.7 shows a photograph of the holder used at SNL. It has 5 co-axial

feedthroughs and one common ground. The TEM window that contains the PCM

cells is glued into an epoxy breadboard and then wire bonded to the contacts.

1.4.3 TEM window manufacturing

Silicon nitride TEM membranes can be purchased directly (i.e., Structure

Probe, Inc.) or fabricated via standard semiconductor fabrication processes67-70. (Note:

David Schoen performed this part of the fabrication at the Stanford Nanofabrication

17

Facility.) For the in-house fabrication, a low stress 50nm silicon nitride film is

deposited on the front of the wafer. This film acted as the TEM transparent membrane

after the Si substrate was removed via a KOH etch. Figure 1.8 shows a schematic of

the finished membrane.

Figure 1.8: Schematic of silicon nitride membrane on silicon substrate. a. Cross-section of silicon substrate. b. Top-view of substrate with gold pads.

To define where the KOH will etch, a second silicon nitride film is deposited on the

backside of the wafer. Part of the silicon nitride film is removed so that the remainder

can act as an etching mask for the KOH etch. KOH results in a highly anisotropic etch

for silicon leading to the geometry depicted in Figure 1.8 for a silicon (100) wafer.

The Au electrodes for electrical contact (Figure 1.8b) are deposited directly onto the

50 nm thin SiNx.

18

1.4.4 History and challenges of in situ TEM technique

Despite the appeal of in situ TEM techniques for studying the microstructural

details of the phase-change mechanism, there have been relatively few efforts in this

direction. The reason for the lack of in situ studies may be the inherent difficulties

that arise. Before discussing these challenges later in this section, we will briefly

mention the prior work that has been done.

An interesting in situ TEM study was presented by Samsung Group at the 9th

Annual Non-Volatile Memory Technology Symposium in Pacific Grove, November

2008. They used a focused ion beam (FIB) to cut a conventional, vertical PCM cell

out of a wafer and observed it in situ TEM while applying voltage pulses with a

specialized scanning tunneling microscope tip. Using this method the group could

observe the reversible phase-transition, but no major findings were reported in the

literature. A problem with this technique is that the Gallium beam used in the FIB can

short out the cell and thereby significantly change its properties. Finally, cutting

individual cells out of a wafer with a FIB is difficult, so the throughput is low.

Besides the work by Samsung, it is difficult to find further reports on in situ switching.

However, it should be noted that other researchers have used the FIB to cut individual

cells from a wafer before and after switching71. This allowed them to obtain snapshots

in time of the cell during its life-cycle.

For an in situ TEM study, the more natural geometry is a horizontal PCM cell,

because it allows for direct observation with the TEM as long as the substrate is

transparent. No further modification to the cell is necessary. However, the challenges

for making an in situ TEM cell are numerous. The fabrication process (outlined in

19

section 2.2.1) requires electron beam lithography (EBL), which is relatively difficult

to do on 50nm thick membranes because they are very fragile. The finished TEM

chips are very small and need to be handled repeatedly for EBL, lift-off, gluing into

the carrier, and wirebonding. The fragile nature of the membranes means that a low

yield is to be expected and must be accounted for by making extra devices. Once the

devices are wire bonded and ready to be inserted into the TEM, they become very

prone to electro-static discharge which also leads to immediate breakage. This can be

overcome by trying to keep the contacts grounded at all times before pulses are

applied.

Figure 1.9: Design challenges for in situ TEM PCM cells.

Finally, there are several requirements for a working PCM cell that lead to

design constraints (Figure 1.9). For one, the repeated melting and high temperatures

during the RESET operation require a strong encapsulation. Otherwise, the material

can evaporate and/or ball up, breaking the cell. Hence, the simplest structure would

20

consist of a TEM membrane, a phase-change layer and an encapsulation layer. This

stack needs to be transparent to the electron beam which means the maximum total

thickness should be kept to less than approximately 200nm. To maximize the TEM

contrast, it is advisable to maximize the thickness of the phase-change material with

respect to the other layers. However, to achieve sufficient cooling rates, the thickness

of the phase-change layer must be kept to a minimum. To increase the cooling rate,

we can add lateral contacts that act as a heat sink and an electrical contact. The final

PCM cell design that we chose to satisfy all these requirements is discussed in Section

2.2.1.

1.5 Vapor-liquid-solid growth

1.5.1 Introduction

The VLS growth20 is a fairly simple method to produce a wide variety of

nanowires. Figure 1.10 shows a schematic of the basic VLS growth set-up employed

in this dissertation.

21

Figure 1.10: Schematic of VLS growth furnace and plot of temperature distribution inside furnace

We used a 1 inch horizontal tube furnace (Lindberg/Blue M) that has a non-uniform

temperature profile. We measured the temperature profile (Figure 1.10) for various

setpoints ranging from 900ºC to 400ºC. While the temperature in the center is close to

the set point, the temperature decreases with distance away from the center. This is

ideal for the VLS growth, because it enables us to evaporate the source material at the

set point and allows us to finely tune the temperatures at the growth substrates simply

by varying the position. With the source material (i.e. GeTe or Bi2Se3) placed in the

hot center region, we flow argon carrier gas through the furnace to transport the source

vapor to the growth substrates. The growth substrates are typically Silicon. They are

prepared with Au nano-particles, which have been shown to be a universal VLS

catalyst for many material systems.

22

For the VLS mechanism to work, the source material must have a eutectic

point with gold. As the vapor streams over the gold catalysts, it is absorbed and

incorporated into the particle. When the composition changes, the catalyst turns to

liquid. As the amount of source material in the liquid increases, the nanowire begins

to nucleate and grow.

Figure 1.11: SEM of early growth stage of GeTe NWs (left: top view; right: side view)

Figure 1.11 shows an SEM of the early growth stages of GeTe NWs with the gold

particles at the tip and the base of the wires extending from the substrate. The gold

particle remains at the tip during the whole growth process and NWs can grow up to

100s of micrometer long.

1.5.2 Growth parameters

The growth parameters used for each synthesis will be discussed in the

corresponding chapters (Sections 3.2 and 5.2). Here, we briefly describe general

500 500

23

considerations regarding the growth parameters for VLS growth. The major growth

parameters include temperature (of source material and substrates), pressure and flow

rate of the carrier gas.

Temperature

Pressu

reFl

ow ra

teTemperature

Pressu

reFl

ow ra

te

Figure 1.12: Schematic of major growth parameters. The growth only works if the parameters are all in a “target range” indicated by the blue ellipsoid.

Generally speaking, there is a certain range for these parameters where growth can

occur successfully (indicated by blue ellipsoid in Figure 1.12). If the parameters are

outside this narrow range, VLS growth will not work: the catalyst may not be active

enough, insufficient (or excessive) material may reach the catalyst, the temperature

may be too hot or cold, leading to evaporation of the nanowire or thin film deposition

of material everywhere, and so forth.

Interestingly, the reported growth conditions for a given system vary widely

between groups. The reason for this is probably the presence of many other

parameters that are typically not discussed in detail because of their complexity and

unknown contribution to growth. These parameters include the flow dynamics inside

the furnace (they are influenced by the geometry of the furnace and position of

substrates, etc.), substrates (cleanliness, oxide quality, etc.), catalysts (size distribution,

24

quality, etc.), precursor (amount, quality, condition, etc.), and even the “seasoning” of

the growth chamber (for example, possible residues from prior growth). We will not

discuss the effect of these parameters in detail, but the reader should be aware of their

importance.

1.6 Topological insulators

1.6.1 Introduction

Topological insulators are materials that are insulators in the bulk but have

conducting surface states. Figure 1.13 shows a schematic representing the band

diagram of a general topological insulator. The diagram looks similar to that of a

regular semiconductor in that the Fermi level is located in the band gap between the

conduction and valence bands. However, topological insulators also have surface (in

3D) or edge (in 2D) states as described by the blue and red lines inside the band gap.

These surface states are due to strong spin-orbit coupling, and exhibit extraordinary

properties such as dissipationless current and the quantum spin Hall effect. A more

detailed explanation of topological insulators is beyond the scope of this dissertation

and we refer the reader to several excellent papers72-78.

25

Figure 1.13: Schematic of simplified topological insulator band diagram.

The short history of topological insulators has been interesting and impressive

in many ways. While most major findings in solid state physics were discovered

experimentally, topological insulators were first conceived theoretically in 2005, and

quickly became one of the hottest topics in physics. Science Magazine voted

topological insulators a runner-up breakthrough of the year. Philip Anderson

(Princeton, 1977 Nobel Prize in physics) believes that “as a technical achievement, or

a series of physics achievements alone, it is pretty spectacular.” Thomas Rieker

(program director for the National Science Foundation's Materials Research Science

and Engineering Center) claims that “this discovery has the potential to transform

electronics, data storage and computing.” Hari C. Manoharan (Associate Professor of

Physics at Stanford) states that work on topological insulators “could lead to

breakthroughs in areas as diverse as spintronics, quantum information and even

particle physics — theorists have predicted that an exotic type of particle that is its

26

own antiparticle (a Majorana particle) might be observed in topological insulators79.”

Stanford Professor of Physics Shoucheng Zhang believes that topological insulators

have the potential to replace the standard CMOS transistors that power today’s

integrated circuits with more efficient switches that dissipate significantly less power.

These statements from leading experts demonstrate the tremendous potential that

topological insulators offer in electronics and other areas. However, the field is still

very young and it is not certain where or when topological insulators will have the

greatest impact.

1.6.2 Materials

While the research for this dissertation was underway, only a handful of

topological insulators were known (BixSb1-x, HgTe, Bi2Se3, Bi2Te3, Sb2Te3). Since

then many other materials80, 81 have been predicted to show topological insulator

behavior. Of the five initial materials, Bi2Se3 was the best choice for our study due to

its fairly simple band structure with the Dirac point located at the Г-point, and its

relatively large band gap of 0.3 eV18, 82, possibly allowing interesting observations at

room temperature. Figure 1.14 shows the crystal structure of Bi2Se3.

27

Figure 1.14: Schematic showing crystal structure of Bi2Se3

It is a layered material that has a rhombohedral phase with the space group

D3d5 (R-3m)83 consisting of planar, covalently bonded quintuple layers linked

predominantly by van der Waals interactions. Each quintuple layer is made up of a

Se-Bi-Se-Bi-Se stack and is approximately 1 nm thick.

1nm

28

2 In situ transmission electron microscopy of phase-

change memory

2.1 Introduction

Phase-change memory (PCM) has been researched extensively as a promising

alternative to flash memory. Important studies have focused on its scalability46-48, 84, 85,

switching speed46, 86, 87, endurance46, 48, 87 and new materials46, 86-88. Still, reliability

issues and inconsistent switching in PCM devices motivate further study of its

fundamental properties. However, many studies treat PCM cells as black boxes;

microstructural changes inside the devices remain hidden. Here, using in situ

transmission electron microscopy (TEM), we observe real-time microstructural

changes in lateral Ge2Sb2Te5 (GST) PCM bridges during switching. We find that

PCM devices with similar resistances can exhibit distinct threshold switching

behaviors due to the different initial distribution of nanocrystalline and amorphous

domains, explaining the variability in switching behaviors of PCM cells in the

literature. Our findings show a direct correlation between microstructure and

switching behavior, providing important guidelines for the design and operation of

future PCM devices with improved endurance and lower variability.

While the atomistic mechanisms involved in the phase transition have been

studied in great detail89-94, few experiments show what the microstructures of the

amorphous or crystalline regions look like before and after switching. However,

knowing the detailed microstructure is important because it relates directly to

29

electrical properties such as the electric field required to induce threshold switching,

sources of variability in switching and the origin of resistance drift95, 96, a common

problem with PCM in which the resistance of the amorphous phase increases over

time. In an effort to study the microstructure, researchers have obtained TEM images

by cutting individual PCM cells out of a chip with a focused ion beam71, 97, 98. This

technique provides a snapshot in time of individual cells, but it does not allow for a

direct correlation between microstructure and electrical behavior during repeated

switching. A variety of in situ TEM techniques have been shown to be very powerful

for understanding the microstructure evolution during dynamic processes such as

electroplating62, nanowire63-65 and nanocrystal66 growth.

Figure 2.1: Device schematic showing silicon substrate with 50nm silicon nitride membrane suspended at the center and GST bridge that is fabricated on the membrane.

We have developed a fabrication process for making single nanostructure

electrical devices on TEM membranes29, 99. In our effort to study the detailed

microstructure during repeated switching, we fabricated 50 nm thick lateral GST

phase-change memory bridge devices on 50 nm thick Si3N4 electron transparent

membranes (Figure 2.1) allowing us to directly correlate electrical behavior with

structural changes by in situ TEM. This geometry forces the current through a narrow

30

region of GST so that sufficient heating to induce the phase change occurs in the

bridge only, while the rest of the device remains crystalline at all times.

2.2 Method

2.2.1 Device fabrication

We used standard 50nm low stress silicon nitride membranes on 200μm silicon

frames as substrates (described in Section 1.4.3). Photo-lithographically patterned

100nm gold contacts were evaporated onto the substrates to allow contact to be made

between the in situ electrical biasing holder and the PCM bridges (Figure 2.1). We

used electron beam lithography and a lift-off technique to pattern the TiN contacts and

Ge2Sb2Te5 cells. Both films were sputtered from stoichiometric targets in an AJA

sputtering system. After lift-off, the devices were encapsulated by a 20nm layer of

sputtered SiO2 (Figure 2.2).

Figure 2.2: Schematic of device cross-section (left) and SEM image of complete device (right). Cross section is taken along the dotted line.

