Towards electron holography of working transistors

Sadegh Yazdi

A dissertation submitted for the degree of

Doctor of Philosophy

of Imperial College London

Supervisors:
Prof. David W McComb
Dr. Alison Harrison

Department of Materials and

London Centre for Nanotechnology

Imperial College London

1
 Statement of Originality

The work presented in this thesis is original and a result of my own research, except where
clearly acknowledged in the text.

Sadegh Yazdi

2
 Abstract

As semiconductor device dimensions are reduced to the deep sub-micron regime, minor
departures from the designed distributions of electrostatic potentials can affect device
performance dramatically. Parameter optimisation in device processing and modelling is
crucial for achieving precise potential profiles. Such optimisation is not possible without
comprehensive feedback from advanced characterisation techniques. The ability to acquire
two- and three-dimensional measurements of potential distributions with high spatial
resolution, high precision and under an applied electrical bias is therefore in great demand.
The technique of off-axis electron holography in the transmission electron microscope (TEM)
promises to fulfil these requirements in two dimensions and can be combined with electron
tomography for three-dimensional measurements.

In this dissertation, the practical challenges that are involved in the application of electron
holography to the characterisation of electrostatic potentials in working MOSFETs are
addressed. A novel method for the application of electrical contacts to a TEM specimen is
developed and applied to a diode structure. Off-axis electron holography measurements are
carried out on a pn junction using both this and an alternative geometry and compared with
simulations. A semi-biased MOSFET is then characterised successfully using electron
holography, suggesting that the examination of working transistors in the TEM is a realistic
prospect. In order to investigate sources of error and practical challenges, the influence of
diffraction contrast on electron holographic phase images is investigated. The effects of
electron beam irradiation on the in-situ characterisation of electrical properties in
semiconductor devices are then assessed using three complementary techniques: off-axis
electron holography and measurements of electron beam induced current in the TEM and
secondary electron dopant contrast in the scanning electron microscope.

3
 Acknowledgments
This thesis could not have been completed without the generous help of many people to
whom I am deeply indebted. Firstly, I would like to express my deepest gratitude to my
supervisors, Prof. David McComb and Dr. Alison Harrison, for their endless support,
encouragement and guidance. I am deeply grateful to Dr. Takeshi Kasama for his assistance
with electron microscopy, invaluable discussions and constructive comments on this thesis. I
am much indebted to Prof. Rafal Dunin-Borkowski for his invaluable advice, supervision and
moreover, using his precious times to read this thesis and give constructive, rigorous and
critical comments about it. I would like to thank Dr. Marco Beleggia, Dr. Beata Kardynal,
Prof. Giulio Pozzi and Dr. Chris Boothroyd for all their help and valuable discussions.

I also thank Dr. Mahmoud Ardakani, Mr. Richard Chater, Dr. James Perkins, Dr. Catriona
McGilvery and Dr. Husain Sebhan at Imperial College London, Mr. Kevin Lee and Dr.
Suguo Huo at University College London and Mr. Adam Fuller, Mrs. Nicole MacDonald, Dr.
Martial Duchamp and Dr. András Kovács at Technical University of Denmark for all their
technical support, training and advice. Thank you to Vadim Migunov for all his help. I would
like to thank Filippo Cavalca for constructing the SEM stub for electrical biasing.

I am very grateful to Robert Pennington, Dr. Shima Kadkhodazadeh and Eric Jensen for
checking for flaws in the text and making valuable suggestions.

Thank you to all the members of the electron microscopy group at the department of
materials at Imperial College London for creating a pleasant working atmosphere. Many
thanks also go to all people at the Centre for Electron Nanoscopy (CEN) of the Technical
University of Denmark for helping me and giving me opportunity to use their excellent
microscopy facilities during my stay in Denmark.

I am grateful to Mr. Markku Tilli at OKMETIC Company and Dr. David Cooper at CEA
LETI for providing silicon pn junction samples. I also acknowledge and thank the EPSRC for
financial support of this project through a Science and Innovation award.

Last, but by no means least, I wish to thank all members of my family, particularly my
parents for their continuous support and unconditional love. I would like to express my
deepest love and appreciation to my wonderful wife, Sepideh, for her steady support, patience
and encouragement. My beloved daughter, Darya, deserves my special thanks for bringing
joy to our hearts and bearing my absence from home.
4
Contents

Chapter 1 .................................................................................................................................. 11

1. Introduction ................................................................................................................... 11

1.1. Thesis outline ................................................................................................................ 12

Chapter 2 .................................................................................................................................. 14

2. Semiconductor devices: physics and characterisation .................................................. 14

2.1. Introduction ............................................................................................................... 14

2.1.1. The trend of semiconductor device development .............................................. 14

2.1.2. Basics of MOSFET operation ............................................................................ 15

2.1.3. MOSFET Scaling ............................................................................................... 18

2.2. Semiconductor physics: basics of pn junctions ......................................................... 22

2.2.1. Introduction ....................................................................................................... 22

2.2.2. Role of doping in semiconductors .................................................................... 22

2.2.3. The pn junction formation ................................................................................. 27

2.2.4. Biasing a pn junction ........................................................................................ 32

2.3. Characterisation of semiconductor devices ............................................................... 37

2.3.1. Scanning Probe Microscopy (SPM) techniques ................................................ 37

2.3.2. 1D-based techniques .......................................................................................... 39

2.3.3. Field ion microscopy.......................................................................................... 39

2.3.4. Chemical selective etching ................................................................................ 40

2.3.5. Electron microscopy techniques ........................................................................ 41

Chapter 3 .................................................................................................................................. 43

3. Electron Holography ..................................................................................................... 43

3.1. Introduction ............................................................................................................... 43

3.2. Off-axis electron holography .................................................................................... 45

3.3. Instrumental requirements for off-axis electron holography..................................... 50

5
 3.3.1. The field emission gun (FEG)............................................................................ 50

a) Temporal coherence ........................................................................................... 50

b) Spatial coherence ............................................................................................... 51

3.3.2. Biprism ............................................................................................................... 54

3.3.3. Lorentz lens........................................................................................................ 55

3.4. Off-axis electron hologram formation....................................................................... 57

3.5. Off axis electron hologram quality............................................................................ 59

3.5.1. Spatial frequency of fringes ............................................................................... 59

3.5.2. Hologram width ................................................................................................. 60

3.5.3. Holographic fringe visibility .............................................................................. 61

a) Effect of exposure time ...................................................................................... 61

b) Effect of biprism voltage ................................................................................... 61

c) Effect of CCD camera ........................................................................................ 62

d) Effect of illumination ......................................................................................... 64

3.6. Experimental measurement of hologram parameters ................................................ 67

3.7. Hologram reconstruction ........................................................................................... 69

3.7.1. Basic reconstruction procedure .......................................................................... 69

3.7.2. Initial filtration ................................................................................................... 71

3.7.3. Selecting the sideband........................................................................................ 72

3.7.4. Reference hologram ........................................................................................... 74

3.7.5. Flattening ........................................................................................................... 77

3.7.6. Phase unwrapping .............................................................................................. 77

3.8. Thickness map ........................................................................................................... 79

3.9. The relationship between the phase and electromagnetic potentials......................... 81

3.10. Mean inner potential .............................................................................................. 83

Chapter 4 .................................................................................................................................. 85

4. Specimen preparation ................................................................................................... 85

6
 4.1. Introduction .............................................................................................................. 85

4.2. Specimen requirements ............................................................................................. 85

4.3. TEM specimen preparation for electron holography ................................................ 87

4.3.1. Focused ion beam principles .............................................................................. 88

4.3.2. Artefacts in FIB-prepared TEM specimens ....................................................... 90

4.3.3. Common FIB-based specimen preparations ...................................................... 92

a) H-bar method ..................................................................................................... 92

b) Ex-situ lift-out .................................................................................................... 94

c) In-situ lift-out ..................................................................................................... 96

4.4. The double lift-out technique ........................................................................................ 98

4.5. Specimen preparation for in-situ electrical biasing ................................................. 102

4.5.1. Introduction ...................................................................................................... 102

4.5.2. Biasing specimen holder .................................................................................. 105

4.5.3. Cube geometry ................................................................................................. 107

4.5.4. Novel specimen geometry which could be applied to a real device ................ 108

A) Preparing the biasing grid ................................................................................ 108

B) Mounting the TEM membrane......................................................................... 109

4.6. In-situ sputter-coating in FIB workstations ............................................................. 111

4.7. A list of the samples used in this thesis................................................................... 113

Chapter 5 ................................................................................................................................ 115

5. A Si pn junction under electrical biasing conditions .................................................. 115

5.1. Results from the cube geometry .............................................................................. 115

5.1.1. The phase-step across the junction as a function of biasing voltage ............... 117

5.1.2. Measuring the electric field and charge density from the phase images ......... 121

5.2. Results from the novel geometry introduced for the biasing of real devices .......... 127

5.2.1. Problems .......................................................................................................... 127

a) Initial biasing experiment ................................................................................ 127

7
 b) Annealing experiment ...................................................................................... 130

c) Repeating the biasing experiment .................................................................... 132

5.2.2. Results from successful biasing experiments with the new geometry ............. 135

5.3. Summary ..................................................................................................................... 140

Chapter 6 ................................................................................................................................ 142

6. Effects of diffraction contrast on the phase profile............................................................ 142

6.1. Introduction ............................................................................................................. 142

6.1.1. Effects of diffraction contrast on the measurement of built-in potentials ....... 144

6.2. Experimental details ................................................................................................ 147

6.2.1. Unbiased pn junction ....................................................................................... 147

a) Systematic tilt................................................................................................... 147

b) Recording holograms ....................................................................................... 150

c) Experimental results......................................................................................... 151

6.2.2. Biased pn junction............................................................................................ 157

a) Specimen specification .................................................................................... 157

b) Recording holograms ....................................................................................... 158

c) Experimental results......................................................................................... 159

6.2.3. Consideration of thickness variation ................................................................ 171

6.3. Summary ................................................................................................................. 173

Chapter 7 ................................................................................................................................ 175

7. Influence of electron radiation .................................................................................... 175

7.1. Introduction ................................................................................................................. 175

7.2. Are FIB prepared surfaces equipotential? ............................................................... 179

7.2.1. Observation of external fringing fields in a FIB prepared pn junction ................ 185

7.2.2. Electric field and charge distribution ............................................................... 187

7.3. The influence of beam current density on the phase profile ................................... 190

7.3.1. Note on using an objective aperture in electron holography ........................... 193

8
 7.4. Electron beam induced current (EBIC) ................................................................... 194

7.4.1. The measurement of EBIC as a function of beam position ............................. 194

7.4.2. The measurement of EBIC as a function of biasing voltage ........................... 196

7.4.3. The measurement of EBIC as a function of beam size......................................... 198

7.4.4. The flow of EBIC through an extraneous resistance can forward bias the
junction 199

7.4.4. Assessing the role of objective aperture .......................................................... 202

7.5. SEM dopant contrast ............................................................................................... 206

7.5.1. Mechanism of dopant contrast ......................................................................... 206

a) Surface states ................................................................................................... 206

b) External fringing field ...................................................................................... 207

c) Metal-semiconductor contact ........................................................................... 208

7.5.2. SEM setup for the observation of dopant contrast ........................................... 209

7.5.3. Biasing experiments in SEM ........................................................................... 210

a) The dopant contrast as a function of biasing voltage ....................................... 210

b) The phase step as a function of biasing voltage ............................................... 212

7.5.4. In-situ metal deposition in SEM ...................................................................... 214

7.5.5. The holographic observation of a metal coated specimen ............................... 216

Chapter 8 ................................................................................................................................ 222

8. Examining a MOS transistor....................................................................................... 222

8.1. Introduction ............................................................................................................. 222

8.2. Sample specification ............................................................................................... 223

8.3. Removing the curtaining effect and providing vacuum path .................................. 225

8.4. Diffraction contrast ................................................................................................. 230

8.5. Charging in a MOS specimen ................................................................................. 233

8.5.1. Holographic observation of specimen charging .............................................. 234

8.5.2. Observation of specimen charging in SEM ..................................................... 236

9
 8.5.3. Carbon coating ................................................................................................. 238

a) Thin layer of carbon (~20nm) .......................................................................... 238

b) Thick layer of carbon (~60nm) ........................................................................ 241

8.6. A semi-biased NMOS transistor ................................................................................. 247

8.7. Summary and discussion ......................................................................................... 254

Chapter 9 ................................................................................................................................ 256

9. Conclusions and future work ..................................................................................... 256

9.1. Conclusions ............................................................................................................ 256

9.2. Future work ............................................................................................................. 261

9.2.1. Surface termination and preparation damage ....................................................... 261

9.2.2. Electron illumination ............................................................................................ 261

9.2.3. Making electrical contacts .................................................................................... 262

References .............................................................................................................................. 263

10
Chapter 1

1. Introduction
Many of the technologies with which almost every aspect of our lives has been revolutionised
rapidly over the last half-century owe, if not their existence, at least their success to the
invention of transistor and its continuous size reduction within integrated circuits. Soon after
the fabrication of the first integrated circuit, it was realised that by reducing the transistor
dimensions a higher performance is achieved at a lower cost. The device size started to shrink
and faster transistors with lower power consumptions were packed in smaller chip areas. The
improvement in the performance along with the cost reduction has been such a strong driving
force for the miniaturisation of transistors that their counts on ICs have been approximately
doubled every two years over the last 40 years. To keep up with this rate many innovative
ideas in the fabrication processes have been developed and implemented in IC
manufacturing. However, as the device dimensions reach the deep sub-micron regime the
fabrication complexities increase drastically, not only due to the device size but also because
of the stringent requirements demanded by the device designers on the precision of the
electrostatic potential distributions across the device. Errors in distributing the electrostatic
potentials within modern miniaturised devices are less tolerable. According to the
international technology roadmap for semiconductors (ITRS), neither are the existing doping
methods controllable enough nor are the potential measurement techniques precise enough to
keep up with next device generation. For the development of a new doping method,
comprehensive feedback is required from a potential measurement technique. This feedback
is also crucial for the parameter optimisation in device modelling. An advanced
characterisation technique capable of providing a quantitative three-dimensional potential
picture of an electrically biased device with high spatial resolution and high precision is
therefore in great demand. The technique of off-axis electron holography in the transmission

11
electron microscope promises to fulfil these requirements in two dimensions and can be
combined with electron tomography for three-dimensional measurements. By using off-axis
electron holography, in principle, the electrostatic potential can be mapped quantitatively
with atomic spatial resolution and better than few tens of millivolts sensitivity under actual
electrical biasing conditions. The practical challenges involved in the application of electron
holography to the measurement of electrostatic potentials in working MOSFET transistors are
addressed throughout this thesis.

1.1. Thesis outline

This dissertation begins with a short introduction to the development trend of semiconductor
devices. Some of the electrostatic potential perturbations posed in MOSFETs as a result of
scaling are described in chapter 2 after explaining the basics of MOSFET operation. The
physics of the pn junction, as the building block of most semiconductor devices, is described
to provide the required background for understanding the results in the later chapters.
Advantages of off-axis electron holography over other alternative techniques in
characterising novel nano-scale semiconductor devices are highlighted when these techniques
are described concisely at the end of this chapter.

Chapter 3 is dedicated to the description of medium resolution off-axis electron holography.

The instrumental requirements for carrying out the technique in a transmission electron
microscope are explained. It is described in detail how a hologram is formed and what
microscope parameters affect the quality of the hologram. The reconstruction procedure used
to extract the amplitude image, phase image and thickness map is described. The relationship
between the phase shift and the electromagnetic potential of the specimen is expressed at the
end of the chapter.

Challenges encountered in preparing TEM specimens for electron holography and in making
electrical contacts for in-situ biasing are discussed in chapter 4. The focused ion beam (FIB)
workstation was used in this study as the primary tool for the specimen preparation, and its
principles, strengths and weaknesses are explained. In this project, the double lift-out method
was developed in order to access to the backside of the FIB prepared TEM specimen and
based on this method making electrical contacts to TEM specimens was facilitated in a novel
geometry. This chapter is confined to the description of specimen preparation techniques
used in this study and the results are reported in subsequent chapters.

12
In chapter 5, the electrostatic potential, electric field and charge distributions extracted from
holographic images of a silicon pn junction under electrical biasing conditions are compared
with those obtained by simulation. Results from two different biasing geometries, including
the one developed as a part of this study, are presented and compared with each other. The
experiments carried out during the development of the novel biasing geometry to identify and
troubleshoot the causes of the failure are described in this chapter.

In chapter 6, the assessments carried out to investigate the influence of diffraction contrast on
the holographic built-in potential measurements are presented. Holographic phase images
acquired from an unbiased and an electrically biased silicon pn junction specimen at different
crystallographic orientations are analysed.

In measuring electrostatic potential distributions using electron holography, emission of

secondary electrons and generation of electron-hole pairs throughout the measurement, as a
result of exposing the specimen to the electron beam, can disturb the original charge
distribution of the sample. In chapter 7, the effects of electron beam irradiation on the in-situ
characterisation of electrical properties in semiconductor devices are assessed using three
complementary techniques: off-axis electron holography and electron beam induced current
measurements in the TEM and secondary electron dopant contrast in the scanning electron
microscope.

In chapter 8, real PMOS and NMOS transistors are examined using off-axis electron
holography. Practical challenges such as curtaining effects, charge accumulations and
diffraction contrast which result from the complex structure of these devices are addressed in
this chapter. Also, the first electron holographic results of a semi-biased MOS transistor are
presented.

In chapter 9, conclusions drawn from this study are summarised. Also, based on these
conclusions a clearer path for moving towards electron holography of working transistors is
shown.

13
Chapter 2

2. Semiconductor devices: physics and characterisation

2.1. Introduction

2.1.1. The trend of semiconductor device development

“The transistor was probably the most important invention of the 20th century.”

Ira Flatow (1)

With the progress in the development of the telephone in the late 19th century, there was a
need for a device capable of switching and amplifying signals. The first electronic
amplification device was the vacuum tube triode, invented in 1906. However, numerous
problems such as high power consumption, large size, costly maintenance, and poor
reliability limited the application of these vacuum tubes. In the 1930s, Mervin Kelly, the
research director of the Bell Labs, recognised that a semiconductor device could be a
promising replacement for the problematic vacuum tube. He formed a department dedicated
to the semiconductor study in Bell Labs.

From that department in 1947, Brattain and Bardeen demonstrated the first working
semiconductor point-contact transistor and three years later Shockley built a more practical
and easier to fabricate pn junction transistor. In 1956, they were awarded a Nobel Prize for
the discovery of the transistor effect.

Very rapidly many products based on the transistor, such as broadcasting radios, were
launched commercially and the designed electronic circuits became more complex. The
number of individual transistors in an electronic circuit soon reached the point that made the
circuits too large to assemble. Time delays in the propagation of electronic signals through

14
these large circuits became a noticeable speed limiting factor. The corresponding solution
was found in 1948 when Jack Kilby and Robert Noyce independently developed a fabrication
process capable of integrating many electronic components into a single chip. For the
invention of integrated circuit (IC), Kilby was awarded a Nobel Prize in 2000. (1; 2; 3)

Without doubt, however, the integration of billions of transistors in today’s microprocessors

and semiconductor memories would have not been possible if the rather easily scalable
MOSFET (metal oxide semiconductor field effect transistor) were not developed in 1960s.
The MOSFET is the most important device for VLSI (very large scale integrated) circuits (4).
The first successful MOSFET was reported by Bell’s scientist John Atalla in 1960 and had a
gate length of 20μm. Only ten years later Intel Corporation released the first commercial
microprocessor, incorporating approximately 2300 MOSFETs with the gate length of 10μm
(5). On the flash-memories launched commercially by Toshiba Corporation in 2010, this
length has been shrunk to 24nm, over 400 times smaller (6). Recently, in February 2011,
Intel Corporation announced that it will construct a $5 billion semiconductor fabrication plant
to reach 14nm node (7).

2.1.2. Basics of MOSFET operation

The structure and working principles of an n-type MOSFET are shown schematically in
Figure 2.1. Two islands, the source and drain, are formed in a p-type silicon substrate.
The central part of the device is a metal-oxide-semiconductor capacitor. The oxide layer
electrically isolates the metal contact, the gate, from the p-type region between the source and
drain.

In Figure 2.1, the source and substrate are connected to the ground as a voltage reference.
When the gate voltage is zero, by varying the drain voltage no current flows between the
source and drain (neglecting the leakage current). The structure can be considered as two pn
junctions connected back to back. By applying a sufficiently large positive voltage to the
gate, called the threshold voltage, the induced electric field in the oxide layer attracts enough
negative charges to locally invert the p-type region between the source and drain into the n-
type. This inversion layer with which the source and drain are connected is called the
channel.

For small drain voltages, the channel acts as a constant resistor between the source and drain,
which can be modulated by the gate voltage. As shown in Figure 2.1a, the drain-source

15
current increases linearly with the drain voltage. A larger voltage on the drain means a larger
electric field between the source and drain and therefore a larger drain current. It also means
a smaller electric field in the oxide layer close to the drain than the source, so the channel
depth decreases across the channel from the source to the drain. And if approximately the
same voltage as the gate voltage is applied to the drain (assume that the threshold voltage is
zero), the channel depth at the drain would be zero (Figure 2.1b). This phenomenon is called
pinch-off. Further increase in the drain voltage pushes the pinch-off point further towards the
source (see Figure 2.1c), but the potential at this point Vsat remains the same. The drain
voltages in excess of Vsat are dropped across the pinched-off section ∆L. If, for the drain
voltages larger than Vsat, the channel length L is assumed to be much larger than ∆L, ,
the electric field across the channel remains almost the same and so does the drain current. (4;
8)

Figure 2.1. Basics of MOSFET operation. a) Small drain voltage

(the IV curve demonstrates the constant resistance
characteristics). b) Onset of channel pinch-off (the drain current
saturates). c) Beyond the saturation (∆L<<L). (4)

16
The MOSFET functionality is described by the distribution of electrostatic potentials across
the device. For four different electrical biasing conditions, the perspective view of this
potential distribution is plotted in Figure 2.2. For simplicity, the p-substrate and the source
are assumed to be connected to the ground. The large potential barriers associated with the
two pn junctions between each of the regions, the source and drain, and the p-substrate are
depicted in Figure 2.2b. As a positive voltage larger than the threshold voltage Vth is applied
to the gate the channel is formed and the two potential barriers are lowered, particularly close

Figure 2.2. Two dimensional distribution of electrostatic potential in a) a MOSFET when: b)

The gate and drain are zero biased. c) A voltage larger than the threshold voltage is applied to
the gate (channel is formed). d) A voltage lower than the saturation voltage is applied to the
drain. e) The channel is pinched-off as a voltage larger than the saturation voltage is applied to
the drain. (9)

to the oxide/substrate interface (Figure 2.2c). The gate voltage modulates the potential
distribution along the x direction, while the drain voltage changes the potential distribution in
the y direction. By applying a small positive voltage to the drain, an electric field is induced
across the channel (Figure 2.2d). In this condition, the electrons in the source region, the

17
majority carriers, only need to overcome a very low barrier between the source and the
channel before the electric field inside the channel accelerates them towards the drain. When
the channel is pinched-off by large drain voltages, the electric field along the channel remains
the same (for long channels) but the field in the pinched-off section becomes larger (Figure
2.2e). (9)

2.1.3. MOSFET Scaling

Smaller MOSFETs switch faster, consume less power and can be packed more densely in a
chip. This means more functionality can be gained in the same chip area. Thus, a higher
performance can be achieved at a lower cost. That is the driving force which has doubled the
transistor count on an IC every two years since 1971 (Moore’s Law) (10).

The difficulties with the scaling of MOSFETs used to be associated only with the fabrication
processes. Fabricating MOSFETs with sub-micron gate lengths indeed is a tough challenge
for the process engineers, but fabrication is not the only problem. In addition to that, several
operational issues also emerge as the device size reaches the sub-micron and deep sub-micron
regimes. These malfunctions can be surmounted in many cases by distributing the
electrostatic potentials across the device more precisely. Accurate control on the potential
distribution can be achieved by modifying the design and structure of the MOSFET. In the
following paragraphs some of these operational difficulties, resulting directly from MOSFET
miniaturisation, and their related solutions are explained.

Many of the device malfunctions are posed by high energetic charge carries. Generally, the
power-supply voltage is not reduced as rapidly as the device dimensions scaled, since it
degrades the frequency response of the device and also causes difficulties for the circuit
engineers (11). Applying the same source-drain voltage to a MOSFET with a shorter gate
length yields an increased lateral electric field between the source and drain. Charge carriers
can gain sufficient energy from this lateral electric field to overcome the potential barriers set
in the MOSFET structure for confining the carrier transport within the channel. These highly
energetic carriers, called hot carriers, can tunnel through the gate oxide barrier and appear,
for instance, as the gate current. Also, they can overcome the pn junction barriers and
transport from the source to the drain throughout the bulk, not the channel. For the same
reason, hot carriers can appear as the substrate current. These degradations in the MOSFET
performance can be reduced either by making the potential barriers larger or by decreasing
the electric field maximum in the channel.

18
The electric field in the channel is maximised at the drain because of the abrupt reverse
biased pn junction between the drain and the channel. To decrease the electric field peak, a
graded or lightly-doped drain extension is fabricated instead of the heavily doped abrupt
junction. Figure 2.3 shows an example of a MOSFET structure with a lightly doped drain
extension and the corresponding simulated electric field distribution expected to achieve in
the drain region under biasing conditions. This profiled lightly doped drain structure is
expected to reduce the maximum of electric field in the channel, set the maximum of electric
field under the gate, separate the highest drain current density path (marked with an arrow in
the figure) from the point where the electric field is maximised and minimise the resistance
that the drain current experiences.

Figure 2.3. The structure of a profiled lightly doped drain extension

and corresponding simulated electric field distribution. In the
simulated electric field, note that the path of the highest current
density, shown by an arrow, has been separated from the highest
electric field contour, 3×105 V/cm. (12)

Another operational difficulty posed by the scaling is the so-called punchthrough. It is an

extreme case of channel length shortening where the depletion regions around the source and
drain merge into a single depletion region, as shown in Figure 2.4. The drain current flowing
in this depleted region under the channel is poorly controlled by the gate voltage as it is far
from the gate. The punchthrough current adds to the off-state current of the device and its
magnitude depends mainly on the potential distribution under the channel. This leakage
current degrades the switching performance, output resistance and power consumption of the
device; and, therefore needs to be minimised.

19
 Figure 2.4. Punchthrough phenomena. The
depletion regions around the source and drain
merge into a single region. (9)

The punchthrough current is expected to be suppressed by implanting p-type (or n-type)

regions in NMOSFETs (or PMOSFETs) under the lightly doped drain extensions. These
punchthrough stopper regions, called “halo”s, are formed by ion implantations at high tilt
angles after defining the gate as shown in Figure 2.5.

Figure 2.5. The p+ halo implantation in an

NMOSFET. (13)

The most important factor in the MOSFET performance is that only the gate voltage controls
the channel. The electric field lines from the source and drain needs to be shielded so that
they do not affect the channel. The extra pn junctions introduced by the halo implantation can
shield these field lines, but in very small devices the required dopant concentration in the halo
regions becomes too high (1019cm-3) (14).

As predicted in the International Technology Roadmap for Semiconductors (ITRS), new

MOSFET structures such as those shown schematically in Figure 2.6 need to be employed for
maintaining the scaling rate (15). These structures are aiming to keep the channel under the
control of the gate voltage at all source-drain voltages. In the conventional bulk structure
shown in Figure 2.6a, the electric field lines from the source and drain affect the channel. In
comparison, in a fully-depleted (FD) silicon-on-insulator (SOI) MOSFET, shown in Figure

20
2.6b, most of the field lines are expected to terminate on the ground layer under the buried
oxide (BOX) layer. This structure can even improve further by introducing additional gates,
double gate (DG), Figure 2.6c, and multiple-gate MOSFETs. The electric field lines under
the device terminate on these additional gates and cannot reach the channel. In other words,
the channel is shielded by its surrounding gate electrode.

Figure 2.6. a) Conventional bulk MOSFET, b) fully depleted silicon on insulator MOSFET and
c) Double gate fully depleted silicon on insulator MOSFET. The arrows show the source-drain
electric field directions. (14)

It is anticipated that by 2013 the conventional bulk MOSFETs will be replaced in almost all
high performance (HP) digital circuits by the FDSOI devices and by 2015 the FDSOI
MOSFETs will reach their limit in HP digital circuits (14). The ITRS 2011 has also predicted
that in 2018 MOSFETs with the new channel materials such as Ge for p-channel and III-V
for n-channel will be produced (15). Perhaps new materials such as carbon nanotubes,
nanowires and graphene will be integrated in the future devices before MOSFET scaling
becomes ineffective (15).

21
2.2. Semiconductor physics: basics of pn junctions

2.2.1. Introduction
As detailed in the previous section, a transistor is an adjustable transport path for the charge
carriers. This path is provided by a controllable distribution of electrostatic potentials. The
building blocks of this electrostatic potential path are junctions which are formed between
materials of different electronic structure: semiconductors, metals and insulators. Insulators
are used to make large potential barriers to confine the carriers within the path. Metals are
used to connect the path to the external power supplies with which the potential distribution
across the path is adjusted. The key building block of almost all semiconductor devices,
including transistors, is the tuneable potential barrier formed at the interface of p-doped/n-
doped semiconductor. The physics of this pn junction is briefed in this section.

2.2.2. Role of doping in semiconductors

One of the main reasons that semiconductors are the most useful material in the fabrication of
electrical and optical devices is that the electrical and optical characteristics of
semiconductors can be modified with ease by controlling the type and amount of impurities
intentionally introduced into them. The process of intentionally adding impurities to a
semiconductor in order to modify its characteristics is called doping.

A perfectly crystalline semiconductor with no impurities is called an intrinsic semiconductor.

At the absolute zero temperature 0K in an intrinsic semiconductor all of the most outer-shell
electrons are engaged in a particular atomic bond. Therefore in these conditions a
semiconductor can be considered as an insulator. By raising the temperature, electrons obtain
the minimum required energy, the bandgap energy, to be excited into the conduction band.
For each electron excited into the conduction band a hole is generated in the valence band,
yielding an electron-hole pair. Thus, in an intrinsic semiconductor, the electron concentration
in the conduction band ni at a given temperature is equal to the hole concentrations in the
valence band pi, i.e.

Eq. (2.1)

This concentration at room temperature 300K is 1. 1010 cm-3 for silicon (4). By adding
a specific type of impurity, this balance can be manipulated towards mostly electrons, n-type,
or mostly holes, p-type, either of which are called extrinsic semiconductors.

22
In order to calculate , two functions should be introduced: f E and E . The probability
that an available energy state at the energy E is occupied by an electron is given by the
Fermi-Dirac function f E : (16)

1
f E E Ef Eq. (2.2)
1 e T

where K is Boltzmann’s constant and T is the absolute temperature. Ef is called Fermi level

and represents the energy with the probability of occupation of . At 0K all the available

energy levels below the Fermi level are occupied and all the available energy levels above the
Fermi level are empty.

Secondly, with the assumption of parabolic structure at the bottom of the conduction band,
the density of allowed energy states in the conduction band for the bulk of a crystal is given
by:

3
2 mn 2
E E Eq. (2.3)
2 2

where mn is the effective mass of an electron and 2 (h is Planck’s constant).

Therefore, at the thermal equilibrium the concentration of electrons in the conduction band
can be calculated by integrating f E E over the entire conduction band as given in the
following equation:

n f E E dE Eq. (2.4)
Ec

Ec is the energy at the bottom of the conduction band. Substituting Equation 2.2 and 2.3 into
Equation 2.4 and assuming E - Ef T yield:

Ec Ef
n Eq. (2.5)
ce T

where c, the effective density of states in the conduction band, is calculated by:

3
2 mn T 2
2 Eq. (2.6)
c 2

23
Similarly, the concentration of holes in the valence band can be calculated by:

Ef Ev
p e T Eq. (2.7)
v

3
2 mp T 2
v 2 2 Eq. (2.8)

where v is the effective density of states in the valence band and mp is the effective mass of
a hole.

Assume that the Fermi level for an intrinsic semiconductor is Ei. According to Equation 2.5
and Equation 2.7 the intrinsic electron and hole concentration are given respectively by

Ec Ei
ce T Eq. (2.9)

Ei Ev
Eq. (2.10)
v e T

As for an intrinsic semiconductor, Ei is given by:

Ec E v T v
Ei ln Eq. (2.11)
2 2 c

The effective density of states in the conduction band and in the valence band of silicon at the
room temperature are c 2. 1019 cm-3 and v 1.0 1019 cm-3 , respectively. Therefore, the
intrinsic Fermi level of silicon is only lowered by 25meV from the middle of the band
gap (Ec Ev ) 2 as the temperature is elevated from absolute zero to the room temperature. On
the other hand, the position of the Fermi level can be tuned with ease by adding dopants into
the crystalline structure of an intrinsic semiconductor.

Adding donors, impurities that simply donate their extra electrons to the conduction band,
moves the Fermi level towards the conduction band and adding acceptors, impurities which
easily accept electrons from the valence band, shifts the Fermi level towards the valence
band. For silicon, elements such as P, As and Sb in the column V of the periodic table are
donors as they have one more valence electron than silicon and make the silicon n-type.
Likewise, elements such as B, Al and Ga in the column III are acceptors and make the silicon
p-type.

24
By simple multiplication it can be shown that in extrinsic semiconductors, the product of the
electron and hole density equals the square of the intrinsic carrier density:

Eq. (2.12)

This means that for instance for a silicon wafer doped with 1018cm-3 phosphorus the electron
density is ~1018cm-3 (assuming that all dopants have been ionised) while the hole density is
only ~200cm-3. Thus in an n-type semiconductor the majority carriers are electrons.

The electron and hole density for an extrinsic semiconductor, given in Equation 2.7 and
Equation 2.9 respectively, can be reformulated easily and expressed with the intrinsic carrier
density and intrinsic Fermi level Ei as:

Ef Ei Eq. (2.13)
e T
Ei Ef
e T Eq. (2.14)
All the above discussions are summarised schematically in Figure 2.7 where the band
diagrams, the density of states, the Fermi-Dirac distributions and the carrier concentrations of
intrinsic, n-type and p-type semiconductors have been shown. (4)

As shown in this figure, the concentration of electrons and holes are equal for the intrinsic
semiconductor and the Fermi level is almost at the middle of the bandgap at a given
temperature, while in the extrinsic semiconductors the Fermi levels are shifted towards the
conduction or valence band.

25
 Figure 2.7. Band diagram, density of states, Fermi-Dirac distribution, and the carrier
concentrations for a) intrinsic, b) n-type, and c) p-type semiconductors. (16)

The semiconductor resistivity , the reciprocal of the conductivity , depends on the dopant
concentration:

1 1 ( μn μp ) Eq. (2.15 )

where μn and μp are the electron and hole mobilities respectively. The graphs in Figure 2.8

show the resistivity of phosphorus- and boron-doped silicon as a function of dopant

concentration. It can be seen in this figure that for a given dopant concentration, the n-type
silicon provides a more electrical conductive path than the p-type since the electron mobility
at a given temperature is larger than the hole mobility. Mapping the resistivity of a
semiconductor wafer is a very common technique in the semiconductor industry for the
determination of the active dopant distribution. (17)

26
 Figure 2.8. Resistivity of boron-doped (p-type) and phosphorus-
doped (n-type) silicon as a function of dopant concentration. (17)

2.2.3. The pn junction formation

After explaining the p- and n-type semiconductor separately, it is now time to find out what
happens when these two types of semiconductors are brought together to form a pn junction.
As other types of semiconductor junctions such as metal contacts and heterojunctions can be
explained similarly (4; 16), here we have limited ourselves to a simple abrupt pn junction
which is the main building block of semiconductor devices.

Due to the large carrier concentration gradient, joining the p-type and n-type regions results
in the diffusion of the carriers from one type to the other one. Diffusion of carriers continues
until enough electric field is built up to compensate the force of the concentration gradient.

27
 Figure 2.9. Energy diagram of uniformly doped p-type and n-type
semiconductors a) before and b) after pn junction formation (16)

As shown in Figure 2.9, the electric field built up across the junction is against the carrier
diffusion and at thermal equilibrium the net carrier transition due to the diffusion and drift is
zero. Accordingly, the drift and diffusion current densities for both electrons and holes must
cancel each other:

Jp Jp drift Jp diffusion 0 Eq. (2.16)

Jn Jn drift Jn diffusion 0 Eq. (2.17)

The drift and diffusion current for the holes are given by,

Jp drift μp Eq. (2.18)

d
Jp diffusion - p dx Eq. (2.19)

T
where p μp is the diffusion coefficient, q is the electric charge, p is the hole

concentration and is the electric field across the junction. Therefore:

28
 d Eq. (2.20)
Jp μp T μp 0
dx
Substituting the first derivative of Equation 2.14, the hole concentration, in Equation 2.20
results in:

dEf
J p μp 0
dx
or
dEf
0 Eq. (2.21)
dx

Likewise, the same result is obtained for the electron current density Jn. This result is one of
the golden rules in the physics of junctions saying that the Fermi level must be constant
throughout the device when the net current flow is zero. To satisfy this rule, a potential
barrier is built across the junction. This built-in potential, Vbi, can be derived by rewriting
d x
Equation 2.20 for the electrostatic potential instead of the electric field, x - :
dx

d x T 1 d x T 1
. . d d
dx x dx

T
bi n p ln
Eq. (2.22)

Where Vn and Vp are electrostatic potentials in n and p regions away from the transition
region, shown in Figure 2.9. and represent the hole concentrations in the n and p

regions respectively. Instead of using the hole concentrations, Equation 2.22 can be rewritten
with the dopant concentrations:

T a d
bi ln 2
Eq. (2.23)
-
where a and d are the ionised acceptor and donor concentrations.

In addition, Poisson’s e uation can be used to relate the space charge, shown in Figure 2.9, to
the electrostatic potential V across the junction.

d2
d a
Eq. (2.24)
dx2

where is the charge density and is the relative permittivity.

29
Neglecting the carriers within the space charge, known as depletion approximation, and
-
assuming complete ionization of the impurities, d d and a a, greatly simplifies the
calculation. Therefore the electric field E(x) can be calculated along the junction as
following:

d 0<x< xno
dE d2
- dx2
dx
- a -xp <x<0
o

d (x - xno ) 0<x<xno
E(x)
- a (x xpo ) -xp <x<0 Eq. (2.25)
o

The electric field distribution calculated by the above equations is depicted in Figure 2.10.
The area of the triangle in this figure corresponds to the potential Vbi. Therefore, the width of
the depletion or transition region W can be calculated by:

1
bi - 2 E0 W
1
bi . a d
W2
Eo Eo Eo 2 a d
a d
W xno xpo
d a a d

2 a d
W bi
a d
Eq. (2.26)

Equation 2.26 shows that the width of the depletion region W varies as a function of the
potential barrier across the junction Vbi.

30
 Figure 2.10. Space charge and electric field
distribution within the depletion region of a pn
junction (4; 16)

It should be borne in mind that in the derivation of above equations the following
assumptions (4; 16; 18) are made:

- the band structure is parabolic near the bottom of the conduction band and top of
the valence band
- the concentration of electrons or holes is low in comparison to the density of states
in the conduction or valence band so that Ef sits well below the conduction band
(~3kT) or above the valence band, so-called non-degenerate condition.
- the transition between the n-type and p-type area is made abruptly (Abrupt
junction)
- the carrier concentration within the transition region at a given time is negligible
(Depletion approximation)
- the effects of semiconductor surface on the properties of the junction are
negligible

The validity of some of these assumptions such as the negligibility of surface effects needs to
be reassessed for TEM studies. Surface conditions, as will be discussed in chapter 7, have a
profound influence on the measured electrostatic potentials using electron holography.

31
2.2.4. Biasing a pn junction
The built-in potential across the pn junction can be tuned by applying external voltages to the
junction. If the p-side of the junction is connected to a higher potential than the n-side, the
external potential will subtract from the built-in potential and lower the potential barrier. This
biasing condition is called forward bias. The balance between the diffusion and drift currents
in Equation 2.16 and Equation 2.17 is perturbed. As the potential barrier decreases, the
diffusion currents increase. The minority carriers are injected to the p- and n- side of the
junction by the opposite sides. The density of minority carriers injected to each side depends
exponentially on the external forward voltage and decays also exponentially with distance as
moving away from the depletion region. This relationship on the n-side and p-side, shown
graphically in Figure 2.11, is expressed by the following equations:

(x xn ) Lp Eq. (2.27)
side e T 1 e

side n e T 1 e( x xp ) Ln
Eq. (2.28)

where and are the hole concentration on the n-side at the thermal equilibrium condition
and after applying the forward voltage of V, respectively. Lp is the diffusion length of holes
(minority carriers). , and Ln are the similar parameters for the electrons on the p-side.

Figure 2.11. Distribution of minority carriers injected to

the n- and p-side of a pn junction as a result of a
forward bias. (4)

The diffusion length of minority carriers is related to the carrier lifetime by the diffusion
coefficient L . Plots in Figure 2.12 show the variation of Ln and Lp as a function of

32
dopant concentration for the bulk silicon. It can be seen in this figure that the diffusion
lengths of minority carriers in silicon for the dopant concentration of 1018-1019cm-3 are larger
than 1μm. However, the minority-carrier diffusion-length measurement carried out on a FIB
prepared TEM specimen in chapter 7 show much smaller lengths. In thinned samples such as
TEM specimens the surface recombination limits the diffusion length of carriers. Many of the
carriers reach the surface and recombine before being able to diffuse to such large distances.

Figure 2.12. Minority carrier lifetime and diffusion-length as a function of dopant concentration in a
p-type (right) and n-type (left) bulk silicon. (19)

To obtain the total current density, the currents due to the diffusion of minority carriers need
to be added together:

J Jp diffusion Jn diffusion
d d
p n
dx xn
dx xp

J Js e T 1 Eq. (2.29)

p n
Js Eq. (2.30)
Lp Ln
where Js is the saturation current density. In the reverse bias condition, where the n-side is
connected to a higher potential than the p-side, the external voltage adds to the potential
barrier and the depletion width increases. Because of the high potential barrier the diffusion
currents greatly reduce and the reverse current density saturates at Js . However, in practice
the current-voltage characteristic, IV curve, of a pn junction only qualitatively follows the
Equation 2.29. In Figure 2.13, the measured IV curve of a bulk Si pn junction is compared
with the ideal one.

33
In the derivation of Equation 2.29 the generation-recombination processes in the depletion
region are assumed to be negligible. However, under the reverse-bias condition, the
generation current in the depletion region becomes dominant over the diffusion current in the
neutral regions. That is why the reverse current, shown as region (e) in Figure 2.13, is
significantly larger than the ideal saturation current. Also, under small forward biases, region
(a), the recombination current in the depletion region results in a larger forward current than
that predicted by Equation 2.29. By increasing the forward bias, region (b), the diffusion
current becomes dominant and the curve follows the ideal behavior. At higher forward
voltages, the density of the injected minority carriers becomes comparable with the majority
carriers, called the high injection condition. In this condition, region (c), the external potential
does not drop exclusively across the junction. An ohmic potential-drop appears outside the
depletion region on the p- and n-side. At higher currents, region (d), the finite resistivity of
the n- and p-side (outside the depletion region) limits the current.

Figure 2.13. Comparison of the ideal current-

voltage IV characteristics of a pn junction with
the one experimentally measured from a real
bulk silicon diode. (4)
Applying a large reverse voltage across a pn junction can break the junction and results in a
large reverse current, as shown in Figure 2.13. The breakdown current is normally induced by
the tunneling and/or avalanche mechanisms. For pn junctions with high dopant
concentrations, where the depletion width is relatively narrow, if the bottom of the
conduction band on the n-side is lowered below the top of the valance band on the p-side via
34
applying a reverse voltage, the carriers would have a large probability to transmit or tunnel
through the potential barrier. For pn junctions with moderate dopant concentrations, the
breakdown current stems mainly from a repeated generation of electron-hole pairs, called the
avalanche process. A carrier is accelerated by the large electric field across the reverse biased
junction to energies sufficient to break the lattice bonds when it collides with atoms. The
electron-hole pair generated as a result of this collision is accelerated again by the same field
and gains enough energy to free other electron-hole pairs. This loop continues, generating a
large reverse current.

The avalanche and tunneling processes are not destructive; however, if the current is not
limited by an external circuit, the induced heat can alter the junction properties irreversibly.
In silicon diodes with a breakdown voltage of less than about 4.5V, the dominant breakdown
mechanism is the tunneling process and when the breakdown voltage is larger than about 7V,
the dominant breakdown mechanism is the avalanche process. The critical electric field at
which the potential barrier across a pn junction starts to breakdown is plotted as a function of
dopant concentration in Figure 2.14. The external reverse voltage required to induce such an
electric field can be calculated from the expression (derived from Equation 2.26):

E2c a d
breakdown bi Eq. (2.31)
2 a d

For a symmetric, abrupt pn junction with the dopant concentration of 5×1018cm-3, the critical
electric field initiating the breakdown is 2×106V/cm and according to Equation 2.31 an
external reverse voltage of Vbreakdown= 4.24V is required to induce such a field.

Figure 2.14. Critical electric field at breakdown as a

function of dopant concentration for Si and GaAs. (4)

35
The IV curve of a pn junction can also be affected by its surface conditions. The presence of
charges on the surface of semiconductors induces a depletion layer on the surface, which can
cause current leakage. For a planar silicon pn junction this current is negligible compare to
the generation current in the depletion region (4), but its effects on a TEM specimen can be
significant. The surface depletion width reaches its maximum when the surface is strongly
inverted. As plotted in Figure 2.15, the maximum width of this surface depletion layer can be
comparable to the thickness of TEM specimens particularly for low dopant concentrations
(20). The effects of surface conditions will be discussed further in chapter 7.

Figure 2.15. Maximum of depletion layer thickness as

a function of dopant concentration for silicon and
GaAs under strong inversion condition. (4)

36
2.3. Characterisation of semiconductor devices

As explained earlier in this chapter, the precise placement of dopants is of vital interest to the
semiconductor industry because the distribution of electrostatic potential is mainly controlled
by the dopants. New metrology techniques are demanded in the International Technology
Roadmap for Semiconductors (ITRS) for mapping the distribution of dopants with spatial
resolution of 1.8nm and concentration precision of better than 2% in three dimensions (15).
Of the numerous methods detailed during the past several years for measuring the 2D/3D
distribution of dopants, none of them has yet fulfilled all the requirements in terms of the
lateral resolution, quantification and precision. However, a few of these methods may be able
to satisfy the requirements proposed by ITRS in the future. In Table 2.1 these methods have
been summarised in five categories, and the most advanced methods of each category have
been chosen for comparison. Each technique is described very briefly and its pros and cons
are addressed below.

Table 2.1. Comparison of dopant/carrier profiling methods (21; 22; 23; 24)
Category Group Resolution Dynamic Quantifiable
representative (nm) range
I) SPM techniques SCM and SSRM 10 - 20 1015 – 1020 cm-3 Limited / yes
II) 1D-based techniques imaging-SIMS 50 Not available Yes
III) Field ion microscopy APT <1 10 – 10 cm
15 21 -3
Yes
IV) chemical etching + other Chemical etch +
10 - 20 1017 – 1021 cm-3 Very difficult
techniques AFM
V) EM techniques EH 1 - 10 1017 – 1020 cm-3 Yes

2.3.1. Scanning Probe Microscopy (SPM) techniques

This category includes techniques such as scanning tunnelling microscopy (STM), Kelvin
probe force microscopy (KPFM), scanning capacitance microscopy (SCM) and scanning
spreading resistance microscopy (SSRM). In all the SPM techniques, scanning a probe
laterally over the sample varies a measurable parameter between the sample and probe such
as electrical resistance, electrical current, electrostatic force or electrical capacitance.
Depending on the type of method, variations in one of these parameters are monitored and
converted to a dopant or carrier map. SCM and SSRM are described briefly in the following
paragraphs. (25)

In standard SCM, a conductive probe tip is scanned on the specimen surface covered with a
thin high quality oxide layer. An MOS capacitor is formed between the conductive tip, oxide

37
layer and specimen. The differential capacitance dC/dV of this MOS capacitor is measured at
each scanning position by applying an AC-voltage to the tip. From the acquired capacitance
image the distribution of carrier concentrations is extracted. This technique has been
successfully applied to an electrically biased MOS transistor (26).

Usually conventional mechanical polishing is applied for the specimen preparation and UV
o
assisted oxidation at low temperature (250-300 C) for the oxide layer formation.
Commercially available highly doped silicon tips are commonly used as the SCM probe. This
technique is so sensitive to the quality of specimen preparation, probe tip and oxide layer that
reliable quantitative information can hardly be extracted from SCM images. Any charges in
the oxide layer or at the specimen/oxide interface can affect the result significantly. Also, the
high lateral force, applied to the tip for the scanning purpose, gradually degrades the tip and
consequently the quality of the SCM image. This latter problem can be overcome to a great
extent by using the tip in the oscillation mode, intermittent contact mode (IC-SCM). If the tip
oscillates with a high frequency throughout the scanning, the lateral force will be
considerably reduced. Consequently in this mode of operation, a high quality oxide can be
grown on the tip instead of the specimen surface. The distribution of dopant concentrations
can be imaged in a wide range from 1015 to 1020 carriers/cm-3 using the IC-SCM technique
(27). This wide dynamic range is achieved at the cost of noise level, increased by parasitic
capacitances between the tip and specimen. (25)

In SSRM, a very sharp conductive wear resist tip is scanned in the contact mode across the
specimen surface. The specimen is biased by applying a DC voltage between the tip and fixed
electrical contact established at the back of the specimen. From the current measurement, the
resistivity of the specimen is determined at each scanning position. As explained in section
2.2.2, the distribution of dopant concentrations can be obtained from this resistivity map.

The SSRM images are quantitatively interpretable and that is the biggest advantage of SSRM
over SCM in dopant profiling. Spatial resolution of better than 3nm has been reported in a
SSRM case study of an MOS transistor (28). Also, a wide dynamic range of 1015 to 1020
carriers/cm-3, the same as SCM, has been demonstrated with SSRM (22).

However, the accuracy of the technique in determining the spread resistance depends strongly
on the quality of the ohmic contacts between the tip and the specimen and also between the
specimen and the backside contact. Therefore, it is difficult to reproduce the result. In
preparing the sample an excellent electrical contact has to be established on the backside.

38
More importantly, the tip needs to be pushed on the specimen surface with a high force,
larger than 1μ , in order to penetrate the native oxide present on the surface of the specimen
and realise a stable ohmic contact with the underlying materials. The tip, which should be
better than 10nm in radius and 0.3nm in RMS roughness, needs to be hard enough that its
sharpness and smoothness remain almost unchanged under such a high force during the
scanning process. Only diamond tips fabricated with the special process can fulfil such
requirements. It should be emphasised that all SPM techniques, including SCM and SSRM,
are restricted to two dimensional analyses as they only provide surface information. (25)

2.3.2. 1D-based techniques

Most of the techniques in this category are based on the widely used secondary ion mass
spectroscopy (SIMS), which is a powerful 1D method for dopant profiling. In the SIMS
technique, a primary focused ion beam is used to sputter away the atoms from the surface of
the sample. As a result of bombarding the sample with primary ions, secondary ions are
ejected from the sample. These secondary ions are directed to a mass spectrometer for
elemental identification.

The ability to detect elements present in the parts-per-billion (ppb) range and to quantify
them with high accuracy has made SIMS a common tool for 1D dopant profiling. The method
is chemically sensitive enough to detect the dopant concentrations of as low as 1015 /cm-3.
Lateral spatial resolution of better than 5nm can be achieved using SIMS, but unfortunately
the depth resolution of the technique is limited to about 50nm. Although SIMS can be applied
to 3D tomographic analysis, the material intermixing caused by the ion beam limits the
spatial resolution of the technique in depth. Lowering the incident beam energy and
decreasing the incident angle can reduce the ion damage but at the same time degrades the
sensitivity of the technique. Normally, in characterising semiconductor devices, the
concentration of active dopants is of interest but in SIMS-based techniques the dopant atoms
are detected directly regardless whether they are active or not. (22; 29)

2.3.3. Field ion microscopy

The most advanced technique in field ion microscopy is atom probe tomography (APT). In
this technique, an electric field of ~ 1010 Vcm-1 is applied between a sharp needle-shaped
specimen (< 100nm in diameter) and an electrode. This electric field is assisted either by
extra electric field pulses or by laser pulses to evaporate the atoms from the specimen apex.
The evaporated ions are chemically identified by time-of-flight mass spectroscopy, and their

39
original positions in the specimen are determined using a position-sensitive detector. (30) The
APT technique can provide a quantifiable 3D elemental map with high precision (~10 atomic
parts per million) and atomic resolution (~0.2nm lateral resolution). The technique has been
recently applied to many semiconductor materials and devices. (31)

Although in principle the precision, resolution and dynamic range of the ATP technique
satisfy the ITRS requirements, in practice several difficulties emerge. The most challenging
part is the specimen preparation since the APT specimen must be in the form of a sharp
needle. Locating the semiconductor device of interest within the specimen apex has only been
possible recently with the use of FIB/SEM dual-beam workstations. The ion beam damage is
still a significant problem particularly because the volume of specimen which can be
analysed with APT is very limited. The field of view is normally restricted to less than
200nm in diameter and 500nm in depth. The technique is completely destructive and after the
APT experiment no further characterisation can be carried out on the specimen. If any other
characterisation method such as TEM analysis is required, it must carry out prior to the APT
measurement. Pre-characterisation is not always efficient since the success rate of APT
technique is not high, usually less than 30%, and depends on the material and specimen.
Semiconductor devices are usually composed of various materials with different field
evaporation rates, which degrade the resolution of the APT technique. All dopant atoms,
regardless of their electrical activity, are measured in the dopant concentration measurement
using APT. (24; 30; 32)

The APT technique is a strong candidate for dopant mapping, but its reproducibility,
accuracy and field of view need to be improved significantly before it can be used routinely
in the semiconductor industry.

2.3.4. Chemical selective etching

In these techniques, the selectivity of chemical etching is employed to map the dopant
concentration to the depth profile. In other words, a chemical etchant which removes material
with a rate proportional to the dopant concentration is used to translate the dopant distribution
to the depth variations. Afterwards, the depth profile can be measured by microscopy
techniques such as AFM, TEM and SEM.

These techniques are useful for junction delineation and if experimental parameters such as
etchant concentration are carefully controlled, then spatial resolution of 10nm can be

40
achieved in the delineating of a junction. However, these methods are very difficult to
quantify and not suitable for finding the dopant distribution. Poor reproducibility is another
drawback of using these techniques. (33; 34; 35)

2.3.5. Electron microscopy techniques

This category includes techniques such as electron-beam-induced current (EBIC), field
emission scanning electron microscopy (FE-SEM), annular dark field scanning transmission
electron microscopy (ADF STEM) and electron holography (EH). Among these techniques
electron holography, which is a TEM based technique, exhibits the best performance and is
very promising when it comes to satisfying the requirements specified by ITRS.

For an electron beam, the electrostatic potential distribution within a TEM specimen can be
viewed as a distribution of refractive indices. The electron beam experiences dissimilar phase
shifts as passes through different refractive index media. This phase shift distribution, which
is proportional to the potential distribution, can be revealed by the means of off-axis electron
holography. In this technique, half of a coherent electron beam is passed through the sample
to record the potential distribution and the other half is used as a reference. By applying a
controllable and symmetric electric field, these two halves are deflected towards each other
and interfere. From the recorded interference pattern, the phase distribution is extracted.

Because of the ability of mapping the electrostatic potential, electron holography has a great
prospect of becoming the most desirable technique for the characterisation of semiconductor
devices. Using electron holography, in principle, the electrostatic potential can be mapped
quantitatively with atomic spatial resolution and better than few tens of millivolts sensitivity,
far better than those demanded by ITRS. In practice also it has been shown that a spatial
resolution of 1nm (36) and sensitivity of ~30 mV (37) are achievable, but these results have
not been verified on device structures. So far in electron holography of MOS transistors, not
better than a spatial resolution of ~6nm and sensitivity of 0.1V have been reported in the
literature (38). Also, in detecting dopant concentrations, a case electron holography study on
a test silicon sample has succeeded to resolve boron concentrations of only larger than 1×1017
cm-3 (39).

It is clear that to fully exploit the capabilities of electron holography for semiconductor
characterisation, further dedicated studies are required. Inherently, semiconductor devices are
integrated within bulk materials and they work under these conditions. In contrast, in electron

41
microscopy the devices are examined within a very thin electron transparent lamella. In
addition to the specimen preparation techniques which might significantly change the
potential map of the device, the specimen surfaces can also introduce regions of different
potential distributions from the bulk. Surface effects and specimen preparation artefacts must
be considered in the interpretation of holographic images. When a semiconductor device is
examined in an electron microscope, another departure from the original device condition is
the fact that the device is irradiated with electrons. The specimen charging and electron-hole
pair generation resulting directly from electron beam irradiation can perturb the potential
distribution (40). Also in electron holography it is assumed that the phase shift of the electron
beam solely stems from the defined refraction index of the material while electron beam
diffractions can also cause similar phase changes in the holographic phase images.

Electron holography has the desirable capability of mapping the electrostatic potential of a
device under actual electrical biasing conditions, which also awaits further development.
Simple pn junctions have been successfully examined by several groups under biasing
conditions, but previous attempts in electron holographic study of biased MOS transistors
have failed (41). The main obstacle is making electrical contacts to the tiny terminals of the
transistor when formed into a TEM specimen.

The above brief explanation of each technique suggests that scanning spreading resistance
microscopy, atom probe tomography and electron holography all have the potential of being
applied as the next generation of semiconductor device characterisation technique. Of these
three techniques, electron holography potentially offers the most advantages. It can also be
combined rather easily with other TEM based techniques such as electron tomography. The
combination of electron holography and tomography gives us the ability to map the profile of
dopants in 3D, which is vital for characterisation of novel nano-scale 3D devices. Successful
tomography-holography of a silicon pn junction has been demonstrated in the literature (42).
Electron holography also has many other applications besides dopant profiling, which has
made it a very powerful characterisation technique (43; 44; 45; 46).

42
Chapter 3

3. Electron Holography

3.1. Introduction

The transmission electron microscope (TEM) can be regarded as the most powerful and
widely used instrument in the characterisation of nanostructured materials. A beam of high
energy electrons interacts with the material as it passes through an ultrathin sample. In this
interaction, information about the internal structure and chemistry of the specimen is
recorded in the amplitude and phase of the electron beam. To extract this information,
different imaging techniques have been developed over many years (47).

Because a TEM image is a spatial intensity distribution, a substantial amount of information

present in the phase of the beam is lost in conventional TEM images. The electron detectors
are only sensitive to the intensity of the input signal, not to the phase. Therefore, retrieving
the phase-shift of the electron beam which has traversed the specimen would not be possible
unless the phase information is translated into the amplitude. The amplitude needs to be
modulated in relation to the phase. This modulation can be carried out by electron
holography, which is an interferometric technique. The Greek word “holo” which means
“whole” is chosen as the entire information of the electron wave (i.e. amplitude and phase) is
recorded in this imaging technique.

The phase shift introduced in the electron wave is responsive to the electrostatic potential of
the sample. Therefore, electron holography, because it reveals this phase shift, is an
unrivalled tool for the measurement of these potentials. Among several modes of electron
holography which have been suggested in literature (46; 48; 49), off-axis electron holography

43
is ideally suited to the characterisation of electrostatic potentials in nanoscale semiconductor
devices.

One reason is that off-axis electron holography is relatively sensitive to a wide range of
spatial frequency variations in the electrostatic potential. Almost all the interesting spatial
frequencies that present in semiconductor samples can be detected relatively equally in the
off-axis electron holography mode. Also, the phase and amplitude information can be
recovered and separated from each other using simple computations. It should be added that
the phase image is almost directly interpretable in this type of holography. These reasons will
be clearer in this chapter where the principles of this technique are explained in a simple way.

44
3.2. Off-axis electron holography

The electron microscope’s detection system is sensitive to the electron intensity which only
represents the amplitude of the electron wave, not its phase. In a very simplified diagram in
Figure 3.1, it is illustrated how interferometry makes it possible to detect the phase of a wave.
Three identical sine waves y1, y2 and yr are shown in this figure. yr is kept as a reference and
the two others are shifted in phase by two different amounts. These phase shifts can be
introduced by passing the waves through different media and/or different path lengths. The
difference between the exit waves, y1 and y2 , is only in their phase. The detector cannot
differentiate between these signals because they have the same amplitude and therefore the
same intensity. However if these waves are interfered separately with the reference wave, yr,
the resulting waves

∆ 1 ∆ 1 Eq.(3.1)
2cos( )
yr y1 cos kx cos kx ∆ 1
2 cos(kx 2 )
Amplitude

∆ 2 ∆ 2 Eq.(3.2)
yr y2 cos kx cos kx ∆ 2
2cos( ) cos(kx )
2 2

∆ 1 ∆ 2
will have different amplitudes: 2cos( ) and 2cos( ). As a result of this interference, the
2 2

amplitude is modulated according to the phase. In other words, the phase information is
coded into the amplitude. The two signals are now distinguishable by the detector.

Figure 3.1. The modulation of the amplitude according to the phase by the
means of interference.

45
The above idea for recording the phase of the electron wave was first proposed by Denis
Gabor in 1948 (50). However, the implementation of it in electron microscopes was
hampered for many years because of two major reasons. The first reason was the lack of a
coherent electron source and the second one was the conjugate image.

A coherent illumination source is a basic requirement for any interference-based techniques.

Despite the fact that holography was originally suggested in electron optics, because of the
lack of an electron source with enough coherency and brightness, many of the basic
principles of today’s electron holography techni ues were first developed in light optics,
where a bright enough coherent source was available, and later adapted in electron
microscopy.

One of the very first of Gabor’s holograms, both recorded and reconstructed using visible
light beams, is shown in Figure 3.2. As shown in Figure 3.1, to form a hologram, the object
wave needs to be interfered with a reference wave. In Gabor’s holograms, the reference wave
is provided by the transmitted part of the beam which has not been scattered by the object.
The simplified ray diagrams in Figure 3.3a and 3.3b demonstrate how such a hologram is
formed and reconstructed respectively.

Figure 3.2. One of the very first holograms recorded and

reconstructed in light optics by Denis Gabor. (51)

46
In Figure 3.3a a point object illuminated by a plane wave is shown. The spherical wave
issuing from the object interferes with the transmitted plane wave acting as the reference
wave. The resulted interference pattern can be recorded at some distance from the object.

Figure 3.3. Ray diagrams showing a) the formation and b) the reconstruction of an inline
hologram from a point object. (52)

For the reconstruction, the recorded hologram is illuminated with a plane wave similar to the
reference wave, as shown in Figure 3.3b. Instead of one spherical wave representing the
object, illuminating the hologram with the plane wave produces two spherical waves. One of
them converges at the point O, as desired, and the other one diverges from Oʹ. As a result, the
reconstructed image is perturbed by a “twin” or “conjugate” image coming from the
divergent spherical wave. The problematic conjugate image appears because, in the formation
of the hologram, the reference and object waves both propagate in the same direction.
Following that, in the reconstruction the illuminated plane wave and two spherical waves
have to move in the same direction.

The conjugate-image problem can be overcome by making an angle between the reference
and object waves during the hologram acquisition, as illustrated in Figure 3.4. Therefore in
the reconstruction, the detector will not see the conjugate image. As shown in Figure 3.4b,
the two spherical waves propagate at different directions. The larger the angle between the
reference and object waves, the farther apart the object and conjugate images.

The holography mode shown in Figure 3.3, where the reference and object waves propagate
in the same direction, is called in-line holography. Correspondingly, the type of holography
where the object and reference waves transmit in different directions is called off-axis
holography, represented in Figure 3.4.

47
 Figure 3.4. Ray diagrams showing a) the formation and b) the reconstruction of an
off-axis hologram from a point object. (52)

The idea that the reference wave does not need to transmit in the same direction as the object
wave does and instead it can be passed through air, off-axis holography, was suggested by
Leith and Upatneiks (53). Figure 3.5 shows how this idea can be implemented in light optics.
A nearly perfectly coherent laser beam was split into two halves using a half-silvered mirror,
called a beam splitter. One of these halves was used as a reference wave and the other half as
an object wave. As can be seen, a simple system of mirrors can be used to recombine these
two waves to give an interference effect.

Figure 3.5. Implementation of off-axis

holography in light optics.

48
However, the implementation of this idea in electron microscopes had to await the invention
and development of the electrostatic biprism and the field emission gun (FEG). The
electrostatic biprism plays the role of the system of the mirrors and the beam splitter in the
above configuration.

49
3.3. Instrumental requirements for off-axis electron holography

3.3.1. The field emission gun (FEG)

Availability of a temporally and spatially coherent electron source is a prerequisite for the
implementation of off-axis holography in electron microscopes. FEGs have satisfied this
requirement to a reasonable extent in modern electron microscopes.

a) Temporal coherence
Instead of having ideal single wavelength λ electrons providing an infinitely long wave-
packet (see Figure 3.6a), due to the uncertainty in the wavelength ∆λ we deal with wave-
packets of limited length (see Figure 3.6b) in the direction of propagation. If two beams

Figure 3.6. Limited coherence length in a non-

monochromatic beam. a) Showing two monochromatic
waves, with wavelength of λ and λ ∆λ, which are in phase at
x = 0 and become completely out of phase at x = lt/2. b)
Showing a wave packet with the length of lt generated from
wavelengths between λ _ ∆λ and λ + ∆λ.

which have been split coherently from the electron beam travel in different paths with the
path length difference of more than , then the interference pattern cannot be produced by

50
superimposing these two beams since there would not be any correlation between them. From
2
the wavelengths producing the sine wave-packet in Figure 3.6b, waves of cos( x)
λ
2
and cos( λ x) are plotted in Figure 3.6a. These waves which are in phase at x 0 interfere
λ

destructively at x 2 since the phase difference ∆ between them reaches rad at this
point.

2 lt 2 λ2
Eq. (3.3)
λ 2 λ λ 2 λ
As can be deduced from Equation 3.3, the wave-packet length is limited by the wavelength
spread ∆λ stemming from the energy spread ∆E of the electron beam. The relationship
between the electron energy E and wavelength λ is given by

Eq.(3.4)
λ 1
E 2
2m0 E(1 )
2m0 c2
Planck’s constant
m0 electron mass
c : speed of light in vacuum

As an example, for an electron beam of 300ke the energy spread of ∆E 1e results in the
temporal coherence length of lt 1μm and the maximum possible number of fringes

of n 10 . Therefore, for electron holography experiments, which use at most,

λ

the temporal coherence of electron sources is more than adequate. (43; 52)

Note that instability in the microscope parameters such as high tension and lens currents
degrades the energy spread of the beam and so its temporal coherence.

b) Spatial coherence
The temporal coherence length is associated with the beam coherence in the propagation
direction. The spatial coherence length refers to the same concept but in the direction
perpendicular to the electron beam. As explained, due to the uncertainty in the wavelength ∆λ
the wave-packet length is limited. Likewise, because of uncertainty in the propagation
direction, α, of emitted electrons the wave-packet width in a plane perpendicular to the
electron beam is also limited.

51
Figure 3.7 represents a monochromatic electron beam propagating in the optical axis
direction with α rad uncertainty or with the beam angle of α. Three representative rays, R, R1
and R2 are shown. This simple diagram shows that from the source to the observation plane,
the distance travelled by the ray R1 is larger than that travelled by the ray R by the path

difference of OP1-OP α. At (half of the wave-packet width or the spatial coherence

2 2

length) away from the optical axis, R and R1 interfere completely destructively as the path

difference reaches .

λ λ Eq.(3.5)
α
2 2 α
This equation shows that the spatial coherence length of a beam with the wavelength of λ
can be increased by decreasing α.

Figure 3.7. Monochromatic electron beam

propagating in the optical axis direction with
α rad uncertainty.

In practice, by using apertures in the illumination system of the microscopes the electrons
emitted at large angles are cut (46). By spreading the beam and selecting its centre part, the
spatial coherence length can be increased. However, the limitation is the beam intensity.

A smaller aperture limits the beam angle and increases the spatial coherence length of the
beam, but at the cost of lower beam intensity. Therefore, in practice smaller apertures cannot
be used to realise larger coherence lengths when the original beam angle is large and the

52
beam intensity is low. Brightness is the electron source property which relates the emission
current density to the beam angle.

Assume an electron source with the radius r and emission current ie which is emitted within
ie
an angle of α. The brightness β of the source is defined as the current density per unit
r2

solid angle α2 (46; 52)

ie Eq.(3.6)
β 2
rα
It can be deduced from Equation 3.5 and 3.6 that for obtaining an electron beam with a large
spatial coherence length and high current density, having a source with high brightness is
necessary.

Among the field emitters and thermionic sources such as hairpin tungsten filaments, LaB6
emitters and Schottky emitters, the cold field emission guns (Cold FEG) provide the largest
brightness because of two reasons: their size and their operating temperature.

Equation 3.6 shows that the smaller the size of the emitting area, the higher the brightness. In
comparison to the hairpin tungsten filaments which have the size of 100μm in radius, the cold
FEGs emit from a sharp tip with the size of less than 0.1μm in radius (52).

In thermionic sources, in addition to the accelerating voltage of the microscope with which
the electrons are accelerated in the direction of the optical axis, the temperature also gives
kinetic energy to the electrons and increases their velocity. But, this velocity has a component
perpendicular to the optical axis, which increases the beam angle α and therefore reduces the
brightness. As an example, a ZrO/W Schottky emitter operating at the temperature of 1800K
and accelerating voltage of 100kV with a tip radius of 0.5-1μm has a brightness
of β 1010 A m2 sr (54), while a cold FEG operating at 300K and the same accelerating
voltage of 100kV with the tip radius of has a brightness of β 1013 A m2 sr (47).
Although the improvement in the brightness is mainly due to the source size, the low
operating temperature of cold FEGs also helps.

It should be added that regardless of the source type, we can increase the brightness by
increasing the accelerating voltage of the microscope. Intuitively the higher accelerating
voltage means that the emitted electrons see a larger electric field in the direction of the
optical axis and deviate less. Therefore, the beam angle α becomes smaller. However,

53
although the brightness increases with the accelerating voltage, the wavelength decreases.
Consequently, lowering the accelerating voltage cannot considerably increase the spatial
coherence length of the beam.

Electron holography requires a beam with a large spatial coherent length and high current
density, therefore sources with high brightness such as cold FEGs or Schottky emitters are a
prerequisite for holography.

3.3.2. Biprism
In order to implement the off-axis holography technique in an electron microscope, in
addition to an FEG source a biprism is required to mimic the role of mirrors and the beam
splitter in Figure 3.5.

The commonly used electrostatic biprism invented by Möllenstedt is a glass fibre coated with
gold or a tungsten wire of about 500nm in diameter, placed between two earthed plates (55).
A positive DC voltage of typically between 50 and 200V is applied between the wire and the
plates as shown in Figure 3.8. Different methods have been suggested in the literature for the
fabrication of the biprism (56). In most cases, it is mounted in the place of one of the
selected-area apertures (57), which has been modified for this purpose. The biprism can be
moved and usually rotated in a plane perpendicular to the optical axis.

Figure 3.8. Biprism configuration: the biprism

filament connected to the DC voltage is placed
almost at the middle of two earthed plates. (52)

As shown in Figure 3.9, when the biprism filament is connected to a positive voltage with
respect to the earthed electrode plates, an almost symmetrical electric field is produced close
to the filament (58). Under the influence of this electric field, the electron waves are deflected
as they pass the biprism. On the right and left side of the filament, the field directions are
opposite. On both sides, the electric field is outwards from the filament, therefore the electron
waves on the right side and those on the left side of the filament are deflected towards each
54
 Figure 3.9. The symmetrical electric field formed close to the
biprism filament, as connected to a positive voltage, deflects the
plane waves propagating on its right and left sides towards each
other to force them to interfere and form an off-axis hologram.

other. They interfere with each other and form an off-axis hologram because, similar to
Figure 3.4 and Figure 3.5, the waves interfering with each other propagate in different
directions.