The as-sputtered Ge2Sb2Te5 layer is amorphous leading to a prohibitively high

threshold voltage that can lead to destructive switching. In order to allow repeated

31

switching, the devices have to be annealed initially to achieve full crystallization. In

later switching, the amorphous region will be confined to the bridge region only,

resulting in smaller threshold voltages. To achieve initial crystallization, we heated

the completed devices on a hot plate at 180ºC for 10 minutes.

2.2.2 TEM characterization of sputtered GST films

To confirm the composition of our films and characterize their crystal structure

we performed careful energy dispersive x-ray spectroscopy (EDS) and diffraction

analysis of sputtered GST films that were annealed on a hot plate at 180°C for 10 min.

Figure 2.3: TEM characterization of annealed GST 225 film after sputter deposition. a, b. Bright-field TEM images of as-deposited GST225 film after annealing on hot plate at 180ºC for 1 min. c. SAD of GST225 film could be accurately indexed to face-centered cubic phase of GST225. No other phase was observed.

Figure 2.3 shows bright-field TEM images of the sputtered GST225 film after

annealing it on a hot plate at 180°C for 1 min. The grain size shows a distribution that

ranges from a few nanometers to 10s of nanometers. The EDS analysis indicated an

elemental composition of 21, 21, 58 at.% of Ge, Sb, Te respectively. GST225 has a

nominal composition of 22, 22, 56 at.%, so the data agrees with GST225 to within the

experimental error of ~2%. We also took selected area diffraction (Figure 2.3c). The

rings could be accurately indexed to the cubic phase of GST225. Figure 2.4

32

summarizes the d-spacings calculated from the measured ring diameters and compares

our values to the theoretically predicted values.

Figure 2.4: Table summarizes measured and theoretical d-spacings of cubic phase of GST225

Each ring agrees very nicely with theory to within approximately 1%. The rings

corresponding to the (331), (531), and (533) planes could not be observed, because

they either had low intensity or due to their close proximity to a different, brighter ring.

Besides the cubic phase of GST225, no other phase was observed. The EDS and SAD

analysis corroborates a composition of Ge2Sb2Te5 to within approximately 2 atomic

percent experimental error.

2.2.3 Electrical testing and TEM observation

The completed devices were glued with carbon paste into a custom electrical

biasing holder. Electrical contact was established via wire bonding between the holder

and the gold pads on the substrate. We used a FEI Tecnai G2 F20 X-TWIN TEM

33

operating at 200kV for in situ observation. Pulsing and IV measurements were

performed with a B1500A parameter analyzer (Agilent Technologies) equipped with a

B1525A semiconductor pulse generator unit. For amorphization pulses, the rise and

fall times were 20ns; for crystallization pulses, the rise and fall times were 500ns.

Resistances were measured at a bias of 0.1V. To ensure that the switching

measurements were not influenced by the TEM’s electron beam, we performed several

control experiments in which we switched repeatedly with the column valve closed

(i.e. electron beam blocked by the column valve). We did not find any significant

difference in switching behavior with the column valve closed or open and conclude

that the electron beam plays no significant role in our measurements

2.3 Switching experiments

2.3.1 Demonstration of in situ switching

To demonstrate the feasibility of in situ switching, we applied short voltage

pulses to devices in the crystalline state inside the TEM. Figure 2.5a shows an

example of a crystalline bridge with typical resistances ranging from 20 kΩ to 80 kΩ.

The bridge consists of a large number of micro-crystalline grains with different

orientation resulting in a granular contrast in the TEM. After applying a 5 V, 400 ns

square pulse, the bridges turned to an amorphous state (Figure 2.5b) and the resistance

increased to between 0.5 and 8 MΩ.

34

Figure 2.5: TEM images and SAD of switching between crystalline and amorphous phase. a, b. TEM image showing a bridge in the (a) crystalline phase and (b) amorphous phase. Typical resistance values for crystalline and amorphous phases are 10-100 kΩ and 1-8 MΩ respectively. The red circles indicate the approximate size and location of the selected area diffraction aperture in (c) and (d) c, d. Selected area diffraction confirming the poly-crystalline (c) and amorphous (d) nature of the bridge.

The amorphous region shows a uniform contrast in the TEM. The edge of the PCM

bridge looks darker, because a thin side-wall of GST remained from the lift-off

fabrication process. Switching back to the crystalline phase was accomplished by

either a longer voltage pulse (2 V, 1-10 μs) or a current scan to approximately 100 μA.

To examine the crystallinity, we obtained selected area diffraction (SAD) of both

states (Figure 2.5c and d). The spot pattern (Figure 2.5c) is due to a polycrystalline

phase and the diffuse ring (Figure 2.5d) is due to an amorphous phase. The SAD

confirms that the phase change was successful.

35

2.3.2 Two types of amorphous phases

After switching back and forth dozens of times, we observed that applying an

amorphization pulse did not always result in a single-phase amorphous domain as

depicted in Figure 2.5b100. Instead, about half the time we obtained a two-phase

amorphous/crystalline domain consisting of an amorphous matrix with nanocrystals

interspersed throughout the bridge. Interestingly, despite the significant differences in

microstructure, in both cases the resistances were on the order of MΩ.

Figure 2.6: TEM images of two-phase amorphous/crystalline (left) and single-phase amorphous (right) domain. The coloring is provided to give an estimate of where the material is amorphous.

The two-phase amorphous/crystalline and single-phase amorphous domain in the same

cell are depicted in Figure 2.6. We will address the consequences of this variability in

the character of the amorphous phase for the switching properties of the cell in the

following sections. However, first it makes sense to look at the dependence of the

phase transition on the applied voltage pulse to understand how the crystalline to

amorphous transition occurs and why we observe differences in the microstructure.

36

2.3.3 Dependence of phase transition on voltage pulse

A typical PCM experiment often performed as a general characterization

method is to measure the resistance as a function of the voltage pulse amplitude. In

such an experiment, a cell in the crystalline state is subjected to progressively higher

pulses. After each pulse, the resistance is measured. If the resistance increased, the cell

is re-crystallized (via a longer pulse or current scan) before the next, larger pulse is

applied. Figure 2.7 shows a schematic of what the data from such an experiment

might look like.

Figure 2.7: Schematic of typical PCM resistance as a function of applied voltage pulse. The crystal structure between crystalline and amorphous is unknown.

At first the resistance remains constant at a low value independent of the applied

voltage pulse, because the cell is completely crystalline and the voltage pulse is too

small to cause sufficient heating to melt the material. When the voltage amplitude

reaches a certain value, the resistance starts to increase. In this transitional region, it is

believed that a small amorphous phase is formed that increases the resistance53, 101: the

higher the voltage pulse, the larger the amorphous region. Once the voltage amplitude

37

becomes large enough, the resistance levels off at a high value, because essentially the

whole region has been transformed to the amorphous state. We indicated the

transitional region in Figure 2.7 with a ‘?’, because it is not clear what the distribution

and size of the amorphous region looks like. However, our in situ TEM method

allows us to observe the detailed microstructural changes during such an experiment.

To observe the details of the crystalline to amorphous transition and to

measure the dependence of resistivity on the voltage pulse, we applied 400 ns pulses

of increasing amplitude to a crystalline device with an initial resistance of 62 kΩ.

Whenever the resistance exceeded 200 kΩ we re-crystallized the device by scanning

the current to 100 μA before pulsing it again. The curve shows typical PCM behavior:

the general trend is that a higher pulse amplitude leads to a higher resistance which

correlates with increasing amorphization (Figure 2.8a). However, we note that the

resistance value depends not only on the pulse amplitude, but also on the history of the

sample. For example, we observed a range of resulting resistances after switching the

same cell from the low resistance state with a 4 V pulse (see multiple data points at 4V

in Figure 2.8a).

38

Figure 2.8: Study of crystalline to amorphous transition. a. Resistance as a function of applied 400 ns voltage pulse. TEM images (b), (c), (d) correspond to labeled points in the graph. The red line connects the average resistance values. b-d. TEM images of phase-change memory bridge. Red ellipses show examples of crystals that survived the amorphization process. Yellow circles point out small differences in microstructure that accompany resistance change from 62 kΩ to 150 kΩ. Green dashed lines surround areas that are predominantly amorphous.

To capture the detailed changes in microstructure as we increased the pulse

amplitude, we recorded TEM images of the device after consecutive 3 V, 4 V, and 5 V

pulses (Figure 2.8 b-d). Figure 2.8b shows that the bridge was predominantly

crystalline with a large spread in crystal size. After a 4 V pulse, the resistance

increased from 62 to 150 kΩ, but the change in crystallization was only minor (Figure

2.8b,c). After a 5 V pulse (Figure 2.8d), the resistance increased to 3 MΩ and a

significant amorphous domain (indicated by green dashed line) interspersed with

crystals was observed. To confirm the crystallinity of the bridge we also obtained

SAD patterns at various resistances after switching (Figure 2.9).

39

Figure 2.9: TEM images and corresponding SADs of bridge in various states of crystallization. a-c. Resistance of bridge in (a), (b), and (c) is 29 kΩ, 115 kΩ and 3.5 MΩ, respectively. The SADs (approximate size and location of SAD aperture indicated by red circles) confirm the crystallinity of the states. In (a), the bridge is essentially completely crystalline as indicated by spot pattern and absence of diffuse ring. In (b), the bridge has some amorphous background, but the SAD still shows significant crystallinity. In (c), the bridge is predominantly amorphous with several crystals remaining embedded in the amorphous matrix.

From these images (Figure 2.8, 2.9) it is evident that resistance is not simply

proportional to the size of the amorphous region, but is rather a function of the

detailed crystalline and amorphous phase distribution that determines available

conduction pathways. During pulsing, the larger crystals (~100 nm) survived the

melting (some examples indicated by red ellipses in Figure 2.8 b-d). Smaller crystals

(<50 nm) tended to disappear (yellow ellipses in Figure 2.8 b,c) after the 4 V pulse

indicating that some degree of vitrification occurred. After the 5 V pulse, significant

melt-quenching occurred, but the pulse was still insufficient to completely melt the

40

larger crystals (Figure 2.8d, area encompassed by green line is mostly amorphous).

The device could be switched back and forth dozens of times with the same crystals

remaining in the amorphous phase. However, after a longer pulse (500 ns), the larger

crystals were more likely to melt completely and in that case the amorphous domain

showed only little crystalline residue even after repeated switching.

Careful study of Figure 2.8 b-d suggests that besides the pulse amplitude and

time, the microstructure of the final high resistance state is influenced by factors such

as the initial size distribution of crystals and available current paths that can lead to

localized heating and melting. While the resistance of the amorphous state was

always in the MΩ range, we expect that its microstructure can have significant impact

on the electrical behavior. For example, the crystals that remain in the amorphous

phase may be a source of resistance drift, as Ostwald ripening and structural relaxation

can further change the conductivity. These detailed observations demonstrate the

value of the in-situ TEM technique and reveal the complex relationship between

voltage pulses and crystallinity as a source of the variability that is often observed in

PCM measurements.

2.4 Threshold field measurements

2.4.1 Significance of threshold field

Since our method allows for an accurate observation of the amorphous phase,

it is ideal for measuring the electric field required to induce threshold switching and

for studying the effect of the microstructure (i.e. presence or absence of a second

phase in of the amorphous domain) on threshold switching. Knowing the exact

41

threshold field is important for the design and operation of PCM cells. Especially

lateral cells have shown a dramatic dependence on this parameter, because large

electric fields can result in current spikes and destructive switching46.

Figure 2.10: Current spikes and destructive switching due to high threshold field. Voltage response graph indicating the point at which breakage can occur (left). TEM image of a sample broken during crystallization attempt (right).

If the electric threshold field is too large, a large threshold voltage must be

applied. Figure 2.10 (left) shows a schematic of a corresponding current scan. As

always, to SET the cell, a voltage larger than Vth has to be applied. If Vth is too large

this leads to a current spike and destructive switching. Figure 2.10 (right) shows an

SEM image after a device broke from a too large voltage pulse.

2.4.2 Measurements

To measure the threshold field, we first switched the bridge to the high

resistance state, then estimated the size of the amorphous region and performed a

current scan while measuring the applied voltage to re-crystallize the bridge. As

expected, we observed two different behaviors in the voltage response to the current

scan corresponding to the presence or absence of a crystalline phase within the

amorphous domain.

42

Figure 2.11: Measurement of threshold switching in high-resistance bridges. a. IV curve showing 120 μA current scan to crystallize the bridge in (c) and to measure the threshold voltage. This bridge exhibits complicated switching behavior. b. IV curve showing 100 μA current scan to crystallize the bridge in (d) and to measure the threshold voltage. This bridge exhibits clean switching behavior. c. TEM image of two-phase bridge before current scan (a). The approximate amorphous domain has been colored blue to assist with visualization. d. TEM image of single-phase amorphous bridge before current scan (b). The approximate amorphous domain has been colored red to assist with visualization.

If there is very little crystalline residue inside the bridge (Figure 2.11d), then the IV

curve shows a very clean snapback52, 96 once the threshold voltage is reached (Figure

2.11b). At this point the entire bridge is in a dynamic conductive state that is still

amorphous with a current of 10 μA (corresponding to a current density of

approximately 0.04MA·cm-2); too low to produce significant heating. As the current is

further increased, the bridge crystallizes. The IV-curve in Figure 2.11b shows that the

voltage in the up-scan is close to the voltage in the down-scan indicating that the

43

resistance in the dynamic on-state immediately after threshold switching is very

similar to the resistance after crystallization has taken place.