3.3.3. Lorentz lens

Although equipping the microscope with an FEG source and a biprism would be enough to
carry out off-axis electron holography, the measurement of electrostatic potentials,
responsible for the functionality of semiconductor devices, requires an additional lens in the
microscope called the “Lorentz lens”. The Lorentz lens is a minilens located within the lower
polepiece of the objective lens, which makes it possible to take a hologram with a wider field
of view.

In Lorentz mode the objective lens is almost switched off and the Lorentz lens is excited. In
comparison to the objective lens, the Lorentz lens has a long focal length which reduces the
magnification and thus increases the holographic field of view. Another important advantage
of Lorentz mode is that the magnetic field at the specimen region is negligible. This feature is
necessary in the study of magnetic materials (59; 44). The large field of view and low
magnetic field at the object plane are obtained in Lorentz mode at the cost of spatial
resolution. Due to the high spherical aberration of Lorentz lenses (~8000mm), the spatial
resolution in this mode is about an order of magnitude worse than that in the normal mode.
Therefore, the Lorentz mode is only used in the medium resolution holography and for high

55
resolution holography the normal mode is used (60; 61). The differences between these two
modes are shown schematically in Figure 3.10 for a Lorentz lens integrated in a typical
Super-Twin objective lens. The configuration shown in this figure for Lorentz mode has been
used to take all of the holograms presented in this dissertation.

Figure 3.10. The differences between the holographic setup in high resolution mode and in Lorentz
mode. (62)

56
3.4. Off-axis electron hologram formation

In Figure 3.11, a simplified schematic diagram of the formation of an off-axis electron

hologram is shown. The specimen is placed in the objective plane in a way that allows nearly
half of the beam to pass through the vacuum and the other half to transmit through the
sample. The half passing through the vacuum without being perturbed by the object is the
reference wave and the other part carrying the sample information is the object wave. The
Lorentz lens focuses both reference and object waves into the back focal plane, then both
waves are deflected towards each other by the biprism and forced to interfere with each other
to give a hologram in the image plane. This is equivalent optically to saying that the object
and reference waves are emitted from two virtual sources on the back focal plane, S1 and S2
respectively.

Figure 3.11. Schematic diagram showing the formation of off-axis

electron holograms.

If the object and reference waves are represented respectively by and r

,

o
Ao ei (2 o. o( ))
Eq.(3.7)

r
ei (2 r. r( )) Eq.(3.8)

the intensity distribution on the image plane would be:

I ( o r
).( o r
)

57
 1 Ao 2 2 Ao cos(2 c
. ∆ ( )) Eq.(3.9)

where ∆ o
- r
and c o- r. The subscripts “o” and “r” refer to the object and reference
waves respectively. and A denote the phase and amplitude of the waves and k refers to the
wave vectors also shown in Figure 3.11. For simplicity, the amplitudes were normalised by
the amplitude of the reference wave.

The periodic distribution of intensity described by Equation 3.9 appears as bright and dark
fringes with which an electron hologram is made. These holographic fringes can be seen in
Figure 3.12 where an off-axis electron hologram, from an InAs nanowire under electrical
biasing conditions, is shown as an example. In addition to the holographic fringes which can
be seen more clearly in the enlarged image in Figure 3.12b, a set of coarse fringes is also
present in the hologram. These coarse fringes, more visible close to the hologram borders, are
Fresnel fringes stemming from the edges of the biprism. These Fresnel fringes do not contain
any information about the specimen.

58
3.5. Off axis electron hologram quality

The following three parameters can be used in assessing the quality of an off-axis electron
hologram.

3.5.1. Spatial frequency of fringes

According to Equation 3.9, the holographic fringes have the carrier spatial frequency of
1
c o- r. Since the wave vectors o and r have the same magnitude o r λ
and the

angle between them β is small, c

can be simplified as:

2 β β
c o - r sin Eq. (3.10)
λ 2 λ

Also from Figure 3.11, the relationship between β and the deflection angle γ, the angle by
which the reference and object waves are deflected under the influence of the biprism electric
field, can be estimated as:

1
β 2
S1S2
tan
2 a b
2aγ
β
a b Eq. (3.11)
1
2
S1S2
tan γ
a
where the distances a and b show the position of the biprism between the back focal plane
and image plane of the Lorentz lens. S1S2 is the distance between the two virtual sources.

By considering the biprism as a line charge placed between two grounded plates, as presented
in Figure 3.9, it can be shown that:

γ γ0 b Eq. (3.12)
Where Vb is the voltage applied to the biprism. The constant γ depends on the geometry of
the biprism and the accelerating voltage of the microscope (63).

Substituting Equation 3.12 and Equation 3.11 in Equation 3.10 results in:

2aγ0
c b Eq. (3.13)
λ (a b)
This equation is important because it shows that for a particular microscope set-up, the spatial
frequency of fringes can be controlled by the biprism voltage with ease. By increasing the

59
biprism voltage the spatial frequency of the fringes increases linearly, or in other words the
1 1
fringe spacing s decreases at the rate of .
c b

Figure 3. 12. a) An example of off axis electron hologram taken from InAs nanowire under
electrical biasing conditions with b) the enlargement showing the fringes inside the wire and
c) the intensity profile showing fringe visibility and spacing.

3.5.2. Hologram width

Another important factor is the hologram width W by which the recorded field of view is
defined. From the simple ray diagram in Figure 3.11, it can be seen that this width on the
image plane of the objective lens (Lorentz lens) will be 2bγ, if the biprism thickness is not
considered. The effect of having a biprism with nonzero thickness 2R can also be taken into
account by subtracting the biprism shadow 2R(a b) a from this value.

2R(a b) 2R(a b) Eq. (3.14)

W 2bγ 2b γ0 b
a a
For the biprism voltage, there is a threshold below which the reference and object waves do
not overlap. This threshold voltage depends on the biprism thickness. This expression also
shows that likewise the spatial frequency of fringes, the hologram width increases linearly by
the biprism voltage.

60
3.5.3. Holographic fringe visibility
In practice, the holographic fringe visibility μ is usually measured from a hologram recorded
in the absence of a specimen by using the expression (46)

Imax Imin Eq. (3.15)

μ
Imax Imin
where Imax and Imin are the maximum and minimum intensities of the holographic fringes
respectively as shown in Figure 3.12c. The fringe visibility μ is of great importance since it
can limit the phase resolution. A decrease in the fringe visibility causes an increase in the
standard deviation of the measured phase and consequently degrades the phase resolution.
The standard deviation of the measured phase is proportional to:

1
E μ2
Eq. (3.16)
e

where DQE is the detection quantum efficiency of the detector and Ne represents the electron
dose (64). To obtain a better phase resolution, the product μ2 e needs to be maximised by
optimising experimental parameters during the acquisition.

a) Effect of exposure time

A longer exposure time is an obvious choice for increasing the electron dose. For an off-axis
electron hologram recorded on a CCD camera, the typical number of electrons per pixel is
100-500 for an acquisition time of 2-8 seconds (45). Even exposure times of longer than
120sec (resulting in the electron dose of over 20000 electrons per pixel) are possible (37), if
the microscope, specimen and biprism are stable enough. Otherwise, increasing the exposure
time degrades the fringe visibility and damages the specimen in the cases of beam sensitive
samples.

Fringe visibility has a more significant influence on the phase resolution than the electron
dose (μ2 compared to e ), so longer exposure time should be avoided if it decreases the
visibility of the fringes.

b) Effect of biprism voltage

Generally, taking a hologram with a wider field of view and higher carrier frequency c
is
desirable. According to Equation 3.13 and Equation 3.14, both of these parameters can be
increased with the biprism voltage. However, in practice the biprism voltage is restrained by
two factors: finite spatial coherent length of the electron beam and limitations of the detection
systems in resolving fine fringe spacings. The biprism voltage can be used to expand the

61
hologram width up to the point where the lateral coherence of the electron beam allows the
reference and object waves to interfere. Also the application of larger voltages to the biprism
would be helpful in achieving finer fringe spacings only as long as the fringe spacings are
large enough to be detected. Because of these two factors, normally increasing the biprism
voltage decreases the visibility of holographic fringes. As described in (46), the fringe
visibility decays approximately exponentially with the hologram width squared W2, in other
2
words with the biprism voltage squared b (see Equation 3.14). Therefore, by increasing the
biprism voltage a larger field of view and higher carrier frequency are obtained, but at the
cost of fringe visibility.

c) Effect of CCD camera

Charge coupled device (CCD) cameras have introduced all the advantages of digital data
manipulation into the field of electron holography where originally photographic films were
used for recording holograms. Characteristics of the CCD camera with which the electron
hologram is recorded have a profound influence on the quality of the holographic fringes.

Linearity of the detector response to the incident electrons is one of the key characteristics
required for quantitative electron holography (46).The nonlinearity of CCD cameras is
normally better than 1%, more than adequate for electron holography (65).

CCD cameras typically can collect up to 3-5×105 electrons per pixel and the standard
deviation of the readout noise, by which the lower limit for the intensity is defined, is ~20e.
Therefore, the dynamic range of CCD cameras is in excess of 1:104, again more than enough
for electron holography. (65; 46)

CCD camera noise can considerably degrade the signal to noise ratio. The performance of the
CCD camera in that respect is expressed by its detection quantum efficiency (DQE) value
defined as:

S R2out
E Eq. (3.17)
S R2in
where S R2out and S R2in represent the squared signal to noise ratio before and after the
CCD camera, respectively. In an ideal detector the DQE is unity while its typical value for
CCD cameras is about 0.8, which means 20% reduction in the signal to noise ratio (65). As
can be seen in Equation 3.16, this parameter reduces the phase resolution in electron
holography.

62
The most important characteristic of the CCD camera, which should be considered carefully
in electron holography, is the modulation transfer function (MTF). The CCD camera acts as a
low-pass filter attenuating the high spatial frequencies. The MTF of a CCD camera describes
the frequency response of this filter. The cut-off frequency of this low-pass filter is mainly
determined by the scintillator.

The scintillator is part of the CCD camera at which the electrons are detected and converted
to photons. The photons generated at the scintillator are transferred to the CCD via fiber-optic
coupling. The size of the area where photons are generated on the scintillator (~ 0 to 100 μm)
is about 2 to times larger than the pixel size (~2 μm) on the CCD. Thus, even if the
electron beam is blocked by an object with a sharp edge generating a high spatial frequency
signal for the camera, the recorded image will not be as sharp as the object. (65)

Assume that the fringe visibility of a hologram before the CCD camera is 100%. The CCD
camera attenuates the fringe visibility and this attenuation depends on the spatial frequency of
the fringes, as illustrated in Figure 3.13. Therefore, when larger voltages are applied to the
biprism to increase the spatial resolution, the fringe visibility decreases not only because of
the finite spatial coherent length of the electron beam but also due to the modulation transfer
function of the CCD camera. Increasing the magnification can be a partial solution. At a
higher magnification, the CCD camera sees fringes with smaller spatial frequency which is
why in practice the fringe visibility improves with magnification. A common rule of thumb is
that the magnification and biprism voltage should be chosen in a way that at least each fringe
is sampled by four pixels (56; 66). However, this improvement in the fringe visibility by the
means of increasing the magnification is achieved at the cost of field of view.

63
 Figure 3.13. Schematic diagram showing that the fringe visibility of interference fringes
with higher spatial frequency is degraded further by the modulation transfer function of
CCD cameras. The interference patterns with 100% fringe visibility are assumed in a) and
c) as the input of a CCD camera. The spatial frequency of input fringes in (c) is higher than
in (a), consequently the fringe visibility of recorded hologram in (d) is lower than in (b).

d) Effect of illumination
Referring back to Figure 3.9, in off-axis electron holography, interference only occurs
between the electrons on the left and right side of the biprism filament. Consequently, it is not
necessary to have a highly spatially coherent beam in all directions. High special coherency is
only required in the direction perpendicular to the biprism filament. Thus, in practice, instead
of using a symmetrical round beam the condenser stigmators are deliberately misadjusted to
form an elliptical beam. In other words, the beam is spread in the direction perpendicular to
the biprism filament to improve the spatial coherence of the beam in this direction, and the
drop of current density due to this spread is compensated by narrowing the beam in the
filament direction where having a high spatial coherence is not necessary. This beam
configuration is shown schematically in Figure 3.14. In order to maximise the coherency, the
major axis of the elliptical beam has to be aligned exactly perpendicular to the biprism
filament. In practice, this requirement is satisfied by maximising the fringe visibility by the

64
use of the condenser stigmators. The minor axis of the beam is normally adjusted to be 2-5
times larger than the hologram width, while the to ratio may be 50 to 100. (45)

Figure 3.14. Schematic diagram showing the

elliptical illumination obtained by misadjusting
the condenser stigmators for improving the
spatial coherence length in the normal
direction to the biprism.

As shown throughout this division, section 3.5, several practical factors can affect the quality
of an off-axis electron hologram. These factors are summarised in Equation 3.18 with which
the phase detection limit in off-axis electron holography is expressed. This equation is the
latest version of Equation 3.16. (43)

μ Eq. (3.18)
μ μ μ μ

In this equation μsc , μinst , μMTF and μinel describe the reduction of fringe visibility due to

- the limited spatial coherency of the electron source

- the instability of the microscope, specimen and environment
- the modulation transfer function of the CCD camera
- the inelastic scattering, respectively.
β
is a selected signal to noise ratio. Icoh -ln( μsc ) is the total coherent current emitted
k2

kαr)2
from a Gaussian source with the degree of spatial coherence μsc (α) e-( (43), brightness
β and wavenumber k. Icoh can be derived from Equation 3.6, and r and α are the same

65
parameter as introduced for that equation. By making the beam elliptical the coherent current
density increases by the factor of ellipticity d12. The total coherent current is multiplied by the
acquisition time ta to obtain the available coherent electrons in the beam. The electron density
on the CCD camera is calculated by dividing Icoh d12 ta by the hologram area W2 where W is
the hologram width. As the hologram is recorded by the CCD camera, this electron density
decreases by the detection quantum efficiency of the detector DQE. In the reconstruction
process, which will be described later, also the electron density decreases because a large part
of the spectrum is masked and only one of the sidebands is selected. If in the reconstruction
process the spatial frequencies larger than are masked, the reconstructed image would
1
have a pixel size of . The average number of electrons per pixel in the reconstructed
2 rec

image, therefore, is given by:

Icoh d12 ta
E Eq. (3.19)
eW2 (2 rec
)2
All parameters in Equation 3.18 play their role before the reconstruction process and during
the acquisition, except . After recording a hologram, the phase detection limit can be
improved by reconstructing a smaller range of spatial frequencies. Phase resolution can be
improved in the reconstruction process at the expense of spatial resolution.

66
3.6. Experimental measurement of hologram parameters

The transmission electron microscope used for all the holographic studies presented in this
thesis is a probe Cs-corrected FEI Titan 80-300ST equipped with a rotatable biprism and
Lorentz lens. The microscope was aligned to run at operating voltages of either 120kV or
300kV. For recording the holograms, the Gatan model 894 2k UltraScan 1000 camera
mounted beneath the viewing screen is used. This camera has a phosphor scintillator and is
1:1 fibre-optically coupled. The pixel size is 1 μm and intensities are digitised using a 16 bits
analogue to digital convertor (ADC). (67)

The parameters introduced earlier for assessing the hologram quality were measured using
the above configuration at the operating voltage of 300kV. Figure 3.15a and Figure 3.15b
show the carrier spatial frequency of the fringes and hologram width W measured as a
function of biprism voltage, respectively. The fringe spacing is also shown in Figure 3.15a.
These show that both the spatial frequency and hologram width increase linearly with the
biprism voltage. The intercept of the line in Figure 3.15b with the horizontal axis shows the
biprism voltage of about 23V required to overcome the biprism shadow.

The effects of biprism voltage and magnification on the fringe visibility are shown in Figure
5.15c. For three different magnifications of x20k, x31.3k and x51.5k, the fringe visibility is
shown at different biprism voltages. As can be seen the fringe visibility decays relatively
exponentially by the biprism voltage for all three magnifications. Also due to the MTF of the
CCD camera the fringe visibility is greater for higher magnifications.

67
Figure 3.15. Measurements of a) interference fringe frequency in pixel-1 and fringe spacing in nm, b)
hologram width in nm and c) fringe visibility in percent at magnifications of x20k, x31.3k and x51.5k
as a function of biprism voltage.

68
3.7. Hologram reconstruction

Holography is a two-step process: recording a hologram and then extracting the amplitude
and phase images from the hologram using a “reconstruction” procedure. Historically,
holograms were reconstructed on an optic bench using optical interferometers such as Mach-
Zehnder one (46). Nowadays, thanks to the development of CCD cameras, fast computers
and digital signal processing techniques, numerical reconstruction is very common. Several
pieces of software are commercially available, in the form of plug-ins for Digital Micrograph
such as HoloWorks (68), Holography (69), HolograFree (70) and ASUHolography (71), for
reconstructing a hologram. The procedure used in these programs to recover the phase and
amplitude is summarised in the following section.

3.7.1. Basic reconstruction procedure

The first and key step in the reconstruction process is to perform a Fourier Transform (FT) on
the hologram in order to produce its frequency spectrum. By considering the fringe visibility
in Equation 3.9, the hologram intensity can be written as:

Ihol ( ) 1 Ao 2 2 μ Ao cos(2 c
. ∆ ) Eq. (3.20)
and its frequency spectrum as:

FT Ihol Eq. (3.21)

FT Ao 2 Central band

μ FT Ao ei ∆ ( - c) Sideband c

Sideband -
μ FT Ao e-i ∆ ( c
) c

The hologram spectrum, such as the one shown in Figure 3.16b, contains three bands. The
central band at 0 comprises the Fourier transform of the uniform intensity corresponding
to the reference wave and the Fourier transform of the intensity distribution of the
conventional bright field TEM image. There is no information about the phase in this part of
the spectrum, so it is not of interest in holography. The two conjugate sidebands at
and - c
are equivalent and they both store the desired phase and amplitude information
relating to the object. The next step is to place a mask around the one of these sidebands to
select it, as shown in Figure 3.16c. Then, applying an inverse Fourier transform on the
selected sideband results in a complex image. The amplitude and phase images, shown in
Figure 3.16e and Figure 3.16f respectively, are calculated from this complex image using the
following simple equations:

69
 Amplitude image Eq. (3.22)
A R2 ( ) I2 ( )

I( )
tan 1 Phase image Eq. (3.23)
R( )
where and are the imaginary and real part of the complex image, respectively.

Figure 3.16. Showing the basic steps of reconstruction process for off-axis electron holography. a)
Representative off-axis electron hologram of an n-type MOSFET. b) The Fourier transform of the
electron hologram showing the central band and sidebands. c) Selecting one of the sidebands by
masking the rest of the spectrum. d) Inverse Fourier transform of the selected sideband in the form of
a complex image (its modulus is shown). e) The amplitude image and f) The phase image.

The above procedure is simple enough to be carried out without any additional plug-in or
code in Digital Micrograph or in any other image processing software such as Matlab, as has
been performed for the hologram in Figure 3.16. However, avoiding the artefacts which some
of them are very clear in this figure requires further sophisticated considerations and more
image processing steps, as explained in the following.

70
3.7.2. Initial filtration
An image can be considered as an infinite data points in two dimensions multiplied by a
rectangular window which truncates the image suddenly at its edges. The spectral leakage of
the rectangular window in the Fourier space is very wide (72) and causes streaks such as
those seen in Figure 3.16b. The appearance of the streaks in the Fourier transform can be
eliminated by applying a soft-edge window such as the Hann window before performing the
Fourier transform. It is normal to filter the hologram by a Hann window of usually order 3
before calculating its Fourier transform. For an N×N pixels image, the Hann window of order
is defined by the expression:

H m,n Hx m .Hy (n) Eq. (3.24)

Where
( 1)j 1 2 jm
Hx m . .sin j .cos 1
2 j Eq. (3.25)
j 1

and

( 1)j 1 2 jn
Hy n . .sin j .cos 1 Eq. (3.26)
2 j
j 1

The Fourier transform of a hologram calculated before and after applying a Hann window of
order 3 is shown in Figure 3.17, as an example. The elimination of the streaks allows the
automated reconstruction procedure to be more accurate when selecting the most intense
pixel as the centre of the sideband.

71
 Figure 3.17. The application of the Hann window. a) A Hann window of
order 3. b) An experimentally acquired hologram and c) its Fourier
transform. d) The product of the hologram shown in (b) by the Hann window
shown in (a). e) The Fourier transform of filtered hologram shown in (d).

3.7.3. Selecting the sideband

The size of the circular mask used to select the sideband, shown in Figure 3.16c, defines the
maximum spatial resolution in the reconstructed image. Assume that the highest frequency
required to be resolved in the reconstruction process is qmax. To keep this frequency in the
selected sideband, the size of the mask has to be at least equal to qmax.

Because of the second order term Ao 2 in Equation 3.20, the information related to the
frequency of qmax also lies at 2qmax away from the centre of the central band and this makes
the size of the central band twice larger than the sidebands (73). Therefore the sideband
centre and the central band centre have to be at least 3qmax apart from each other, as shown in
Figure 3.18. In other words, the biprism voltage has to be chosen in a way that the carrier
spatial frequency of the fringes qc becomes 3 times larger than the desired maximum spatial
frequency ( c
3 max
). However, in practice because of dealing with weak phase objects and
rather large noise levels normally the contributions from the central band at 2qmax become

72
buried in noise and the effective size of the central band becomes almost the same as the
sideband. Thus, in most cases, it is sufficient to increase the biprism voltage so that qc
becomes slightly larger than 2qmax (74).

Figure 3.18. Schematic diagram showing the

minimum required distance between the
central band and the sidebands in order to be
able to resolve features with the spatial
frequencies of less than qmax in the
reconstruction.

It should be noted here that the electrostatic potential normally varies so slowly in the
specimen that often the spatial resolution is not a limitation. The spatial resolution is
determined either by the pixel dimensions in the reconstructed image or by the size of the
selected sideband, whichever is larger. Since the phase resolution is more critical, choosing a
smaller mask size in selecting the sideband is somewhat preferential since it removes the high
frequency noise. However, care must be taken not to cut too many frequencies as it can
introduce artefacts in the images (73).

It is important that the centre of the selected sideband is placed exactly at the centre of the
new image. Otherwise, an artificial phase ramp is introduced in the reconstructed phase
image. The size of this new image defines the dimensions of the reconstructed phase and
amplitude images. If the size of this image is chosen to be considerably larger than the mask
size, the reconstructed phase and amplitude images would be very noisy. The same position is
chosen for the sidebands of the sample and reference holograms to remove spurious phase
ramps due to lack of sub-pixel choice of origin of sideband centre.

73
For the same reason of minimising the spectral leakage, explained above for the application
of the Hann window, instead of using a sharp mask in selecting the sideband it is preferred to

Figure 3.19. A centred sideband a) before and b)

after the filtration by c) an 8th order Butterworth
filter with the cut off frequency of 1/10nm.

apply a diffused one. A Butterworth filter is commonly used for this purpose (74). The
application of a Butterworth filter of order 8 with a cut off frequency of 1/10nm is shown in
Figure 3.19, as an example. The streak in this figure can also be seen in Figure 3.16b between
the central peak and sidebands, and corresponds to the Fresnel fringes. If they cause problems
in the interpretation of reconstructed images, these kinds of streaks also can be filtered out to
some extent at this step.

3.7.4. Reference hologram

Figure 3.20 shows a reconstructed phase image corresponding to a hologram taken in the
absence of any specimen. In this situation it is expected that the reference and object waves
have the same amplitude and phase, and consequently the phase difference between them is
zero. However, it can be seen in this figure that in practice the phase image is not uniform
and phase variations in the image are obvious.

74
 Figure 3.20. a) An empty hologram and b) its reconstructed
phase image showing the phase variation across the field of
view due to the geometric distortion.

As explained previously, in off axis electron holography the phase information is extracted
from the lateral displacement of the interference fringes. Along the microscope column, in
addition to the specimen which causes a phase shift in the object wave and accordingly
displaces the fringes laterally, there can be several other undesirable factors effectively
introducing similar displacements in the recorded holographic fringes. Examples of these
undesired factors are distortions stemming from the projection lenses and from the CCD
camera, inhomogeneities in the charge distribution along the biprism filament, thickness
nonuniformities in the biprism filament and spurious electric fields from the charging of the
microscope components such as fixed apertures under the electron illumination (45).

The spurious phase changes due to these factors can be recorded in the reference hologram,
and later in the reconstruction process can be subtracted from the phase image. The reference
hologram is acquired after each hologram by removing the specimen carefully from the field
of view without changing any parameters of the microscope. In the reconstruction process,
the recorded phase distortions are subtracted, by dividing the complex image obtained from
the sample hologram by the one obtained from the reference hologram.

In the automated reconstructed procedures, the true centre of the sideband can be determined
within ±1 pixel accuracy if the reference hologram is used. In the Fourier space of a reference
hologram, the pixel with the highest intensity in the sideband corresponds to the centre of the
sideband, while that is not necessarily true for an object hologram. This is another advantage
of acquiring a reference hologram. The position of the sideband centre which in principle
should be the same for both the object and reference holograms is determined from the latter

75
one. This advantage is more appreciable in the holography of strong electric or magnetic
fields where both the reference wave and the object wave are modulated strongly by the field.

The difficulty in finding the sideband centre without the reference hologram is illustrated in
Figure 3.21. We applied -80V to a sharp tungsten needle which was positioned at the distance
of about 150nm away from a grounded gold electrode. The produced electric field was strong

Figure 3.21. a) A hologram taken from the strong electric field induced between a sharp tungsten
needle biased at -80V with respect to a gold electrode positioned ~150nm farther. b) The hologram in
the Fourier space. The selected sideband centre c) without and e) with using a reference hologram and
d) and f) their corresponding phase images respectively.

enough to result in a field emission current of ~0. μA from the tungsten tip. As can be seen,
in the Fourier transform of the hologram shown in Figure 3.21b, the sideband is not
symmetric. Therefore, the correct centre of the sideband cannot be selected by searching for
the pixel with the maximum intensity in the sideband area. The result of centring the sideband
with such a search and its corresponding phase image are shown in Figure 3.21c and Figure
3.21d respectively. The properly centred sideband using the reference hologram, however, is
shown in Figure 3.21e together with its corresponding phase image in Figure 3.21f. The
extension of phase contours around the tungsten tip is more symmetric and follows the
expected electric field more closely in this phase image (Figure 3.21f) in comparison to the
one reconstructed without a reference hologram (Figure3.21d). However, due to the influence

76
of the electric field on the reference wave, proper simulation is still necessary for quantitative
interpretation.

3.7.5. Flattening
Although the employment of the Hann window and reference hologram in the reconstruction
procedure improves the accuracy of finding the sideband centre, this accuracy cannot be
better than ±1 pixel because of the discrete nature of images. Assume that the sideband centre
determined in the reconstruction process is off by a vector q from its exact position. This
error occurring in the Fourier space is equivalent of multiplying the complex image
with e2 i .
, which means adding the phase of 2 q.r to the phase image. This artificial phase
appears in the phase image in the form of a plane phase ramp. If the hologram has an
unperturbed vacuum region, the coordinates of this plane can be found by fitting a plane to
this area. Following that the plane phase ramp can be removed from the entire phase image.
This process is called flattening. (46)

3.7.6. Phase unwrapping

As explained earlier in the reconstruction process, the phase image is obtained by applying
the arctangent function, in the form of Equation 3.23, to the reconstructed complex image.
Due to the fact that output values of the arctangent are limited between (- , ), phase changes
larger than 2 results in discontinuities in phase images. In such phase images, it is not
possible to distinguish between the phases of and 2 . Phase unwrapping is the
procedure by which these phase discontinuities are resolved and the phase image is converted
into a continuous image which is more convenient in many cases to interpret.

In some cases, although the actual phase variation is smaller than 2 , one phase jump appears
in the phase image. This phase discontinuity occurs because the phase range compromises
one of the (2k 1) points. This phase jump can be removed easily from the region of interest
by adding an appropriate phase constant to the complex image before applying Equation
3.21. Removal of such a phase jump from a region close to a pn junction is demonstrated in
Figure 3.22. The phase jump which has occurred almost at the pn junction in Figure 3.22a,
marked by an arrow in the image, was removed by adding about 1.3 rad to the complex
image.

Nevertheless, the phase usually varies more than 2 across the region of interest and therefore
more sophisticated algorithm is required for the phase unwrapping. Several algorithms have

77
been proposed for the implementation of phase unwrapping (75). One of the simplest ones
involves searching the phase image to find the adjacent points with a phase difference greater
than a specific value close to 2 , for instance 4 or 5. Then 2 is added to or subtracted from
subsequent points. The difference between the unwrapping algorithms is often related to
which method is used in finding the phase jumps. When the image is noisy, it is challenging
for automated unwrapping algorithms to find the true phase jumps.

Figure 3.22. a) Demonstration of a phase wrap (pointed by an

arrow) close to a pn junction and b) its removal after adding a
phase constant of about 1.3 rad to the complex image.

For qualitative interpretation, it is not always necessary to unwrap phase images since
sometimes in a wrapped image, such as Figure 3.21e, the trend of 2 jumps expresses more
clearly how the phase changes. Other forms of displaying the phase images such as an
equiphase contour map or a cos n which amplifies the phase by a factor of n and displays it
in a form of contour map are also common (73). Examples of these forms of phase images
can be seen throughout this thesis.

78
3.8. Thickness map

The amplitude image is roughly the square root of an energy-filtered version of the bright
field TEM image intensity. As explained earlier in this chapter, temporal coherence is a
prerequisite for the formation of interference patterns. Now, assume that the electron beam
has an energy spread of ∆E 1e . In off-axis holography, the part of the beam passing
through the vacuum as a reference wave does not lose energy while the part passing through
the specimen, the object wave, loses some energy due to inelastic scattering. Only those
electrons that have lost less than 1eV will interfere strongly with the reference wave (73).
Also in the reconstruction process we only use the sideband. Most of the inelastically
scattered electrons concentrate close to the central band, not sidebands (76). Thereby the
reconstructed amplitude image can be considered as a pure zero loss filtered image

(<10-1 e ) (43). The intensity in the vacuum region of the amplitude image can be treated as
the total intensity under the electron-energy loss spectrum (EELS), It . Likewise, the intensity
in the specimen region can be related to the area under the zero loss peak of the EELS, I0 .
This analogy is illustrated in Figure 3.23.

Figure 3.23. An analogy between the reconstructed amplitude image and the
EELS spectrum.

I0
From electron energy loss spectroscopy, it is well known that the ratio can be related to the
It

specimen thickness t by (77):

I0
t λi ln( ) Eq. (3.27)
It

79
where λi is the mean free path for inelastic scattering. This equation is valid for the
holographic amplitude image as well (78). When the reference hologram is acquired in
t
addition to the object hologram, the λ map can be calculated by the following expression:
i

t Ao
2 ln( ) Eq. (3.28)
λi Ar
where Ao and Ar are the amplitude of the object hologram and the reference hologram,
respectively (78).

80
3.9. The relationship between the phase and electromagnetic potentials

Electron holography is a powerful characterisation technique because it reveals the phase

shift of the electron wave that has passed through a sample, in addition to its amplitude. The
shift in the electron beam phase stems from the electric and magnetic fields of the sample. To
understand how electromagnetic fields introduce phase shifts to the high energy electron
beam, the Schrödinger equation corresponding to such a system has to be solved.

For an electron beam, with energy E, passing through a static electromagnetic field, with the
electrostatic potential V and magnetic vector potential A , the Schrödinger equation would be:

1 2
. i eA e E Eq. (3.29)
2m
where m, e, and are the electron mass, the electron charge, the Planck constant and the
wave function, respectively. In solving this equation it is assumed that inelastic scattering
events are negligible. Also, dynamical diffraction effects are not considered.

2 iz
If the incoming electron beam is given in the form of 0
(z) e λ , which is the plane wave
solution of the unperturbed Schrödinger equation

2
2m
2
E 0 Eq. (3.30)

propagating along the optic axis z, the solution of the Equation 3.29 can be considered in the
form of (WKB approximation (79))

(x,y,z) o
(z) (x,y) Eq. (3.31)
where
x,y ei (x,y) Eq. (3.32)

Now, if Equation 3.31 is substituted in Equation 3.29 and the terms 2

, A2 and A. are
neglected (the phase object approximation) the phase of the electron can be derived as (58)

2 e
x,y x,y,z dz Az x,y,z dz Eq. (3.33)
λE
where the integration is performed along a straight path l parallel to the optic axis z. x,y,z
and Az x,y,z are the electrostatic potential and z component of the magnetic vector potential
respectively. To include relativistic effects, Equation 3.33 needs to be modified in the form of

81
 2 e
x,y CE x,y,z dz Az x,y,z dz Eq. (3.34)

where

2 2m0 c e a Eq. (3.35)

CE 2 e )
λ a (m0 c a
m0 electron mass
c : speed of light in vacuum
Va: accelerating voltage of the microscope

CE is a constant depending only on the accelerating voltage of the microscope. The values of
CE calculated for three common accelerating voltages are given in Table 3.1.

Accelerating voltage 120 200 300

Constant 0.00856 0.00729 0.00653

Table 3.1. Value of CE calculated for three different accelerating voltages

To explore the pure electrostatic case, as is the subject of this study, the second term in
Equation 3.34 can be ignored. Also, if the electrostatic potential is invariant along the
specimen thickness in z direction, the integration in Equation 3.34 simplifies to

x,y CE x,y t(x,y) Eq. (3.36)

where t(x,y) represents the specimen thickness profile. If the thickness is known, the
electrostatic potential map can be calculated from the phase image.

82
3.10. Mean inner potential

Normally the main proportion of the phase shift in electron holography is coming from the
contribution of the mean inner potential. A volume average of the atomic electrostatic
potentials within the material is called the mean inner potential (MIP) (46). Stemming from
this definition the mean inner potential Vmip can be expressed by:

1
mip r dr Eq. (3.37)

where is the volume of the unit cell for a crystalline solid or the volume of material for a
disordered solid, and r is the crystal potential. Nonetheless, r is not known and an
approximation has to be made to calculate the Vmip.

The simplest approximation used to estimate Vmip is to ignore the bonding effects between
the atoms and assume that the solid is made of an array of neutral atoms:

2
mip fj (0) Eq. (3.38)
2 m0 e
j

where fj (0) are the atomic scattering amplitudes for forward scattering of electrons ( 0 ).
The summation is performed over the j atoms present in the unit cell. (46)

There are two problems with this approximation. Firstly, in the calculation of the atomic
scattering amplitudes fj ( ), the energy of the electron beam affects the value of fj ( ) while

mip is a material characteristic not an electron beam property. Secondly, since redistribution
of valence electrons due to the binding is ignored in this approximation, Vmip calculated by
Equation 3.38 is normally higher than its real value. For silicon this upper limit of the MIP
would be 13.84V if the scattering amplitude of fj 0 0. 7 3 nm calculated by Rez et al.
(80) is used without any relativistic corrections for the mass of the high energy electrons.

On the other hand, a possible lower limit of mip can be calculated by assuming that crystal
is composed of an array of ions at the lattice points with uniformly distributed perfectly free
electrons in the valence band. The following equation was proposed by Bethe to calculate the
lower limit

3 ep
mip Eq. (3.39)
10 0 r0

83
where p denotes the number of free electrons and r0 represents the radius of a sphere with the
same volume as the crystal atom. This lower limit for Si is 11.47V as calculated by Radi (81).

Although the beam phase shift is mainly due to the MIP, it will be clear in the following
chapters that for the measurement of built-in potentials across homojunctions, the absolute
value of the MIP is not important, so long as the specimen thickness is known.

In addition to the MIP and the built-in potential, other parameters such as charging and
dynamical diffraction also shift the phase of the electron beam. Discussion on these
parameters which are less directly related to the technique of off-axis electron holography
will be carried out throughout next chapters.

84
Chapter 4

4. Specimen preparation

4.1. Introduction

The quality of the information deduced from TEM observations is limited by both the
microscope and the specimen. Many constraints previously limiting TEM have been removed
in contemporary aberration-corrected microscopes. These state-of-the-art microscopes,
equipped with highly coherent and bright FEG sources, are normally stable enough that a
hologram acquisition time of over one minute is possible without a significant decrease in the
fringe visibility. However, these advances in the microscope side require stringent
requirements from the TEM specimen.

For all TEM techniques, the specimen must be electron-transparent. Off-axis electron
holography imposes additional restrictions on the specimen and specimen imperfection is
even less tolerable when the electrostatic potential map of a nanoscale device is demanded.

4.2. Specimen requirements

As described earlier in chapter 3, an electron hologram is formed by the interference of two

waves: the reference wave and objective wave. The reference wave is provided by passing a
part of the electron beam through vacuum, a region which has not been perturbed by the
electromagnetic field of the specimen. The vacuum must be close enough to the region of
interest that the reference and object waves overlap as they are deflected towards each other
by the biprism electric field. Therefore, either the region of interest has to be adjacent to the
specimen edge or a hole has to be prepared in the immediate vicinity of that region. The
maximum distance between the region of interest and the specimen-vacuum interface is

85
limited by the hologram field of view. In general, this distance should not be larger than
500nm when medium resolution holograms are taken.

Another consideration in preparing a TEM specimen for potential profiling is the specimen
thickness. In most TEM techniques such as HRTEM and EELS (47), samples with a
thickness of less than 100nm are preferable, usually with no minimum limitation. In contrast,
in electron holography, the phase shift is proportional to the specimen thickness; the thicker
the specimen, the larger the phase shift. Nevertheless, inelastic scattering also increases with
the thickness and degrades the coherency of the transmitted beam. The trade-off between
enhancing the phase signal and fading interference fringes by inelastic scattering, both
resulting from the thickness increment, suggests that the specimen thickness needs to be
optimised for the maximum signal-to-noise-ratio (SNR). Thicknesses between 200nm and
500nm are used in previous works for electron holography of Si samples.

Also thickness non-uniformity significantly affects the potential sensitivity of electron

holography. Thickness variation results in a phase shift which could cause ambiguity in the
interpretation of the phase images because it is then not possible to distinguish between phase
shift stemming from electrostatic potential variation and thickness variation. Therefore,
unknown changes in the thickness of the sample appear as noise in the measured electrostatic
potential. Suppose, for example, that sensitivity of 0.1V in the potential profile is required for
a Si pn junction sample with a thickness of 300nm. The mean inner potential of Si is roughly
12V and the built-in potential of pn junction is about one order of magnitude less than that;
therefore, 0.1V sensitivity in the measured built-in potential means better than 1% precision
in the specimen thickness. For a 300nm thick specimen, the thickness must be known to
within 3nm along the p-type and n-type regions.

Targeting a specific nano-scale device is another challenge in preparing TEM specimens

from semiconductor devices. The preparation technique, in many cases, needs to be site-
specific to less than one micron.

86
4.3. TEM specimen preparation for electron holography

Various approaches, each with its own pros and cons, have been taken in the literature in
preparing TEM specimens from semiconductor devices for holographic analysis. Among
these approaches, all conventional techniques based on mechanical polishing and broad ion
milling struggle to fulfil the specimen requirements for electron holography. A very fragile
wedge-shaped specimen which has most likely missed the region of interest with significant
amount of debris left and rather rough surfaces often results from techniques relying entirely
on the mechanical thinning. However, the strong desire for surface free from ion damage has
been the motivation for using mechanical wedge polishing in (82) for preparing a TEM
specimen of an MOS transistor. Although having no ion damage is a tempting advantage, the
numerous disadvantages of mechanical wedge polishing have convinced many microscopists
who prefer the conventional specimen preparation to introduce the broad Ar ion milling at the
final stage of their sample preparation. Mechanical thinning followed by the broad Ar ion
milling has been used by few groups (38; 83; 84; 85) to successfully detect the electrostatic
potential of device specimens. The thickness non-uniformity, due to preferential milling
between different materials, limits the accuracy of the electrostatic potential measurement in
these specimens (86).

Moreover, compared to the size of the semiconductor devices, the mechanical thinning
followed by the broad Ar ion milling is a blind technique. This technique can never be
considered as a site-specific specimen preparation method for devices which are scaling
down every day. There should be a large number of devices in the specimen without
preference for any of them, so that by chance few devices with a suitable thickness locate
close to the edge of the hole provided by ion milling.

The difficulty of providing a path for the reference wave in the conventional Ar ion milling
methods can be seen in (87), where TEM specimens from different transistors have been
prepared with dimpling and Ar ion milling. To overcome the blindness of the technique, they
employed the milling and imaging ability of a focused ion beam (FIB) in making holes close
to the transistors.

The milling and depositing capabilities of FIB workstations besides their imaging ability have
made the FIB tools very appealing in preparing the TEM specimens. By incorporating an
SEM in modern dual-beam systems, the imaging ability of FIB is even boosted. Particularly

87
in preparing specimens from semiconductor devices for the purpose of electrostatic potential
mapping using electron holography, FIB-based specimen preparation techniques have many
advantages over the conventional methods.

An individual device located at a specific position in a bulk wafer can be targeted in the
preparation of TEM specimen if dual-beam systems are employed. Even if the position of the
device is given with about ±10μm uncertainty in each direction, X, Y and Z, it is possible to
locate and make a TEM specimen from the device.

Moreover, in comparison with conventional methods, the reference wave path can be
provided in the specimen with ease in FIB-based techniques. Also, the desired specimen
thickness for electron holography can be obtained without difficulty as thickness
measurement with reasonable accuracy is possible during preparation.

In addition to the above advantages, the FIB can be used for making electrical contacts.
Therefore, in this project, FIB has been chosen as the primary specimen preparation
technique and the results presented in this dissertation are from FIB-prepared specimens.

4.3.1. Focused ion beam principles

By applying a large electric field, positively charged Ga ions are extracted from the tungsten
tip wetted by the heated Ga. After passing through different electrostatic lenses, apertures and
deflectors, these ions form a focused ion beam (FIB), which can be scanned across a sample.
In terms of TEM sample preparation concerns, the consequences of interactions of these ions
with the sample can be summarised into four categories.

a) Sputtering away the substrate atoms:

The milling ability of FIB workstations comes from heavy and energetic beam ions
sputtering the atoms away from the surface of the sample. This is widely used in
making, thinning and polishing TEM samples. In FIB workstations, the milling rate
can be controlled by the beam current. High currents (a few nano-amps) are used in
the first digging steps, medium currents (a few hundred pico-amps) in the thinning
steps and low currents (a few tens of pico-amps) in the final polishing steps of TEM
sample preparation (88).
b) Generating secondary charged particles:
As a result of specimen-ion interactions, secondary ions and electrons are emitted
from the sample. These charged particles can be collected by a detector for the

88
 imaging purpose. To avoid damaging the sample in the imaging mode and to obtain a
better resolution, low beam currents are used normally for imaging. Also the
secondary ions can be used to characterise the sample.
c) Chemical reaction:
Interaction of high energy ions with loosely bonded molecules of some gases can
result in breaking the bonds and releasing the atoms. This property is used in FIB
workstations for maskless depositing of materials. An appropriate precursor gas is
injected over the surface of the sample using a fine nozzle while the ion beam is
scanned across the desired area. Molecular cracking occurs as a result of interaction
between high energy ions and the gas molecules. The desired atoms are released,
depending on the precursor gas. Then, as shown in Figure 4.1 these atoms are
deposited on the surface of the sample and any volatile by-products are pumped out.
Deposition of a wide range of materials using the FIB has been reported in the
literature. However, for TEM sample preparation, deposition of tungsten or platinum
is usually used.

Figure 4.1. Principle of FIB

induced deposition (89)

d) Ion implantation and damaging:

The other consequences of ion-sample interaction are ion implantation and induction
of damage into the sample. This damage particularly appears as an amorphous layer in
the case of crystalline samples. This change in the material structure shows that TEM
samples prepared by FIB suffer significantly from ion damage.

All three abilities of FIB (milling, imaging and depositing) are used to prepare the TEM
specimens while ion implantation and ion damage are the main drawbacks of methods based
on FIB.

89
4.3.2. Artefacts in FIB-prepared TEM specimens
As mentioned previously, despite all the advantages of preparing TEM samples using FIB,
FIB-prepared specimens suffer from damages and artefacts induced by high energy ions.

It is known that the distribution of the current density in the ion beam follows the Gaussian
distribution close to the centre of the beam and decays exponentially away from the centre
(90). Consequently, instead of having parallel surfaces, FIB prepared samples have a slight
wedge shape. This wedge effect can be problematic in phase image interpretation. The wedge
angle or the angle between the top and bottom surfaces of the sample increases with the ion
beam current. Tilting the specimen relative to the beam direction and proportional to the
beam current together with using well aligned low beam currents at the final thinning stages
is an effective way of minimising the wedge angle (91).

Because FIB milling is based on removing atoms, it can be applied to a wide range of
materials. However, the sputtering rate is not constant for all materials and more tightly
bound atoms are likely to be removed at a lower rate. Differences in the sputtering rate of
different materials result in uneven surfaces on samples with complex topography. For
instance, in Figure 4.2, a TEM micrograph of a FIB prepared specimen from an integrated
circuit with different metallisation layers is shown. Thicker and thinner regions in this sample
can be distinguished by the dark and bright intensities respectively. Layers located beneath

Figure 4.2. The curtaining effect in a FIB prepared

specimen. Red arrows point to the materials with a
lower milling rate, which protect the regions below
from ion beam milling. Darker regions are thicker (92).

the low sputtering rate materials, marked by the arrows in Figure 4.2, have been protected
from the Ga ions and have remained thicker than other areas. This variation in the thickness,
90
which looks like a curtain in TEM images, is problematic for quantitative electron
holography, and should be avoided.

Milling the samples from a direction such that Ga ions do not see the changes of the sample
topography is an effective solution. For example, in preparing a FIB-sample from an
integrated circuit if the specimen is milled from the substrate side, Ga ions would not see the
complex metallisation structure located above the region of interest. Although milling the
specimen from the substrate side seems very appropriate, this requires a geometry which
gives the ion beam access to the backside of the sample. Different geometries have been
suggested in the literature for this purpose (93; 94; 95; 96; 97).

Generally in FIB-based specimen preparation, Pt or W bar is deposited on the interesting area

in order to protect the region beneath it from the ion damage during the next steps. However,
this step itself damages the surface of the sample. In the case of a crystalline sample, an
amorphous layer is formed beneath the Pt or W bar as a result of the ion damage during
deposition. This near-surface layer can be tens of nanometres thick (98). Where the region of
interest is very close to the surface of the sample, coating the sample with carbon or a thin
layer of metal such as gold prior to FIB deposition would protect the surface of the sample
completely. In dual-beam systems, this protective metal or carbon layer can be deposited
using the electron beam induced deposition (EBID) technique (99; 100; 101; 102).

The most severe problem with the FIB-specimens is the damage left by the energetic Ga ions
on the top and bottom surfaces of the TEM specimens after preparation. This damaged layer
is generally referred as an amorphous layer and/or electrically dead layer in the literature. The
amorphous layer is believed to result from FIB ion damage. Amorphisation resulting from ion
implantation is a well-known phenomenon in the semiconductor industry (9). In the ion
implantation process, a minimum dose is required to form an amorphous layer. Conversely,
an amorphous layer is formed in the FIB milling process regardless of what current is used.
The ion beam interacts with atoms many times before sputtering occurs (103). Thus, even
during a low current milling the delivered ion dose is more than the minimum dose required
for amorphisation. To put it simply, even in crystalline materials, sputtering occurs after
amorphisation.

Redeposited materials are also usually amorphous. Although most of the milled atoms are
pumped out, there is still a significant chance that the milled materials redeposit on the

91
freshly milled surfaces. Redeposited materials contain much more Ga (>18%) compared to
directly amorphised layers (104; 105).

The amorphous layer is a physical layer with a distinctive thickness, while the crystalline
dead layer is more a hypothetical layer defined to explain to some extent the discrepancy seen
between theoretical and experimental results. Generally, in ion milled specimens, holographic
measurements show lower potentials than theoretical calculations. To reduce this difference,
a crystalline layer in which the phase step across the junction is negligible is assumed to be
on top and bottom of the specimen. This layer is called “dead layer” because the measured
built-in potential in this layer is zero. The dead layer thickness has been previously noted to
be dependent on the dopant concentration and specimen preparation details (106). For a
higher dopant concentration, the dead layer is usually thinner (106).

During the FIB preparation many of the Ga ions are implanted into the specimen. These Ga
ions can potentially act as dopants (Ga is a p-type dopant for silicon) and change the
electrostatic potential distribution of the specimen.

Using a low energy beam in the final thinning and polishing stages of the FIB preparation is
reported by many groups to be very effective in reducing ion damage. Low energy Ar ion
millers, such as a Fischione Nanomill Ar ion miller, also can be used for post-FIB processing
to reduce the damage layer (91; 107; 108; 109; 110; 111; 112). Both amorphous and dead
layer thicknesses decrease with reduced ion beam energy (113). Low temperature annealing
of FIB-prepared specimens is also reported as a way of re-crystallising the specimen surface
and removing the implanted Ga atoms (114).

4.3.3. Common FIB-based specimen preparations

The H-bar technique (98), ex-situ lift-out (115) and in-situ lift-out (116) are three common
FIB-based methods used for electron holography of semiconductor devices.

a) H-bar method
When using the H-bar method, the bulk sample is thinned using mechanical polishing to form
a <100μm thick slab, with the region of interest approximately at its centre. As shown in
Figure 4.3a, the prepared slab is glued to half of a TEM washer using silver epoxy.

92
 Figure 4.3. a) Schematic diagram showing H-bar specimen
preparation. b) A bright field TEM image of a specimen prepared
using the H-bar method.

After protecting the region of interest from the ion beam which is normal to the sample
surface by depositing a 1μm-thick platinum strip, two large trapeziums are milled on either
side of the protective stripe using a high ion beam current of 6.5nA. Then the region of
interest is thinned down to 2μm using a medium beam current of 3nA. Further thinning is
carried out using much lower beam currents 300pA is used down to the thickness of 1μm
and 100pA beam for the final thinning. Before the final thinning, cuts are made in the
membrane to provide the vacuum region for the reference wave. To reduce the thickness of
the damage layer, a low energy beam of 2keV is used for polishing the surface of the
specimen at the small angle of ±3o . In order to compensate for the shape of the ion beam and
to obtain a parallel lamella, the specimen is tilted slightly depending on the beam current.
Figure 4.3b shows the low magnification TEM image of a final specimen.

One of the main draw-backs of the H-bar geometry is that the milling process is constrained
to be from only the front-side of the sample, so avoiding the curtaining effect by using back-

93
side milling is not possible. Also in this geometry, tilting the specimen in the TEM is very
limited as the beam can be blocked by the thick part of the sample even at a small tilt angle.

b) Ex-situ lift-out
Transferring a FIB-thinned lamella from the substrate to a TEM grid, known as the “lift-out”
method, was first proposed by Overwijk et al. in 1993 (117). Compared to the H-bar

Inside the FIB Outside the FIB

a) Two trenches are e) A sharp and freshly
milled on both sides of prepared glass needle is
the protective Pt strip. brought close to the
lamella using a micro-
manipulator.

b) At tilt, the f) The needle is swept

lamella is detached over the lamella.
from the sides and
bottom, leaving the
two joints.

c) At tilt, the g) The lamella is

lamella is thinned attached to the glass
down to the desired needle using the
thickness. electrostatic force.

d) The two joints are h) By sweeping the

milled and the lamella needle on a TEM
is detached from the carbon grid, the lamella
bulk completely. is transferred to the
grid.

Figure 4.4. Steps involved in the ex-situ lift-out TEM specimen preparation.

technique which requires pre-thinning, the lift-out method requires almost no prior
preparation before loading the sample into the FIB workstation. The steps involved in the ex-
situ lift-out method are shown in Figure 4.4. After depositing the protective Pt strip, the
material on both sides of the strip is removed by digging two trenches. The amount of
material that needs to be milled away in this stage is much smaller than the H-bar method.

94
 o
Afterwards, the lamella is thinned down to 1μm. By tilting the sample relative to the ion
beam, the bottom and two sides of the lamella are milled, cutting through the region between
the two trenches and leaving the lamella attached to the bulk only by two thin joints (marked
in Figure 4.4b). Then the sample is tilted back to the original position and the lamella is
thinned to the required final thickness using a low beam current. The last preparation step
using the FIB workstation is to cut the two joints and release the lamella. The process of
lifting out and transferring the membrane to a TEM grid is carried out outside the FIB, which
is why this techni ue is called “ex-situ lift-out”. A freshly prepared sharp glass needle is
brought towards the lamella very carefully while watching the approach with an optical
microscope. The electrostatic force at the tip of the needle is strong enough to attract the
lamella when the needle is swept back and forth about 1μm above the membrane. Then the
lamella is carried by the needle toward a TEM carbon grid. By sweeping the needle over the
carbon film, the lamella is then transferred to the grid.

From the semiconductor industry point of view, the ability to prepare a very site-specific
TEM specimen from an unbroken wafer is considered as a great advantage for the ex-situ lift-
out, particularly because many steps of this method can be automated. However, this
technique is not ideal. As with the H-bar method, curtaining is inevitable as backside milling
is not an option. Further thinning of the sample, which is sometimes required, is not possible
after putting the sample on the carbon grid. Redeposition is more problematic in the ex-situ
lift-out technique, particularly because when cutting the joints as shown in Figure 4.4d the
milled materials are likely to redeposit on the final polished surfaces of the membrane. Also
the success rate in the ex-situ lift-out highly depends on the user’s skill.

95
c) In-situ lift-out
By installing a micromanipulator inside the FIB chamber, the lift-out procedure can be
carried out in-situ. Steps involved in the in-situ lift-out are depicted in Figure 4.5.

a) Trenches are dug on d) The joint point is

both sides of the milled away and the
protective Pt stripe. lamella, bonded to the
needle, is lifted out.

b) At 45o tilt, the lamella e) The lamella is attached

is detached from the sides to a TEM grid in a similar
and bottom except one way that the needle and
point labelled “Joint”. lamella were bonded.

c) By depositing Pt in the f) The joint point between

tiny gap between the the needle and lamella is
needle and lamella, they milled away. The sample
are bonded together. is ready for further
thinning.

Figure 4.5. Steps involved in in-situ lift-out TEM specimen preparation.

Several steps are involved in the in-situ lift-out procedure. After digging two trenches on
either side of the Pt stripe protecting the region of interest, the sample is tilted usually by
relative to the ion beam to partially detach the lamella from the substrate. At this step, the
lamella becomes detached from the bulk in all three sides except one point, labelled “Joint” in
Figure 4.5b. Then the sample is tilted back to the 0o tilt position relative to the ion beam.
Similar to the ex-situ lift-out, the needle is brought towards the lamella gradually. During the
approaching, the needle is being watched carefully with the SEM (in dual-beam system) and
with the FIB in the imaging mode. In the single beam FIB systems, the shadow of the needle
on the lamella appears when the needle is close enough to the lamella. In dual-beam systems,
the right position of the needle can be found with much less effort because imaging can be
carried out at two different angles using FIB and SEM. After parking the needle in the
immediate vicinity of the lamella, the needle and lamella need to be bonded together.
Depositing Pt or W fills up the remaining gap between the needle and the lamella, and bonds
them together. Afterwards, the joint point is milled away. At this point, the specimen, freed
from the substrate, is transferred to a TEM grid. The same procedure used to attach the needle

96
to the lamella can be applied to attach the lamella to the TEM grid. After attaching the
lamella to the TEM grid, it is safe to mill away the joint between the needle and the lamella.
At this point the sample is ready for further thinning and polishing.

In-situ lift-out offers many advantages over other FIB based specimen preparation
techniques. If necessary, the specimen can be further thinned and polished after initial TEM
observation. Compared to the H-bar method, the specimen can be tilted to larger angles
without shadowing problems. The in-situ lift-out technique is not destructive and the
specimen can be prepared from an unbroken wafer. Success rate of 100% is achievable after
some initial practices. Compared to the H-bar and ex-situ lift-out methods, the lamella is not
surrounded by any materials throughout the thinning and polishing steps, therefore the
redeposition has less chance to occur. Moreover, backside milling is possible by modifying
the technique.

97
4.4. The double lift-out technique

As discussed previously, the in-situ lift-out is the most suitable TEM specimen preparation
technique available for mapping the electrostatic potential distribution of semiconductor
devices using electron holography. However, curtaining is still present when using the
standard in-situ lift-out technique.
The specimen may become electrically charged under the electron beam. This matter will be
discussed in chapter 7. In the TEM pn junction specimens studied in the literature, generally
p-type and n-type layers have had dissimilar electrical paths to ground. Dissimilarity in the
grounding paths has been a concern for some researchers, because it can be the cause of
charging in pn junction specimens (114; 118). In pn junction specimens prepared using the
regular in-situ lift-out method, the grounding paths are not similar for the n- and p-layers. The
layer located on top of the sample is directly in contact with a metal layer, either the
protective Pt/W layer deposited by FIB or a metal layer coated on the sample before being
loaded into the FIB platform, while the other layer has a different path to the ground.
The double lift-out technique was developed as a part of this project. Both the curtaining
effect in MOS device samples and the dissimilar grounding paths in pn junction samples are
avoided with this method because the ion beam has access to the specimen back-side. The
double lift-out method is illustrated schematically in Figure 4.6. The Z axes of the coordinate
systems shown in Figure 4.6 are pointing to the direction of the ion beam. Similar to the
standard lift-out, a free-standing lamella is FIB-milled and lifted out by the tungsten needle
(Figure 4.6a). One concern with the in-situ lift-out is the damage that the ion beam causes
when the bottom of the lamella is cut from the substrate. In the standard method, the bottom
o
cut is made when the sample is tilted by with respect to the ion beam. The Ga ions
penetrate and damage further into the lamella at this step. To minimise this damage instead of
o
one cut at two cuts are made by tilting the sample only ±10o . The lifted-out lamella,
therefore, has a wedge shape with a wedge angle of about . The lamella is intentionally
milled to about μm. Thus the final TEM membrane, only a thin part within the centre of the
lifted out lamella, is protected from the Ga ions throughout the whole procedure (from (a) to
(f) in Figure 4.6, until the last thinning step in Figure 4.6h).
After lifting out the rather thick and tall lamella, instead of bonding it to an erect TEM grid
like the standard in-situ lift-out, it is attached to a horizontally mounted TEM grid, as shown
in Figure 4.6b. The standard Omniprobe grid or the regular half-cut mesh grid can be used.
After attaching the lamella, the grid is removed from the FIB instrument and mounted

98
Figure 4.6. Steps involved in the double lift-out technique. The Z axes of the coordinate systems are pointing to the direction of the ion beam.

99
vertically, as in the regular in-situ lift-out, and then placed back into the FIB. This means the
ion beam would be perpendicular to the surface of the lamella. By milling most of the wedge-
shaped substrate away, the sample is trimmed at this step. Also, as shown in Figure 4.6d, in
order to provide the electron holography reference path in the case of MOS transistor
specimens, the materials above the transistors (including the protective Pt layer) are milled
away in this step. For pn junction specimens, only the substrate is trimmed off. This needs to
be performed with caution, since the ion beam is perpendicular to the specimen surface and
thus can cause damage deep inside the lamella. Although the specimen is very thick and the
Ga ions cannot reach the central part of the lamella, the FIB was not used for imaging the
specimen in this step. By precisely aligning the coincident points of the ion beam and
electron beam (in the dual beam system), ion imaging was completely avoided in this step.
After trimming the sample, the grid is removed from the FIB and once more mounted
horizontally and reloaded into the FIB. This time the grid is flipped 1 0o relative to Figure
4.6b, such that the substrate side faces upwards as shown in Figure 4.6e. The ion beam now
has access to the substrate side of the sample. After depositing a ~3μm thick Pt protective
layer, the sample can be lifted out for the second time. Then, the sample is attached to an
erect Omniprobe grid for the final thinning and polishing step, as shown in Figure 4.6g.

The specimen is thinned down to the desired thickness using low-current beams (starting
from 300pA and concluding with 50pA). After thinning, the specimen is polished by a 2keV
Ga ion beam for 10min on each side to remove the damage layer left from the 30keV beams.
The effect of 2keV Ga ion milling is shown in Figure 4.7. To measure the thickness of the
amorphous layer, an oblique FIB-cut was made at the edge of this sample to form a wedge
shape with the wedge angle of 30o . The amorphous layer at the edge of the sample thinned
using 30keV beam in Figure 4.7a is ~13nm, and after polishing the sample using 2keV ion
beam this thickness is reduced to less than 1nm, as shown in Figure 4.7b.
 Figure 4.7. TEM images from the edge of a wedge-
shaped FIB-prepared specimen with a wedge angle
of 30o, showing the effectiveness of low energy ion
milling. a) After preparing the sample with 30keV
ion beam and b) after polishing the sample surface
with 2keV beam for 10min on each side.

The double lift-out approach introduced here for accessing the backside of the sample can be
compared with that of Schwarz et al. (119), Sato et al. (94), and Han (41).

101
4.5. Specimen preparation for in-situ electrical biasing

4.5.1. Introduction
In-situ electrical biasing of a semiconductor device in TEM makes it possible to map the
electrostatic potential distribution of the device under working conditions using electron
holography. Electrostatic potential maps measured experimentally from a semiconductor
device electrically biased at different operating points would be of great interest to device
engineers. Comparing these maps with process simulations can be used in optimising the
device design and fabrication processes. This, however, is not the only reason why in-situ
electrical biasing of the TEM specimens has drawn considerable attentions from the electron
holographers.
The very first observation of pn junction in TEM was carried out under electrical biasing
conditions (120). They applied reverse bias to the junction not to study a working device, but
to increase the electric field across the junction and thus enhance the electron-beam phase-
shift. Improving the phase signal has been one of the main motivations for electron
holographers when applying an electrical bias to TEM specimens. This is why reverse biasing
conditions have been often chosen in previous holographic studies. Also examining a forward
biased pn junction is more difficult because of the effects of series resistance.
In addition, holograms taken at different biasing voltages can be used to eliminate the effects
of diffraction contrast. This contrast, arising in many cases from the strain in the specimen,
causes ambiguity in the phase image interpretation, which will be discussed in chapter 6.
Since the diffraction perturbs the phase images recorded at different voltages equally, taking
the difference between these phase images would compensate for diffraction contrast.
Likewise, the contribution of the mean inner potential to the phase image can be compensated
for, which is more important in hetero-junctions where the two sides of the junction have
different MIPs. Also in homo-junctions, removing the phase shift due to the MIP mitigates
the effect of thickness non-uniformity (see chapter 8 for the explanation).
There is a discrepancy in the literature regarding the ability of in-situ biasing to reactivate the
dopants which were deactivated presumably as a result of FIB damage. After a series of
biasing experiments on a FIB-prepared pn junction, Han et al. measured the built-in potential
of the junction under a zero volt electrical biasing condition (121). The measured potential
was in agreement with their measurement from a mechanically thinned specimen of the same
junction. To explain the absence of FIB damage, they postulate that biasing cycles anneal the

102
specimen and consequently the dopants which were passivated by FIB milling are being
reactivated. Twitchett et al. came to a similar conclusion when their measurements showed
that the charge density in the depletion region, instead of remaining unchanged, increases by
the reverse bias voltage (122). Conversely, Cooper et al. have reported no measurable
changes in the phase shift across the pn junction before and after biasing (123).
Making electrical contacts to a semiconductor device in a TEM specimen makes it possible to
compare the current-voltage characteristics between the device after the TEM specimen
preparation and under electron beam illumination with the intact unprepared device within
the bulk material. This comparison helps evaluate how specimen preparation and electron
illumination affect the electrical characteristics.
Despite these benefits from in-situ biasing, the only semiconductor device which has been
examined under electrically biased conditions in TEM is a simple pn junction. Specimen
preparation is the main obstacle. Making electrical contacts to the nanoscale terminals of a
device which has been made into a TEM specimen is challenging, even for a pn junction.
Three different approaches are reported in the literature for biasing a pn junction.
In preparing a suitable TEM specimen for biasing a pn junction, Frabboni et al. chemically
thinned a series of junctions fabricated using a custom procedure (124). The structure of their
specimen is shown in Figure 4.8. The electron transparent region close to the hole made by
chemical etching consists of several parallel pn junctions. Far away from this thin area, all the
p-type wells are electrically shorted by a TiAg layer. The patterned SiO2 layers isolate the
TiAg from the n-substrate. In this geometry, by applying external voltage between the TiAg
layer and the back of the sample, the electron transparent pn junctions close to the hole can be
biased. By applying 4V reverse bias to a pn junction, Frabboni et al. amplified the electric
field across the junction and managed for the first time to record the leakage of this field into
the vacuum using electron holography.

103
 Figure 4.8. The geometry used by Frabboni et al. to apply an in-situ electrical
bias to a pn junction in TEM. (58)

The approach Twitchett et al. took in 2004 for the electrical biasing of a FIB prepared pn
junction involves micromachining a thin membrane on one corner of a cleaved cube (125).
The electrical contacts are made to the top and bottom of the cube, so that the voltage can be
applied between the two sides of the sample. This technique is very convenient when a p-type
(or n-type) layer is grown on an n-type (or p-type) substrate, but cannot be applied to more
complex structures. This method is described further later in this chapter.
The structure shown in Figure 4.9 was developed by Han et al. to bias a pn junction with the
original intention to apply it to a real device (121). After forming the pn junctions by
diffusing the n+-wells in a p-type substrate, they applied the H-bar method to prepare
specimen for TEM. To protect the surface of the sample from Ga ion beam, they coated the
sample with Cr before mechanical thinning. In this geometry, the mechanically thinned
sample is attached to a half-cut Cu grid using silver epoxy. In that way, the electrical contact
to the p-type substrate is made via the conductive epoxy, which is large and far from the
electron transparent area. The electrical contact to the n+-well is made via a movable tungsten
needle as shown in Figure 4.9c. The needle is moved in to touch the FIB deposited Pt layer
which is connected to the n+-well via the Cr film. The patterned oxide film (used in the
fabrication process as a mask) isolates the Cr from the p-type substrate.

104
 Figure 4.9. a) Schematic diagram and b) FIB image showing the
geometry used by Han et al. for applying an electrical bias to a
pn junction TEM specimen. This geometry is based on the H-
bar technique. c) Diagram of the sample structure. (reproduced
from (121) )

4.5.2. Biasing specimen holder

Figure 4.10 shows the specimen holder employed in the biasing experiments described in the
next chapters (126). In the last two biasing experiments, explained above, the same holder
was used. This single tilt holder has the ability to tilt the sample up to ±70o from the
horizontal and contains two types of removable cartridges which can be mounted at the end
of the TEM holder rod as shown in Figure 4.10. One of the main benefits of using this holder
is that after loading the sample in the cartridge, it can be safely transferred between the TEM
and FIB workstation.
In order to bias the same TEM specimen in the FIB/SEM and the TEM, a stub shown in
Figure 4.11 was developed during this work, which is compatible with these cartridges and
provides required wiring for them. Both cartridges have two electrical contacts which are
connected to the TEM holder and the SEM stub. The TEM holder has a nanoprobe which can
be positioned in three-axes (X, Y, Z) using piezo-drives. This movable probe is used as the
third electrical contact in biasing experiments.

105
Figure 4.10. Three contact biasing TEM holder and its movable cartridges. In cartridge A the
specimen is clamped vertically while in cartridge B it is clamped horizontally. These two
cartridges allow different specimen geometries to be examined. Both cartridges contain two
electrical contacts. For the third contact, a sharp needle can be positioned in x, y and z direction
using micrometers for coarse movements and piezo-drives for fine ones.

Figure 4.11. The SEM stub designed to mount the TEM cartridges in
FIB/SEM workstations and apply an electrical bias to the TEM specimen in
SEM. Three pins are attached to mount the stub in different directions.

106
4.5.3. Cube geometry
The reliable and straightforward specimen preparation shown in Figure 4.12 has been applied
in many of biasing experiments presented in next chapters. This geometry suggested first by
Twitchett et al. is referred to as the cube geometry in this work.
To prepare a specimen in the cube geometry, a large piece of wafer (1cm 1cm) was cleaved
from the pn junction samples and mechanically thinned down to the thickness of ~ 0μm.
While the thinned sample was still glued to the grinding disc, the backside of the sample was
gently scribed, as shown in Figure 4.12. After dissolving the mounting wax, gentle pressure
on the backside of the sample using a tweezer produced many 90o cleaved wedge specimens
of ~ 2mm×2mm. Then, the cleaved sample was mounted into the cartridge. The two
relatively large electrodes clamp the sample firmly, so that reliable electrical contacts are
made on the top (n-layer in our sample) and bottom (p-layer) of the cleaved sample. Next, a
thin parallel-sided membrane was micromachined from the corner of the cleaved cube using
the FIB.

Figure 4.12. Specimen preparation for the cubic geometry. The sample is mechanically
thinned down to ~ 0μm and cleaved to pieces of ~2mm 2mm. The 0μm 2mm 2mm
cube is mounted in cartridge A (see Figure 4.11) and loaded in the FIB. A thin lamella is
micromachined at the corner of the cube. Then the cartridge can be transferred to the
TEM holder.

107
4.5.4. Novel specimen geometry which could be applied to a real device
A suitable TEM specimen preparation approach for applying an electrical bias to a real
device, such as a transistor, should have site specificity not only in specimen preparation but
also when electrical contacts are made to tiny terminals of the device. In the approaches
proposed in the literature (and described here earlier) for biasing the specimen, the biasing
voltages are conducted to the electron transparent region by the bulk of the sample. The
electrical contacts are far from the thin area and thus applied voltages might be dropped at
other junctions, also as a function of distance between the contacts and the region of interest.
The limitation of existing electrical biasing geometries in making site-specific electrical
contacts to TEM specimens has also been an obstacle for the recent study carried out by Liu
et al. on a GaN light emitting diode (LED), where several attempts have been made to use the
cube geometry for the application of electrical biases to two relatively large terminals of the
LED (127).
The specimen geometry illustrated in Figure 4.14f allows for the application of electrical
biases directly to the thin p-type and n-type regions. This geometry is suitable for biasing 2D
devices in TEM. The steps involved in preparing the specimen are as follows:

A) Preparing the biasing grid

A glass slide is ground and polished down to a thickness of less than ~ 0μm. The thinned
slide is then sandwiched between two 1mm Al sheets using G1 epoxy. As with conventional
cross-sectional TEM specimen preparation, the sandwich is cut into thin pieces using a
diamond saw, and mechanically thinned down to a thickness of ~30μm. The grid is loaded in
the cartridge and silver paint is used to fix the contacts if necessary. To avoid short-circuiting
the contacts by the cartridge clamp, a very thin piece of glass is glued to the grid as shown in
Figure 4.13c.
In the next step, the FIB is used to prepare the grid for mounting the TEM membrane. The
location where the TEM membrane needs to be mounted is prepared by cutting a window
using the FIB. The window is milled mainly on the Al side rather than on the glass; as a
result, the electron beam does not hit the glass part of the grid during holography, prevents
the grid from charging during holography experiments. Before mounting the TEM
membrane, two tungsten lines from the Al pads to the glass edge of the window are deposited
using the FIB, as shown in the SEM images in Figure 4.13e and 4.13f.

108
 Figure 4.13. Steps involved in preparing the biasing grid. a) A mechanically thinned glass slide,
with a thickness of less than 0μm, is sandwiched between two Al pieces. b) The sandwich is
cut and thinned down to ~30μm. c) The grid is mounted in the cartridge. To avoid the electrical
short-circuit caused by the cartridge clamp, a small glass piece is glued to the grid. d) A window
is milled using FIB as shown in the optical photo. SEM images in e) and f) show the required
wiring to the Al pad 2 and 1 carried out by depositing W lines.

B) Mounting the TEM membrane

The double lift-out procedure described earlier in this chapter is used to prepare the TEM
membrane. After the second lift-out, the specimen is mounted on the biasing grid. The FIB
induced tungsten deposition is applied to connect the contacts on the membrane to the
tungsten lines on the biasing grid as shown in Figure 4.14. After finishing the FIB deposition
process, the tungsten deposition over sprays around the tungsten lines need to be removed
using FIB milling, to avoid an electrical short-circuit. To ensure that tungsten deposition
during the attaching of the membrane does not short-circuit the contacts, a hole is milled at
the interface between the specimen and glass as shown in Figure 4.14d.