In contrast to the threshold switching for a pure amorphous phase, the required

voltage to switch a mixed phase (Figure 2.11c) is much lower (Figure 2.11a) and the

IV curve does not show a clean snapback. These differences in electrical behavior are

due to the complicated crystalline-amorphous mixture that can lead to conductive

percolation paths when the threshold field is exceeded locally. Once such a

conductive filament exists, it causes local heating and crystallization resulting in

continuously decreasing resistance of the bridge (Figure 2.11a). The complexity of the

IV behavior suggests that several processes, including threshold switching, localized

heating and crystallization, may be occurring simultaneously. When switching back

and forth several times, we found that the threshold voltage is nearly constant at about

2V regardless of the resistance for a mixed amorphous phase with crystalline residue

(Figure 2.11a). Figure 2.12 shows two more examples of threshold switching

correlated with crystal structure.

44

Figure 2.12: Measurement of threshold switching in high-resistance bridges. a. IV curve showing 120 μA current scan to crystallize the bridge in (c) and to measure the threshold voltage. This bridge exhibits complicated switching behavior, because it is in a mixed phase consisting of nanocrystals interspersed in an amorphous matrix resulting from repeated switching. b. IV curve showing 100 μA current scan to crystallize the bridge in (d) and to measure the threshold voltage. This bridge exhibits complicated switching behavior, because it consists of nanocrystals interspersed in an amorphous matrix. c. TEM image of mixed phase bridge before current scan (a). d. TEM image of bridge before current scan (b). The bridge consists of nanocrystals that are due to annealing the as-deposited GST225 film after sputtering at 180°C for 10 min. Despite the high density of nanocrystals, the resistance of the bridge is still very high indicating that no connected current paths exist. The nanocrystals appear as dark dots in TEM images due to the diffraction contrast. Amorphous regions exhibit uniform contrast.

Direct correlation between the IV behavior and the detailed microstructure of

the amorphous phase using in situ TEM can explain the source of variability that is

observed in conventional current scans. With this knowledge one may form

conclusions about the microstructure of the amorphous domain simply by studying the

current scan response.

45

2.4.3 Discussion

If the amorphous phase is relatively pure as shown in Figure 2.11d, the

threshold voltage scales nearly linearly with the resistance. This is expected, since the

resistance and required threshold voltage should be dominated by the smallest

separation distance between the two crystalline regions at the contacts (Figure 2.13b).

By measuring the shortest distance between crystal paths in the pure amorphous

domain, damorph, and comparing it to the threshold voltage, we estimate the required

threshold field to be on the order of 4 V·μm-1. This number appears to be largely

independent of the cross-sectional area of the sample for the geometries studied here.

The plot of the threshold voltage as a function of the estimated size of the amorphous

domain provides a near linear relationship for the pure amorphous phase (Figure

2.13a). Since the mixed amorphous phase is interspersed with nanocrystals and

depends on percolation paths rather than the electric switching of a clearly defined

region, it is not possible to measure a corresponding damorph.

Figure 2.13: Dependence of threshold voltage on amorphous region parameters. a. Threshold voltage as a function of the size of the amorphous region. b. Threshold voltage as a function of resistance in the amorphous state. We observe two different behaviors depending on the microstructure of the amorphous phase.

46

Interestingly, the threshold field we measured for the pure amorphous GST

phase is about 1 order of magnitude below that reported by other groups who have

investigated PCM bridges46, 102. While the origin of this disparity is still under

investigation, we note that the lower threshold field was necessary for reversible

switching. Groups who measured larger threshold fields found the resulting threshold

voltage for Ge2Sb2Te5 prohibitively high as it prevented switching from the

amorphous to the crystalline state for all but very short bridges46. This observation

further corroborates the need for careful studies that can pinpoint the variables that

affect the threshold field.

Factors contributing to measured threshold field variability between research

groups include differences in materials composition, processing, cooling-rates during

melt-quenching and the difference between melt-quenched and as-deposited

amorphous phases. In our study, the correct Ge2Sb2Te5 composition was confirmed to

within experimental error via energy dispersive x-ray spectroscopy and diffraction

analysis, as detailed in Section 2.2.2. Since etching can change the material properties

of GST, we avoided etching and used a lift-off process instead. Similarly, we took

precautions to limit exposure of the phase-change bridge to solvents. Further studies

are needed to find the root cause of the variation in threshold fields. Our in-situ

technique may be a useful tool in this investigation, because it allows a direct

correlation between the electrical and structural properties. A better understanding of

the contributing factors may allow the material to be engineered to obtain threshold

fields that result in desirable threshold voltages for a given device geometry.

47

2.5 Control of quenching rate via device geometry

The above discussion of differences in amorphous phases and threshold

switching shows that the microstructure of the high resistance state is very important

for device performance. It is likely that a single-phase amorphous domain will show

less resistance drift and longer data retention, because it offers fewer nuclei that can

grow or be affected by Ostwald ripening. However, the mixed phase can be switched

with less energy, because it requires a smaller volume to be programmed. In an effort

to control the microstructure of the amorphous phase, we manipulated the quenching

rate by changing the device geometry.

Figure 2.14: Switching of high-aspect-ratio bridge with lower cooling rate. a. Crystalline bridge obtained after 60μA current scan; resistance is 18 kΩ. The red circles show examples of large crystals embedded inside the bridge. b. The same bridge after applying a 2V, 300ns pulse. The red circles show that the crystals in (a) disappeared, indicating that the bridge melted completely as a result of the pulse. However, the bridge did not turn completely amorphous but re-crystallized due to the lower cooling rate. c. The same bridge after 60μA current scan; resistance is 29 kΩ.

48

We created a narrower (200 nm), longer (1.5 μm) and thicker (100 nm) bridge. This

results in a slower quenching rate since the main heat dissipation is through the

contact leads. Figure 2.14a shows the bridge in a low resistance (18 kΩ) crystalline

state. After a 2 V, 300 ns voltage pulse, the resistance increased to 176kΩ. While the

domain is still highly crystalline (Figure 2.14b), the disappearance of the larger

crystals indicates that the complete bridge melted and then partially re-crystallized due

to the extended cooling time (red circles in Figures 2.14a,b show examples of the re-

crystallization). The bridge could be returned to its low resistance state by applying a

60 μA current scan. As expected, the bridge could not be switched to a single-phase

amorphous domain by increasing the voltage pulse. Figure 2.15 shows the complete

switching sequence of the NW shown in Figure 2.14.

Figure 2.15: Repeated switching of high-aspect-ratio bridge with lower cooling rate. The figure shows TEM micrographs of the bridge after repeated switching. 60 μA current scans were used for crystallization and various voltage pulses were applied to achieve a high-resistance state. The series of TEM images with the resistance shows that the bridge could not be turned completely amorphous by a 2V, 300ns pulse. Increasing the pulse further resulted in the bridge breaking.

49

Increasing the duration of the 2V pulse did not lead to complete amorphization, and

increasing the pulse amplitude resulted in sample breakage. This experiment

demonstrates that by changing the bridge geometry we can decrease the cooling rate

and thereby increase the crystal residue of the high resistance state.

Similarly, we can achieve pure amorphous phases by using bridge geometries

designed for higher cooling rates (Figure 2.16).

Figure 2.16: TEM micrograph of a short bridge that can achieve a single-phase amorphous domain without crystalline residue due to its fast cooling-rate.

Accurate control of the cooling-rate is necessary to achieve optimized switching

performance. Our experiments show that changing the device geometry provides

control over quenching rate, which in turn determines the level of amorphization

achieved. Further studies are needed to optimize these parameters.

2.6 Conclusions

In situ TEM studies of PCM cells during switching enable observation of the

phase-change with unprecedented accuracy. In this chapter we presented the first

50

observation of in situ switching of lateral phase-change cells inside the TEM. Using

this technique, we found significant variability in the microstructure of the high-

resistance phase, which explains the observed variability in PCM switching

measurements. In particular, we observed two distinct high-resistance states: a single-

phase amorphous domain and a two-phase amorphous/crystalline domain. The

technique we developed will allow further investigation of current topics in PCM such

as the source of resistance drift, scaling behavior, properties of other phase-change

materials, failure mechanisms, and multi-state switching. The observations presented

here as well as future studies using this technique can assist in the design of improved

PCM cells.

As mentioned in chapter 1, it is not clear yet which material will ultimately

show the best properties for PCM. Interestingly, studies on NWs have shown

exceptional performance48. Hence, the next chapter will discuss the synthesis of

phase-change NWs and our in situ study of these NWs will be covered in chapter 4.

51

3 Synthesis and characterization of GeTe phase-

change nanowires

3.1 Introduction

3.1.1 Motivation and overview

While PCM is gaining increasing momentum in academia and industry, the

phase-change materials that make up the active region of the device are still

extensively researched because materials optimization is a straightforward way to

improve the performance of any PCM cell. In this search for ever-better materials,

nanowires can play an important role: synthesis via the VLS growth is relatively

simple, they can be made into nanoscale dimensions without the need for multi-

million dollar equipment, and they offer high quality single crystalline geometries

without etching damage or other defects. Hence, NWs are an ideal platform for

fundamental studies. In this chapter, we report the synthesis and characterization of

GeTe and Sb2Te3 nanowires via the vapor liquid solid growth.

3.1.2 Background

Since the early 1990s, the vapor-liquid-solid (VLS) mechanism has been

demonstrated to be a general method for producing one-dimensional IV, II-VI, and III-

V semiconducting NWs19, 20, 103-107.6-12 Gold nanoparticles are often used as universal

catalysts to define the growth of these materials into wire-like structures. As part of

this dissertation research, we were the first group to report the VLS growth of Sb2Te3

52

NWs. The VLS synthesis of GeTe NWs was developed by us and by a group at

Harvard that worked on the same topic independently around the same time. Since the

development of the growth, phase-change NWs have been investigated by several

groups. We will briefly discuss their progress in section 3.5.

3.1.3 Materials

Here we exploit the VLS growth of GeTe (GT) and Sb2Te3 (ST) phase-change

NWs with Au nanoparticles as catalysts. The family of Ge-Sb-Te (GST) materials is

one of the most important systems for phase-change memory devices due to their rapid

reversible change between crystalline and amorphous states at temperatures

compatible with those of device applications (for Ge2Sb2Te5, melting point 610 °C,

glass transition temperature 350 °C)44. Germanium telluride and ST are two important

members of this family of materials and function as the basis to fabricate GST ternary

alloys. They have melting points of 716 and 630 °C108 and amorphous phase

crystallization temperatures of 145109 and 77 °C110, respectively.

Figure 3.1: Quasi-binary phase diagram between Au and GeTe (Sb2Te3). The shape of the drawing represents the phase diagram between Au and GeTe. However, the general shape for Au and Sb2Te3 is very similar and the diagram can be used for showing the temperature values and general shape.

53

To investigate the fundamental possibility of a VLS growth with Au as the catalyst,

we have looked into the phase diagrams of the materials involved. As shown in Figure

3.1, GT and ST have a relatively simple bulk quasi-binary phase diagram with Au.

The eutectic temperatures are at 480 and 454 °C, respectively108. The phase diagram

suggests that it is possible to use Au nanoparticles for VLS growth of GT and ST

phase-change NWs. Due to the small size of Au nanoparticles, the eutectic

temperature might be lower than the bulk value111.

3.2 Synthesis

Nanowires were grown on silicon (100) substrates with a native oxide that

were functionalized with 0.1% w/v aqueous poly-L-lysine solution (Ted Pella) and

dipped into 50 nm diameter Au colloid solution (Ted Pella). The negatively charged

Au nanoparticles stick to the positively charged poly-L-lysine. The substrates were

positioned downstream in a 1 in. diameter horizontal tube furnace (Lindberg/Blue M)

with the source material, finely ground GeTe (Alfa Aesar, 99.999%), placed in the hot

center region. A 5% H2 in N2 gas mixture acted as a carrier gas to transport the vapor

to the colder furnace region. Before each experiment, the quartz tube of the furnace

was evacuated to <100 mTorr and flushed with the carrier gas repeatedly to decrease

oxygen contamination. NWs were grown at ~50 Torr and ST NWs were grown at 1

atm pressure. Typical parameters for the NW synthesis were as follows: temperature T

= 450 °C, duration time t = 1-6 h, and carrier gas flow rate = 150 sccm. The best

growth was achieved in the colder region within 5-10 cm of the furnace end with a

temperature of ~370 °C. Samples were also grown on 50 nm thick Si3N4 membranes

54

for further studies with a TEM.

3.3 Characterization

3.3.1 Scanning electron microscopy

The NW surface morphologies were examined in an FEI Sirion scanning

electron microscope. The SEM is the first step in characterizing any growth product,

because it allows for visual inspection of the as-grown NWs. Figure 3.2 shows two

SEM images of GeTe NWs at various magnifications to show the density and size

distribution of the NWs. We found that the NWs can grow 10s of micrometer long

with a diameter distribution ranging from approximately 20-300 nm.

20 μm 2 μm20 μm20 μm 2 μm2 μm

Figure 3.2: SEM image of as-grown GeTe nanowires, showing high density and a wide size distribution with diameters ranging from roughly 20-300 nm.

The thicker NWs (~200 nm diameter) can even grow 100s of micrometer long (Figure

3.3) if the growth time is increased sufficiently (~ 4hrs). The fact that the diameter of

55

the NWs is significantly larger than the starting catalyst size is believed to be due

catalyst growth from Ostwald ripening or the absorbance of GeTe precursor112.

Figure 3.3: SEM image of long GeTe nanowires.

The large size distribution observed here is good for switching experiments,

because it allows for the study of scaling properties49. However, for future device

fabrication, it is important to obtain good control. Progress has been made by other

groups112.