109
 Figure 4.14. Images showing the procedure applied for attaching the specimen to a
biasing grid. a) The sample, lifted-out using the “double lift-out” techni ue, is bonded
to a biasing grid using the FIB induced W deposition. b) The contacts that are made
between the W lines on the specimen and on the grid. c) The connections that are made
between the W lines using the FIB-induced W deposition. d) To avoid electrical short-
circuits between the W lines, a hole is milled throughout the specimen. e) The
specimen thinned down to the desired thickness. f) Schematic diagram of an
electrically wired specimen.

The results from biasing experiments on this geometry will be discussed in chapter 5.

110
4.6. In-situ sputter-coating in FIB workstations

To protect the surface of specimen from oxidation after the FIB preparation, a thin layer of
gold/palladium was sputter-coated in-situ in FIB. This metallic layer also provides a
conductive path for charges to be dissipated during electron holography and has other
advantages which will be discussed in chapter 7. To perform sputter coating in the FIB
chamber, material was lifted out from a gold/palladium wire using the standard in-situ lift-out
method. By milling the surface of the Au/Pd as a target in close proximity of the sample,
sputtered Au/Pd atoms deposit on the sample. Deposition rate depends on the milling rate,
distance and place of the sample with respect to the target. As Ga ions are used for sputtering
the target, the deposited film will contain Ga. Other materials can also be deposited using this
method.
To find suitable conditions for coating the surface of TEM specimens, the Au/Pd target was
milled with different ion beam currents and at different distances from the sample. When the
target and sample were face-to-face, the deposition rate was high. The SEM dopant contrast
was completely masked after milling of the target for 60sec with the low ion beam current of
93pA at a distance of more than 10μm from the sample. Figure 4.15b shows the SEM image,
taken at 2o , of the Au Pd piece parked 10μm away from the pn junction sample before
deposition. The dopant contrast which can be seen in Figure 4.15b disappears after the
deposition, seen in Figure 4.15c. The FIB image in Figure 4.15a shows the position and
distance of the Au/Pd target relative to the sample.

Figure 4.15. a) FIB image (top view) showing the positions of the Au/Pd target and the specimen. b)
SEM image (tilted view) before sputter coating and c) after sputter coating.

The suitable conditions given in the following were found by means of trial and error. The
Au Pd target was parked μm from the sample as seen in the SEM image in Figure 4.16. The
surfaces of the target were milled with the 93pA/30keV Ga ion beam for 10min on each side.

111
The size and position of the window used for the milling are shown in the FIB image in
Figure 4.16. These conditions were applied in preparing the specimen characterised in
chapter 7.

Figure 4.16. The FIB image (left), from the top view and the SEM image
(right) from the side view show the position of the target relative to the
specimen. The dashed box in the FIB image was used to mill the target for
10min with a 30keV/93pA ion beam.

112
4.7. A list of the samples used in this thesis

The results presented in this dissertation are from four different samples: two different pn
junctions and two types of MOS transistor: p-type and n-type.

- Sample A (pn junction): The Si pn junction in this sample is formed by growing a

1μm-thick phosphorus doped (n-type) layer followed by the same thickness of boron
doped layer (p-type) onto a lightly doped p-type substrate using reduced-pressure
chemical vapour deposition. The junction is symmetric and abrupt with the dopant
concentration of 2 101 cm3. The dopant concentration profile in this sample
measured using secondary ion mass spectroscopy (SIMS) (128) is shown in Figure
4.17.

Figure 4.17. SIMS profile acquired

from the sample A. (128)

- Sample B (np junction): This silicon sample was generously provided by OKMETIC
Company (129). The pn junction comprised a μm-thick arsenic doped n-type layer
which had been grown epitaxially onto a (100) oriented boron-doped p-type substrate
using molecular beam epitaxy (MBE). The nominal dopant concentration on both
sides was in excess of 1020 cm-3 . However, the measured resistivity using a four-point
probe was 0.01 ±0.00 -cm and 0.00 ±0.000 -cm for p-type substrate and n-type
epitaxial layer respectively. These measurements correspond to active dopant
concentration of ~6 101 cm-3 for both p- and n-sides, and a built-in potential
of bi 1 . Some of the dopant atoms presumably form interstitial or clustered defects
in the silicon crystal. Also, controlling the epitaxial growth conditions is difficult for
high dopant concentrations (130).

113
 - Sample C (PMOS transistors): the examined device sample contained closely packed
0.1 μm p-type silicon metal-oxide-semiconductor-field-effect-transistors (MOSFET)
with a dopant concentration of 3 101 cm-3 . The theoretical built-in potential for these
MOSFETs is ~1
- Sample D (NMOS transistors): The structure of this sample is almost the same as
sample C with the same dopant concentrations. In chapter 8, these two samples are
characterised in details and differences between them are explained.
TEM specimens were prepared in a FEI Helios Nanolab 600 dual-beam FIB/SEM system
equipped with gas injection systems for both tungsten and platinum deposition and with an
in-situ Omniprobe for micromanipulations. The FIB can be operated at different accelerating
voltages between 500V and 30keV.

114
Chapter 5

5. A Si pn junction under electrical biasing conditions

In principle, characteristic properties of a pn junction are measurable with high accuracy
using off-axis electron holography. The phase shift introduced in an electron wave while
passing through a sample is proportional to the electrostatic potential within that specimen.
Electron holography reveals this phase shift, so the distribution of electrostatic potential
associated with a pn junction can be mapped from holographic phase images. Thus,
according to the Poisson equation, the distribution of the electric field and charge density can
be calculated by taking the first and second derivative of the potential. These characteristic
properties should respond appropriately to external biasing voltages in a working diode, a pn
junction.

pn junction specimens under bias voltages are analysed in this chapter. In the first section, a
specimen with the cube geometry discussed in the previous chapter is investigated and in the
second section, the novel geometry introduced in the preceding chapter for the electrical
biasing is examined.

5.1. Results from the cube geometry

To apply external voltages to a pn junction, the cube geometry was used to prepare a TEM
specimen from sample B described in chapter 4, the pn junction with an active dopant
concentration of ~6 101 cm-3. The crystalline thickness of the specimen was determined to
be 0±10nm using convergent beam electron diffraction (CBED) under a two beam
condition (47). The external voltage was swept from 0V to 2V reverse bias and from 0V to
1V forward bias in intervals of 0.2V. To characterise these biasing conditions, off-axis
electron holograms were acquired.

115
The holograms were acquired using an FEI Titan TEM 80-300 operated at 120kV in Lorentz
mode. By setting the biprism voltage at f 70 and magnification at M = 18600, holograms
with a field of view of ~0.6 2. μm2 and 6 pixels per holographic fringe were acquired. A
representative object and reference hologram recorded from the specimen under 0V biasing
conditions is shown in Figure 5.1. The reconstructed phase and amplitude images from this
hologram pair (with the reconstruction mask radius of 1/14nm) are also shown in Figure 5.1c
and 5.1d respectively. The p- and n- regions are clearly discernible in the phase image. The
initial orientation of the specimen in the single tilt biasing holder was approximately 6o away
from the <001> zone axis. The pn junction was oriented exactly edge-on with respect to the
electron beam by tilting the specimen to the 040 Kikuchi band centre line where at this
orientation it was 5.2o away from the <001> zone axis. No diffraction contrast is seen in the
amplitude image. The effect of diffraction contrast on the phase image will be discussed
further in the next chapter.

Figure 5.1. A representative electron hologram pair: a) the object hologram, b)

the reference hologram and their corresponding reconstructed c) phase and d)
amplitude images.

116
5.1.1. The phase-step across the junction as a function of biasing voltage
As explained in chapter 2, when a p-type and n-type semiconductor are brought into contact
(in thermal equilibrium where no currents exist and no external voltage is applied) a potential
barrier, called the built-in potential, is formed between the p-side and n-side. This potential
barrier appears as a phase step in phase images such as the one shown in Figure 5.1c. The n-
side has a higher potential than the p-side; therefore, the electron beam experiences a larger
phase shift, according to Equation 3.36, as it passes through the n-side than it does through
the p-side.

By applying an external potential between the p- and n- regions, the height of the potential
barrier changes accordingly. The observation of this change in the phase image indicates that
the pn junction within the TEM specimen responds to the external voltage. In a reverse biased
pn junction, the potential barrier V is the sum of the built-in potential and the applied
voltage . So, the phase step across the junction is expected to increase
linearly with the applied reverse voltage.

∆ CE bi app t Eq.(5.1)

For measuring the phase steps, the following sigmoidal equation with 4 parameters A, B, C,
and D was fitted to the phase profiles.

B Eq.(5.2)
y A x–C
1 e
As shown in chapter 2, a pn junction potential profile can be described with two simple
quadratic equations within the depletion region and two straight lines outside of it. In
commercial device simulators, however, more sophisticated models are used to include pn
junction non-idealities such as generation-recombination in the depletion region, high-level
injection and series resistance effects (see chapter 2). To evaluate how well Equation 5.2 can
predict the potential profile of a symmetric pn junction, it was fitted to a wide range of
simulated potential profiles from “P Junction Lab” software (a PA RE-based software
(131)). Potential profiles were simulated under the same dopant concentrations and biasing
conditions used in this work. As examples, the curve fitting results for 2V, 1.2V and 0V
reverse bias and 0.8V forward bias are shown in Figure 5.2. From these graphs and their
corresponding r-squared values, shown for each fit, it is clear that the curve fitting error
which might be introduced by Equation 5.2 is negligible.

117
 Figure 5.2. emonstrate the “goodness of the fit” of E uation .2 to the simulated potential
profiles obtained from PN Junction Lab software. (NA = ND = 6×1018 cm-3)

For the specimen under examination, the phase step measured across the pn junction is
plotted in Figure 5.3a as a function of applied reverse bias. An example of fitting Equation
5.2 to an experimental phase profile is shown in Figure 5.3b. As can be seen in Figure 5.3a,
the phase step increases linearly with the applied reverse bias, which shows that the junction
has responded as expected to the external voltages. By comparing the best-fit line to the
phase steps, as shown in Figure 5.3a, with Equation 5.1 the specimen thickness t and built-in
potential bi were measured to be 500±10nm and 0.9±0.1V respectively. Dividing the slope
by CE yields the specimen thickness and dividing the intercept by the slope results in the
built-in potential.

∆ .3 app 3.92
t 00nm, bi 0.9
∆ CE t app CE t bi

This measurement looks very promising since, within the experimental margin of error, the
measured built-in potential is in agreement with the theoretical calculation. The theoretical
value of built-in potential across a pn junction with the symmetric dopant concentration of

118
6 101 cm-3 is 1.02V. The difference between the crystalline thickness measured using
CBED and the thickness measured from the above plot of phase step versus applied voltage is
50±10nm. This difference can be explained by considering a 25±5nm thick crystalline layer
on top and bottom of the specimen, which has been electrically deactivated because of the
FIB damage. The crystalline dead layer has been widely assumed in the literature for the
explanation of holographic results (86; 125; 128).

Figure 5.3. a) The phase step across the pn junction as a

function of applied reverse bias. Note that, as expected, the
phase step increases linearly with reverse biasing voltages. b)
A representative phase profile taken at 0V biasing condition
across the pn junction. The phase steps used in (a) were
measured by fitting a curve to the phase profiles, as
demonstrated in (b).

119
However, the graph of phase step versus applied voltage in Figure 5.3a can also be
interpreted from another perspective. In the above calculation of built-in potential, it was
assumed that the total applied voltage is dropped across the junction and nowhere else in the
circuit. While, for example, it is likely that some of the applied voltage is dropped on the
contacts between the cartridge electrodes and the specimen. The slope of the best-fit line in
Figure 5.3a is 8% smaller than 4.71radV-1 (CEt = 0.00856 radV-1nm-1×550 nm) expected from
a 550nm thick crystalline specimen. Instead of relating this difference to the crystalline dead
layer, it can be assumed that 8% of the applied voltage is dropped somewhere else in the
circuit, not across the junction. With this assumption the built-in potential of 0.84V is
calculated from the phase step versus applied voltage graph in Figure 5.3a.

120
5.1.2. Measuring the electric field and charge density from the phase images
To directly extract the electric field and charge density distribution from a phase image, the
noise level in the phase image needs to be low. With each differentiation to determining the
electric field and charge density, the original noise in the phase image is amplified. One
approach to overcome this problem is to fit a curve to the phase profile and differentiate the
fitted curve. This approach was used by Twitchett et al. in (122). In most of the phase images
presented in this thesis, the signal to noise ratio was large enough that even the second
derivative of the phase images was clear. Curve fitting was not necessary to determine the
electric field and charge density.

Figure 5.4 shows the potential, electric field and charge density images obtained directly
from differentiating the phase images at different biasing voltages. The phase and amplitude
images at 0V biasing are shown on the left of the figure. The region marked by the dashed
box on the phase image is used for the calculation of the electric field and charge density
images on the right.

To translate the phase image into an electrostatic potential map the entire 550nm crystalline
thickness of the sample was considered. The potential map V(x, y) was calculated from:

(x, y)
x, y
CE t
-1
(x, y) phase (rad); CE 0.00 6 rad. .nm-1 ; thickness t 0nm.
Then, the electric field E(x, y) was calculated by taking the first derivative of the potential:
E(x,y) - (x,y).

As can be seen, the electric field is maximised at the junction. Qualitatively, the electric field
images show that the junction has responded appropriately to the external voltages. Under
forward bias, the depletion width decreases and under reverse bias it increases as expected.

To determine the charge density , the second derivative of the potential was calculated.

2 -
( 0 . 10-1 F cm, Si 11.7)
Si 0

On the n-side the charge is positive and on the p-side it is negative, as seen in the calculated
images. For quantitative analysis, profiles taken from the same regions marked on the images
in Figure 5.4 by dashed boxes are compared with the simulation. For the simulation, “P
Junction Lab” software developed by Vasileska et al. was used (132).

121
Figure 5.4. Demonstration of the 2D potential, electric field and charge density distributions obtained directly from the holographic phase
images. Representative phase and amplitude images are shown on the left. The dashed white box on the phase image shows the area of
which the potential, electric field and charge density images are illustrated on the right. The images are not displayed in their full field of
view because the details would be less visible. These images are shown at different biasing voltages, from +0.8V forward-bias to -2V
reverse bias with intervals of 0.4V. Profiles taken from the red dashed boxes on the images are used for quantitative comparisons.

122
The simulated and experimentally measured electric fields across the junction for different
biasing voltages are shown in Figure 5.5a and 5.5b, respectively. As this shows, the general
trend of experimental data is consistent with simulation. The depletion width and maximum
electric field increase with reverse-bias voltages and decrease with forward-bias. However,
quantitatively, there are large differences between the experiment and simulation.

Figure 5.5. a) Simulated electric fields across a symmetric and

abrupt pn junction with p and n dopant concentrations of
6 101 cm-3 at different biasing voltages. b) Electric field
profiles taken from the area shown in Figure 5.4. Note the
asymmetry in these profiles particularly on the p-region
indicated by the arrow. The horizontal scales are different
between the experimental and simulated graphs.

123
The simulated and measured electric fields at the junction are plotted in Figure 5.6 as a
function of biasing voltage. The measured electric fields are about seven times smaller than
the simulated ones. Considering the electrically dead layer only increases the electric field by
10%.

-200

-400
Max of Electric Field (kV/cm)

-600

-800

-1000

-1200

-1400 Simulation
Measurement
-1600

-1800
-2.0 -1.6 -1.2 -0.8 -0.4 0.0 0.4 0.8
Applied Voltage (V)

Figure 5.6. The simulated and measured maximum electric field at the pn
junction as a function of biasing voltage.

Another important discrepancy between the simulated and measured electric fields in Figure
5.5 is related to the depletion region. The measured depletion widths are about 5-10 times
larger than the simulated ones. Moreover, the depletion region is asymmetric. This
asymmetry shows itself mainly on the p-side and increases by reverse voltages. The arrow in
Figure 5.5b highlights this asymmetry. The area under the electric field graphs represents the
built-in potential and should be equal on the p-side and n-side for a symmetric junction. This
area is plotted as a function of bias voltages for the p- and n-region separately in Figure 5.7.
Although the total area (the sum of the area on the p- and n-side) is approximately consistent
with the predicted potential barrier, the contribution of the p-side is always larger than the n-
side. For example at 2V reverse bias, the area under the p-side represents ~1.8V while the
area under the n-side is only ~1.2V. The sum of these two values is in agreement with
~2×1.5V potential barrier expected to be built up across the junction under 2V reverse bias.

124
 Figure 5.7. The area under the electric field profiles in Figure 5.5b on the p-side and n-side.
These areas should theoretically be the same on the p- and n-side for a symmetric pn junction.
The green triangle symbols represent the area under the p-side (or n-side) of the simulated
electric field graphs in Figure 5.5a.

All the above discrepancies in the electric field profiles appear also in the profiles from the
charge densities, as plotted in Figure 5.8. The measured charge density is more than order of
magnitude smaller than the dopant density of 6 101 cm-3 for this sample. Another
inconsistency between theory and measurement, shown in Figure 5.8, is the increment of
charge density with biasing voltages. The charge density, which is expected to remain
unchanged under biasing voltage variation, increases with the reverse voltages. Assuming
that the measured charge densities are correct, Twitchett et al. concluded that the electrical
biasing of the specimen reactivates some of the dopants which have been passivated during
the specimen preparation (122). However, by repeating the experiment after many biasing
cycles we obtained the same results. If the deactivated dopants were reactivated by applying
external voltages, they would have remained active after removing the bias voltage.
However, the same charge density was measured at 0V before and after the biasing cycle.

125
 Figure 5.8. a) Simulated charge densities across a symmetric abrupt
pn junction with the dopant concentration of 6 101 cm-3 at different
biasing voltages. b) Experimental charge density profiles taken from
the area shown in Figure 5.4. Note that, unlike the simulation, the
charge density increases with reverse voltages in these profiles. The
horizontal scales are different between the experimental and
simulated graphs.

Despite the fact that the phase step across the junction unequivocally responds to the biasing
voltages, the experimental phase variation in the depletion region are not consistent with
theory. Considering a uniformly thick dead layer on top and bottom of the specimen might
help recover the built-in potential, but it does not explain the other discrepancies mentioned
here.

126
5.2. Results from the novel geometry introduced for the biasing of real

devices

In developing the new geometry, introduced in the previous chapter to apply electrical biases,
two steps were involved. The first step, explained in detail in chapter 4, was the design and
preparation of the specimen. In the second step, the prepared specimen needed to be tested
electrically in the microscope. First attempts were not successful in applying an electrical bias
to this specimen geometry, as the phase step across the junction did not response to the
external biases. The experiments carried out to identify the causes of the failure are described
in this section. In the following section, the results from this geometry are presented after it
responded to the external voltages.

5.2.1. Problems

a) Initial biasing experiment

The low-magnification TEM image in Figure 5.9 shows a prepared TEM specimen. The Pt
lines deposited using FIB for applying voltages are marked on the image.

Figure 5.9. A low magnification TEM image

of a pn junction specimen prepared with the
new geometry. FIB-deposited Pt lines are
marked on the image.

The crystalline thickness of the specimen was measured to be 350±10nm using CBED. The
phase and amplitude images shown in Figure 5.10 are reconstructed from the holograms

127
taken before applying any voltages. The phase profile across the junction showed a phase
step of 1.5rad.

Figure 5.10. The amplitude and phase images reconstructed from the
hologram taken before applying any voltages to the specimen. The
line profile taken from the region indicated on the phase image shows
1.5rad phase step across the junction.

The same biasing voltages used in examining the cube geometry, 0V to 2V, under both
forward and reverse bias, were applied to the specimen. The phase step did not change.
Therefore, we increased the bias voltage gradually and took holograms. Up to 10V, no
change was observed in the phase image whereas at ~10V the specimen started to drift. The
specimen drifted quickly and, in less than one minute, it cracked at the platinum-silicon
interface, as shown in Figure 5.11c. Immediately after the appearance of the crack, the
biasing voltage was turned down to zero. Surprisingly, the phase-step measured across the

128
junction afterwards was 40% larger. Not only did the phase step increase from 1.5rad to
2.1rad, but also the phase profile was more symmetric, which can be seen in Figure 5.11.

Figure 5.11. a) The amplitude and b) phase images reconstructed from a hologram taken under
0V biasing condition and after the application of 10V reverse bias for 1min. c) A low
magnification TEM image of the specimen showing the crack formed after the 10V bias. The
droplets marked on the image are formed after the biasing. d) The phase profile from the area
indicated on the phase image shows the phase step of 2.1rad. Note that this phase profile is more
symmetric than the one before biasing shown in Figure 5.10.

To investigate whether after this specimen modification occurred during the application of
biasing voltage the phase step responses to the external voltages or not, bias voltages were
applied again. By starting from small voltages and increasing slowly, no change was
observed. Neither the phase step changed nor the specimen drifted. Presumably, the electrical
contacts were broken after the formation of the crack, although the crack was only limited to
a small part of the specimen.

129
b) Annealing experiment
When the large voltage of 10V was applied in the initial biasing experiment to the new
geometry, the specimen started to drift. This observation suggested that the specimen
temperature increases as a result of biasing. Studies carried out by Cooper et al. of annealing
FIB prepared pn junction specimens showed that the phase step across the junction improves
after annealing (114). Also here, in the initial biasing experiment, a larger phase step across
the junction was measured after the experiment. The phase step increment and specimen drift
suggested that the specimen was annealed during the biasing experiment.

Because in this biasing geometry the lamella is attached to the biasing grid (see chapter 4),
the specimen does not fit into commercially available heating holders. To investigate the
effects of annealing on the specimen, a specimen prepared by the double lift-out method was
used. As explained in chapter 4, double lift-out specimens have the same geometry as the
biasing specimens except that they are attached to the Omniprobe grid, so they can be
annealed in-situ in TEM using available heating holders. The same procedure used in
preparing the biasing specimen was also applied, in preparing this sample. The crystalline
thickness of this specimen was also measured to be 350±10nm.

For annealing the sample, a Gatan 652 double tilt heating holder (with a SmartSet Hot Stage
Controller) was used. The specimen was annealed at 00o C for one hour and then the
temperature was reduced to room temperature. The annealing temperature and duration were
taken from (118), where effects of in-situ annealing on the phase images of FIB prepared pn
junctions were studied.

Before and after annealing, off-axis electron holograms were acquired under the same
acquisition conditions. The phase profiles taken across the junction before and after the
annealing are plotted in Figure 5.12. The phase step increased from 1.4rad before annealing
to 2rad after, similar to the biasing experiment. Compared to the phase profile in Figure 5.11,
the depletion region in this specimen is wider and less symmetric, presumably because no
electrical contact is applied to the region of interest.

130
 0.2
0.0 Before annealing
o
-0.2 After 500 C annealing
-0.4
-0.6

Phase profile (rad)

-0.8
-1.0
-1.2
-1.4
-1.6
-1.8
-2.0
-2.2
-2.4

-700 -600 -500 -400 -300 -200 -100 0 100 200 300 400 500 600 700
Position (nm)

Figure 5.12. Phase profiles measured across the pn junction

before and after annealing. The specimen was annealed in-situ
for one hour at a nominal temperature of 00o C. Both profiles
are taken at room temperature.

This annealing experiment indicates that heating has been the source of the phase step
improvement in the initial biasing experiment. The increase in the specimen temperature
during the biasing is expected to be due to Joule heating at the electrical contacts. The fact
that the specimen was cracked at the Pt-Si interface during biasing indicates that the contact
resistance between the FIB-deposited Pt and doped silicon was large. The heat due to the
biasing appears localised to the contact region, which is why the specimen cracked in that
area approximately 1min after the specimen was biased at 10V.

Despite the biasing experiment, in the above annealing experiment the source of heat (the
holder furnace) is about 1-2 millimetres from the thin region of the specimen. The actual
specimen temperature is expected to be lower than the 00o C value measured by the
thermocouple mounted near the holder furnace. During and after annealing for one hour, no
cracks were observed in the specimen. However, when the specimen was examined in an
SEM after annealing, the Pt surface was rough. Many droplet-like features were observed on
the surface. Figure 5.13 shows an SEM image taken from the specimen after annealing.
Similar features to the droplets seen on the Pt surface in this image can also be recognised in
the TEM image of the biased sample in Figure 5.11c. The black arrows in Figure 5.11c are

131
pointing to these features which have formed close to the Pt-Si interface. From this, it appears
that heating changes the structure of the FIB-deposited Pt.

Figure 5.13. An SEM image of the specimen after

being annealed in the TEM for one hour. Note the
rough surface of Pt as indicated by the arrow.

c) Repeating the biasing experiment

The biasing experiment was repeated with a thicker sample. The non-electron-transparent
part of the specimen was kept thick during the FIB preparation in order to have a larger
contact area and, consequently, decrease the contact resistance.

As with the initial biasing experiment, small reverse voltages were applied initially, but no
changes in the phase step across the junction were observed. The voltage was increased
gradually until the sample started to drift at approximately 15V. The biasing voltage was
turned down to zero almost immediately after specimen drift started. The sample did not
crack, but the Pt layers expanded. As with the initial biasing experiment, holograms taken
afterwards showed that the phase step across the junction increased as a result of heating. The
phase step also did not respond to the biasing voltages.

Then, the voltage was increased to 15V for a second time and the specimen behaviour was
monitored. The Pt structure started changing completely. Figure 5.14 shows TEM images of
the specimen before applying any bias voltage, after applying 15V for a short time (less than
30sec) and after applying 15V for a long time (~ 3min). Also in this figure, SEM images of
the specimen after FIB preparation and after being kept under 15V bias for the long time are
presented.

132
 Figure 5.14. Low magnification TEM images of the specimen a) before applying any bias
voltage b) after applying 15V for less than 30sec and c) after applying 15V for
approximately 3min. SEM images of the specimen d) after specimen preparation and e)
after applying 15V during TEM observation. Droplets which can be seen clearly in e)
were started to form as 15V was applied to the specimen. The first two droplets observed
in TEM are marked in b).

Figure 5.14e shows the droplets that started forming under 15V bias voltage. The two
droplets marked in Figure 5.14b are the first ones that formed after applying 15V bias
voltage. Additional droplets rapidly formed, and they covered the entire surface of the Pt
layers. These droplets were forming, the specimen was drifting rapidly, which indicates that
heating was likely involved. These droplets are likely a Pt-Ga compound. Reguer et al.
reported the formation of similar droplets on the surface of FIB-deposited Pt nanowires, when
electrically biasing the wires (133). They identified the composition of these droplets as
PtGa2 and Pt.

As mentioned in chapter 4, FIB deposited layers are the result of metal-containing precursor
gas decomposition, which does not deposit a pure metal. The deposited material contains
undesired impurities such as carbon and oxygen from the precursor and Ga from the ion
beam. The Pt deposited layers, according to (88), should be composed of, for example,

133
platinum (45%), gallium (28%), carbon (24%) and oxygen (3%). Gallium and carbon
constitute a significant percentage of FIB deposited Pt layers. The formation of Pt and PtGa2
droplets on the surface of Pt layers means depleting the inside of these layers of conductive
Pt. Also, the conductive droplets on the surface of the sample are isolated from each other as
seen in Figure 5.14e. Thus, the resistance of the FIB-deposited Pt layer increases as a result
of the formation of these droplets. That is presumably why the pn junction biasing
experiments failed in the previous sections.

Unlike FIB deposited Pt which loses its conductivity under electrical bias, the conductivity of
FIB deposited tungsten (W) according to Dallaporta et al. studies improves with electrical
biasing (134). Similar droplets to those observed in this work on the Pt surfaces also form on
the FIB-deposited W nanowires. For both Pt and W, the temperature and possibly the electric
field are likely driving forces for droplet formation.

In contrast to Ga and Pt for which many different stable compounds have identified (135), Ga
and W are believed to be immiscible, as no stable compound has been found for them outside
of high-pressure regimes (136). Consequently the droplets on the FIB deposited W are likely
to be composed of only Ga. The formation of these Ga droplets purifies the W and increases
its conductivity.

Another advantage of FIB-deposited W as compared to Pt is the percentage of metal present

in the deposited material. Compared to the FIB-deposited Pt which contains only 45%
platinum, the FIB deposited W is composed of 75% tungsten (Ga: 10%, C: 10% and O: 5%)
(137). The electrical conductivity of W wires deposited using FIB was reported to be twice
that of FIB-deposited Pt wires.

Simply by using FIB deposited W instead of Pt when preparing a specimen in the new
specimen geometry, an electrical bias was applied successfully to a pn junction. The pn
junction also responded clearly to low biasing voltages. There was no need to apply high
voltages to overcome contact resistance between the FIB-deposited W and the p- and n-doped
regions. Voltages were applied between 2V reverse bias and 1V forward bias. At these
voltages, no changes in the structure of the FIB-deposited W layers were observed. The
results from this biasing experiment are analysed in the following section.

134
5.2.2. Results from successful biasing experiments with the new geometry
Figure 5.15 shows the specimen used in this section for electrical biasing experiments. FIB-
deposited W was used both for the attachment of the specimen to the biasing grid and for the
electrical connections.

Figure 5.15. The characteristic IV curve of the specimen. The inset TEM image shows the pn
junction specimen whose as-measured IV curve is plotted in blue. Removing the slope of this
curve, shown by the dashed line, yields the red IV curve. This slope can be considered as the
effect of a resistor in parallel with the pn junction, shown schematically on the right-down side
of the figure.

The blue curve in Figure 5.15 shows the experimental characteristic current-voltage (IV)
curve of the specimen. The positive voltages in this graph correspond to forward bias
voltages and negative voltages to reverse bias voltages. The exponential growth of current
under forward bias indicates that the pn junction is responding to the applied voltage. There is
no direct electrical connection between the two W lines connecting the p- and n- regions to
the Al pads. However, current leakage between these two lines cannot entirely be ruled out
although special care was taken to avoid this current leakage by milling away the W over
sprays around the W lines after the deposition and by digging a hole where the lamella is
135
attached to the biasing grid. The redeposited material and the FIB-induced damage layers
provide an additional path for the current. From the slope of IV curve in the reverse bias, the
electrical resistance of this path can be estimated to be R 0k . This leakage can be
considered as a resistor in parallel with the pn junction, as schematically shown in Figure
5.15. The red IV curve in Figure 5.15 was obtained by subtracting the contribution of this
resistor from the as-recorded blue IV curve.

Current leakage on the surface of the specimen due to redeposited material and FIB damaged
layers is not unique to this geometry and needs to be considered for all FIB-prepared TEM
specimens. Figure 5.16 shows the IV curves of a pn junction specimen prepared with the
cube geometry. The red curve was measured from the cleaved sample before FIB milling and
the blue one was measured after completion of the sample preparation. A similar slope to the
one seen in Figure 5.15 is also present here (compare the two IV curves measured before and
after FIB milling).

1000

800 After milling

Before milling
600

400

200
Current (A)

-200

-400

-600

-800

-1000
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0
Voltage (V)

Figure 5.16. IV curves from a pn junction specimen with the cube

geometry. The red curve was measured before FIB milling and the
blue one was measured after the completion of sample preparation.

It should be added that contact resistances between the p-region and W line, n-region and W
line, W lines and Al pads, Al pads and cartridge contacts and the resistance of W lines and
other components of the circuit, which are in series with the junction, also affect the IV
curve.

136
Phase images of the specimen in Figure 5.15 show the effect of bias voltage. Phase images
reconstructed from holograms taken at 1V reverse bias, 0V and 1V forward bias are
presented in Figure 5.17a, b and c respectively. These images show, the pn junction
responding appropriately to the external voltage. Applying a reverse bias increases the
potential barrier across the pn junction. This can be seen in the phase image taken under 1V
reverse bias, Figure 5.17a, when compared with the phase image at 0V, Figure 5.17b. The
voltage step across the junction in the phase image is almost invisible in Figure 5.17c where
1V forward bias decreases the potential barrier.

Figure 5.17. Phase images of the specimen in Figure 5.15 under a) 1V reverse bias, b) 0V bias
and c) 1V forward bias. It can be seen that the contrast across the pn junction is increased with
the reverse bias and decreased with the forward bias.

The measured phase step across the pn junction as a function of reverse bias is plotted in
Figure 5.18. The phase step increases linearly with the voltage. The fitted line to the
experimental data in Figure 5.18 has the slope of 2. 6 rad and the intercept of 2.52rad:

∆ 2. 6 app 2. 2
The comparison of this equation with Equation 5.1 yield an active thickness of
-1
t 0±10nm and the built-in potential of 0.9±0.1 (CE 0.006 3 rad. .nm-1 ). The
crystalline thickness of this specimen was measured to be 500±10nm. The values measured
from this geometry are consistent with the ones obtained from the cube geometry.

137
 Figure 5.18. The phase step across the pn junction specimen in Figure 5.15 as
a function of reverse bias voltage. The phase step increases linearly with
reverse bias.

The electric field and charge density profiles shown in Figures 5.19a and 5.19b respectively
were extracted directly from the phase images at different reverse voltages. The same
behaviour observed in the cube geometry specimen, in Figures 5.5 and 5.8, is also seen in the
pn junction specimen prepared using this geometry.

The maximum of electric field at the junction and the depletion width increases with reverse
bias. The asymmetry seen in Figure 5.5 in the electric field profiles on the p-side of the
junction is also visible in Figure 5.19a, indicated by an arrow. The charge density obtained
from this specimen is also comparable to the one extracted from the cube geometry specimen.
The measured charge density is smaller than the theoretical calculation and increases with
reverse bias voltage.

For both electrical biasing geometries, similar discrepancies between the simulated data and
experimental measurements were observed. Discussions about possible origins of these
discrepancies are carried out in chapter 7 where the influence of electron irradiation on the
measurement of built-in potentials is investigated. Some of these discrepancies can be related
to external fringing fields above and below the specimen.

138
Figure 5.19. a) Electric field and b) charge density profiles of the pn junction
specimen. These profiles obtained from the new geometry specimen are
comparable with the ones from the cube geometry in Figures 5.5a and 5.8b.

139
5.3. Summary

In the first part of this chapter, a pn junction TEM specimen under a wide range of in-situ
electrical bias conditions has been studied. The junction properties measured from
holographic phase images were compared with simulated junctions. The two-dimensional
distribution of electrostatic potential, electric field and charge density across the junction
were extracted directly from the holographic phase images. The trends of these experimental
results were broadly consistent with simulations. In agreement with classical pn junction
theory, the potential barrier, depletion width and electric field across the junction increased
under reverse bias and decreased under forward bias. Also the measured built-in potential
was consistent with the theoretical values, if a 25nm crystalline inactive layer is considered
on both the top and bottom of the specimen.

However, the measured electric fields were about seven times smaller than the simulated
ones. The measured charge density was approximately order of magnitude smaller than the
expected value, and not consistent at different biases. The electric field was asymmetric and,
on the n-side, the measured charge density was larger than that on the p-side. These
discrepancies cannot be explained by the dead layer concept and require further studies.

The second part of the chapter discusses applying a different geometry for electrical biasing,
in which direct electrical contacts are made to the thin regions of the specimen using FIB
deposited metals. The electrostatic potential, electric field and charge density maps obtained
from this geometry were consistent with the results from the cube geometry.

While developing this geometry, it was found that FIB-deposited Pt is not a good candidate
for making electrical contacts to silicon. The phase step across the pn junction did not
respond to biasing voltages in the specimen that used FIB-deposited Pt for electrical contacts,
although the phase step increased irreversibly after applying a large voltage (10V). This
irreversible increment in the phase-step is very likely due to heating effects because
annealing an analogous specimen resulted in a similar change. Also both annealing and
biasing experiments resulted in the formation of droplets on the Pt surfaces, presumably a
Ga-Pt compound. It is speculated that because of Joule heating at the Pt-Si contacts, Pt and
Ga atoms form a compound that migrates to the surfaces, reducing the electrical conductivity
of FIB-deposited layers.

140
The phase step across the junction responded to biasing voltages when FIB-deposited W
instead of Pt was employed for the electrical contacts and wiring. To detect changes across
the junction under biasing conditions, it was not necessary to apply a large bias voltage and at
lower bias voltages no changes were observed in the W layer. The percentage of desired
metal in FIB-deposited W is significantly larger than that in FIB-deposited Pt (according to
(137)), which likely results in a smaller contact resistance between W and Si than between Pt
and Si.

141
Chapter 6

6. Effects of diffraction contrast on the phase profile

6.1. Introduction

The inner potential of a crystal V(r) is periodic and can be expanded in the form of a Fourier
series:

0 exp i2 . Eq. (6.1)

where g is the reciprocal lattice vector and r is a three-dimensional vector in reciprocal space.
The zero-order Fourier coefficient V0 is the mean inner potential (MIP) of the crystal (138).
The MIP of most materials lies between 5V and 30V and fluctuations around V0 are
described by the Fourier coefficients Vg (139). These Fourier coefficients are estimated to be
a few volts for low index g (138).

In many materials problems, precise quantitative information cannot be obtained from

electron holography without accurate knowledge of V0. The most accurate method currently
available for measuring the MIP is based on the quantification of phase changes in an
electron beam using off-axis electron holography. (46)

According to Equation 3.36, if an electron beam passes through a matter with inner potential
of V0 (instead of V(r)), it experiences the phase shift of

∆ CE 0t Eq. (6.2)
relative to an electron beam travelling in a vacuum. As explained in chapter 3, this phase shift
can be measured using off-axis electron holography. If the specimen thickness t is known, the
inner potential V0 can be extracted. The accuracy of this method is limited mainly by

142
experimental factors such as uncertainties about the specimen thickness, surface conditions
and dynamical diffraction.

The specimen thickness is linked to V0 in Equation 6.2 and any error due to the uncertainties
about the specimen thickness appears directly in the measured MIP. To reduce this error,
wedge samples or samples with known geometry and thickness such as spheres and
nanowires can be used. For example, the MIP of crystalline silicon has been measured using
electron holography from both cleaved-wedge specimen (V0 = 9.26±0.05V is reported in
(139) and V0 = 12.52±0.71V in (140)) and spherical particles (V0 = 12.1±1.3V) (141).
Although crystalline materials are cleaved along specific planes which can be used to
determine the wedge angle and specimen thickness, it is possible that the specimen has a
variation from the wedge angle at the very edge of the specimen. As discussed in (45), this is
likely to be the main reason for the above disagreement between the different measurements
of silicon mean inner potential. This example clearly emphasises the importance of
independent measure of specimen thickness.

Surface conditions and specimen charging are other factors which need to be evaluated and
taken into account when determining the MIP. Depending on the surface states, substantial
electrical dipole layers can be present on the surface of the specimen, which affect the
measurement of MIP (142). Specimen charging under the electron beam should also be
assessed as another source of error in measuring the MIP (45).

The linear relationship between the phase shift ∆ and specimen thickness t in Equation 6.2
is valid only if effects of the second term in Equation 6.1 can be neglected, the kinematical
condition. Periodic and drastic change in the electrostatic potential within a crystalline
specimen is responsible for dynamical diffraction of the electron beam. The electron beam is
diffracted multiple times, depending on the specimen thickness, by the three dimensional
periodic potential V(r), before coming out of the crystal. From the point of view of the
electron beam, the local variation in the electrostatic potential is maximised when the
specimen is oriented along a low order zone axis. A highly non-uniform potential distribution
is seen by the electron beam at this orientation. It is possible to make the distribution of
electrostatic potential more homogenous in the electron beam’s point of view by tilting the
sample away from the zone axis. Contributions of residual dynamical diffraction to the
transmitted beam phase can be calculated numerically (143). Bloch wave calculations (144)
by Gajdardziska-Josifovska et al. showed that dynamical contributions to the measured MIP

143
can be reduced to less than 5% by tilting the specimen away from zone-axis orientations and
major Kikuchi bands (139). Recent study carried out by Lubk et al., however, suggests that
the effects of dynamical diffraction have been underestimated in the calculation of MIP
(145).

The phase shift due to dynamical diffraction not only influences the measurement of MIP, but
also affects the built-in potential measurements.

6.1.1. Effects of diffraction contrast on the measurement of built-in potentials

From the applications point of view, there is more interest in the measurement of built-in
potentials across junctions rather than in the MIPs of single phase materials. Internal
electrostatic fields across various junctions such as homogenous pn junctions, III-V
semiconductor hetero-junctions and grain boundaries are responsible for the functionality of
almost all electronic devices. Both the MIP and built-in potential can be measured using off-
axis electron holography, but there are differences between them which should be considered
in their measurements.

Built-in potentials are usually about an order of magnitude smaller than the MIPs. For
instance, the MIP of silicon is approximately 12V while the built-in potential across a pn
junction is in the order of 1V. Accordingly the phase shift due to the built-in potential, for the
same specimen thickness, is smaller and more sensitive to any perturbation than the signal
due to the MIP.

The MIP is related to the charge distribution within the material on the atomic scale, while
the built-in potential normally is a change in the electrostatic potential in a larger scale. With
respect to the resolution of the microscope, the inner potential is a periodic and non-
homogeneous distribution of potentials while the built-in potential is generally homogeneous
and non-periodic. The diagram in Figure 6.1a illustrates the potential function of a single
crystal; the MIP is defined as an average of this electrostatic potential over the unit cell of the
crystal. Also, the electrostatic potential across a uniformly doped pn junction is shown in
Figure 6.1b for comparison. The width of the depletion layer even for a high dopant
concentration of 1018cm-3 is larger than 30nm which is equivalent to ~60 unit cells of silicon.
As mentioned earlier, by tilting the sample far enough away from the zone axes the potential
function of a single crystal appears more homogeneous in projection.

144
 Figure 6.1. a) Potential function of a one-dimensional single crystal. b)
Electrostatic potential through a uniformly doped pn junction.

Strength and homogeneity are two characteristic features which make the built-in potentials
distinct from the inner potentials. The MIP is usually measured from single phase materials
without any interfaces or complex structures in the sample. While, the built-in potential
measurement has to be carried out across interfaces buried generally within complex
structures of electronic devices. The fabrication process of electronic devices normally
involves growing various layers with different mechanical properties on top of each other.
The final device can be described as a complex distribution of different materials with
different mechanical and thermal properties. This inhomogeneity induces strains in different
layers, resulting in local changes of diffraction conditions in the TEM specimen prepared
from the device. The presence of strongly diffracting area when the homogeneous area of the
specimen is tilted to a weakly diffracting condition is problematic in characterising the built-
in potentials using electron holography. Local variation in diffraction conditions and small
built-in potential signal are factors highlighting the contribution of diffraction contrast on the
measurement of built-in potentials.

Many studies have been carried out over years to understand the effects of dynamical
diffraction on holographic phase images (143; 146; 147; 148). These studies address the
effects of dynamical diffraction on the measurement of MIPs. However, the effects of
diffraction contrast on the measurement of built-in potentials are less studied. “Tilting the
sample away from the zone axis to minimise the diffraction contrast” is the statement which
can be found in most of the built-in potential measurement studies (see for example (115;
38)). There is only one systematic study presented in literature regarding the effects of
diffraction contrast on the built-in potential measurement (149). In that study, Formanek et al.

145
rotated a MOS transistor viewed along the <110 direction about two systematic rows <001
and <110 . Along each row, four angles were suggested as favourable tilt angles. However,
Formanek et al. found more than 10% deviation between the built-in potentials measured
along these two different rows. (149)

One of the main concerns in the measurement of built-in potentials is how much local
variation in the diffraction conditions affects the potential measurement (150). Reviewing the
literature shows that local damping in the amplitude due to the diffraction contrast is visible
in the amplitude image in many cases even when the specimen is tilted away from the zone
axis to minimise dynamical diffraction. For example, Figure 6.2 shows the phase and
amplitude images of an AlGaN/AlN/GaN heterostructure. Diffraction contrast can be seen in
the amplitude image. The local variation in the diffraction condition, visible in the amplitude
image, has changed the phase of the electron beam. Another example is shown in Figure 6.3,
where despite careful tilting of the sample the diffraction contrast is visible close to the Si pn
junction.

Figure 6.2. a) Phase and b) amplitude Figure 6.3. Diffraction contrast appeared in the
images showing an AlGaN/AlN/GaN amplitude image despite carful tilting of the
heterostructure. The white arrow in the sample to a minimum dynamical diffraction
amplitude image points to the condition. The effect of diffraction contrast can be
diffraction contrast visible in the phase seen in the phase image close to the pn junction.
image. (151) The arrows are pointing to the diffraction contrast.
(149)

In the rest of this chapter experiments carried out to understand the effects of diffraction
contrast on the built-in potential measurement are described.

146
6.2. Experimental details

6.2.1. Unbiased pn junction

To investigate the influence of diffraction contrast on the potential measurement, a FIB
prepared pn junction from sample A (see chapter 4) was systematically tilted at different
angles relative to the electron beam direction. The crystalline thickness of the specimen was
measured using CBED to be 360±10nm. The phase and amplitude images reconstructed from
the holograms taken at each tilt were compared with each other.

a) Systematic tilt
In order to precisely tilting of the specimen, the position of the direct beam (000-beam) and
diffracted beams were recorded on the Kikuchi map for each individual hologram. The
description of the procedure used for achieving the precise specimen tilt is given in this
section.

Figure 6.4 shows the schematic diagram of the FIB prepared specimen used in this study. The
pn junction is grown on the (100) substrate. The cross-sectional TEM lamella is prepared
with the intention of orienting the surface of the lamella perpendicular to the <001 direction.
Therefore, the coordinate system shown in Figure 6.4a is used for measuring the specimen
tilt. To achieve any desired orientation, a double tilt holder is used. In loading the sample in
the holder, special care was taken to rotate the grid so that the tilting axes of the holder are
aligned to the <100 and <010 directions. This alignment allows tilting of the sample along
or perpendicular to the systematic row, without needing to use both tilt axes. Using only one
tilt axis at a time ensures that the sample is being tilted along a straight line. The <010
direction is aligned by the α-tilt axis of the microscope compustage and the <100 by the β-
tilt axis of the double-tilt holder. Therefore, keeping the β-tilt constant and changing the α-tilt
results in movement perpendicular to 040 band on the Kikuchi map. Moving parallel to 040
band is possible by changing the β-tilt.

In this experiment the sample is tilted along two paths: path A perpendicular to the 040 band
and path B parallel and at the centre of 040 band, as shown in Figure 6.4b. The width of the
040 Kikuchi band for Si is equivalent to 0.83o of α-tilt when a 300keV electron beam is used.
For the crystalline Si the lattice constant is a = 0.543nm and therefore the (040) planes are d =
0.136nm apart from each other. For a 300keV beam with the wavelength of λ 0.00197nm
the Bragg’s angle B is:

147
 o Eq.(6.3)
2d sin B nλ B 0. 1
which results in the Kikuchi band width of W040 2 B = 0.83o.

Figure 6.4. a) Schematic diagram of the FIB prepared specimen and coordinate system used for
addressing the specimen orientation. b) Simulated Kikuchi pattern of Si around the <001> zone axis.
Double arrows A and B point to the path along which the specimen was tilted.

For the path A, this narrow width needs to be sampled with fine steps. In this path, for each
point above the centre of the Kikuchi band there is a corresponding point below the centre of
the band. In terms of diffraction conditions, these two points are exactly the same. In other
words, tilting the sample ±α1 degrees results in the same diffraction conditions. To relate
these pair of points, a high precision in the determination of tilt angle is required. The read
out of tilt angles on the microscope consoles normally does not have high enough accuracy
due to the compustage backlash. Although, this accuracy can be improved by tilting the
sample only in one direction, either increasing or decreasing the angle, the accuracy will not
be better than 0.1o. The highest accuracy in determination of tilt angle can be achieved by
comparing the CBED pattern with the simulation.

In order to have a reasonably large field of view which is necessary for observation of
devices, electron holograms are taken in the Lorentz mode (see chapter 3). Despite giving a
large field of view in image mode, the Lorentz lens limits the field of view in diffraction
mode to 42mrad = 2.4o. Observing the diffraction pattern in Lorentz mode is similar to

148
looking at a zoomed-in map. Navigation using a zoomed-in map is only useful when the
place of interest is close. Normally, a zoomed-out map is used to get close to the region of
interest and then to find the exact place, the map is zoomed-in. That is the strategy used in
this experiment.

First, in the normal mode (Objective lens on and Lorentz lens off) the sample is tilted to the
<001 zone axes using the Kikuchi lines in the diffraction pattern. Then the microscope is
switched to Lorentz mode by switching the Twin-objective lens off and using the Lorentz
lens as an objective lens. Normally the incident angle of the beam is slightly different in these
two modes. Therefore, after switching to Lorentz mode the sample should be tilted slightly to
be oriented to the <001 zone axis. Now, as the specimen orientation relative to the electron
beam is known, the Kikuchi lines in the zoomed-in diffraction map in Lorentz mode can be
used to move along path A and B. For each tilt, before acquiring a hologram the CBED
pattern is recorded. By comparing the Kikuchi patterns with simulation, the orientation of the
sample is measured with better than ±0.01o accuracy.

As an example, it is shown in Figure 6.5 how the tilt angle can be determined using Lorentz
mode CBED pattern. The image shown in Figure 6.5a is a CBED pattern taken in Lorentz
mode. By choosing a narrow width of the intensity histogram, the contrast is stretched for
better observation of Kikuchi lines in Figure 6.5b. The white circle on the image represents
the maximum field of view in this mode (42mrad). The Kikuchi map shown in Figure 6.5c is
the simulated pattern using JEMS software. A montage of these images is shown in Figure
6.5d, which can be used for determining the angle. In path A when α 0 the direct beam
(000) sits at the middle of ±040 Kikuchi band and 220 reflection lies on 040 line. In Figure
6.5, by measuring the distance of reflected spots from 0 0 line, α is determined to be
0.25±0.01o. The CBED patterns of two points which are only 0.04±0.01o apart from each
other are shown in Figure 6.6. This resolution in measuring the tilt angle cannot be achieved
by relying on the compustage read out.

149
Figure 6.5. Demonstration of tilt angle determination in the Figure 6.6. Demonstration of
Lorentz mode. a) CBED pattern taken in the Lorentz mode b) high precision in determination
stretched contrast for better observation. The white circle shows of tilt angle in the Lorentz
the field of view which is 42mrad. c) Simulated Kikuchi pattern mode. Patterns shown in a) and
using JEMS software. The white arrow directs towards the b) are from two points which
<001> zone axis. d) Montage of a, b and c. are only 0.04±0.01o apart. Note
the displacement of 220
reflection relative to Kikuchi
lines.

b) Recording holograms
According to the specimen geometry (as shown in Figure 6.4) the pn junction is extended up
to the edge of the specimen. Therefore, to capture the junction in a hologram, not a wide
overlap region (hologram width) was required. However, normally the first ~50nm from the
specimen edge contains a highly damaged layer. A region at least 100nm away from the
specimen edge is desirable for this investigation. Accordingly the magnification and biprism
voltage were optimised to give ~0. 2μm2 field of view. With this setup, the whole width of
the p-type region, almost the entire n-type region and the edge of the protective tungsten layer
can be acquired in a single hologram. Also ~500nm width gives enough room to average the
phase profiles across the junction over a large width in order to reduce the noise. Setting up
the biprism voltage at Vb=120V, magnification at M = 16600 and exposure time at texp=16sec

150
results in the above mentioned field of view, sampling rate of 4 pixels per hologram fringe
with the fringing spacing of 3.5nm and reasonable signal to noise ratio. As changing the
illumination conditions such as beam spot size and brightness can potentially affect the
specimen charging, all the microscope conditions were kept unchanged throughout the
experiment. For taking the reference holograms the specimen was always brought to the same
position. Also at each tilt, at least 3 holograms were taken in order to diminish the noise by
averaging. One of the phase images reconstructed from the hologram taken under these
conditions is shown in Figure 6.7. Only the region inside the dashed lines contains phase
information (hologram width).

Figure 6.7. Representative phase

image reconstructed from a hologram
taken under the biprisim voltage of
120V, magnification of 16600 and
exposure time of 16sec.

c) Experimental results
In order to quantify the effect of diffraction contrast on the measurement of built-in potential,
the phase step across the pn junction was measured at each tilt point. Phase profiles were
extracted from the 170×1000nm2 box shown in Figure 6.7. The box was centred on the
junction and 200nm away from the specimen edge. Averaging over 25 pixels, equivalent to
170nm, reduces the noise level. Where necessary, the effect of specimen charging, which
shows itself as a slope on the p and n regions, was removed by subtracting a straight line
from the phase profile (Charging effects are discussed in chapter 7). Then Equation 5.2 was
fitted in each profile (See chapter 5) for measuring the phase step across the pn junction. A
representative curve fitting is shown in Figure 6.8.

151
 Figure 6.8. Demonstration of a curve fit to the experimental data
obtained from the phase image shown in Figure 6.7.

In the calculation of built-in potential, we followed the convention of assuming a uniform

electrically inactive layer on the p- and n- regions. As discussed in chapter 5, considering an
electrically inactive layer is not completely true. However, as in this experiment the built-in
potential values obtained from the same sample are being compared, the error in the
calculation of built-in potential due to this assumption should be consistent in all of our
calculations. A dead layer thickness of 2×30nm is used in this experiment. According to the
linear relationship between the phase step ∆ and active thickness ta, the built-in potential Vbi
is calculated using the following equation:

∆
bi
C E ta
∆
rad bi Eq.(6.4)
CE 0.006 3 1.96
nm

ta 360nm 60nm 300nm

The measured built-in potentials are summarised in Table 6.1 and Table 6.2. Data in Table
6.1 is associated with tilting the sample along the systematic row by keeping α 0 o and
changing β. This path is shown in Figure 6.4b as path B. Measured built-in potentials as a
function of tilt angle are also shown as a graph in Figure 6.9.

152
Table 6.1. Summary of measurements across path B shown in Figure 6.4.

Figure 6.9. Built-in potentials measured as a function of tilt angles

along path B.

At β 0o the specimen was exactly at <001> zone axis and the last point was 1.59 o away
from the zone axis and at the centre of ±040 Kikuchi band. As mentioned before, at each tilt
at least three holograms (at some points 9 holograms) were taken. After reconstructing the
phase and amplitude from each three holograms, the phase and amplitude images were
averaged to increase the signal to noise ratio. This averaging is helpful mainly for the
amplitude images as they are relatively noisy. The built-in potentials given in the tables are
calculated from the averaged phase images. For evaluating the effect of noise, potential was
also calculated from the phase images before averaging. The standard deviation was less than
±0.02V which is shown as error bars in Figures 6.9 and Figure 6.11. The third rows of the
tables represent how much amplitude drops as the beam passes through the sample. The
amplitude images were normalised by the amplitude in the vacuum region. Then, normalised
amplitude was averaged over the same region (refer to Figure 6.7) where phase step

153
measurement was carried out. As can be seen in Table 6.1 when the specimen is tilted close
to the zone axis the amplitude drops significantly due to strong dynamical diffraction. At the
zone axis the average of the amplitude is less than 6% of the beam amplitude in the vacuum
region. The phase images were noisy and error in measuring the built-in potential from
different holograms was larger than 0.02 at β0 and 1, so the built-in potentials regarding to
these points are excluded in the table.

For tilts larger than 4, measured built-in potentials were consistent and statistical evaluation
of these potentials in the form of (mean ± standard deviation) results in the built-in potential
of 0.79 ± 0.02V. This value is shown graphically in Figure 6.9. The amplitude images shown
in Figure 6.10 show no measurable diffraction contrast at any of these tilts (except barely at
4).

Figure 6.10. Amplitude images reconstructed from the holograms taken at tilt
angle β2 to β14. [Black = 0, White = 1]

The consistency in the measurement of built-in potential for tilts larger than β4 is in
agreement with the absence of diffraction contrast in the amplitude images. However, for the
tilts β2, β3 and β4 which resulted in significantly smaller potential measurement, the amplitude
images also show no diffraction contrast. If the consistency in the measurement for tilts larger
than β4 were only due to the absence of diffraction contrast, then the same potential should
have been measured at tilts β2, β3 and β4 because no diffraction contrast can also be seen at
these tilts.

Instead of relating the consistency of measurement to the absence of diffraction contrast, one
can explain it by the absence of dynamical diffraction, which also explains why the measured
potentials are not the same at tilts β2, β3 and β4. The tilts β2, β3 and β4 are closer to the zone
axis so the holograms at these tilts suffer from stronger dynamical diffraction in comparison
to the larger tilt angles. On the other hand, the tilt angles at β5, β6 etc. are still small and close

154
to the zone axis, so the same trend should be expected in the measurement of potentials as the
specimen was tilted further away from the zone axis. In other words, tilt angle β 4= 0.42o is
too small to be assumed as a threshold angle for suppressing the effects of dynamical
diffraction. From tilt β4 to β7 the averaged amplitude is increasing, confirming a large change
in dynamical diffraction, but the built-in potential does not follow the same trend.

Assuming that β4= 0.42o is the threshold for the dynamical diffraction effect, then in path A,
which is much further away from the zone axis, no variation (or a smaller variation) in the
potential measurements is expected. Also, if the behaviour observed in Figure 6.9 is only due
to dynamical diffraction, the measured potentials should vary symmetrically when the
specimen is tilted along path A. Note that path A is symmetric relative to the centre of 040
band. For each point above the centre of the Kikuchi band, there is a corresponding point
below the centre, which has the same dynamical diffraction conditions.

By holding β at 1.66o and tilting α with fine steps, 1 points were sampled along path A.
These points and measured potentials corresponding to them are shown in Table 6.2 and
Figure 6.11.

Comparing the graph in Figure 6.11 with the one in Figure 6.9 reveals a larger scattering in
the measured potentials along path A than B. Because the tilt range along path A
o
(β 1.660 , -0.3 α 0. 1o ) is smaller and further away from the zone axis than path B
α 0o , 0o β 1. 9o , it was expected to see less variation in the potential measurement in
path A. In addition, no symmetry can be seen in Figure 6.11. For instant, both α5 and α-5 are
equally as far away from the centre of the Kikuchi band, but there is a 0.14V difference
between the measured potentials at these tilts (0.84V and 0.7V respectively).

Table 6.2. Summary of measurements across path A shown in Figure 6.4

155
 Figure 6.11. Built-in potentials measured as a function of tilt angles
along path A.

These observations could be interpreted to mean that dynamical diffraction does not play a
significant role in the measurement of built-in potentials. However, the appearance of
diffraction contrast in the amplitude images at some tilts stops us from coming to a firm
conclusion. The tilts where diffraction contrast was observed are marked by * in Figure 6.11.
The amplitude images of all tilts are shown in Figure 6.12.

Figure 6.12. A montage of amplitude images reconstructed from holograms taken along path B.
[Black = 0, White = 1]

Diffraction contrast perturbs the phase image, therefore the large scattering and lack of
symmetry in Figure 6.12 might be due to the presence of diffraction contrast. On the other
hand, diffraction contrast is only present clearly at α2, α-2, α8 and α9. The mean ± standard
deviation of measured potentials even without these tilts is 0.8±0.05V. The standard

156
deviation, which shows how much variation there is from the average, for the data from this
path is two-and-a-half times larger than that calculated for data from path A (0.02V).

Now, the question is whether this large statistical dispersion is caused only by residual
diffraction contrast or there are other factors which have a larger effect on the phase. The best
approach to answer this question is to remove the effects of dynamical diffraction and
diffraction contrast completely to investigate whether consistent potential can be measured at
different tilts or not. The effects of dynamical diffraction and diffraction contrast on the
potential measurement can be eliminated by applying an electrical bias to the junction.

6.2.2. Biased pn junction

It is possible to keep diffraction conditions the same and change the phase step across a pn
junction by applying electrical biases to the junction. In other words, by applying different
voltages across the junction it is possible to remove the effect of diffraction contrast. This
approach is used in the following experiment to assess the effects of diffraction contrast and
dynamical diffraction on the built-in potential measurements.

a) Specimen specification
For the electrical biasing experiment, the TEM specimen was prepared from sample B. The
pn junction comprised a μm thick arsenic-doped n-type layer which had been grown
epitaxially on to a boron-doped p-type substrate using molecular beam epitaxy. In this sample
low resistivity of the substrate made it possible to make good electrical contacts to the p- and
n- regions using cube geometry (See chapter 4 for the details of sample and specimen
preparation). Also, the junction was at the depth of μm from the sample surface. Therefore
strains and damage resulting from the FIB deposition of the W protective layer would be far
away from the region of interest. In the previous experiment it was observed that many of the
diffraction contrasts started to form close to the interface between the FIB deposited W and
Si. This contrast at some tilt positions was extended to the junction. As in this sample the
junction is far away from the surface of the sample, it can be assumed that any diffraction
contrast which might be observed in this experiment would not be from the strain in the
interface between W and Si. Also, the thickness of the specimen was chosen to be relatively
thick to reduce the chance of specimen bending. The crystalline thickness of the specimen
was measured to be 570±10nm using convergent beam electron diffraction (CBED)
technique. The sample geometry and its orientation with respect to the electron beam are
shown in Figure 6.13. For tilting the sample and determining the tilt angle, the same

157
procedure that was explained for unbiased specimen was used. A limitation in this
experiment, relative to the previous one on an unbiased pn junction, is that the biasing
sample-holder is a single tilt holder. In preparing the sample using FIB and mounting the
sample in the holder special care was taken to make the tilt axis of the holder parallel to the
o
<010 direction. The sample was tilted along path C (β 6. ) perpendicular to the 040
Kikuchi band. This path is shown in Figure 6.13.

Figure 6.13. a) Schematic diagram of the biasing specimen and coordinate system used for addressing
the specimen orientation. b) Illustration of path C on a simulated Kikuchi pattern around the <001>
zone axis.

b) Recording holograms
The same microscope conditions which were used in the previous experiment were used in
this study. Furthermore, at each tilt position, the pn junction was electrically biased. The bias
voltage was swept from 0V to 2V reversed bias with the step size of 0.2V at each tilt
positions. For each voltage, three holograms (three object and three reference holograms)
were taken. Also, at each tilt by keeping the position of specimen unchanged and altering the
biasing voltage from 0V to 2V reverse bias with the step size of 0.5V, five more holograms
(only the object hologram) were taken. In analysing these holograms, the 0V hologram was
used as a reference hologram. Like in the previous experiment, at each tilt a diffraction
pattern was also recorded. The whole experiment was carried out in three days without taking
the specimen out of the microscope and without changing the condition of the microscope.

158
c) Experimental results
As an example, the reconstructed phase image at the tilt position of α1 = 0.52mrad and reverse
bias voltage of 1.2V is shown in Figure 6.14a. The corresponding phase profile extracted
from the 200×1200nm2 box shown on the phase image is plotted (black squares) in Figure
6.14b. As can be seen, the phase profile on the p- and n- regions shows different slopes.
Presence of slope in the phase image can be explained by the error in finding the centre of the
sideband during the reconstruction process (This artefact is explained in chapter 3). By
flattening the phase image it is possible to remove this slope. However, the difference
between the slopes of the n- and p- regions cannot be explained by any reconstruction
artefacts.

In order to quantify the phase step across the junction it is necessary to remove these slopes.
To remove the slopes on both n- and p- sides, first the phase image was flattened by fitting a
plane in the n-type region of the sample (the curve plotted in green triangles in Figure 6.14b).
Then, after extracting the phase profile, the slope on the p-type region was removed by fitting
a straight line in the p-region. The phase profile of the junction after removing the slopes is
plotted in blue circles. By fitting Equation 5.2 in the completely flattened profile, the phase
step was measured for each tilt position and each biasing voltage.

159
Figure 6.14. a) Demonstration of a phase image reconstructed from 1.2V reversed bias junction tilted
at α1 = 0.52mrad. b) Phase profile as extracted, after flattening on the n-region and after removing the
slope difference are plotted in black squares, green triangles and blue circles respectively. Fitted curve
using Equation 5.2 is plotted in red (solid line).

Results of the phase-step measurements as a function of tilt angle and biasing voltage are
shown in Figure 6.15. Similar to the unbiased experiment, a large scattering in the phase-step
measurements is seen for all the biasing voltages. Note that the trend of phase-step variation
as a function of tilt angle is repeated for all the biasing voltages. It can be deduced that
electrical biasing has had an insignificant effect on the scattering of measurements. Also,
Figure 6.15 shows that the pn junction has been biased properly throughout the experiment.

160
Figure 6.15. Step in phase measured across the junction as a function of tilt angle and biasing voltage.

From the amplitude images, it was clear that some of the large deviations in the measurement
are due to the diffraction contrast. Tilt positions highlighted by the blue boxes in Figure 6.15
are the tilts where diffraction contrast appeared in the amplitude images. At other positions
the contrast in the amplitude image was almost uniform. For example, the phase and
amplitude images corresponding to the tilt position α4 are shown in Figure 6.16. The phase
image is related to the reversed biasing voltage of 2V. Emergence of strong diffraction
contrast about 400nm away from the junction on the n-side resulted in a large drop of
approximately 2rad at the same position in the phase image. Although the drop in the
amplitude only appeared in a relatively narrow band, the reduction in the phase due to the
diffraction contrast did not recover within the field of view. In other words, the diffraction
contrast has appeared in the form of a valley in the amplitude image while in the phase image
it is in the form of a step. Despite significant effect on the phase image, the phase step across
the pn junction seems to be unaffected by the diffraction contrast at this tilt angle. For
measuring the phase step at this tilt position, the curve fitting was carried out within the range
of -250nm to 850nm.

161
Figure 6.16. Demonstration of diffraction contrast effect on the phase image. a) Phase image b)
amplitude image and c) phase profile extracted from the box shown in the phase image. Arrows are
pointing to the same position where diffraction contrast appears. Note 2rad phase change due to
diffraction contrast.

In contrast to the tilt angle α4, where the effect of diffraction contrast on the phase step is
insignificant, the diffraction contrasts at the tilt angles α-7 and α7 substantially influence the
phase step measurements. As shown in Figure 6.17, the phase difference between the p-side
and n-side of the pn junction at the tilt position exceeds 14rad which is more than 50%
larger than that at α4. Perhaps the distance of the diffraction contrast from the junction is
responsible for this disparity between the two cases. At the tilt positions α-7 and α7, the phase
steps were measured manually from the phase profiles.

It is important to mention that the reconstructed amplitude images were invariant for different
applying voltages. Consistency of the amplitude image for different biasing voltages
reassures that the diffraction conditions remains unaffected for different biasing voltages.
Because diffraction contrast affects the phase image, electron holographers tilt the specimen
until finding the orientation that contrast variation disappears in the amplitude image. At this
condition it is assumed that the effect of diffraction contrast is negligible. However, this study
shows that even when there is no measurable variation in the amplitude image, different
phase images are obtained at different angles.

162
Figure 6.17. a) Phase image b) amplitude image and c) phase profile extracted from the box shown in
the phase image from tilt position α-7. Arrows are pointing to the same position where diffraction
contrast appears.

The amplitude images at the tilt positions α0, α1, α2, α3, α5, α6, α8 and α-1, α-2, α-3, α-5, α-6, α-8
where amplitude images show almost no diffraction contrast are shown in Figure 6.18.
Although these amplitude images are very similar, there are minor differences between them.
For example faint variation in the contrast at tilt position α-6 and α-3 can be seen. Now the
question is whether these faint differences among the amplitude images are responsible for
relatively large changes in the phase image or there are other factors affecting the phase
images.

Figure 6.18. Amplitude images corresponding to the tilts where almost no diffraction contrast appears.
[Black = 0, White = 0.3]

The phase profiles at α-1, α0, α1 and α6 are shown in Figure 6.19 for comparison. Although
the amplitude images at these positions are very similar, comparing the phase profiles reveals

163
a relatively large variance. For example, the phase difference between α0 and α-1 exceeds
2rad within the field of view, while the angle between these two positions is less than 1mrad.

Relationship between diffraction contrast and the electron beam phase shift is not linear.
Also, although the amplitude images at α-1, α0, α1 and α6 are very similar, they are not exactly
the same. Therefore, it is tempting to ascribe the anomalous results to the residual diffraction
contrast. On the other hand, variation in the slope of the phase profile without major changes
in the shape could hint at the presence of different specimen charging.

Figure 6.19. Phase profile across the junction for tilt positions α-1,
α0, α1 and α6 which show no diffraction contrast in their amplitude
images.

To entirely exclude the effect of diffraction contrast, holograms were recorded at different
bias voltage without the vacuum reference. It should be emphasised that the microscope
conditions were kept unchanged throughout the experiment. Even the sample position was
not changed to exclude the chance of including any unknown factor. Only the power supply
was swept from 0V to 2V reverse bias with the step of 0.5V. Holograms taken at zero volts
were used as reference holograms for the reconstruction.

164
 Figure 6.20. Phase profile across the junction for tilt positions α-1, α0, α1 and
α6. The 0V hologram is used as a reference hologram for the reconstruction.

Phase profiles from the same set of tilt positions used in Figure 6.19 are chosen again in
Figure 6.20 for comparison. Phase steps in Figure 6.19 are larger than Figure 6.20 because in
the latter, zero volts bias holograms have been used as the reference holograms for the
reconstruction. Therefore, the phase steps in Figure 6.20 reveal only phase changes arising
from the 2V reverse bias. In other words, the difference between the phase profile in Figure
6.19 and 6.20 is due to the built-in potential and diffraction contrast effects. A very similar
behaviour is seen in both of the graphs. The slope difference between the p-side and n-side,
visible only on the p-side in Figure 6.19 and 6.20 because of flattening, changes in the same
order in both figures. This consistency in the behaviour for different sets of holograms, with
and without vacuum reference, confirms that the large variation in the phase profile is not due
to noise. Moreover, it verifies that diffraction contrast is not the only responsible factor for
the significant amount of scattering in the phase step measurements.

165
 Figure 6.21. Phase step as a function of tilt angle at the bias voltage of 2V. Holograms taken
from 0V bias specimen and vacuum are used as a reference in the reconstruction procedure for
red curve and blue curve respectively.

Phase steps measured using vacuum as reference and 0V hologram as reference, in the
reconstruction procedure, are plotted in Figure 6.21 for all tilt positions. The phase step
corresponds to 2V reverse bias. Scattering in the measurement is considerably smaller when
the 0 hologram is used to remove the diffraction contrast. Particularly at tilt positions α7 and
α-7 where the effect of diffraction contrast on the phase image were significant, removing the
diffraction contrast caused the phase step to fall within the phase step range of the other
positions.

166
The phase and amplitude image of position α-7 reconstructed using vacuum as reference and
the 0V hologram as reference are shown in Figure 6.22a and 6.22b respectively. The
diffraction contrast in the amplitude image, Figure 6.22a, has almost disappeared in Figure
6.22b where the 0V hologram has been used for the reconstruction, although the image is

Figure 6.22. Phase and amplitude images reconstructed using a) vacuum and
b) 0V hologram as a reference for removing the distortion. c) Phase profile
corresponding to the phase images shown in (a) and (b).

slightly noisy. Phase profiles extracted from the same region in Figure 6.22a and 6.22b are
shown in Figure 6.22c. After removing the phase variation due to the built-in potential and
diffraction contrast, similar behaviour in the phase profile is seen in comparison to other tilt
positions like α-6 where there is no diffraction contrast (compare the phase profile in Figure
6.22 with those in Figure 6.20). Also, for other biasing voltages (0.5V, 1V and 1.5V) similar
behaviour was observed in the phase profiles.

The above analysis reveals that besides diffraction contrast there are probably other factors
responsible for the large scattering in the measured phase steps. The slope-difference between
the p- and n- region is presumably due to these factors. As the phase images are flattended on

167
the n-side, the phase difference due to the slope appears on the p- region. The slope changes
by applying voltage Vapp and specimen tilt α. The angle α is the same angle defined for tilting
the sample, which shows how much sample has been tilted away from the centre of the 040
Kikuchi band. To investigate how the slope-difference ∆S changes as a function of α, it is
necessary to remove the effect of diffraction contrast. Diffraction contrast affects the shape of
the phase profile (refer to Figure 6.17). Therefore, to assess the relationship between ∆S and
α, the phase images reconstructed using the 0 holograms were used. After flattening the
phase image using the n-side, the slope on the p-side of the pn junction was measured by
fitting a straight line. An example of this line fitting in the p-region of the phase profile is
illustrated in Figure 6.23.

Figure 6.23. Demonstration of line fit to the p-side of the

phase profile.

The measured slope-difference between the n- and p- region ∆S is plotted as a function of tilt
angle for the 2V reverse bias in Figure 6.24. Also the corresponding phase step is plotted in
the same figure for comparison. Correlation between the two graphs confirms that in the
absence of diffraction contrast the factor causing the inconsistency in the phase step
measurement is the different slope on the p- and n-region. The slope difference is maximised
when the pn junction is exactly parallel to the beam. When the pn junction makes an angle
with the electron beam the slope difference decreases. The same behaviour was observed for
other reverse voltages.