Interestingly, using the same method we could also grow GeTe helical

nanowires (Figure 3.4) without significantly changing the growth parameters. Straight

and helical structures can coexist in the same NWs as shown in the Figure 3.4

(bottom), suggesting that they have the same chemical composition and crystal

structure.

56

Figure 3.4: SEM image of as-grown GeTe helical and straight NWs (top). The nanowires can randomly change from straight to helical growth (bottom).

Both left- and right-handed helical NWs can be found with equal probability. Similar

helical structures have been observed in many other systems such as ZnO113, Si114,

SiO2115, boron carbide116, amorphous carbon117 and ZnGa2O4

118. Several mechanisms

have been proposed that can form helical structures including growth influenced by

the surface tension between the catalyst and nanowire, the presence of internal

structural defects, electrostatic attractions between polar surfaces, and a

57

crystallographic superlattice structure119. However, since the helical NWs are not

important for our phase-change memory investigations, they are mentioned here for

completeness only and will not be discussed further.

3.3.2 Transmission electron microscopy

To study the crystal structure and composition of the NWs we transferred them

from the growth substrate onto carbon film supported by Cu grids by carefully

touching the two substrates together. Alternatively, we sometimes directly grew NWs

on Si3N4 membranes. To characterize the chemical compositions and structures, we

carried out EDX, HRTEM and electron diffraction (ED) studies. As expected, the

straight and helical NWs show the same chemical composition and crystal structure.

Figure 3.5: TEM image of a straight GeTe NW. The dark spot at the tip corresponds to a gold nanoparticles. Insets show a high resolution image of the lattice fringes and SAD pattern.

Figure 3.5 shows a TEM image of a GeTe NW. The dark spot at the tip

corresponds to a gold nanoparticle suggesting that the growth was indeed via the VLS

mechanism. Most nanowires have the catalyst directly at the tip120; Figure 3.5 shows

58

another common configuration where the particle is at the side of the NW. It is likely

that during growth the catalyst is directly at the tip and thereby defining the diameter

of the NW. During cool-down of the furnace, the conditions inside the growth

chamber change significantly and the catalyst may move to the side of the wire and

stay there upon solidification. The movement of the catalyst to the side of the

nanowire may offer some insights into the mechanism to form helical nanowires, but

further studies for this are necessary.

The HRTEM and SAD (inset of Figure 3.5) can be indexed to the

rhombohedral crystal structure of GeTe with a growth direction of [-111]. However,

we also found [-102] as a growth direction in other GeTe NWs120.

Figure 3.6: TEM Energy Dispersive X-ray Spectroscopy of GeTe NW.

The EDX analysis confirmed a composition of 50% Ge and 50% Te, each within 2%

experimental error, for both straight and helical NWs (Figure 3.6). The Cu and C

signals in the EDX spectrum are from the TEM grids.

59

3.4 Sb2Te3 nanowires

We also developed the VLS growth method for Sb2Te3 NWs, because it is

another technologically relevant material: doped Sb2Te3 has shown excellent

switching properties in lateral PCM cells46 and it can be regarded as a building block

of GST. Figure 3.7 shows a typical SEM image of as-grown Sb2Te3 NWs.

Figure 3.7: SEM image of as-grown Sb2Te3 NW

The growth conditions were similar to the ones used for GeTe (Section 3.2), but

Sb2Te3 grew preferentially within 10 cm of the hot center region of the furnace, where

the temperature was ~440 °C. The as-grown NWs have diameters ranging from

approximately 50-100 nm and lengths of approximately 1-2 μm. We also confirmed

the composition of these NWs with EDX elemental mapping120.

3.5 Recent developments and future outlook

Since the development of the VLS technique as a simple synthesis method for

60

phase-change NWs, the field has been moving rapidly thanks to the contributions of

several groups researching phase-change NW structures. This includes work on the

improvement of growth control112, study of phase-change nanowire switching

properties48, 49, new materials such as GST48 and In2Se3121, and studies of threshold

switching53 and drift122. The progress in this field makes it likely that phase-change

NWs will remain an important platform for fundamental study in the near future.

However, it is not yet clear whether they will ever be important for technological

applications. The IC industry is still looking beyond the scaling of their standard

transistor designs, and nanowires are considered one possible option to replace current

transistors. If a paradigm shift towards nanowires and self-assembled electronics

comes to pass, phase-change nanowires may also play an important role.

Figure 3.8: Schematic of phase-change memory that is based on NWs. The phase-change NW is

grown on top of a p-n junction for selective addressing.

For example, if Silicon NWs were integrated into future circuits one could easily

incorporate phase-change nanowires to make essentially self-assembled memory

elements. Figure 3.8 is a schematic that outlines a simple fabrication process for

nanowire based memory. In this process the phase-change material (i.e. GST) is

61

grown directly on top of a p-n junction to form a hetero-structure NW that acts as a

selection device and memory element. While potential applications like these are still

many years away, it will be exciting to see what the future holds beyond the scaling of

regular CMOS structures.

3.6 Conclusion

We have successfully developed the VLS synthesis of single crystalline GeTe

NWs and Sb2Te3 NWs, paving the way for many future studies on scaling and other

current issues in PCM research.

62

4 Electrical characterization and in-situ TEM of

phase-change nanowires

4.1 Introduction

4.1.1 Background and motivation

Over the past several years, phase-change memory (PCM) has shown promise

as a replacement for flash because of its excellent scalability84, 85, high-speed

switching, long lifetime, and nonvolatility123. PCM operates by reversibly driving a

small volume of material through a reversible crystalline-amorphous structural

transition. The phase-change is actuated electrically by joule heating, and the two

states can be easily distinguished due to the substantially lower resistivity of the

crystalline phase compared to the amorphous phase. Cells usually have a vertical

architecture, where the PCM material is sandwiched between two electrodes44. This

structure can be integrated into a CMOS architecture. However, vertical cells do suffer

from one significant drawback for fundamental study: since the active material is

buried inside several other materials, it is difficult to directly probe the structural

transformation inside a working device.

Cells have also been fabricated in a horizontal architecture, where the PCM

material is patterned into a narrow bridge between two electrodes46, 47. This open

structure is much more amenable to characterization. The bridge is not buried under an

electrode, so it is accessible from above with probe microscopy. A careful choice of

63

substrate124 can allow even for direct observation of the structure in a TEM. PCM

NWs developed recently by our group120 and others48, 125, 126 are good candidates for

use as the active material in horizontal line cell studies. NWs present several

advantages over patterned thin films: they have smooth well-defined surfaces, they are

initially in a high quality single crystalline state, and they can be deposited on a

substrate in a single solvent-free room temperature step124. Single crystalline GeTe

NWs are particularly interesting, because their diameter is smaller than grains in

crystallized thin films that are on the order of several micrometers127. The electrical

switching properties of single NWs have been studied and excellent scaling behaviors

have been shown48, 49, but so far few studies have taken advantage of the horizontal

architecture of NW devices for in-depth characterization of the details of the structural

transformation. Of particular interest are the size and location of the amorphous

domains of the devices in the off state, the growth morphology and kinetics of the

crystalline domains when switching to the on state, and other behaviors that may limit

device reliability or dictate the power required to actuate the devices.

In this chapter, we examine phase-change GeTe single-nanowire devices using

ex situ and in situ transmission electron microscopy techniques to directly correlate

nanoscale structural transformations with electrical switching. The results are

surprising: instead of a crystalline-amorphous transformation, the dominant switching

mechanism during multiple cycling appears to be the opening and closing of voids in

the nanowires due to material migration, which offers a new mechanism for memory.

During switching, compositional changes and the formation of banded structural

defects are observed in addition to the expected crystal-amorphous transformation.

64

Our method and results are important for phase-change memories specifically, but also

for any device whose operation relies on a small scale structural transformation.

4.1.2 Device fabrication

For this study we grew single crystal GeTe NWs (explained in chapter 3) and

fabricated a large variety of devices on NWs ranging in diameter from 60 to 200 nm.

Our as-grown NWs were single crystalline with the equilibrium rhombohedral phase

of GeTe.

Figure 4.1: Device fabrication steps outlined with SEM micrographs.

Figure 4.1 outlines the typical fabrication steps used to make single nanowire devices.

Silicon (100) substrates were pre-patterned with photolithographically defined lines

65

(Figure 4.1a). These lines were connected to larger pads, allowing us to make

electrical contact with a probe station or via wire bonding. Next we used electron-

beam lithography to deposit small numbers in the center region of the larger contacts

(Figure 4.1a,b). These numbers acted as a coordinate system that allowed us to

accurately locate each individual nanowire. The nanowires were typically deposited

via a dry transfer. Gently touching the growth substrate with the measurement

substrate was then often sufficient to cause nanowires to break off and adhere. Figure

4.1a and 4.1b show the density distribution of nanowires after such a dry transfer. An

alternative method of nanowire deposition was to sonicate the growth substrates in

solvent. The resulting solution could be deposited on the measurement substrate by

spin coating or drop casting. After depositing the NWs, we determined their position

with the SEM and use electron beam lithography to deposit contacts on the NWs.

Figure 4.1c shows the finished devices. For switching measurements, we also

deposited an encapsulation layer to prevent the nanowires from evaporating. For this

purpose, we used 20-30nm films of Al2O3 or SiO2 deposited via sputtering or atomic

layer deposition.

Figure 4.2: SEM micrographs of GeTe NW with Pt contacts deposited by FIB

66

As an alternative to depositing contacts with EBL, we also used Pt deposition

with a focused ion beam (FIB). The advantages of the FIB method are that it does not

require designing of contacts, and the ion beam can break through a potential oxide on

the surface of the NW. The disadvantage of the FIB is that it is limited to deposit a

low quality Pt that has significant C content. Further, the FIB also causes some Pt

“overspray” that can lead to shorting of the device if the separation is too short. To

prevent shorting, an electrode separation of at least 2 μm was used. An example

device is pictured in Figure 4.2.

In addition to making single NW devices on Si substrates, we used the

methods described in this section to make devices on 50 nm thick Si3N4 membranes

for our in situ observations (Figure 4.6 shows example).

4.1.3 Instruments and measurements

Platinum deposition was performed with an FEI Strata 235DB dual-beam

FIB/SEM with a 10 pA Ga+ ion beam and an accelerating voltage of 30 kV. Scanning

electron microscope (SEM) images were obtained with the same FEI strata or an FEI

XL30 Sirion SEM with FEG source. The EFM image was obtained with an Asylum

MFP-3D AFM. The scan was performed in a noncontact mode at 200 nm tip-sample

separation with one electrode grounded, the other electrode held at 2 V and the tip at -

1 V. The ex situ TEM studies were carried out on an FEI CM20 operated at 200 kV,

and the in situ TEM studies were carried out in a JEOL 3010 operated at 300 kV at the

National Center for Electron Microscopy at Lawrence Berkeley National Laboratory.

Resistance measurements in the lower resistive states were done at a low bias of 0.2 V,

67

the highly resistive states (R > 1 GΩ) were measured at a bias of 2 V.

4.2 Conventional (black-box) switching experiment

4.2.1 Electrical characterization

Before trying to switch the nanowires we characterized their electrical

properties and compared them to those of the bulk material. Figure 4.3 shows an SEM

image of a NW with four contacts on Si-SiO2 substrate and the corresponding

measurement results. Two-point and four-point probe measurements show linear IV

behavior; however, the lower resistance obtained by the four-point probe measurement

reveals that even in the crystalline state, the majority of the resistance comes from the

contacts.

Figure 4.3: Four point probe GeTe nanowires device with corresponding measurement results.

The two-point probe resistance of 26 kΩ is about 50 times higher than the four-point

resistance of 530 Ω, suggesting that the contact resistance is much larger than the

intrinsic NW resistance. The resistivity of the NWs as calculated from the 4-point

68

measurement is approximately 4 × 10-4 Ω·cm, which is consistent with the bulk value

of GeTe128. Our measurements conducted on approximately 50 devices show that the

contact resistance is 5-50 times that of the NW resistance. Occasionally, the contact

resistance is similar to the NW resistance.

Even though our TEM examinations did not show a significant surface oxide

on the NWs (chapter 3.3.2), we believe that the contact resistance is in part due to the

existence of a thin amorphous oxide layer between metal leads and NWs129. To

examine this point further, we conducted in situ TEM heating experiments.

Figure 4.4: TEM image of GeTe NWs during heating experiment.

Interestingly, when heating the NWs we observed sublimation of the GeTe only out of

the NW ends rather than the surface. After sublimation, a thin amorphous shell

remained. From EDX measurements we concluded that the shell was germanium

oxide. This in situ TEM heating study provides direct proof that the GeTe NWs tend

to have a thin surface oxide that can increase contact resistance.

69

4.2.2 Nanowire switching

The NWs can be switched from a low resistance to a high resistance state by

applying a short voltage pulse, as in previous studies49, 125. Despite the intrinsic,

amorphous germanium oxide of 2-5 nm thickness on the NW surface, switching back

to the low resistance state was not observed unless the NWs were coated in an extra

encapsulation layer. Figure 4.5 shows the resistance as a function of switching cycle

for a NW that is 200 nm in diameter and 5 µm in length encapsulated with 20 nm of

SiO2.

Figure 4.5: Cycling behavior. Nanowires were switched to a high resistance state with a 10 V, 200 ns pulse, and switched back to the low resistance state with a 100 mV/s scan up to 5 V.

The NW with an initially low two-point resistance state of 44 kΩ switched to a high

two-point resistance state of 8 MΩ after a short (200 ns) voltage pulse with 10 V

amplitude was applied. Switching back to the low-resistance was accomplished by a

voltage scan to 5 V at a rate of 100 mV/second.