168
 Figure 6.24. Phase steps across the junction due to the 2V reversed bias
measured using the 0V hologram as reference (shown in red square). Slope
difference between the p-side and n-side (shown in blue square).

The next step would be to find the relationship between the slope difference ∆S and biasing
voltage. To extract this relationship, the holograms taken at different biasing voltages (0V to
2V reversed bias with the step size of 0.2V) with the vacuum reference holograms were used
for reconstructing the phase images. In order to exclude the effect of diffraction contrast, tilt
positions α0, α1, α2, α3, α5, α6, α8 and α-1, α-2, α-3, α-5, α-6, α-8 which showed no major contrast
in their amplitude image were analysed (refer to Figure 6.18).

Plotting the slope difference ∆S as a function of biasing voltage app revealed that the slope
difference changes linearly with the biasing voltage. Figure 6.25 shows the slope difference
as a function of biasing voltage for α0. It was found that the slope-difference changes linearly
with the biasing voltage at the rate of 0. ±0.3rad.μm-1.V-1 for the selected tilt positions. The
large error of ±0.3rad.μm-1.V-1 is because this linear relationship did not show a constant rate
at all the tilt positions.

The graphs in Figure 6.24 reveal that the phase-step measurement is directly affected by the
slope difference . The difference in the slope of the phase, on the p- and n- region, varies
with tilt and biasing voltage. The large scattering in the phase-step measurement, observed
even in the absence of diffraction contrast during the tilting experiments, can be explained by
the variation of ∆S with tilting angle.

169
 Figure 6.25. Demonstration of the linear variation of slope difference
with biasing voltage
In contrast to the phase-step versus tilt angle plot (e.g. Figure 6.24) which shows a large
scattering in the data, in the phase-step versus applied voltage plot no notable deviation from
a straight line was found. Such a plot is depicted in Figure 6.26 for the tilt angle α0. As both
the phase step ∆ and slope difference ∆S change linearly with the biasing voltage, the effect
of ∆S variation cannot be seen in the phase-step versus applying voltage plot.

Figure 6.26. Demonstration of linear variation of the phase step with

biasing voltage.

170
6.2.3. Consideration of thickness variation
Thickness variation is one of the concerns in measuring the built-in potentials using electron
holography. The presence of thickness variations in the form of wedge-shapes across the pn
junction results in different slopes in the phase profile on the p- and n- regions. A cross-
section of a pn junction specimen with a wedge shape is shown in Figure 6.27, where the
vertical and horizontal axes are parallel to the electron beam direction and the object
coordinate x, respectively. Below this schematic, the electrostatic potential across the pn
junction is illustrated. The measured phase (x) varies across the pn junction both with the
thickness t(x) and electrostatic potential V(x)

x CE x t(x) Eq. (6.5)

Therefore, the phase slope on the n-side and p-side would be:

t
CE n 2 CE n tan( )
x n side x
t
CE p 2 CE p tan( )
x p side x
and the slope difference ∆S can be written as:

S 2 CE ( bi app ) tan( ) Eq. (6.6)

x n side x p side

The slope difference ∆S is plotted as a function of wedge angle for the built-in potential of
1V under 0V and 2V reverse bias in Figure 6.28. These graphs highlight the fact that
thickness variation due to a wedge shaped specimen cannot be removed effectively from the
phase profile by subtracting a straight line, because of the slope difference. For example the
o
phase difference across 300nm gate of a transistor which has a wedge angle of would
be more than 0.3rad which is equivalent to 0.15V in a 300nm thick sample. In very large
wedge angle specimens (θ>15o) the effect of slope difference on the symmetry of the phase
profile inside the depletion region also should be considered.

In the biasing experiment, according to Figure 6.24 the slope difference at 2V reverse bias is
on average 1.7rad.μm-1. This slope difference corresponds to the wedge angle of θ = 2.5o in
Figure 6.28. This large wedge angle is equivalent to 350nm thickness difference across the
μm length of the specimen. In order to investigate the thickness uniformity of the specimen,
the converged electron beam was scanned across the sample to the depth of 8μm, and at each
position a CBED pattern was acquired. The thickness of specimen in all positions was found

171
to be 570±10nm. Therefore, the slope difference in the phase image in this experiment
certainly is not due to the thickness variation. Moreover, thickness variation cannot explain
the variation of the slope difference with specimen tilting.

Figure 6.27. a) Cross-section of a wedge-shape pn

junction specimen with a wedge angle of . b)
Electrostatic potential across a pn junction.

Figure 6.28. Slope difference between the p- and n-side

of a pn junction in a wedge shaped specimen as a
function of wedge angle for a pn junction with 1V
built-in potential.

172
6.3. Summary

The influence of diffraction contrast on the measurement of phase step across a Si pn junction
using off-axis electron holography has been assessed carefully via employing the ability of
in-situ electrical biasing to alter the phase step without changing the diffraction contrast
effect. The phase profile across the junction has been measured at over 50 tilt angles
distributed in three straight paths. 15 tilts were sampled from the straight path starting at the
<001> zone axis and ending 1.59o farther in the middle of the 040 Kikuchi band (see Figure
6.4), 18 tilts were sampled from the path covered the entire 0.83 o width of the 040 Kikuchi
band and laid perpendicular to the band and 1.66o away from the <001> zone axis (see Figure
6.4), and 23 tilts were sampled from the path similar to the previous one but 6.85o away from
the zone axis (Figure 6.13). For the first two paths, the same specimen was tilted using a
standard double-tilt holder whereas a different specimen with different geometry which
allows in-situ electrical biasing was used for sampling of the last path.

At the first path, parallel to the Kikuchi band, where the junction was exactly edge-on at all
tilts the scattering in the measurements was not severe apart from the tilts very close to the
zone-axis. Consistency of the measurements in this path which experiences a large variation
in the dynamical diffraction conditions suggests that the effect of dynamical diffraction on
the built-in potential measurement is not significant. This implication was also supported by
the lack of symmetry in the potential measurements along the second path despite the fact
that for each tilt in this path there was a corresponding point which had the same dynamical
diffraction condition. On the other hand, appearance of diffraction contrast in the amplitude
images at some tilts made us suspicious of the possibility of residual dynamical diffraction
being the cause of inconsistent measurements at the tilts with the same dynamical diffraction
condition. Electrical biasing of the pn junction in the third path made it possible to remove
the effect of diffraction contrast from the phase images.

The biasing experiment confirms that the effect of diffraction contrast which appears in the
amplitude image can be significant in the measurement of built-in potentials. By removing
the effect of diffraction contrast the phase steps measured at the tilts with strong diffraction
contrast laid within the range of measurements from other tilts which showed no diffraction
contrast in their amplitude images. This verifies that electrical biasing has been successfully
employed in removing the diffraction contrast effect from the phase image.

173
Although removing the diffraction contrast reduced the scattering of measurements, this
reduction was only effective at the tilts where diffraction contrast was observed in the
amplitude image. This reveals that lack of consistency in the measurement of phase step at
different tilts is not solely due to the diffraction contrast. It was found that variation in the
phase step measurement is in agreement with the variation in the slope difference of the
phase profile on the p- and n- region. The slope difference varies with the tilt angle and
biasing voltage. The relation between the slope difference and biasing voltage was found to
be linear. This linear relationship explains why no notable deviation from a straight line has
been seen in this experiment and similar electrical biasing studies. It seems that the spread of
measurements at different tilts, observed in these experiments, is due to the different slope of
the phase profile on the p- and n- regions rather than residual diffraction contrast. It is
speculated that in measuring the built-in potentials across the heterojunctions, particularly
where the MIP on two sides of the junction is different, the effect of dynamical diffraction is
more severe.

In the next chapter where the effect of specimen charging is investigated, the origin of the
slope on the phase profile and possible reasons for having different slopes on the p- and n-
regions are discussed.

174
Chapter 7

7. Influence of electron radiation

7.1. Introduction

Electrostatic potentials, which are responsible for the functionality of electronic devices,
modulate the phase of the electron beam as it passes through the TEM specimen. Electron
holography is used to demodulate the electron beam and extract the electrostatic potential
signal. Unfortunately, besides being modulated by the potentials present in the specimen, the
electron beam may also alter the potentials. In other words, the distribution of electrostatic
potential within a specimen can be changed due to exposure to the electron beam. Emission
of secondary electrons and generation of electron-hole pairs as a result of exposing the
specimen to the electron beam are the two most important factors which can disturb the
charge balance within the specimen.

Emission of secondary electrons from an insulating region of a specimen leads to the

accumulation of positive charge. The positively charged region builds a potential hill. The
hillslope normally extends into the vacuum and covers the hologram field of view (152).
Appearance of hill-shape phase profile within the specimen and electrostatic fringing field in
the vacuum region are the two well-known symptoms of specimen charging (153). The latter
is more problematic because the presence of electrostatic field in the vacuum region violates
the assumption of having a field-free reference wave, while this assumption is responsible for
the direct interpretability of the phase image.

Leakage of electric field into the vacuum can be mitigated by covering the surface of the
specimen with a conductive layer as a shield. Carbon coating is used by many researchers for
providing such a shielding layer. For instance, in studying MOS transistors, which charged

175
under the electron beam, Dunin-Borkowski et al. showed that carbon coating is effective in
clearing the reference wave from the electric field (93). By comparing phase images before
and after carbon coating, McCartney et al. observed that p-type and n-type semiconductors
behave differently adjacent to a charged oxide (154) (153). They examined two different
silicon pn junction structures: one had an n-type/oxide interface, and the other had a p-
layer/oxide interface. Only in the latter, was the phase-step across the junction different
before and after carbon coating.

Reports on different behaviour of p-type and n-type semiconductors under electron beam
illumination are not limited only to the cases where they are close to an insulator. Cooper et
al. have found that even in the absence of a distinct oxide layer in the specimen, p-doped
layers and n-doped layers react differently under electron illumination (106). Figure 7.1
shows the results from (155) acquired from a GaAs pn junction. The phase profiles across the
junction were acquired using different beam intensities before and after carbon coating. Since
the slope difference between the n-doped and p-doped regions varies as a function of the
beam intensity (also known as spot size) (155) concluded that this behaviour is due to
specimen charging during electron beam illumination, despite the absence of any intentional
insulator in the highly doped and conductive GaAs specimen. Cooper et al. also disscussed
the asymmetry of the pn junction phase profiles (155; 106). It is still not clear whether or not
this asymmetry is due to the charging from the secondary electron emission.

Figure 7.1. Phase profile across an abrupt GaAs pn junction as a function of beam spot
size a) before and b) after carbon coating.( From (155))

Another discrepancy between theory and experiment which was explained by specimen
charging is related to the absence of external fringing fields in pn junction specimens. The
theory predicts that the pn junction electric field leaks from the specimen into vacuum. This

176
leakage should be visible in holographic phase images (156). Nevertheless, to the best of our
knowledge the external fringing field corresponding to the built-in potential of a pn junction
has never been observed in electron holograms. Even when an almost damage-free cleaved
specimen was examined by Harrison et al., a fringing field only emerged after reverse biasing
the junction (125).

By considering the influence of electron beam irradiation via introducing a positively charged
oxide layer on the surfaces of the specimen, Beleggia et al. explained the absence of fringing
fields in unbiased junctions and the emergence of these fringes under a reverse bias condition
(157). Figure 7.2 shows 2D simulations of electrostatic potential distribution and phase shift
with and without consideration of a surface charge layer, carried out by Beleggia et al. It can
be seen that the presence of a charged layer affects the electrostatic potential distribution and
results in equipotential surfaces. Also, the contribution of the electrostatic field around the
specimen to the phase shift varies depending on the charge density on the surface.

Figure 7.2. Simulation of potential distribution and beam phase shift with and without
considering a surface charge. Note the following changes after introducing a surface
charge density of 2.5×1013 e.c./cm2: 1) The contribution of external field to the phase
shift decreases from 90% to 30%. 2) The depletion width decreases by about 50%. 3) The
potential becomes constant on the surface at 0V. 4) The external field appears at 3V
reverse bias.(adapted from (157))

If the influence of the electron beam was only limited to the formation of a uniform surface
charge, then, according to the above simulation, a narrower depletion width should be
observed in phase images. However, a larger depletion region is usually measured from the
phase images as compared to standard abrupt pn junction theory. In addition, consideration of

177
a uniformly charged layer does not help explain the asymmetry of the phase profile observed
across symmetrically doped pn junctions.

Besides highlighting the artefacts caused by the electron beam, the above examples from the
literature and other similar studies emphasise three points:

a) The standard theory of doped semiconductors fails to describe the results obtained on
pn junctions illuminated with high energy electrons (158). Although modifying the
theoretical model by including the influence of the electron beam in the form of
charged surfaces has been helpful, this does not explain many features of the
experimental data.
b) The experimental methods used for mitigating the influence of the electron beam on
the potential measurement should be revised. Carbon coating, as the most common
method used, might prevent the leakage of electric field from the specimen to the
vacuum, but other subtle manifestations of specimen-beam interaction still can
adversely affect the potential measurement. (159; 160; 161)
c) The influence of the electron beam on the sample is not limited only to specimen
charging, which can be explained by the emission of secondary electrons from the
non-conductive parts of the sample and accumulation of immobile charges. It is also
necessary to consider the effect of secondary electron emission from the
semiconductor specimen and electron-hole pair generation within the internal electric
field of the device (40; 162).
In order to gain insight into the consequences of the electron-beam/specimen interaction for
the measurement of built-in potentials, a series of experiments were performed. The
experiments described in this chapter provide more information about the artefacts of electron
beam irradiation. The obtained information could be used to develop a more extensive
theoretical framework for pn junctions that can cope with specimens illuminated by an
electron beam.
The measurement of IV characteristics, electron-beam-induced-currents and SEM dopant
contrast as well as the phase images obtained using electron holography of the same
specimen make it possible to consider different aspects of electron beam irradiation.

178
7.2. Are FIB prepared surfaces equipotential?

As discussed earlier, in theory both specimen charging and internal electric fields can result
in the leakage of the electric field into the vacuum and the appearance of fringing fields in
phase images. However, in practice, the absence of external fringing fields in FIB-prepared
specimens, particularly in semiconductor samples, has led microscopists to conclude that
FIB-prepared surfaces are equipotential (125; 163). Consistently, external fringing fields
were not observed in many of the FIB prepared pn junction specimens examined in this
study. In Figure 7.3, a representative phase image and its four times amplified cosine are
shown, indicating that despite the presence of an electric field across the junction, no electric
field has leaked into the vacuum. So, it is assumed here that whole surface of the specimen or
at least the edge of the sample is equipotential shielding the electric field.

Figure 7.3. Representative a) phase image and b) four-times-amplified cosine of the phase
image showing the absence of fringing fields in the vacuum region.

However, surprisingly, we observed external electrostatic fringing fields in the specimens

prepared using the FIB-lift-out technique, in which part of the substrate was lifted out along
with the pn junction. Figure 7.4a shows a reconstructed phase image of a representative
specimen. The corresponding eight-times-amplified phase contours depicted in Figure 7.4b
clearly shows the presence of a fringing field in vacuum. Different numbers of contours on
the p- and n-region in the specimen, parallel to the junction, show the difference in phase-
slope between these two regions. These contours come together dramatically at the specimen

179
edge because the phase step across the junction is lower close to the specimen edge. This
matter is discussed further later in this chapter.

Figure 7.4 Representative a) unwrapped phase image and b) eight-times amplified cosine
of the phase image showing the appearance of a fringing field in the vacuum region.

The source of the fringing field in Figure 7.4 was not clear at this magnification. The
magnification was reduced and the whole 13μm width of the specimen was recorded in a
single hologram. This hologram and its corresponding reference hologram are shown in
Figure 7.5a and Figure 7.5b respectively. In the reconstructed phase image shown in
Figure7.5c, the 4μm-thick protective Pt layer deposited by FIB does not contain any
interpretable phase information as it is not electron transparent. Additionally, the bottom of
the specimen is highly damaged by FIB and is covered to a great extent with re-deposited
material during the lift-out process, thus, the phase in this region cannot be extracted. The
sample was composed of a 1μm-thick layer of 2×1018 cm-3 phosphorus-doped Si followed by
a 1μm-thick layer of 2×1018 cm-3 boron-doped on a lightly-doped p-type substrate, as marked
on the phase image.

180
 Figure 7.5. A representative low magnification hologram covering the whole width of the
specimen. a) The object hologram. b) The reference hologram. c) The reconstructed phase
image. d) Equi-phase lines spaced radians apart overlaid on the phase image. Phase

values in the protective Pt layer and in the highly FIB damaged layer are not valid.

As it can be seen, the vacuum region in the phase image is not flat. Equi-phase lines spaced
2
radians apart overlaid on the phase image in Figure 7.5d clearly show the external fringing
field. Also the hologram taken from another specimen with slightly larger width (1 μm),
shown in Figure 7.6, confirms the reproducibility of the results.

181
 Figure 7.6. a) Phase image and b) Equi-phase lines obtained under the same conditions as
Figure 7.5 from a slightly larger sample showing the reproducibility of the result.
Direction of the electric field in the vacuum is shown by the red arrow.

In these phase images, the larger the phase shift of the electron wave, the brighter the phase
image. Therefore, both phase images, Figure 7.5c and Figure 7.6a, show that the phase shift
in the vacuum is larger near the lightly doped substrate than near the highly n- and p-doped
regions. Since the phase shift is proportional to the electrostatic potential, the larger phase
shift in the vacuum corresponds to a larger potential. In other words, the equiphase lines
mapped in Figures 7.5 and 7.6 are interpretable as equipotential lines, particularly in the
vacuum, where there is less complexity due to the absence of mean inner potential and
thickness variation effects. According to this explanation, the direction of the electric field
(the potential gradient) outside the specimen is parallel to the specimen edge and from the
substrate to the Pt layer as indicated by the red arrow in Figure 7.6b. If the specimen was
charged up positively and uniformly, then the external electric field should be perpendicular
to the specimen edge. In Figures 7.5 and 7.6, it appears that only the substrate has charged up
positively under the electron beam and not the entire specimen, presumably because of the
high resistivity of the substrate. In these holograms the reference wave is perturbed by the
external fringing fields and this is likely why the phase contours in vacuum turn back towards
the specimen (58).
When the magnification was reduced and therefore, a larger area of the specimen was
illuminated, the specimen charged up differently. The hologram and reconstructed phase
image under this condition, from the same sample as in Figure 7.4 and Figure 7.5, are shown
in Figure 7.7. It seems that as the whole surface of the specimen was illuminated, the entire

182
specimen charged up positively. The gradient of the potential in this case is perpendicular to
the edge of the specimen. Differences in the specimen charging with magnification have been
also reported previously in (164). Frost et al. attributed these differences to the electrons
backscattered from the biprism (165).

Figure 7.7. Very low magnification hologram: a) object hologram b) reference hologram
c) reconstructed phase image d) Equi-phase lines. Note the perturbation, due to the
charging, in the holographic fringes occurring when the specimen is brought to the
hologram field of view in the object hologram; compare it with the reference hologram.
This perturbation is not seen in Figure 7.5. Decrease in the potential further away from
the specimen can be seen in the vacuum region in the phase image, as a hill-shape
potential.

183
The electric fields outside the specimen in Figures 7.6 and 7.7 can be compared intuitively
with the electric fields induced by two point charges. The electric field generated by two
equal positive charges, as shown in Figure 7.8a, is similar to the field in Figure 7.7d and the
electric field generated by two opposite charges, shown in Figure 7.8b, is similar to Figure
7.6b.

Figure 7.8. The electric field in vacuum from a) two equal positive point charges and b)
two opposite but equal point charges, present on the surface of the specimen. Compare
them with the electric field direction in Figure 7.7 and Figure 7.6.
After milling away the substrate and depositing an additional layer of W using the FIB, no
fringing fields were observed in the vacuum. Figure 7.9 shows the phase image and its
corresponding equiphase contour map after milling the substrate. As predicted the source of
specimen charging was the almost intrinsic Si substrate.

Figure 7.9. a) The phase image and b) the equi-phase contour map after removing the
substrate and depositing an additional W layer. Compare them with images in Figure 7.5
and 7.6.

184
7.2.1. Observation of external fringing fields in a FIB prepared pn junction
The previous experiments clearly revealed that FIB prepared specimens do not necessarily
have equipotential surfaces. Therefore, it should be possible to observe external fringing
fields arising from the internal electric fields of a pn junction in a FIB prepared specimen, if
the specimen is not strongly charging under the electron beam and if the internal field is large
enough to leak through the FIB-damaged surfaces.
Figure 7.10 shows reconstructed phase images obtained from a FIB-prepared specimen as a
function of reverse bias. The cube geometry explained in chapter 4 is used for electrical
biasing. The sample comprised a highly doped n-type layer with a resistivity of
0.00 ±0.000 -cm and grown on a highly doped p-type substrate with a resistivity of
0.01 ±0.00 -cm. The low resistivity of the substrate eliminates the problem of specimen
charging. In order to minimise the FIB damage, the bottom, top and edge surfaces of the
specimen were polished for a few minutes with 2kV Ga ions (See chapter 4 for more
information about the sample, geometry and specimen preparation).

Figure 7.10. Reconstructed phase images and their eight-times-amplified phase contours
as a function of reverse biasing voltage, showing the fringing field in the vacuum. The
number of fringes in the vacuum increases with the reverse-voltages.

185
In Figure 7.10b, eight-times amplified phase contours clearly show the leakage of the pn
junction electric field into the vacuum. By applying a larger reverse voltage the electric field
across the junction increases, correspondingly the number of fringes in the vacuum increases.
This clear response of fringing fields to the applied voltages verifies that the fringing fields in
these images are from the pn junction electric field, and not from the specimen charging.
However, according to the classical electrostatic theory, the maximum of the leaked field
along the edge of the specimen should occur at the position where the inner field is
maximised. In other words, the position where the density of phase contours is maximised
should match the position of the junction inside the specimen. Phase contours in Figure 7.10
show that the maximum of the external electric field has shifted towards the p-region.
However, the external fringing field has not been observed in previous works on FIB-
prepared pn junctions, so this shift in the fringing field cannot be compared directly with
other studies. In the study of chemically thinned pn junction carried out by Frabboni et al.,
the position of the external fringing field is also not matched the junction position (124).
Also, in a holographic study of cleaved Si pn junctions, Twitchett et al. could not reconstruct
the phase shift inside the specimen, due to the rapid variation in the thickness. Therefore it is
not clear whether the external fringes in that study are shifted or not (125). In the phase image
reconstructed from the hologram taken by Cooper from a cleaved reversed biased GaAs pn
junction, shown in Figure 7.11, it can be seen that the fringing field also has shifted towards
the p-region (118).

Figure 7.11. Phase image and its cosine acquired from

a cleaved GaAs pn junction. (118)

It is important to mention that the same shift in the external field was also observed when the
new biasing geometry (see chapter 4) was examined. This observation verifies that the shift
in the external field is not related to the geometry of the sample or the specimen holder.

186
7.2.2. Electric field and charge distribution
If the signal to noise ratio is good enough, the electric field and charge distribution can be
determined respectively from the first and second derivatives of the phase image divided by
-CE t (see chapter 2). The electric field distribution across a pn junction and in the vacuum
region is shown in Figure 7.12 for 2V reverse bias. The entire 550±10nm crystalline
thickness of the specimen measured by CBED is considered in the calculations. Note that in
the vacuum the same thickness of 550±10nm has been used for the conversion of the phase to
potential, which might have resulted in rather large underestimation of the electric field in
this region. Colouring in the vacuum and the specimen represent different values as shown in
the figure.

Figure 7.12. Calculated electric field from the first derivative of the phase image. The
colour codes inside the specimen and in the vacuum are different as shown on the right. In
the conversion of phase to potential in the vacuum region, 550nm crystalline thickness of
the specimen is used. The very edge of the specimen in this image is covered (by the gray
strip) because the phase wraps along this edge renders uninterpretable phase in this region.

A close inspection of Figure 7.12 provides information about the distribution of the electric
field within the specimen. First, the electric field drops dramatically close to the edge of the
specimen. From ~275kV.cm-1 at ~200nm away from the specimen edge, the electric field
decreases to ~35kV.cm-1 less than 50nm from the edge. Moreover, the maximum of the
electric field bends towards the p-region closer to the edge. Unfortunately, the steepness of
the phase gradient at the specimen edge prevents recovery of the phase very close to the edge.
This region is covered with a gray stripe in Figure 7.12. Consequently, continuity of the
electric field cannot be traced in this figure between the specimen and vacuum. However, the
direction of bending in the maximum of the electric field inside the specimen is in agreement
with the place where the electric field has leaked.

187
Another feature of Figure 7.12 is the asymmetry of the electric field across the junction,
despite the junction being symmetric and abrupt. The electric field gradient on the p-side is
smaller than the n-side electric field gradient. This behaviour can be seen more clearly in
Figure 7.13, where the calculated charge density from the second derivative of the phase
image is shown across the junction. The maximum charge density determined on the n-doped
side, 3.3×1017cm-3, is larger than the p-doped side, 2.2×1017cm-3. The depletion region is also
wider on the p-side than the n-side. As it is discussed later in this chapter, presumably the p-
and n-side of the junction are charged differently under the electron beam. This might be the
reason for the asymmetry of the electric field.

Figure 7.13. Calculated charge density from the second derivative of the phase image.
Note that the charge density declines close to the edge of specimen and reaches zero near
the edge.
The phase image in previous examples records the projected phase shift of the beam along
the beam direction. Figure 7.12 and 7.13 show the projection of the electric field and charge
density along the specimen thickness. Therefore, the rapid decline in the charge density
observed at the edge of the specimen should also be considered close to the top and bottom
surfaces of the sample. This can explain why the determined charge density (~1017cm-3) is

188
significantly smaller than the expected value (~1018cm-3) for this specimen. Any explanations
such as FIB damage and surface termination for the drop of charge density close to the
specimen edge are also applicable to the top and bottom surfaces. Not only should the
decreased charge density observed at the edge of the sample be assumed also to be present on
the top and bottom of the specimen, but the leaked electric field in the vacuum surrounding
the pn junction should also be considered above and below the sample. The shift of the
fringing fields to the p-region above and below the specimen explains the expansion of the
depletion region to the p-side and therefore the asymmetry of the junction.

189
7.3. The influence of beam current density on the phase profile

The current density of the electron beam affects the measured built-in potential. To
investigate the influence of current density, Cooper et al. previously measured the phase
profile across a pn junction for different beam intensities (155). They found that a lower
beam current density results in a larger phase step across the junction. Because of the
geometry of the sample, they related this to specimen charging. In their geometry, the H-bar
geometry, the p-type or n-type layer is electrically floating so it can charge up.
In order to investigate if charging of the specimen is the origin of this behaviour, several
holograms were taken using different current beam densities from a pn junction sample with
an additional W layer. The crystalline thickness of the specimen was measured by CBED to
be 320±10nm (see chapter 4 for specimen preparation details). In the reconstructed phase
image from the low magnification hologram the vacuum region was flat, so no external
fringing fields due to charging were observed.
The current density of the electron beam on the specimen depends on different parameters
such as lens excitations and the size of the apertures used in the illumination system. The
monochromator focus, like other lenses in the illumination system, affects the current density
of the beam on the specimen even though the monochromator is in unfiltered mode (166). To
change the beam current density, the excitations of the first and second condenser lenses
(known as spot size) were varied. However, to compare the phase profiles under different
beam intensities we did not rely on the spot size value. The current density on the sample was
measured by converting the number of electron counts on the CCD camera to the current
density. It is worth mentioning that after changing the spot size and before taking each
hologram, the orientation of the specimen was examined. The specimen was kept at the same
orientation throughout the experiment. When changing the spot size, the beam might slightly
tilt due to the changes in the value of lenses. This tilt was corrected by tilting the specimen
back to the same orientation, to keep the same diffraction condition throughout the
experiment.
Figure 7.14 shows the phase profiles measured across the junction as a function of beam
current density (J). The phase profiles were extracted after flattening the phase image on the
n-region. This figure shows that the absence of an external fringing field (as the most well-
known symptom of specimen charging) does not prevent the phase step across the junction
from decreasing when an electron beam of higher current density illuminates the specimen.

190
 Figure 7.14. Phase profiles measured across a pn junction for different beam current
densities.

The above experiment suggests that charging is at least not the main cause of phase step
reduction at high beam intensities. However, in assuming that the specimen is not charged up,
we did not confine ourselves to considering the absence of fringing field in the vacuum
region. After the above experiment, a ~30nm thick layer of carbon was evaporated on the
specimen. The same experiment was repeated. Figure 7.15 shows the phase profiles after
carbon coating as a function of beam current density. No major changes in the phase profile
were observed after carbon coating, which confirms the absence of severe charging in the
specimen.

191
 Figure 7.15. Phase profiles measured across the junction from the same sample in Figure
7.14, after coating with a layer of carbon.

In addition to carbon coating, the experiment was also repeated for different objective
aperture sizes. It is shown previously that charging can be reduced (or even sometimes
prevented) by using an objective aperture when acquiring high-resolution images (159).
Phase profiles obtained after inserting a 0μm objective aperture are depicted in Figure 7.16.
Again no significant changes relative to Figure 7.14 can be seen in the profiles. Repeating the
experiment for an aperture size of 10μm, 30μm and 100μm resulted in similar behaviour of
the phase profiles.
In summary, the above experiments reveal that the primary reason for the reduction of phase
step with increased beam intensity is unlikely to be the specimen charge.

192
 Figure 7.16. Phase profiles measured across the junction from the same sample in Figures
7.1 and 7.1 , before carbon coating and after inserting 0μm objective aperture.

7.3.1. Note on using an objective aperture in electron holography

Acquiring holograms in the presence of very small objective apertures, such as a 10μm-
diameter aperture, results in the reduction of the beam intensity recorded on the CCD camera
in the specimen region. To put it simply, the normalised amplitude images reconstructed from
the hologram taken with small objective apertures are darker in the specimen region than

those with larger objective apertures or no objective aperture. Therefore a map obtained

when a small objective aperture is inserted would show a thicker specimen. The small
objective aperture below the specimen cuts off some of the diffracted beams and therefore
decreases the intensity of the beam in the specimen region, which emerges as larger

thicknesses in the map.

193
7.4. Electron beam induced current (EBIC)

The measurement of electric field within the material using electron holography is necessarily
accompanied by a current passing along the electric field. Electron-hole pairs (EHPs)
generated by the incident beam are separated in response to the electric field, if they
happened to be generated within the field. The movement of generated holes in the direction
of the electric field and electrons in the opposite direction produces a current which flows in
the direction of the internal electric field. The electrical contacts made for applying voltages
to the junction made it possible to measure this current induced by the electron beam (EBIC).
A TEM specimen prepared using the cube geometry from the symmetric pn junction with a
dopant concentration of 6×1018cm-3 was used in this experiment for measuring the EBIC. The
crystalline thickness of the specimen was measured to be 550±10nm using CBED. See
chapter 4 for further details of materials and specimen preparation.

7.4.1. The measurement of EBIC as a function of beam position

The schematic diagram presented in Figure 7.17a illustrates the procedure applied in this
experiment. An electron beam with a diameter of 80nm (FWHM) and a current of 2nA
(measured from CCD) was positioned at different places across the junction as shown in
Figure 7.17a.
To ensure that the beam is away from the declined electric field region (shown in Figure
7.12), the beam was scanned 500nm away from the sample edge. The induced current
measured at each position is plotted as a function of beam distance from the junction in
Figure 7.17b. In this experiment, the instrument used for measuring the current (Model
2602A Keithley SMU) was connected such that positive current corresponded to the forward
current of the diode (from p-type to n-type).
Note that the induced current always flows from the n-side to the p-side, regardless of the
beam position. The excess carriers generated as a result of electron beam illumination diffuse
from where they were generated. The carriers that reach the space-charge region of the
junction drift across the depletion region. Due to the direction of the electric field within the
depletion region, electrons always drift towards the n-doped regions and holes towards the p-
doped regions. Therefore, the EBIC flows from the n-region to the p-region. The total
number of generated EHPs is expected to be the same in all cases examined here, since the
size of the electron beam was kept constant throughout the experiment and the specimen

194
 195
Figure 7.17. a) Schematic diagram showing the experiment procedure. A beam of 80nm in diameter is scanned across the junction and the induced current
is measured. b) The EBIC measured as a function of beam position. Inset graphs show the exponential function fitted to the measured currents.
thickness was measured to be uniform using CBED. Consequently, the magnitude of the
current at different positions represents the relative number of excess carriers diffused or
initially generated within the electric field of the junction. The current is maximised when the
beam is centred upon the space charge region, because in that case the maximum number of
EHPs can potentially contribute to the EBIC because they are separated by the field. The
current measured at this position, about 100nA, shows approximately 50 times amplification
of the 2nA electron beam used in this experiment.
As seen in Figure 7.17b, the current decays exponentially when the electron beam moves
away from the space charge region. The drop of current is steep such that fitting an
exponential function for the determination of the diffusion length (167) results in a diffusion
length of Ln=85±15nm and Lp=74±15nm for the electrons on the p-side and holes on the n-
side, respectively. These diffusion lengths are considerably shorter than that measured for
bulk silicon (a few microns) (168). However, by carrying out a similar measurement on a
mechanically prepared p-type silicon TEM specimen, Han et al. recently reported a diffusion
length of Ln=150nm (169). The short diffusion length is attributed to the high surface
recombination velocity in thin TEM specimens. The majority of generated excess carriers
recombine on the surface of the specimen rather than diffuse to the space charge region. It is
also known that ion milling increases surface recombination (170), which explains the shorter
diffusion length in the FIB prepared specimen compared to the 150nm diffusion length
reported by Han et al. for similar dopant concentration (169).

7.4.2. The measurement of EBIC as a function of biasing voltage

In addition to measuring the EBIC at zero bias voltage, the IV curve from the junction was
also recorded when the beam was on the junction (thus, the maximum of EBIC current) and
when the TEM column valves were closed (no illumination). These IV curves are shown in
the inset of Figure 7.18, while the graph in this figure shows the difference between these two
cases. To put it simply, this graph represents the electron beam induced current as a function
of biasing voltage. As can be seen, at 0V bias, the measured current is ~100nA in the reverse
direction. By reverse biasing the junction up to 0.5V, the current increased gradually with the
voltage, which can be attributed to the extension of the space charge region. However, the
variation of the space charge region with the biasing voltage cannot explain the behaviour of
the curve in the reverse bias voltages larger than 0.5V and in the forward biases.

196
 Figure 7.18. The measured EBIC as a function of biasing voltage for the electron beam
80nm in diameter centred at the space charge region. The inset shows the IV curves of the
junction with (red) and without (black) electron beam on the junction. The difference
between the two IV curves is the EBIC current.

By forward-biasing the junction, the space-charge width decreases and, thus, less induced
current is expected, which agrees with the trend of the curve at low forward voltages.
However, very rapid dying of the current and changing of its direction to the forward current
cannot be attributed to the changes of the junction electric field. Increasing the induced
current with the external voltage and in the same direction as voltage is applied, both in the
forward biasing and in the reverse biasing (when the reverse voltage is higher than 0.6V),
indicates that the external voltage not only changes the electric field across the junction, but
also induces electric field in other regions of the specimen. Figures 7.12 and 7.13 show that
close to the edge of the specimen the electric field and charge density significantly drop.
These regions, which are very likely present also on the top and bottom of the specimen, are
presumably the layers in which the external voltages induce electric fields.
The very simple schematic diagrams in Figure 7.19 illustrate the direction of the electric field
in these layers and across the pn junction under both forward and reverse bias. The black
arrows show the direction of the electric field. Under forward bias, the electric field in these

197
layers is in the opposite direction of the electric field of the junction. Therefore, generated
EHPs in these layers are separated in the inverse direction relative to those generated within
the space charge region. Therefore, the induced currents in these layers and in the space
charge region cancel each other rapidly by increasing the voltage. Narrowing the space
charge region under forward bias condition also expedites this process. Very quickly, the
current induced in the same direction as the bias voltage becomes dominant and increases
with the bias voltage.

Figure 7.19. Schematic diagram showing the direction of the electric field in the damaged
surface layers and the space-charge region in a FIB-prepared pn junction in a) forward
bias and b) reverse bias. Arrows show the electric field direction.