There are several interesting observations. First, the resistance of the device in

the on state drops slightly from the initial on state after the first cycle. This can

probably be explained by annealing of the contacts via joule heating leading to a lower

70

contact resistance. Second, since the resistivity in the crystalline state after the first

cycle is 11 kΩ, the following 10 V pulses correspond to approximately 0.9 mA of

current (or a current density of 2.9 × 106 A/cm2), which is 4 times lower than

previously reported values for a 100 ns pulse49. Lastly, and most significantly, the on-

off ratio of our NW devices after the first cycle is increased with cycling to nearly 107.

While this ratio is similar to what has been found in thin film studies130, a previous

NW study has revealed a rather large diversity of values for the on-off ratio125, from

103 to 107, and our values are at the top of this range.

4.2.3 Investigation of high resistance in off state via SPM

A naive model for the resistivity of the off state might simply take into account

the volume of transformed material. If a cylindrical plug of the NW is transformed to

the amorphous state, one would expect the resistance to scale linearly with the length

of this plug. Given that the value of the resistivity of the amorphous phase is between

103 and 104 Ω·cm128, 130, a measured resistance of 100 GΩ in a roughly cylindrical

wire of diameter 200 nm would require that the wire be more than 30 µm long, even

taking the lower value for the reported resistivity. This observation is interesting,

because our devices have an electrode separation less than 10 µm. Even though the

resistivity of the amorphous GeTe devices in our NWs may be somewhat different

than that reported for the thin films, one would assume that at least the entire device

under measurement must transform to the amorphous phase to generate the measured

resistance. In order to test this prediction, devices switched to a highly resistive state

were examined via electric force microscopy (EFM).

71

Figure 4.6: SEM, AFM, and EFM images of the same single nanowire device in a high resistance state. There is a 2 V difference between the light and dark color in the EFM image.

Figure 4.6 shows three images of a device taken with SEM, atomic force

microscopy (AFM), and EFM (from left to right). The contrast in the EFM image is

due to the local electric potential and topographic features. If the whole NW had

transformed to the amorphous phase as expected, the EFM should show a gradual

change in contrast from one contact to the other as the potential slowly drops through

the uniformly resistive region. The observation in Figure 4.6 is quite different however

as the entire voltage appears to drop over a region tightly confined to the top contact,

suggesting that the structural transformation is localized there, contrary to the

expectation based on the naive model and the measured switching data.

4.3 Direct TEM Observation of NWs during switching

4.3.1 TEM observation of off state

The inconsistency between the predicted size of the amorphous domain and the

72

physical situation motivated us to fabricate the NW devices on electron transparent

silicon nitride membranes (Figure 4.7), making possible direct observation of the

NWs’ structure by TEM25 in the high and low resistance state.

Figure 4.7: Low-magnification SEM image of a silicon nitride membrane (dark) with gold contact pads (light).

Our initial work focused on discovering the structure of the off state after pulse-

induced switching in air without extra encapsulation. During TEM observation, care

was taken to avoid e-beam induced damage.

Figure 4.8: Nanowire before (lower left inset) and after pulsing (main panel); the switching mechanism is identified as void formation at the top contact (upper right inset).

73

The bottom left inset of Figure 4.8 shows a TEM image of a NW device before

switching. The initial two-point resistance was 8 kΩ. The NW was very stable with

respect to 200 ns square pulses with voltage below 10 V. After a 10 V pulse in air,

which resulted in a current of approximately 1.3 mA, the two-point resistance was

changed to 1012 Ω.

Simultaneously, a dramatic structural change was observed close to one of the

contacts (bright contrast in NW, Figure 4.8 (main panel and inset)). We observed a

thin amorphous hollow tube of GeO2 confirmed by EDS, indicating that a void formed

in the NW. GeO2 is highly resistive and even a small segment close to the metal

contact can increase the measured NW resistance up to 1012 Ω, suggesting that void

formation could be a switching mechanism for the devices we measured on regular

silicon chips.

Considerable effort was invested to apply pulses just large enough to raise the

resistance of the NWs without forming voids. A 20 µs voltage pulse of 3.5 V applied

to a device with a large electrode separation of 15 µm and resistivity of 12 kΩ,

resulted in very little change. Only some hillock formation and small pinholes were

observed (Figure 4.9).

74

Figure 4.9: a. TEM image of nanowire after a 3.5V, 200ns pulse shows only minor changes including a bulge and small pinhole evaporation. b. Higher magnification image of bulge and pin hole c. Higher magnification image of pinhole close to contact.

When the pulse was raised to 4.0 V, the resistance increased by a factor of 104 to 100

MΩ and a surprising variety of structural transformations were observed (Figure

4.10a).

Figure 4.10: a. Low-magnification TEM image of nanowire after a long low magnitude pulse. b. Banded planar defect structure. c. Void and amorphous region formation. Area I is amorphous and Area II, like the rest of the wire, is crystalline (SAED inset).

Bulging and small voids are formed as before, but we also observe a periodic set of

planar defects in one section of the NW (Figure 4.10b) that extends over several

75

microns. These defects are likely twin boundaries formed by cooling GeTe from its

higher temperature cubic phase back to the room temperature rhombohedral phase131

and can also be epitaxial misfit upon re-growth of material with different compositions.

Further studies are required to clarify these details. A region that was transformed to

the amorphous state was also observed (Figure 4.10c); its structure was confirmed by

selected area electron diffraction of the slightly brighter area (amorphous, top inset)

and the darker area (crystalline, bottom inset). The amorphous region has also

undergone a composition change as shown in the comparative EDS spectrum in Figure

4.11 (right). The amorphous region has been partially depleted of tellurium, most

likely due to the higher vapor pressure of tellurium at temperatures sufficient to melt

the material. The tellurium may escape the GeO2 shell through small defects formed

during the heating process, since no tellurium rich areas have been observed. It should

be noted that in this case the transformations occurred over large parts of the NW,

because its resistance during the pulse is similar to the contact resistance resulting in

less localized heating. Figure 4.11 (left) shows the voltage pulse and current response

in the NW indicating an initial resistance of only 3.7 kΩ.

Figure 4.11: Pulse applied to NW shown in Figure 4.10a; blue is the applied voltage, and red is the current through the device (left). EDS spectrum of areas I and II from Figure 4.10c (right).

76

These experiments dramatically illustrate the fundamental mechanisms that render

reversible switching in these systems impossible without a high quality encapsulating

shell, but the large number of structural transformations observed in the ex situ study

makes it very difficult to determine which structural transformation is chiefly

responsible for the measured resistance change during reversible switching.

4.3.2 Real-time observation of dynamic behavior: opening and closing of voids

In order to determine which of the observed behaviors was the primary

contributor to the resistance change, the real-time dynamic behavior of the NWs was

observed via in situ electron microscopy of the devices under a voltage scan. Figure

4.12 and 4.13 summarize the surprising results of these measurements with a NW

coated with 20 nm sputtered silicon oxide (Figure 4.12 inset). Figure 4.12 shows the

electrical and structural behavior of the NW beginning with its as-fabricated state with

a low initial two-point resistance of 3.7 kΩ. As in the ex situ observations, the initial

increase in resistance between points I and II in Figure 4.12a was correlated with the

formation of a small void.

Figure 4.12: a. Voltage scan on a single nanowire device (inset). The blue square in the inset shows the location of the TEM observation in (b). b. TEM images taken in situ during the voltage scan in panel, at times I, II, III, and IV. Note the correlation of resistance with void size.

77

As more current was applied, the resistance of the device continued to rise, correlated

with the growth of the void. A thin shell of GeTe material on the inside of the

encapsulating oxide tube appeared to provide a current path through the device around

the void. A blue rectangle shows where the initial structural changes occurred due to

the voltage scan as shown in Figure 4.12b. At the end of the scan the NW’s resistance

had increased to 42 kΩ (~10 times the original resistance), and the void had grown to

almost 1 µm in length. Similar opening of voids could also be achieved by pulsing the

nanowire, but there tended to be less GeTe remaining resulting in higher final

resistances.

Figure 4.13: TEM micrographs showing void formation via voltage pulse.

For example, Figure 4.13 shows the same nanowire before and after applying a 4V

square pulse. The pulse was strong enough to move the GeTe material and leave a

large void inside the nanowire.

Remarkably, a new mechanism for returning the device to a low resistance

state was observed (Figure 4.14). When a small voltage was applied, the NW remains

in a high resistance state, and large void regions were observed by TEM. As the

voltage was scanned past 1.5 V, movement in the remaining material was observed

(SEM sequence in Figure 4.14). Between points ‘S’ and ‘F’, the TEM images suggest

78

that the material became molten, and the shell was suddenly refilled with material.

This closing of the void was correlated with a large increase in the measured current,

carrying it over the compliance set for the measurement (Figure 4.14).

Figure 4.14: In situ TEM observation showing closing of void and corresponding voltage scan.

The final resistance measured was similar to the resistance of the NW in its original

state before any void formation had been observed. Though counter-intuitive, this void

formation mechanism is the best candidate for explaining the extremely large on-off

ratios observed in the many NW devices fabricated and tested on regular substrates.

79

Figure 4.15: Material movement in void induced switching. a. SEM after multiple cycles. b. Schematic showing the movement of material inside the NW.

SEM images of the NW after several cycles of this behavior are shown in

Figure 4.15a, in which the bright (dark) contrast on the NW indicates the presence

(absence) of GeTe materials. Since part of the NW outside the contact appears to be

empty (dark), it is evident that the material in the NW outside the contacted region can

be a source for refilling voids. A schematic (Figure 4.15b) illustrates the observed

flow of material that was observed: the NW region outside the contacts acted as a

reservoir allowing more material to enter the NW and close voids. Material could

escape through small cracks or pinholes in the oxide shell.

The exact mechanisms that lead to movement of material are not yet

understood, but all our TEM studies have shown that the structural transformations are

localized to rather small regions of the NW. These regions are usually, but not always,

located close to the contacts. To better understand the significance localized heating

plays in the motion of material, my collaborator, David Schoen, made a simple model

with commercially available software (COMSOL Multiphysics). The model confirms

that localized heating can play a major role in the behavior of the NW99. The

80

temperature distribution in the NW is highly influenced by contact resistances or other

hot spots (such as voids). The actual pattern of transformation in a NW can be very

complicated and narrow regions can move. Hence, one might expect the heating

during a voltage scan to be highly dynamic. Since voids can form on either or both of

the positive and negative electrode after a pulse or scan, and material can move in

either direction in the same NW during the same scan, electromigration can probably

be ruled out as a dominant mechanism. Instead, capillary forces and expansion of

material due to melting and pressure buildup in the oxide encapsulation due to the

evaporation of tellurium seem the most likely candidates to explain to the observed

motion of liquid PCM material.

4.4 Conclusion

Our in situ TEM study revealed the mechanism responsible for the very high

resistance state in GeTe NWs. Surprisingly, rather than the expected crystalline to

amorphous transition, we found that the reversible switching mechanism was

dominated by the opening and closing of voids inside the NW. This new switching

mechanism could potentially be useful in devices that require a very high resistance in

the OFF state. However, using this mechanism in a useful switch (i.e. one that can

switch more often in a well controlled manner) would require a significant amount of

engineering. The device geometry presented here is not ideal for such a purpose,

because switching endurance is limited to dozens of cycles. It might be possible to

engineer a switch using a larger reservoir of material so that switching could be

repeated many times.

81

While our study showed interesting behavior in phase-change NWs, one

important question remains: why did we not observe the expected crystalline to

amorphous transition? Since we saw melting, but never a pure amorphous phase, the

most likely explanation is that we could not achieve cooling rates sufficient to vitrify

GeTe. In order to achieve the crystalline to amorphous transition, it may be necessary

to make design changes to achieve higher cooling-rates. However, very good

switching performance has been observed in GST NWs48. Hence, it would be

interesting to use this in situ technique to observe the details of switching in GST

NWs and compare it to our observations presented in Chapter 2.

82

5 Topological insulator nanoribbons

5.1 Introduction and motivation

Topological insulators, a new class of materials which are insulating in the

bulk but have conductive surface states, have generated great interest since their

theoretical discovery in 200514-18, 82, 132-134. Theory predicts many fascinating

electronic properties in topological insulators, including spin-polarized dissipationless

current. However, few transport measurements have shown signatures of the

topological states.

Figure 5.1: Schematic of (a) bulk sample of Bi2Se3 and corresponding (b) band diagram.

The reason for this discrepancy is that bulk samples tend to have many intrinsic

defects that act as electron donors, effectively lowering the resistance of the materials.

Figure 5.1a shows a schematic of a bulk sample that serves to illustrate the difficulty

of measuring the surface states. A bulk crystal is typically 1mm thick, consisting of

approximately 1 million quintuple layers. If the bulk layers contribute to the

conduction mechanism, it is very difficult to measure the contribution of the surface

states because of the small signal-to-noise ratio (1 million bulk layers compared to 2

83

surface layers). The band structure of a bulk crystal looks similar to that of a

degenerately doped semiconductor with the Fermi level in the conduction band

(Figure 5.1b). In order to decrease the contribution of the inner layers to conduction,

we synthesized and measured topological insulator nanoribbons. Figure 5.2 depicts a

schematic of a nanoribbon (NR) that consists of only 20-100 quintuple layers, greatly

increasing the surface-to-volume ratio. In nanoribbons, the bulk contributes much less

to the overall conduction, producing a more favorable signal-to-noise ratio (20-100

bulk layers compared to 2 surface layers) for the study of the topological states.

Figure 5.2: Schematic of topological insulator nanoribbon

The work for this chapter was a fruitful collaboration with Hailin Peng, Keji

Lai, Desheng Kong and Yulin Chen. All contributed to the experiments and to the

analysis of the results. I was heavily involved in the synthesis, characterization, device

fabrication, measurements and analysis discussed in this chapter over the span of one

year, but the work is a result of our close collaboration.

5.2 Synthesis of topological insulator nanoribbons

From our experiments with gallium selenide135 we learned that these layered

compounds grow into both nanowires and sheet-like structures called “nanoribbons”.

Since Bi2Se3 has a layered structure, we expected that it could grow into nanoribbons

84

as well. Wide nanoribbons have the advantage of allowing for important experimental

techniques such as the Hall effect measurement, an important tool in the

characterization of these materials. We synthesized Bi2Se3 NRs in the same VLS

growth furnace that was set up for GeTe NW growth (chapter 2).

The range of possible growth parameters for Bi2Se3 turned out to be relatively

large. For a typical synthesis136 we placed the source material, Bi2Se3 powder

(99.999%, 0.2 g per growth from Alfa Aesar), in the hot center of the furnace and we

put Si (100) substrates, functionalized with 0.1% w/v aqueous poly-L-lysine solution

(Ted Pella) and coated with 20 nm Au nanoparticles (Sigma Aldrich) in the

downstream side of the furnace at distances ranging between 6 to 12 cm from the

center. We initially pumped down the tube to a base pressure of less than 100 mTorr

and flushed with Ar gas several times to remove O2 residue. We then heated the

furnace to 450-580 °C (setpoint at hot center of the tube) and maintained this high

temperature for 1-5 h while flowing 30 standard cubic centimeters (sccm) Ar carrier

gas. Finally, we allowed the system to cool down over the time span of a few hours.

After the synthesis, a gray layer covering each substrate over a 1 cm2 area could be

observed by the unaided eye.

Figure 5.3: SEM of as-synthesized Bi2Se3 nanoribbons.

85

Our SEM observations (Figure 5.3) confirmed that Bi2Se3 could grow in a

nanoribbon or sheet-like form. Indeed, we observed sheets with lateral dimensions of

over 100 μm for the longer growth times. However, the thickness of the sheets tended

to range between 20-100 nm.

Figure 5.4: SEM and AFM of Bi2Se3 NR device. Graph shows height measurement of AFM cross section.

To confirm the thickness of the ribbons we conducted AFM experiments. As an

example, Figure 5.4 shows an SEM and AFM image of a single nanoribbon device.

The AFM cross section measurement shows that the thickness of the ribbon is only 22

nm.

To characterize the crystal structure, growth direction and chemical

composition of the NRs, we performed TEM and EDX characterization. Figure 5.5

depicts a TEM image of a nanoribbon. The SAD pattern and HRTEM (Figure 5.5

86

insets) shows a hexagonal pattern confirming that the top face is perpendicular to the

[001] direction. The growth direction is equivalent to a [11-20] direction, which

agrees with our observation of other nanowires135, 137 with a layered structure.

Figure 5.5: SEM and AFM of Bi2Se3 NR device. Graph shows height measurement of AFM cross section.

The HRTEM image shows hexagonal lattice fringes with a lattice spacing of 2.1 Å

between (11-20) planes, in agreement with the literature value. As a result of their

crystal structure and growth direction, nanoribbons have a rectangular cross section

perpendicular to the growth direction. The EDX analysis confirmed that the

composition was Bi2Se3 to within ~2% experimental error. Further AFM and scanning

tunneling microscopy measurements showed that the NRs had atomically flat surfaces

with 1 nm step edges136. Our characterization of the Bi2Se3 NRs confirmed their

suitability for transport measurements due to their very high surface-to-volume ratio

and compatibility with Hall measurements. These measurements are discussed in the

following sections.

87

5.3 Electrical characterization of nanoribbons

In order to characterize the electronic properties of the as-synthesized Bi2Se3

NRs, we fabricated devices via the electron beam lithography method discussed in

Chapter 4.1.2. Figure 5.6 shows a schematic of a device that allows for 4-point probe

and Hall measurements on the same NR.

Figure 5.6: Schematic of NR device. Setup allows for 4-point probe (green V-) and Hall (blue V-) measurement on the same ribbon.

The resistivity of an n-type doped semiconductor is given by ρ = 1e ⋅ μ ⋅ n

, where e is

the electron charge, μ is the electron mobility and n is the electron density. Similarly,

the Hall resistance is RH =−1

n ⋅ e. Hence, measuring the resistance and Hall resistance

allows us to deduce the carrier concentration and the mobility as long as we know the

dimensions of the ribbon. Low-frequency (1,000 Hz) and d.c. standard four-probe

magnetoresistance measurements were carried out in a Quantum Design PPMS-7

instrument. The temperature range is 2-300 K and the magnetic field can be applied up

to ±9 T.

Figure 5.7 shows a basic device for a narrow ribbon. These ribbons are not

wide enough to make Hall Probe contacts, but they are useful for measuring the 4-

88

point probe resistance. To make Ohmic contacts we evaporated either Cr/Au or Ti/Au

contacts with a thickness of approximately 5/195 nm. After measuring dozens of

devices we found that the contact resistance was typically negligible and the sheet

resistance of the devices at room temperature was typically on the order of a few 100

Ω, demonstrating that the nanoribbons are still quite conductive, similar to bulk

crystals.

Figure 5.7: SEM of Bi2Se3 NR device at 30° rotation. Even though the NR appears bent it is lying flat on the substrate and the contacts are on top.

Figure 5.8 shows the four-terminal resistance of a device from room

temperature down to 2K. The resistance decreases with temperature and reaches a

minimum value below 20 K. This behavior agrees with that of heavily doped

semiconductors. The doping in Bi2Se3 NRs is believed to be due intrinsic defects such

as Se vacancies138.

2 μm

89

Figure 5.8: Resistance as a function of temperature for Bi2Se3 NR.

These measurements indicate that the conduction mechanism in NRs is still dominated

by bulk carriers, despite the single crystalline nature and high quality of the ribbons.

The Fermi level appears to remain inside the conduction band as depicted in Figure

5.1.

To measure the mobility, we fabricated devices that allow for 4-point and Hall

measurements (Figure 5.9) on the same NR. For the Hall measurement, a known

current is driven through the device (indicated by I+ and I- in Figure 5.5) and the

voltage is measured between electrodes that are aligned perpendicular to the flow of

current while a magnetic field is applied perpendicular to the top face of the NR.

600700800

90010001100120013001400

150016001700

Res

ista

nce

(Ohm

)

0 50 100 150 200 250 300Temperature (K)

90

Figure 5.9: SEM Bi2Se3 NR device. This setup allows for 4-point probe and Hall measurements on the same device.

We measured over 10 Hall bar devices and generally observed two types of behavior

(Figure 5.10). Some devices showed very linear behavior with a relatively small slope

corresponding to a high carrier area density (in this case 1.5×1014 cm-2). Other

devices had a larger slope corresponding to a lower density (in this case 2.6×1013 cm-

2) at low magnetic fields. Interestingly, in the low density samples, we could observe

a difference in slope between the low and high magnetic field measurement.

91

Figure 5.10: Hall measurements of two different devices showing different behaviors.

This change in slope is typically due to the presence of two types of carriers with

different mobility. In this case, it is likely that the slope change is due to the surface

states that were masked by excessive bulk carriers in the high density samples. All the

measurements showed clear n-type behavior by the sign of the Hall slope. The

variation in carrier density is probably due to variations in the defect concentration in

the NRs. To study this further, it would be interesting to perform in situ TEM studies

of the NRs allowing a correlation between structure and electrical parameters.

However, in our regular TEM observations we did not see a large variability in the

morphology or composition for most ribbons. The measured mobility was on the

order of 2000 cm2 V-1 s-1.

The measurements in this section show that NRs, despite their favorable

surface-to-volume ratio, are still dominated by bulk conduction. In order to find more

experimental proof of the surface states, we conducted magneto-resistance

92

measurements to look for Aharonov-Bohm oscillations32. These measurements are

discussed in the following section.

5.4 Aharonov-Bohm interference in topological insulator

nanoribbons

5.4.1 Introduction to Aharonov-Bohm effect

Since the conduction in Bi2Se3 is dominated by bulk carriers originating from

crystal defects and thermal excitation, it is difficult to perform transport measurements

on the surface states. To prove the existence of surface states we set out to measure a

transport signature of the topological states via the Aharonov-Bohm Effect.

The Aharonov-Bohm (AB) effect is strongly related to one of the fundamental

mysteries of quantum mechanics: the double slit experiment (Figure 5.11). When

more than one path is available to an electron, the electron can interfere with itself,

creating an interference pattern.

Figure 5.11: Schematic of double slit experiment.

93

In the AB effect, the electrons also have multiple available paths; however, the

electrons also pass through a magnetic field that changes their quantum phase. Figure

5.12 shows a schematic of a conducting ring structure that provides two pathways for

the electrons. A magnetic field is applied perpendicular to the electron flow. When

the magnetic field strength is varied, the change in quantum phase gives rise to

constructive and destructive interference between the electrons139. This interference

can be measured as oscillations in the resistance of the ring structure. The

characteristic periodicity in the magnetic field is given by ∆B = Φ0/S, where Φ0 = h/e

is the flux quantum, S is the area enclosed by the electron pathways, h is Planck’s

constant and e is the electron charge.

Figure 5.12: Schematic of Aharonov-Bohm experiment.

Further, the AB effect has been measured in conducting cylinders such as carbon

nanotubes140, In2O3/InOx core-shell NWs141 and other NWs that have special surface

states142, 143. In these cases, the same periodicity in resistance fluctuations is observed

if the magnetic field is aligned along the NW and S is taken to be the cross-sectional

area of the NW. It is important to note that the AB effect is only observed when just

94

the surface of the NW is conducting or when the surface has distinct states that do not

significantly interfere with the bulk conduction.

5.4.2 Aharonov-Bohm measurement

AB interference can only be observed in nanostructures if the conduction

electrons remain phase coherent after completing closed trajectories. Hence, the AB

effect is more likely to be seen in narrow ribbons than in very wide ones. In order to

look for the AB effect in the Bi2Se3 NRs, we fabricated devices of narrow NRs and

aligned the magnetic field parallel to the ribbons. The samples were measured at 2K

to decrease phonon contributions.

Figure 5.13: Magnetoresistance and SEM of Bi2Se3 NR

Figure 5.13 shows an SEM of a narrow NR device and the corresponding resistance

measurement as a function of magnetic field strength. A pronounced drop in resistance

1250

1260

1270

1280

Res

ista

nce

(Ohm

s)

-10 0 10Magnetic Field (T)

B

95

at zero field due to the weak anti-localization effect144 is evident, consistent with the

presence of strong spin–orbit coupling in Bi2Se3. For magnetic fields between -9T to

9T, clear periodic oscillations can be observed in the resistance. To find the

periodicity, we first take the derivative dRdB

to separate the oscillatory part from the

slow-varying background145, and then take the Fourier transform.

Figure 5.14: Fourier transform of dR/dB data derived from Figure 5.X

The Fourier transform shows a sharp peak corresponding to a periodicity of 0.62T

(Figure 5.14). This measurement agrees well with the expected period of ∆B = Φ0/S =

0.63T. The cross section, S, was determined via SEM and AFM measurements.

5.4.3 Discussion and conclusion

The observation of the AB effect in Bi2Se3 NRs may represent the first

transport measurement of the topological surface states in this material system146.

Further, we can reach several conclusions based on our observations. First, though the

majority of the conduction still occurs through the bulk of the nanoribbons, we can

clearly observe the AB oscillations. These considerations strongly suggest that there

96

is little interaction between the bulk and surface electrons. This is to be expected,

because if the Fermi level is close to the bottom of the conduction band, the surface

states and bulk states have different momentum (Figure 5.1b) making interactions

unlikely. Second, since the AB effect requires the electrons to move around the whole

perimeter of the NR, our results demonstrate that the surface states exist on all the

surfaces of the NR, not just on the basal plane. Since the side walls expose dangling

bonds and the quintuple layers are only linked by weak van der Waals forces, this

observation provides further evidence for the robustness of the surface states.

Current work in this area focuses on further lowering the bulk contribution to

conduction via doping and gating the NRs147, 148. Achieving pure surface conduction

will open up new experimental possibilities for probing the surface states in these

exciting materials.

5.5 Conclusion

Topological insulators are a new class of materials with exciting prospects in

electronics, spintronics, and other applications. In this chapter we presented the

synthesis and characterization of topological insulator Bi2Se3 NRs. The high surface-

to-volume ratio helped suppress bulk conduction inside the nanoribbons, which

allowed for the observation of Aharonov-Bohm interference. This interference

measurement was the first transport measurement to provide evidence of the

topological states in this material system. Rapid progress in the young field of

topological insulators is producing a wealth of new knowledge; we can expect to see

more breakthroughs soon.

97

6 Conclusion

Though we could only focus on a few materials, the research for this

dissertation gave a glimpse of the varied and fascinating properties of chalcogenide

materials. We concentrated on chalcogenides as phase-change memory and

topological insulator materials, two applications with great significance for electronics.

Phase change memory is on the verge of commercialization and could eventually

replace flash memory, while topological insulators have the potential to transform

current electronics by offering a new paradigm for computer design. The right

geometry was needed to reveal the most interesting properties of our phase-change

and topological insulator chalcogenide materials. Nanostructures provided high

crystal quality, TEM compatibility and a large surface-to-volume ratio. These

properties made possible in situ TEM observation of phase-change memory cells and

measurement of Aharonov-Bohm interference in topological insulators. The studies

yielded several interesting and often unexpected results.

We developed in-situ TEM switching of phase-change materials, enabling

direct correlation of electrical switching behavior with structural changes. Our in situ

study of lateral phase-change cells showed significant variability in the microstructure

of the high-resistance phase; even in the same cell after applying the same

amorphization pulse. We observed a pure amorphous phase as well as a mixed

amorphous phase containing crystalline residues, both with similar high resistances

but different threshold switching behavior. This finding can assist in the design of

better PCM cells through control of the microstructure of the OFF state. Our in situ

98

study of phase-change GeTe nanowires allowed us to explain the very high resistances

observed in the OFF state, and we discovered a new switching mechanism: reversible

void formation. In the future, this mechanism could be exploited to create devices

with a highly resistive OFF state, though careful engineering would be required to

achieve sufficient device endurance. Our in-situ studies of the phase-change cells and

nanowires during switching provided important insights into the detailed mechanisms

of the phase-change. The in situ TEM technique applied to PCM studies can be an

effective tool to elucidate important topics in phase-change memory such as the origin

of resistance drift, scaling and differences in materials.

In our topological insulator study, we synthesized single-crystal nanoribbons

because they offer a large surface to bulk conduction ratio. These nanoribbons

enabled us to measure the AB effect, and hence to observe one of the first electrical

signatures of the topological states. Doping and gating may help to further suppress

the bulk conduction and to enable future spintronics devices.

99

References

1. Schock, H.-W. & Noufi, R. CIGS-based solar cells for the next millennium. Progress in Photovoltaics: Research and Applications 8, 151-160 (2000).

2. Chu, T. L. Thin film cadmium telluride solar cells by two chemical vapor deposition techniques. Solar Cells 23, 31-48 (1988).

3. Pelusi, M. et al. Photonic-chip-based radio-frequency spectrum analyser with terahertz bandwidth. Nature Photonics (2009).

4. Ho, N. et al. Single-mode low-loss chalcogenide glass waveguides for the mid-infrared. Optics Letters 31, 1860-1862 (2006).

5. Lamont, M. R., Luther-Davies, B., Choi, D., Madden, S. & Eggleton, B. J. Supercontinuum generation in dispersion engineered highly nonlinear (γ = 10 /W/m) As2S3 chalcogenide planar waveguide. Opt. Express 16, 14938–14944 (2008).

6. Paivasaari, K., Tikhomirov, V. K. & Turunen, J. High refractive index chalcogenide glass for photonic crystal applications. Optics Express 15, 2336-2340 (2007).

7. Jha, R. & Sharma, A. K. High-performance sensor based on surface plasmon resonance with chalcogenide prism and aluminum for detection in infrared. Optics Letters 34, 749-751 (2009).

8. Zogg, H. et al. IV-VI mid-IR tunable lasers and detectors with external resonant cavities. Proc. of SPIE 7453, 74530R (2009).

9. DiSalvo, F. J. Thermoelectric Cooling and Power Generation. Science 285, 703-706 (1999).

10. Kim, Y. et al. Structural and thermoelectric transport properties of Sb2Te3 thin films grown by molecular beam epitaxy. Journal of Applied Physics 91 (2002).

11. Zhang, T. et al. High Speed Chalcogenide Random Access Memory Based on Si2Sb2Te5. Jap. Journ. App. Phy. 46, L247–L249 (2007).

12. Wuttig, M. & Yamada, N. Phase-change materials for rewriteable data storage. Nature Materials 6, 824-832 (2007).

13. Kane, C. L. & Mele, E. J. Quantum spin Hall effect in graphene. Phys. Rev. Lett. 95, 226801 (2005).

14. Bernevig, B. A., Hughes, T. L. & Zhang, S.-C. Quantum Spin Hall Effect and Topological Phase Transition in HgTe Quantum Wells. Science 314, 1757-1761 (2006).

15. Fu, L., Kane, C. L. & Mele, E. J. Topological insulators in three dimensions. Phys. Rev. Lett. 98, 106803 (2007).

16. Moore, J. E. & Balents, L. Topological invariants of time-reversal-invariant band structures. Phys. Rev. B 75, 121306 (2007).

17. Qi, X. L., Hughes, T. L. & Zhang, S. C. Topological field theory of time-reversal invariant insulators. Phys. Rev. B 78, 195424 (2008).

18. Zhang, H. et al. Topological insulators in Bi2Se3, Bi2Te3 and Sb2Te3 with a single Dirac cone on the surface. Nature Physics 5, 438-442 (2009).

19. Morales, A. M. & Lieber, C. M. A laser ablation method for the synthesis of crystalline semiconductor nanowires. Science 279, 208-211 (1998).

100

20. Wagner, R. S. & Ellis, W. C. VAPOR-LIQUID-SOLID MECHANISM OF SINGLE CRYSTAL GROWTH. App. Phys. Lett. 4, 1753975 (1964).

21. Duan, X., Wang, J. & Lieber, C. M. Synthesis and optical properties of gallium arsenide nanowires. App. Phys. Lett. 76, 1116-1118 (2000).

22. Guichard, A. R., Barsic, D. N., Sharma, S., Kamins, T. I. & Brongersma, M. L. Tunable Light Emission from Quantum-Confined Excitons in TiSi2-Catalyzed Silicon Nano Lett. 6, 2140 (2006).

23. Hochbaum, A. I. et al. Enhanced thermoelectric performance of rough silicon nanowires. Nature 451, 163 (2007).

24. Boukai, A. I. et al. Silicon nanowires as efficient thermoelectric materials. Nature 451, 168 (2008).

25. Y., C., Wei, Q., Park, H. & Lieber, C. M. Nanowire Nanosensors for Highly Sensitive and Selective Detection of Biological and Chemical Species. Science 293, 1289 (2001).

26. Patolsky, F. & Lieber, C. M. Nanowire nanosensors. Mater. Today 8, 20 (2005).

27. Chan, C. K. et al. High-performance lithium battery anodes using silicon nanowires. Nature Nanotech. 3, 31 (2008).

28. Schoen, D. T. et al. Phase transformations in one-dimensional materials: applications in electronics and energy sciences. J. Mater. Chem. 19, 5879–5890 (2009).

29. Peng, H. L., Xie, C., Schoen, D. T. & Cui, Y. Large Anisotropy of Electrical Properties in Layer-Structured In2Se3 Nanowires. Nano Lett. 8, 1511-1516 (2008).

30. Lai, K. et al. Nanoscale Electronic Inhomogeneity in In2Se3 Nanoribbons Revealed by Microwave Impedance Microscopy. Nano Lett. 9, 1265-1269 (2009).

31. Merget, F., Kim, D. H. & Haring Bolivar, P. Lateral phase change random access memory cell design for low power operation. Microsyst. Technol. 13, 169-172 (2007).

32. Aharonov, Y. & Bohm, D. Significance of electromagnetic potentials in the quantum theory. Phys. Rev. 115, 485-491 (1959).

33. Raoux, S., Welnic, W. & Ielmini, D. Phase Change Materials and Their Application to Nonvolatile Memories. Chem. Rev. 110, 240-267 (2010).

34. Terao, M., Morikawa, T. & Ohta, T. Electrical Phase-Change Memory: Fundamentals and State of the Art. Jap. Journ. Appl. Phy. 48, 080001 (2009).

35. Jedema, F. Phase-change materials: Designing optical media of the future. Nature Mater. 6, 90-91 (2007).

36. Ovshinsky, S. R. Reversible electrical switching phenomena in disordered structures. Phys. Rev. Lett. 21, 1450-1453 (1968).

37. Kim, K. & Koh, G. H. in 24th International Conference on Microelectronics 377-384 (2004).

38. Bez, R. & Cappelletti, P. in Symposium on VLSI Technology 84-87 (2005). 39. Akerman, J. Applied Physics: Toward a Universal Memory. Science 308, 508

(2005).

101

40. Scott, J. F. & Paz de Araujo, C. A. Ferroelectric Memories. Science 246, 1400 (1989).

41. Segal, M. Ferroelectric memory: Slim fast. Nature Nanotech. 4, 405 (2009). 42. Yang, Y. C., Pan, F., Liu, Q., Liu, M. & Zeng, F. Fully Room-Temperature-

Fabricated Nonvolatile Resistive Memory for Ultrafast and High-Density Memory Application. Nano Lett. 9, 1636-1643 (2009).

43. Wuttig, M. Nature Mat. 4, 65 - 266 (2005). 44. Hudgens, S. & Johnson, B. Overview of phase-change chalcogenide

nonvolatile memory technology. Mrs Bulletin 29, 829-832 (2004). 45. Solis, J. & Afonso, C. N. Ultrashort-laser-pulse-driven rewritable phase-

change optical recording in Sb-based films. Appl. Phys. A: Mater. Sci. & Process. 76, 331-338 (2003).

46. Lankhorst, M. H. R., Ketelaars, B. W. S. M. M. & Wolters, R. A. M. Low-cost and nanoscale non-volatile memory concept for future silicon chips. Nature Mater. 4, 347-352 (2005).

47. Chen, Y. C. et al. Ultra-thin phase-change bridge memory device using GeSb. Int. Electron. Devices Meet. Tech. Dig., 777-780 (2006).

48. Lee, S. H., Jung, Y. & Agarwal, R. Highly scalable non-volatile and ultra-low power phase-change nanowire memory. Nature Nanotech. 2, 626-630 (2007).

49. Lee, S. H., Ko, D. K., Jung, Y. & Agarwal, R. Size-dependent phase transition memory switching behavior and low writing currents in GeTe nanowires. App. Phys. Lett. 89 (2006).

50. Yoon, S.-M. et al. Electrical Characterization of Nonvolatile Phase-Change Memory Devices Using Sb-Rich Ge-Sb-Te Alloy Films. Jpn. J. Appl. Phys. 46, 7225-7231 (2007).

51. Adler, D., Henisch, H. K. & Mott, S. N. The mechanism of threshold switching in amorphous alloys. Rev. Mod. Phys. 50, 209 (1978).

52. Pirovano, A., Lacaita, A. L., Benvenuti, A., Pellizzer, F. & Bez, R. Electronic Switching in Phase-Change Memories. IEEE Trans. Elect. Dev. 51, 452-459 (2004).

53. Yu, D., Brittman, S., Lee, J. S., Falk, A. L. & Park, H. Minimum Voltage for Threshold Switching in Nanoscale Phase-Change Memory. Nano Lett. 8, 3429-3433 (2008).

54. Warren, A. C. Reversible thermal breakdown as a switching mechanism in chalcogenide glasses. IEEE Trans. Electron Devices 20, 123-31 (1973).

55. Adler, D., Shur, M. S., Silver, M. & Ovshinsky, S. R. Threshold switching in chalcogenide-glass thin films. J. Appl. Phys. 51, 3289-309 (1980).

56. Emin, D., Seager, C. H. & Quinn, R. K. Small-polaron hopping motion in some chalgogenide glasses. Phys. Rev. Lett. 28, 813-16 (1972).

57. Lai, S. Current status of the phase change memory and its future. IEDM Tech. Dig. 29.6 (2003).

58. Lee, S. H. et al. Full Integration and Cell Characteristics for 64Mb Nonvolatile PRAM. Proc. Symp. VLSI Tech. Dig., 20-21 (2004).

59. Chen, C.-F. et al. in Memory Workshop, 2009. IMW '09. IEEE International 1-2 (2009).

102

60. Klein, A. et al. Changes in electronic structure and chemical bonding upon crystallization of the phase change material GeSb2Te4. Phys Rev Lett 100, 016402 (2008).

61. Shportko, K. et al. Resonant bonding in crystalline phase-change materials. Nat. Mater. 7, 653-658 (2008).

62. Williamson, M. J., Tromp, R. M., Vereecken, P. M., Hull, R. & Ross, F. M. Dynamic microscopy of nanoscale cluster growth at the solid-liquid interface. Nature Mater. 2, 532-536 (2003).

63. Kodambaka, S., Tersoff, J., Reuter, M. C. & Ross, F. M. Germanium Nanowire Growth Below the Eutectic Temperature. Science 316, 729-732 (2007).

64. Stach, E. A. et al. Watching GaN Nanowires Grow. Nano Lett. 3, 867-869 (2003).

65. Wen, C. Y. et al. Formation of Compositionally Abrupt Axial Heterojunctions in Silicon-Germanium Nanowires. Science 326, 1247-1250 (2009).

66. Zheng, H. et al. Observation of Single Colloidal Platinum Nanocrystal Growth Trajectories. Science 324, 1309-1312 (2009).

67. Molzen, W. W., Broers, A. N., Cuomo, J. J., Harper, J. M. E. & Laibowitz, R. B. Materials and technique used in nanostructure fabrication. J. Vac. Sci. Technol. 16, 269-272 (1978).

68. Enquist, F. & Spetz, A. The fabrication of amorphous SiO2 substrates suitable for transmission electron microscopy studies of ultrathin polycrystalline films. Thin Solid Films 145, 99-104 (1986).

69. Morkved, T. L., Lopes, W. A., Hahm, J., Sibener, S. J. & Jaeger, H. M. Silicon nitride membrane substrates for the investigation of local structure of polymer thin films. Polymer 39, 3871-3875 (1998).

70. Grant, A. W., Hu, Q.-H. & Kasemo, B. Transmission electron microscopy 'windows' for nanofabricated structures. Nanotech. 15, 1175-1181 (2004).

71. Gibby, A. M. A layered chalcogenide phase change memory device. Ph.D. Thesis. Stanford University (2008).

72. Zhang, S.-C. Topological states of quantum matter. Physics 1 (2008). 73. Kane, C. L. & Mele, E. J. A New Spin on the Insulating State. Science 314,

1692-1693 (2006). 74. Qi, X. L. & Zhang, S.-C. The quantum spin Hall effect and topological

insulators. Physics Today, 33-38 (2010). 75. Moore, J. E. The birth of topological insulators. Nature 464, 194-198 (2010). 76. Nagaosa, N. A New State of Quantum Matter. Science 318, 758-759 (2007). 77. Kane, C. L. An insulator with a twist. Nature Phys. 4, 348-349 (2008). 78. Moore, J. An insulator's metallic side. Nature 460, 1090-1091 (2009). 79. Manoharan, H. C. Topological insulators: A romance with many dimensions.

Nature Nanotech. 5, 477-479 (2010). 80. Chadov, S. et al. Tunable multifunctional topological insulators in ternary

Heusler compounds. Nature Mater. 9, 541-545 (2010).

103

81. Lin, H. et al. Half-Heusler ternary compounds as new multifunctional experimental platforms for topological quantum phenomena. Nature Mater. 9, 546-549 (2010).

82. Xia, Y. et al. Observation of a large-gap topological-insulator class with a single Dirac cone on the surface. Nature Phys. 5, 398-402 (2009).

83. Wyckoff, R. W. G. Crystal Structures (Krieger, 1986). 84. Hamann, H. F., O'Boyle, M., Martin, Y. C., Rooks, M. & Wickramasinghe, K.

Ultra-high-density phase-change storage and memory. Nature Mater. 5, 383-387 (2006).

85. Milliron, D. J., Raoux, S., Shelby, R. & Jordan-Sweet, J. Solution-phase deposition and nanopatterning of GeSbSe phase-change materials. Nature Mater. 6, 352-356 (2007).

86. Hong, S.-H., Bae, B.-J. & Lee, H. Fast switching behavior of nanoscale Ag6In5Sb59Te30 based nanopillar type phase change memory. Nanotech. 21, 025703 (2010).

87. Lee, S. et al. Demonstration of a Reliable High Speed Phase-Change Memory Using Ge-Doped SbTe. Jour. Electrochem. Soc. 156, H612-H615 (2009).

88. Lencer, D. et al. A map for phase-change materials. Nature Mater. 7, 972-977 (2008).

89. Hegedus, J. & Elliott, S. R. Microscopic origin of the fast crystallization ability of Ge-Sb-Te phase-change memory materials. Nature Mater. 7, 399-405 (2008).

90. Kolobov, A. V. et al. Understanding the phase-change mechanism of rewritable optical media. Nature Mater. 3, 703-708 (2004).

91. Sun, Z. M., Zhou, J. & Ahuja, R. Unique melting behavior in phase-change materials for rewritable data storage. Phys. Rev. Lett. 98 (2007).

92. Welnic, W. et al. Unravelling the interplay of local structure and physical properties in phase-change materials. Nature Mat. 5, 56-62 (2006).

93. Akola, J. & Jones, R. O. Structural phase transitions on the nanoscale: The crucial pattern in the phase-change materials Ge2Sb2Te5 and GeTe. Phys. Rev. B 76, 235201 (2007).

94. Sun, Z., Zhou, J. & Ahuja, R. Structure of Phase Change Materials for Data Storage. Phys. Rev. Lett. 96, 055507 (2006).

95. Pirovano, A. et al. Low-Field Amorphous State Resistance and Threshold Voltage Drift in Chalcogenide Materials. IEEE Trans. Elect. Dev. 51, 714-719 (2004).

96. Ielmini, D., Lacaita, A. L. & Mantegazza, D. Recovery and Drift Dynamics of Resistance and Threshold Voltages in Phase-Change Memories. IEEE Trans. Elect. Dev. 54, 308-315 (2007).

97. Song, S. A., Zhang, W., Jeong, H. S., Kim, J.-G. & Kim, Y.-J. In situ dynamic HR-TEM and EELS study on phase transitions of Ge2Sb2Te5 chalcogenides. Ultramicroscopy 108, 1408-1419 (2008).

98. Yoon, S.-M. et al. Nanoscale observations of the operational failure for phase-change-type nonvolatile memory devices using Ge2Sb2Te5 chalcogenide thin films. App. Surf. Sci. 254, 316–320 (2007).

104

99. Meister, S., Schoen, D. T., Topinka, M. A., Minor, A. M. & Cui, Y. Void Formation Induced Electrical Switching in Phase-Change Nanowires. Nano Lett. 8, 4562-4567 (2008).

100. Meister, S., Kim, S., Cha, J. J., Wong, H.-S. P. & Cui, Y. In situ TEM observation of microstructural changes in phase-change memory bridges during electrical switching. (submitted to Nature Nanotech.) (2010).

101. Krebs, D. et al. Characterization of phase change memory materials using phase change bridge devices. Jour. App. Phy. 106, 054308-7 (2009).

102. Krebs, D. et al. Threshold field of phase change memory materials measured using phase change bridge devices. App. Phys. Lett. 95 (2009).

103. Wu, Y. & Yang, P. Germanium Nanowire Growth via Simple Vapor Transport. Chem. of Mater. 12, 605-607 (2000).

104. Duan, X. & Lieber, C. M. General Synthesis of Compound Semiconductor Nanowires. Advanced Materials 12, 298-302 (2000).

105. Duan, X. & Lieber, C. M. Laser-Assisted Catalytic Growth of Single Crystal GaN Nanowires. Journ. Amer. Chem. Soc. 122, 188-189 (1999).

106. Bjork, M. T. et al. One-dimensional heterostructures in semiconductor nanowhiskers. Appl. Phys. Lett. 80, 1058-1060 (2002).

107. Kuykendall, T. et al. Crystallographic alignment of high-density gallium nitride nanowire arrays. Nature Mater. 3, 524-528 (2004).

108. Prince, A., Raynor, G. V. & Evans, D. S. Phase Diagrams of Ternary Gold Alloys (ed. Metals, I. o.) (Brookfield, VT, 1990).

109. Yamada, N., Ohno, E., Nishiuchi, K., Akahira, N. & Takao, M. Rapid-phase transitions of GeTe-Sb2Te3 pseudobinary amorphous thin films for an optical disk memory. Journ. Appl. Phys. 69, 2849-2856 (1991).

110. Damodara Das, V., Soundararajan, N. & Pattabi, M. Electrical conductivity and thermoelectric power of amorphous Sb2Te3 thin films and amorphous-crystalline transition. Journal of Materials Science 22, 3522-3528 (1987).

111. Zhang, Z. & et al. Modelling for size-dependent and dimension-dependent melting of nanocrystals. J. Phys. D: Appl. Phys. 33, 2653 (2000).

112. Jennings, A. T., Jung, Y., Engel, J. & Agarwal, R. Diameter-Controlled Synthesis of Phase-Change Germanium Telluride Nanowires via the Vapour-Liquid-Solid Mechanism. J. Phys. Chem. C 113, 6898-6901 (2009).

113. Gao, P. X. et al. Conversion of Zinc Oxide Nanobelts into Superlattice-Structured Nanohelices. Science 309, 1700-1704 (2005).

114. Tang, Y. H. et al. Morphology of Si nanowires synthesized by high-temperature laser ablation. J. Appl. Phys. 85, 7981-7983 (1999).

115. Zhang, H.-F., Wang, C.-M., Buck, E. C. & Wang, L.-S. Synthesis, Characterization, and Manipulation of Helical SiO2 Nanosprings. Nano Lett. 3, 577-580 (2003).

116. McIlroy, D. N., Zhang, D., Kranov, Y. & Norton, M. G. Nanosprings. Appl. Phys. Lett. 79, 1540-1542 (2001).

117. Chen, X. et al. Mechanics of a Carbon Nanocoil. Nano Lett. 3, 1299-1304 (2003).

105

118. Bae, S. Y., Lee, J., Jung, H., Park, J. & Ahn, J.-P. Helical Structure of Single-Crystalline ZnGa2O4 Nanowires. J. Am. Chem. Soc. 127, 10802-10803 (2005).

119. Korgel, B. A. MATERIALS SCIENCE: Nanosprings Take Shape. Science 309, 1683-1684 (2005).

120. Meister, S. et al. Synthesis and Characterization of Phase-Change Nanowires. Nano Lett. 6, 1514-1517 (2006).

121. Yu, B. et al. Indium selenide nanowire phase-change memory. Appl. Phys. Lett. 91, 133119-3 (2007).

122. Mitra, M., Jung, Y., Gianola, D. S. & Agarwal, R. Extremely low drift of resistance and threshold voltage in amorphous phase change nanowire devices. Appl. Phys. Lett. 96, 222111-3 (2010).

123. Wuttig, M. & Yamada, N. Phase-change materials for rewriteable data storage. Nature Mater. 6, 824-832 (2007).

124. Sedgwick, T. O., Agule, B. J. & Broers, A. N. Novel Method for Fabrication of Ultrafine Metal Liners by Electron-Beams. J. Electrochem. Soc. 119, 1769-& (1972).

125. Yu, D., Wu, J., Gu, Q. & Park, H. Germanium telluride nanowires and nanohelices with memory-switching behavior. J. Amer. Chem. Soc. 128, 8148-8149 (2006).

126. Lee, J. S., Brittman, S., Yu, D. & Park, H. Vapor-liquid-solid. and vapor-solid growth of phase-change Sb2Te3 nanowires and Sb2Te3/GeTe nanowire heterostructures. J. Amer. Chem. Soc. 130, 6252-6258 (2008).

127. Chopra, K. L. & Bahl, S. K. Amorphous versus Crystalline GeTe Films. I. Growth and Structural Behavior. J. Appl. Phys. 40, 4171-4178 (1969).

128. Mikolaichuk, A. G., Kogut, A. N. & Ignativ, M. I. Electric Properties of Telluride and Germanium Selenide Thin Films. Izv. Vyssh. Uchebn. Zaved., Fiz., 103-& (1970).

129. Sun, X., Yu, B., Ng, G. & Meyyappan, M. One-Dimensional Phase-Change Nanostructure: Germanium Telluride Nanowire. J. Phys. Chem. C 111, 2421-2425 (2007).

130. Bahl, S. K. & Chopra, K. L. AMORPHOUS VERSUS CRYSTALLINE GETE FILMS .3. ELECTRICAL PROPERTIES AND BAND STRUCTURE. J. Appl. Phys. 41, 2196-& (1970).

131. Stoemenos, J. & Vincent, R. Twinning Faults in Epitaxial Films of GeTe and GeTe-SnTe Alloys. Physica Status Solidi (a) 11, 545-558 (1972).

132. Konig, M. Quantum spin hall insulator state in HgTe quantum wells. Science 318, 766-770 (2007).

133. Chen, Y. L. et al. Experimental realization of a three-dimensional topological insulator, Bi2Te3. Science 325, 178-181 (2009).

134. Fu, L. & Kane, C. L. Topological insulators with inversion symmetry. Phys. Rev. B 76, 045302 (2007).

135. Peng, H., Meister, S., Chan, C. K., Zhang, X. F. & Cui, Y. Morphology Control of Layer-Structured Gallium Selenide Nanowires. Nano Lett. 7, 199-203 (2006).

106

136. Kong, D. et al. Topological Insulator Nanowires and Nanoribbons. Nano Lett. 10, 329-333 (2009).

137. Peng, H., Schoen, D. T., Meister, S., Zhang, X. F. & Cui, Y. Synthesis and Phase Transformation of In2Se3 and CuInSe2 Nanowires. J. Am. Chem. Soc. 129, 34-35 (2006).

138. Hyde, G. R., Beale, H. A., Spain, I. L. & Woollam, J. A. Electronic properties of Bi2Se3 crystals. J. Phys. Chem. Solids 35, 1719-1728 (1974).

139. Aronov, A. G. & Sharvin, Y. V. Magnetic flux effects in disordered conductors. Rev. Mod. Phys. 59, 755 (1987).

140. Bachtold, A. Aharonov-Bohm oscillations in carbon nanotubes. Nature 397, 673-675 (1999).

141. Jung, M. et al. Quantum Interference in Radial Heterostructure Nanowires. Nano Lett. 8, 3189-3193 (2008).

142. Nikolaeva, A., Gitsu, D., Konopko, L., Graf, M. J. & Huber, T. E. Quantum interference of surface states in bismuth nanowires probed by the Aharonov-Bohm oscillatory behavior of the magnetoresistance. Phys. Rev. B 77, 075332 (2008).

143. Richter, T. et al. Flux Quantization Effects in InN Nanowires. Nano Lett. 8, 2834-2838 (2008).

144. Hikami, S., Larkin, A. I. & Nagaoka, Y. Spin-Orbit Interaction and Magnetoresistance in the Two Dimensional Random System. Prog. Theor. Phys. 63, 707-710 (1980).

145. Huber, T. E., Celestine, K. & Graf, M. J. Magnetoquantum oscillations and confinement effects in arrays of 270-nm-diameter bismuth nanowires. Phys. Rev. B 67, 245317 (2003).

146. Peng, H. et al. Aharonov-Bohm interference in topological insulator nanoribbons. Nat. Mater. 9, 225-229 (2010).

147. Kong, D. et al. Few-Layer Nanoplates of Bi2Se3 and Bi2Te3 with Highly Tunable Chemical Potential. Nano Lett. 10, 2245-2250 (2010).

148. Cha, J. J. et al. Magnetic Doping and Kondo Effect in Bi2Se3 Nanoribbons. Nano Lett. 10, 1076-1081.