Under reverse bias, the electric fields in these two layers and in the space charge region are in
the same direction. Under small reverse bias voltages, the induced current increases slowly
with the voltage. This increase is due to expansion of the space charge region. By further
increasing the voltage, the current induced in the top and bottom layers becomes dominant,
likewise the forward bias, and increases with bias voltage.
Increasing the EBIC with the voltage under forward and reverse bias occurs with different
rate, which is probably related to changes in the thickness of the top and bottom layers with
applied voltage. These layers do not have a constant thickness, which explains the nonlinear
behaviour of the EBIC curve with applied voltage.

7.4.3. The measurement of EBIC as a function of beam size

As shown in the previous experiment, most of the induced current is coming from the part of
the electron beam which impinges upon the space charge region. Therefore, by decreasing the
size of the electron beam while monitoring the induced current, the width of the depletion
region can be estimated.

198
Figure 7.20 shows that by decreasing the beam size, the EBIC increases. When the beam
diameter reached about 60nm, further condensing of the beam did not change the induced
current. 60nm is an overestimation for the depletion width. With this method the width
measured is that in which all generated EHPs are collected rather than the depletion width.
This width can be called the “collection width”. ue to the short diffusion length, the
measured width can be considered as a rough estimate of the depletion width.

Figure 7.20. The measurement of EBIC as a function of e-beam diameter. The diameter
shown is the full width of the beam at half its maximum intensity (FWHM).

7.4.4. The flow of EBIC through an extraneous resistance can forward bias the
junction
The large induced current and high surface recombination velocity revealed in the above
experiments raise two important considerations for electron holography of FIB-prepared pn
junctions. First, the current induced by the beam in electron holography is not negligible. If
induced currents go through an extraneous resistance, the drop in voltage forward biases the
pn junction, reducing the measured phase step. In practice, this drop in voltage is likely when
the induced currents flow through the re-deposited materials on the surface of FIB-prepared
specimens. In unbiased specimens, in which the p- and n-doped regions are not grounded

199
properly, the induced currents pass through a higher resistance and reinforce the forward bias
the pn junction.
Second, the short diffusion length in FIB-prepared pn junction TEM specimens means that
only the part of the beam which is on the space charge region contributes to the EBIC. In
electron holography, an elliptical beam is applied to improve beam coherence (see chapter 3).
When the pn junction is perpendicular to the vacuum region, the elongated beam covers a
large part of the space charge and induces a larger current. As shown in Figure 7.21, the beam
might illuminate part of the specimen which is much thicker and induce a large current.

Figure 7.21. Schematic diagram showing an elliptical beam

illuminating a FIB prepared pn junction specimen. Note that in this
arrangement where the junction is perpendicular to the vacuum
region, almost half of the beam covers the space charge region.

Earlier in this chapter, it was shown that the phase step across the junction decreases with
increasing the electron beam current density. Assuming that the phase step reduction in
Figure 7.15 is due to the forward biasing of the pn junction, in the following the load required
to cause this forward-bias voltage is estimated.
The resistance Rd shown in Figure 7.22 represents the resistance due to FIB damage and
material re-deposition. The incident electron beam induces the current IR in the reverse bias
direction. Because of the direction of IR, the voltage VR=Rd IR forward biases the pn junction.
To calculate Rd, both VR and IR are required. By varying the electron beam current density it
is possible to change the EBIC IR; and changes in can be measured from the phase profiles
in Figure 7.15.

200
 Figure 7.22. Schematic diagram illustrating a
forward bias VR induced by an EBIC IR as it
passes through the resistance Rd.

During electron holography, only the part of the electron beam which impinges on the
collection width contributes to IR. This part of the electron beam is marked in Figure 7.22.
The current in this part of the electron beam, Ieffective, can be calculated by multiplying the
area of this region, A, by the e-beam current density Je.
Ieffective = A × Je Eq.(7.1)
Assuming the collection width of about 100nm, for an elliptical beam with a long axis of
20μm the area A would be ~1μm2. Only half of the beam passes through the specimen; the
other half passes through vacuum. (A 10μm 0.1μm 1μm2)
For each phase profile in Figure 7.15, Ieffective can be calculated (the e-beam current density
for each phase profile is given in Figure 7.15); and by taking the current amplification factor
of 50, the current would be:
IR = 50 × Ieffective 0 1μm2 × Je Eq.(7.2)
Changes in the phase steps, ∆ , can be converted to voltage variation, ∆VR, using the
equation:
∆ CE ∆ R ×t Eq.(7.3)
The arrow shown in Figure 7.15 illustrates the method used for measuring the phase step. In
measuring the phase step changes ∆ , the phase profile from the beam that had the lowest
current density was used as a reference. For the thickness t, the crystalline thickness was
used, 320±10nm as measured by CBED. The plot of VR as a function of induced current IR is
shown in Figure 7.23. The slope of this plot gives Rd 6M .

201
 Figure 7.23. An estimation of the possible resistance which could cause the phase
variations shown in Figure 7.15. Variations of the phase step across the junction with the
electron beam current density are converted to voltage changes ∆VR.

7.4.4. Assessing the role of objective aperture

The use of an objective aperture is generally known to be helpful in reducing the beam
induced charging seen in some specimens. However, there is no clear picture of how the
objective aperture influences specimen charging. It has been speculated that the emission of
secondary electrons from the aperture, generated by the scattered beam, causes this behaviour
(159). In order to address this speculation experiments were conducted to measure EBIC
before and after insertion of an objective aperture.
A beam of 500nm in diameter was centred on the junction so that the induced current was
maximised. The measured EBIC before inserting an aperture was 30nA in the reverse bias
direction. As illustrated in Figure 7.24 the cube geometry was used for the EBIC
measurement in this experiment.
After inserting the objective aperture, the EBIC dramatically increased. The measured current
was sensitive to the size and position of the aperture. For each aperture-size the current was
minimised when the aperture was centred with respect to the beam, as shown in Figure 7.25a.

202
The measured current was so sensitive to the position of the aperture that it could be used to
calibrate the centre of the aperture. By moving the aperture away from the centre, the current
increased and, when the beam was blocked by the aperture, the current was maximised.

Figure 7.24. a) Cube geometry used for detecting the secondary electrons SE. b) The
schematic diagram shows the emission of secondary electrons from the objective aperture
as hit by a diffracted beam.

Figure 7.25. a) When the objective aperture was centred relative to the beam, the EBIC
was minimised. b) Slightly moving the aperture away from the centre resulted in an
increase in the EBIC. c) When the beam was completely blocked by the aperture, the
EBIC was maximised.

The EBIC measured for each aperture size when the aperture was centred, the minimum
current, is plotted in Figure 7.26. This figure shows that the current increases with decreased
aperture diameter. For all of the objective apertures, when the electron beam was completely
blocked, the EBIC was increased to 850nA, which is indicated as the zero-size aperture in
Figure 7.26.

203
 Figure 7.26. The EBIC measurement as a function of objective aperture size. When the
beam was completely blocked by the aperture is indicated as the zero size.

As illustrated in Figure 7.24, when the objective aperture is inserted, the diffracted beam hits
the aperture, generating many secondary electrons. In the cube geometry, a large part of the
cleaved cube is not irradiated by the ion beam during the preparation, so its undamaged
surface is a good detector for secondary electrons. The current measured by the external
circuit is the sum of the electron beam induced current (EBIC) and secondary electrons
induced current. The former current is constant if the electron beam is not changed, thereby it
does not vary with objective aperture size. By decreasing the aperture size or moving it away
from the centre, more diffracted beams hit the aperture, generating more secondary electrons.
The emitted secondary electrons hit the cleaved surface of the sample, generating EHPs. Due
to the large minority carrier diffusion length in the undamaged part of the sample, the
majority of excess carriers diffuse to the space-charge region, so they are swept out,
producing the current.
A difference of about an order of magnitude between the current induced by the secondary
electrons and the current induced by the electron beam reveals that many secondary electrons

204
are produced in the vicinity of the specimen when part of the transmitted electron beam hits
the objective aperture. These secondary electrons are energetic enough to travel the few-
millimetres distance to the specimen surface and generate EHPs.
This experiment suggests that the number of secondary electrons generated when the
scattered or diffracted part of the electron beam hits the objective aperture is large enough to
neutralise the positive charges accumulated on the specimen. However, if the electric field
around the positive charges is not strong enough or if it is being shielded by a conductive
layer like a carbon film, then the secondary electrons would not be attracted by the field
generated by the charges. Thus, the positive charges will not be neutralised despite the
presence of sufficient electrons in the environment around the sample.
Also, the EHPs generated within the sample could be responsible for the charge reduction
when an objective aperture is inserted. In this case, EHP generation in the specimen could
increase the conductivity of the material, making easier for the charges to be dissipated. In
this case, the energy of the secondary electrons can play an important role. The secondary
electrons should have enough energy to generate sufficient EHPs to increase the conductivity
of the material to a level that allows the charges to be dissipated. In ref (159) and (171), these
possibilities are discussed.

205
7.5. SEM dopant contrast

As with EHP generation, secondary electron (SE) emission is also unavoidable when electron
holograms are taken from the specimen. From the specimen charging point of view,
secondary electrons (SEs) are undesirable. SEs are generally not used in TEM, thereby there
is normally no TEM SE detector and a direct measurement of secondary electrons is not
feasible in most TEMs. Comparing the secondary electron images taken using a scanning
electron microscope (SEM) with holographic phase images of the same FIB-prepared pn
junction specimen gives a clearer picture of the junction under electron illumination.

7.5.1. Mechanism of dopant contrast

SE images from pn junctions show that the intensity of SEs is higher on the p- doped side
than the n-doped side. This contrast in the image reveals that the effects of illuminating the
pn junction specimen by an electron beam are not the same on both sides of the junction. The
reason why the p-sides of the pn junctions appear brighter than the n-sides in the SE images
is still a matter of debate (172; 173; 174; 175; 176). Three main models have been proposed
in the literature to explain SEM dopant contrast.

a) Surface states
The presence of surface states and defects on the surface of the specimen results in surface
band bending. The band diagram bends differently close to the surface of the p-type and n-
type material. An example band diagram showing this effect is shown in Figure 7.27. Due to
these surface states, a depletion region is formed on the surface of the sample. On the n-side,
the direction of the electric field within the depleted surface is towards the vacuum while
on the p-side it is towards the material. Consequently, secondary electrons formed within the
depleted surface (region 1 in Figure 7.27) are accelerated on the p-side while they are
decelerated on the n-side. Also, according to these surface electric fields, the vacuum level
bends differently close to the surface on the n-side and p-side, as shown in the diagram.
Therefore, the accelerated secondary electrons on the p-side see a lower surface barrier than
the decelerated SEs on the n-side. Those secondary electrons formed in the deeper region of
the specimen (region 2 in Figure 7.27), regardless of the surface electric field, will see a
larger barrier on the n-side than p-side. Perovic et al. ascribed the measured SEM dopant
contrast to this difference between the band diagram energy on the n-type and p-type (177).

206
 Figure 7.27. The band diagram bends differently close to the
surface of the p-type and n-type material (adapted from (177)).

b) External fringing field

The direction of the electric field leaking to the vacuum from the junction, the external
fringing field (also called patch field) accelerates the escape of secondary electrons above the
p-region and decelerates it on the n-region. Sealy et al. attribute the dopant contrast to this
external fringing field (172).

Figure 7.28. The electrostatic potential distribution in the

vacuum and inside an abrupt symmetric pn junction with p-
and n-dopant concentrations of 1018cm-3. Arrows show the
component of the electric field which is perpendicular to the
specimen 50nm above the specimen surface. The surface state
density of ~1010cm-2 is considered. (adapted from (178))

Figure 7.28 shows a simulated electrostatic potential distribution in vacuum due to the
fringing field, carried out by Chee et al (178). The arrows in this figure represent the

207
amplitude and direction of the surface-normal component of the external electric field 50nm
above the sample. Above the p-doped region, the field is towards the specimen, assisting the
escape of secondary electrons, while, the upward direction of the electric field above the n-
region hampers SE emission.

c) Metal-semiconductor contact
The formation of a metal-semiconductor contact on the surface of the sample is described by
El-Gomati et al. as the cause of dopant contrast (176). Carbon contamination is polymerised
under electron illumination, forming a graphite layer. Contact between the graphite layer and
silicon (p-type and n-type) results in the energy band diagram shown in Figure 7.29. The
workfunction of graphite ( c
.1e ) is larger than of intrinsic silicon ( si
. e ).

Therefore, the barrier height of the graphite-silicon contact ( ( c - si

)) would be larger than
0.25eV for the n-type silicon and smaller than 0.25eV for the p-type (in the absence of
surface states), creating an ohmic contact on the p-side and a Schottky contact on the n-side.
The potential barrier due to the Schottky contact decreases SE emission from the n-type
region.

Figure 7.29. The energy band diagram of graphite-silicon contact for the p-type and
n-type. (adapted from (176))

The intention behind this concise review of these SEM dopant-contrast models is not to
assess their validity, but to discuss how they can be considered in the electron holography of
pn junctions. The presence of surface states, fringing fields and metal-semiconductor contacts
all refer to the specimen surface structure. Any effects this might have on the observation of a
bulk SEM sample would be intensified in a TEM specimen, because of the TEM specimen’s
top and bottom surfaces. Therefore, it is likely that the number of SEs escaping from the p-
type regions is greater than the number escaping from the n-type region when electron
208
holograms are taken from a pn junction. The charging problem in TEM, normally attributed
to the emission of SEs, should be more severe, if present, on the p-side of the specimen than
on the n-side.
However, dopant contrast in SEM disappears at high beam energies. Thus, in TEM it might
assume that there is no difference between SE emission from the p-side and n-side.
According to El-Gomati et al. the disappearance of the dopant contrast with increased beam
energy is due to “the dominance of the energetic backscattered electrons and their
respectively generated SEs” (176). In other words, the detected secondary electrons do not
originate where the primary beam hits. Therefore, at high energies the SE image is not
representative of the dopant distribution, although the probability that a SE could escape from
the p-region would be higher for all electron beam energies. Reducing the beam energy
confines the SEs to the illuminated area, so the SE image represents the dopant concentration
at these energies.
In addition to the beam energy many other experimental parameters such as the working
distance, beam current, scanning speed, vacuum condition, detector type, energy of secondary
electrons which are detected and the specimen tilt can affect the measured SEM dopant
contrast (179). These parameters are of secondary order and most of them are related to the
SEM detection system or hydrocarbon contamination (173; 178).

7.5.2. SEM setup for the observation of dopant contrast

Before carrying out the following experiments, the microscope conditions were optimised to
maximise dopant contrast. These conditions were systematically used, ruling out the effect of
operational parameters on dopant contrast. The conditions used for recording the SE images
in the following experiments are given in Table 7.1.

Table 7.1. Conditions used for observing the dopant contrast

209
7.5.3. Biasing experiments in SEM
This section discusses experiments comparing SE images and electron holographic phase
images of the same pn junction under electrical biasing conditions. For this purpose, a TEM
specimen in the cube geometry was prepared using the FIB. After preparing the specimen, the
biasing experiment was carried out first in SEM and then in TEM. After the experiments, the
crystalline thickness of the specimen was measured to be 550 ± 10nm using CBED.

a) The dopant contrast as a function of biasing voltage

The specimen geometry is shown in the low magnification SE image in Figure 7.30a. The
FIB-deposited Pt bar marked in Figure 7.30a is only for the protection of the specimen during
the milling and is irrelevant for the biasing. The dopant contrast was measured from the thin
region of the sample, the region which is electron transparent in TEM and is used to take the
hologram. This region is indicated in Figure 7.30a by a dashed rectangle.

Figure 7.30. a) A low magnification SE image showing the specimen geometry. b) SE

images showing the enhancement of the dopant contrast by applying reverse bias to the pn
junction.

As demonstrated in Figure 7.30b, the secondary electron images taken at 0V, 1V and 2V
reverse biases clearly show that the dopant contrast increases with the pn junction potential.
By increasing the applied voltage in the reverse bias, the electrostatic potential across the
junction increases, and as can be seen in Figure 7.30b the dopant contrast is also enhanced.
The graph in Figure 7.31 shows the measured contrast as a function of the reverse biasing
voltage. The biasing voltage was swept from 0V to 2V reverse bias with the intervals of
0.2V. The dopant contrast was measured using the following formula (adapted from (180)):
Ip In
Cpn Eq.(7.4)
Ip I0

210
Where Ip and In are the SE intensities from the p-side and n-side, respectively. I0 is the
intensity obtained from the vacuum region, which is the dark region in Figure 7.30a.

Figure 7.31. Dopant contrast across a FIB prepared pn junction as a function of reverse bias
voltage. The dopant contrast increases linearly with the reverse bias voltage.

As can be seen in this graph, the dopant contrast increases linearly with the reverse voltage
Vapp. If it is assumed that the dopant contrast is proportional to the potential across the
junction, the following formula can be written:
Cpn (0 ) bi Cpn (0)
Cpn ( app )
Cpn app app Cpn (0) Eq.(7.5)
bi app bi

where Cpn ( app ) is the contrast across the junction at the applied reverse voltage Vapp; and
Vbi is the built-in potential. For the graph in Figure 7.31, dividing the intercept of the line
fitted to the experimental data by its slope results in the built-in potential of Vbi 0.9 .
intercept
bi 0.9 Eq.(7.6)
slope

211
b) The phase step as a function of biasing voltage
Figure 7.32a shows a reconstructed phase image of the same specimen as in Figure 7.30
taken at 0V bias voltage. The area selected by the dashed rectangle in the phase image is
compared at 0V, 1V and 2V reverse bias. The line profile from the dotted line in Figure 7.32a
is used to measure the phase step across the junction. Due to the high signal to noise ratio,
averaging was not necessary.

Figure 7.32. a) A reconstructed phase image taken from the same specimen
shown in Figure 7.30. b) The phase images at 0V, 1V and 2V reverse bias,
extracted from the dashed box in (a), for the comparison.

Figure 7.33. The phase step across the pn junction shown in Figure
7.32 as a function of biasing voltage. The blue dashed line shows the
theoretical values.

212
The phase step across the junction is plotted as a function of reverse bias voltage; similar to
the experiment as carried out in SEM, the bias voltage was swept from 0V to 2V reverse bias
with an interval of 0.2V. As explained in chapter 5, from the slope and intercept of the line
fitted to the experimental data, shown in Figure 7.33, the junction built-in potential and the
specimen thickness would be:
intercept
bi 0.9
slope
slope 3
t 60nm
CE 0.006 3
For this sample, with the crystalline thickness of ~550nm and built-in potential of ~1V, the
expected line is plotted for comparison, as the blue dashed line in Figure 7.33.

213
7.5.4. In-situ metal deposition in SEM
When preparing TEM specimens using a dual-beam FIB/SEM, found that the dopant contrast
is enhanced when FIB-milled materials redeposit on the specimen surface. During final steps
of the specimen preparation, when a low-current ion beam is used to gently thin the specimen
down to electron-transparent thicknesses, a very thin layer is re-deposited on nearby surfaces,
which improves the dopant contrast. Figure 7.34 shows a secondary electron micrograph
taken during the final steps of the specimen preparation. The darker and thinner region close
to the edge has been thinned using a 93pA Ga ion beam. The re-deposition on the adjacent
region has improved the contrast across the junction from 7% to 10%, as measured from the
indicated boxes.

Figure 7.34. A representative SE image of a FIB-

prepared pn junction showing enhancement of dopant
contrast where the milled materials re-deposit.
The consistency of this observation motivated us to use the re-deposition process to passivate
the surface of the specimen immediately after FIB preparation. A chunk of Au/Pd (60:40)
was lifted out from a bulk sample and brought close to a pn junction specimen as a target, as
shown in Figure 7.35. By milling the surface of the target, a thin layer of Au/Pd/Ga, is
formed on the specimen (See chapter 4 for details).
Figure 7.35a and 7.35b show the pn junction specimen before and after the re-deposition.
These micrographs are SE images taken under the same conditions as given in Table 7.1.
Figure 7.35a and 7.35b show that the dopant contrast has significantly improved after
deposition. The intensity profile before and after the deposition is plotted in Figure 7.35c.
The dashed arrow in Figure 7.35a shows the direction of intensity profiles. The dopant
contrast before the deposition is ~14% while after the deposition the dopant contrast

214
increases by more than four times (~58%). Note that the contrast has increased mainly due to
an increase in the intensity on the p-side and there is only a small decrease on the n-side.

Figure 7.35. The dopant contrast in a FIB prepared pn junction a) before and b) after
depositing a layer of Au/Pd. c) The intensity profiles taken from the images in (a) and (b).

Also, the contrast measured from a freshly cleaved sample was considerably lower than that
from the Au/Pd deposited one. Figure 7.36a shows the secondary electron image taken from
the cleaved sample. The contrast measured along the dashed box marked on the SE image is
plotted in Figure 7.36b. As expected, the contrast in the cleaved sample (~20%) is larger than
the 14% contrast obtained from the FIB prepared surface in Figure 7.35a. As also suggested
by other groups (180; 181), this difference is probably due to Ga ion damage on the surface
of the sample.

215
 Figure 7.36. a) The secondary electron image taken from a freshly cleaved pn junction. b)
The intensity profile taken from the region marked by the dashed box in (a).

However, this contrast is lower than that measured from the Au-coated surface in Figure
7.35b. This difference highlights the importance of surface conditions. As explained by El-
Gomati et al., and discussed earlier in this chapter, the contrast enhancement after Au
deposition is likely related to the metal-semiconductor contact (176).

7.5.5. The holographic observation of a metal coated specimen

Figure 7.37 illustrates the electron holographic phase image obtained from the sample shown
in Figure 7.35 after being thinned and coated with Au/Pd. The hologram was taken at 120kV
and the thickness of the specimen was measured to be 530 ± 10nm using CBED.

Figure 7.37. a) A reconstructed phase image from the Au coated specimen. b) The phase
profile extracted from the box in (a) without flattening.

The phase profile shown in Figure 7.37b is directly extracted from the reconstructed phase
image without applying flattening or any other image processing. Although the profile covers

216
the entire field of view, there is no sign of any slope on either side of the junction. Both the n-
side and p-side phases are completely flat along the 1.6μm field of view, and the profile is
remarkably symmetric. The absolute value of the phase shown in the profile does not
represent the mean inner potential because of the unwrapping problem at the entrance of the
specimen from the vacuum. The phase step across the junction is 3.75rad which corresponds
to a built-in potential of Vbi 0. 2 if whole 30nm crystalline thickness of the specimen is
considered.

3.7 rad
∆ CE bi t bi rad 0. 2
0.00 6 30nm
nm

Flatness of the phase profile on both the p- and n-side in this metallic-layer-coated sample
suggests that leakage of electric field to the vacuum has presumably been responsible for the
slopes observed in phase images previously. The metallic layer on the surfaces of the
specimen prevents the internal electric field from leaking to the vacuum. Thus, not only is the
vacuum region near the edge of the specimen free of electric field, but also the vacuum
region above and below the specimen. This hypothesis that presence of electric fields outside
the specimen is mainly responsible for the slopes appearing in the phase profile is also
supported by the observations in previous chapter. Figure 6.25 shows that the slope
difference between the p- and n-region increases linearly under reverse bias voltages. A
larger reverse voltage means a larger electric field across the junction and consequently a
larger field in vacuum. Also Figure 7.12 shows that the fringing fields are not coincident with
the position of the junction. The maximum of the electric field in the vacuum is shifted
slightly towards the p-region. This is also consistent with the asymmetry of the phase profile.
The shift of the electric field to the p-side is speculated to be related to the larger emission of
secondary electrons from the p-side. A native oxide layer likely forms on the surface of the
specimen as exposed to the air before being loaded into the TEM. Due to the emission of
secondary electron, this layer charges up (which is the base of the simulation carried out by
Beleggia et al. (157)). Because the number of secondary electrons emitted from the p-type
region is larger than the n-type region, the n-side and p-side presumably charge differently
under electron beam irradiation. The difference in the amount of charge between the two
sides of the junction could also be the reason for the shift of the electric field in the vacuum.
As the surface of this specimen, covered with a thin layer of gold, is protected from

217
oxidation, the difference in the number of secondary electrons escaping from the p-side and
n-side does not cause the charging on the surface of the specimen.
However, the slope in the phase profiles is observed even in the phase images in which the
external fringing fields are not present. Considering this argument, the absence of the electric
field in the vacuum region close to the specimen edge does not necessarily mean that there
are no electric fields above and below the specimen. It should be emphasised that the leakage
of the pn junction electric field was only observed in the specimens which not only their top
and bottom surfaces, but also their edges have been polished with a low energy ion beam for
a few minutes (see chapter 4 for specimen preparation). The edge of the specimen is normally
exposed to the FIB for a longer time than the top and bottom surfaces, so the thickness of the
damaged layer at the specimen edge is expected to be thicker than that on top and bottom of
the sample. The electric field can leak through the thin damaged layer to the vacuum above
and below the sample while it is being blocked by the thick damaged layer at the edge.
Further inspection of the phase image shown in Figure 7.37 provides more useful
information. The interpretation of this phase image is less complicated as the contribution of
the specimen charging and external field to the phase image have been ruled out. The graph
in Figure 7.38 shows single-pixel-wide line profiles extracted from the phase image at

Figure 7.38. One pixel line profiles from the phase image in Figure 7.37a at different
distances from the specimen edge.

218
different distances from the edge. Except from the three lines closest to the edge which
suffered from phase unwrapping problems, the rest of the line profiles could be compared. In
this graph, only the line profiles taken up to the distance of 216nm from the edge are shown,
as the rest were very similar. From this graph it is clear that the phase step across the junction
decreases as approaching the specimen edge. Also, surprisingly this graph reveals that most
of the variation in the phase step occurs on the n-side. The phase variation on the n-side is
almost three times larger than on the p-side. We speculate that this is related to Ga
implantation during milling. In this sample, on the n-side the dopant atoms are arsenic
(atomic mass of 75) which is close to the Ga atomic mass (71). On the other hand the dopant
atoms on the p-side are boron with an atomic mass of 10. It is probably easier for the Ga ions
to cause damage to the crystalline structure of silicon which is highly doped with arsenic
rather than boron. Also Ga is a p-type dopant for silicon; therefore, implanted Ga on the n-
side can potentially change the silicon from n-type to p-type (182).
For each line profile from 60nm to 480nm into the specimen, the phase step was measured by
fitting to the equation 5.2. The measured phase steps as a function of distance from the edge
are shown in Figure 7.39. By extrapolating these data, it is predicted that, in the first 40nm
from the specimen edge, the phase step is zero.

Figure 7.39. Phase steps across the junction in Figure 7.37 as a function of
distance from the specimen edge.

219
Further from the specimen edge, the phase step starts to increase until approximately 250nm
away from the edge, after that point the phase step is almost invariant. In theory a constant
phase step is expected because the specimen thickness is constant. But the phase step drops
close to the edge, presumably due to the FIB-damage and surface termination. The conditions
at the edge are very similar to the top and bottom surfaces. Therefore, it can be assumed that
effects of the top and bottom surfaces are the same as the edge.
The schematic diagram in Figure 7.40 illustrates the behaviour of phase step along the
specimen thickness based on this assumption.

Figure 7.40. Predicted phase step along the

specimen thickness based on the measurement at
the specimen edge.

From the curve in Figure 7.39, it is not easy to calculate the contribution of different surfaces
(top, bottom and edge) to the phase step. Instead, the model in Figure 7.40 can be modified
into the following model in Figure 7.41.

Figure 7.41. Considering both surfaces of the

specimen on one side.
This model postulates that the potential varies as a function of thickness instead of having a
constant built-in potential all along the specimen thickness. The potential across the junction
at the thickness of t would be:
t bi f(t) Eq.(7.7)
where f(t) represents the fraction of built-in potential which has survived after specimen
preparation at thickness t. This fraction, f(t), can be calculated by normalising the graph in
Figure 7.39. Therefore, the phase step measured with electron holography of a FIB-prepared
pn junction is:

220
 t0 t0
∆ CE t dt CE bi f t dt Eq.(7.8)
0 0

As shown in Figure 7.42, the integral of f(t) for this specimen over the thickness of 530nm is
429nm, therefore:
∆ 3.7 rad
bi t 1.02
CE 0 0 f t dt rad
0.00 6 .nm 29nm

For an abrupt symmetric pn junction, this built-in potential corresponds to the dopant
concentration of 4×1018cm-3.

Figure 7.42. Fraction of built-in potential survived after

FIB specimen preparation at different thicknesses.

221
Chapter 8

8. Examining a MOS transistor

8.1. Introduction

Due to their complex architectures, transistors have not been good candidates for being used
as test structures in establishing a reliable characterisation technique for potential profiling.
Instead, the simple pn junction has been employed in the way of developing a quantifiable
method for measuring the potential distribution within semiconductor devices. However, to
prove that the concept is extendable to a real device, MOS transistor, as the most important
device used in integrated circuits, is an appropriate choice.

After 30 years from the first attempts (120) in detecting the electric field of the pn junction
using electron interferometry techniques, Rau et al. in 1999 demonstrated that off-axis
electron holography has incontrovertibly the capability to be employed in two-dimensional
mapping of the potential distribution within MOS transistors (86). Since then, a reasonable
amount of electron holography studies have been carried out on a wide range of MOS
transistors (82; 83; 98; 115; 183; 184; 185; 186). Reviewing these studies reveals that in
addition to the issues associated with the pn junction, which were addressed in the previous
chapters, the transistor structure itself raises more problems. Therefore, it is even more
challenging to quantify the results obtained from a transistor than a pn junction, and only few
quantitative data can be found in the literature (38; 84; 85; 116; 187).

Here, based on the lessons learnt from examining FIB prepared pn junctions, we attempt to
address some of the issues regarding the electron holography of FIB prepared MOS
transistors.

222
8.2. Sample specification

The samples examined in this study are 0.1 μm p- and n-type silicon metal-oxide-
semiconductor field-effect-transistors (MOSFETs) with the dopant concentration of
3×1018cm-3. The schematic diagram and micrographs shown in Figure 8.1 and Figure 8.2
illustrate the structure of the PMOSFET. The NMOSFET sample had the same structure.

As shown in Figure 8.1, transistors are buried beneath metallisation layers, which are covered
by an oxide film. Although only two metallisation layers are present in this sample, including
the tungsten plugs, and the 0.18μm node is not cutting-edge technology, the structure above
the devices is still complicated. Knowing all the details about the structure of the layers above
the device is not necessary for mapping the potential, but it can help us to prepare better
specimens and in some cases ease the interpretation of TEM images.

Figure 8.1. Schematic diagram showing the structure of the PMOSFETs examined in this
study.

In order to investigate the structure of the sample, two trenches, A and B, were dug using FIB
after protecting the surface of the sample with FIB-deposited Pt, as shown in Figure 8.2a. Via
these trenches the cross-section of the sample can be examined in two directions. SEM

223
images of the cross-sections exposed by these trenches are shown in Figure 8.2b and Figure
8.2c. Furthermore, images of the Al metallisation and tungsten plugs viewed from the top, by
lifting out the block covered with Pt in Figure 8.2a and removing the protective Pt and oxide
layers, are shown in Figures 8.2d and e. Some of the features marked in these images will be
used later in this chapter to explain TEM observations. Reasons for the formation of defective
features such as the centre-line seam, keyhole and void during the fabrication process can be
found in ref (188).

Figure 8.2. SEM micrographs showing the structure of the sample. a) Represents the area
where two trenches, A and B, are dug using FIB milling. b) Image taken from trench A. c)
image taken from trench B. d) Illustrates the Al metallisation lines after removing the
protective Pt and oxide layers. e) Shows the cross-section of the W plugs surrounded by
silicon oxide.

224
8.3. Removing the curtaining effect and providing vacuum path

Figures 8.3a and b show STEM annular dark field (ADF) images taken from a specimen
prepared by the conventional in-situ lift-out method. The intensity variation in the substrate
part of the specimen in both images clearly shows the curtaining problem. For a single
transistor in Figure 8.3b, the thickness variation beneath the gate is apparent. The intensity
profile from the box in Figure 8.3a is shown in Figure 8.3c. The projection of all the features
above the silicon substrate along the ion beam can be traced in this profile which is

Figure 8.3. Annular dark field images from the specimen prepared using conventional
lift-out method. a) Showing the structure of the specimen. Note the curtaining effect in
the bulk. The white arrow shows the direction of the Ga ion beam. b) Illustrates the
curtaining beneath the gate of a single transistor. c) Intensity profile taken from the
dashed box in (a). The minimum of the intensity corresponds to the area beneath the
voids. Decay of intensity from W1 to W is consistent with the shrinkage of W plugs.

proportional to the thickness of the specimen. Since the material beneath the voids is
removed more easily by the Ga ion, these regions appear darker in STEM-ADF images,
which means thinner regions. On the other hand beneath the Al and W layers, the specimen is

225
much thicker, since these layers protect the materials beneath them from Ga ions. Also a
gradual decrease in the intensity profile in Figure 8.3.c from right to left (which shows that
the thickness of the substrate below the tungsten plugs decreases in this direction) is
consistent with the decrease in the width of the W plugs from W1 to W5.

The reason the W plug width is decreased from W1 to W can be found in Figure 8.2e. It can
be deduced from this figure that instead of being located on the corners of a rectangle net, the
W plugs are placed on the corners of a parallelogram net. The position of the TEM slice we
made with respect to these plugs is illustrated schematically in Figure 8.4; the circles
represent the W plugs. This diagram explains why W plugs become narrower as moving to
the left of the Figure 8.3a.

Figure 8.4. Schematic

diagram showing the
place from where the
TEM slice in Figure 8.3 is
prepared with respect to
the W plugs. The circles
represent the W plugs.

Also from the curtaining effect in Figure 8.3, the direction of Ga ion beam can be determined.
The white dashed arrow in the figure shows the ion beam direction. From this image it can be
deduced that the beam has not been completely perpendicular to the specimen. Despite the
care taken during the lift-out and mounting process to ensure milling perpendicular to the top
surface of the sample a tiny angle (less than 1o) is unavoidable. Even this tiny angle shows
itself in the curtain formed in the substrate region. All in all, the message is that the thickness
variation is enormously sensitive to the structure that the ion beam sees along its path.

As mentioned also in chapter 4, in electron holography due to the dependency of phase shift
both on the electrostatic potential and specimen thickness, it is easy to misinterpret the effect
of specimen thickness variation as changes in the electrostatic potential. One simple approach
to alleviate the effect of thickness corrugation on the phase image, as proposed by Dunin-
Borkowski et al. (93), is to use the phase profile from the substrate part of the sample, where
no information about the doping distribution exists, to subtract the contribution of thickness
variation to phase in the region of interest. This approach is useful in enhancing the phase

226
contrast due to the dopant distribution, since it removes large part of the phase changes which
is due to the thickness non-uniformity. The phase profile in the specimen is a function of
built-in potential, mean inner potential and thickness:

∆ x CE . dopant min . t(x) Eq. (8.1)

Since mean inner potential Vmin is about an order of magnitude larger than dopant potential
Vdopant, by removing the contribution of mean inner potential CEVmint(x) from the phase
image, the error due to thickness variation CEVdopantt(x) in the phase becomes negligible.
However, in practice, due to the presence of noise and thickness variation along the direction
of the Ga ion beam this method is not desirable.

Figure 8.5. a) and b) showing the annular dark field images from a specimen prepared
using double lift-out method. No curtaining effect can be seen in the images.
Compare them with images in Figure 8.3. The intensity profiles along the direction A
and B are shown in c) and d) respectively.

227
A better approach would be to mill the specimen from its substrate side, in order to avoid
thickness corrugations. By using the double lift-out technique explained in chapter 4, we
managed to completely avoid the curtaining effect. Moreover, a uniform thickness along the
Ga ion direction was achieved by depositing an additional Pt layer on the backside (see
chapter 4 for details). Figure 8.5 shows an ADF image taken from the specimen prepared
with the double lift-out method. The intensity profiles taken along the transistors, direction A,
and along the ion beam, direction B, both show a uniform intensity which confirms the
thickness uniformity of the specimen. In order to provide a vacuum path for the reference
wave, the first protective Pt layer, oxide layer and Al metallisation were removed before the
second lift-out (see chapter 4). A low magnification bright-field (BF) TEM image of the
finished specimen which will be referred to later in this chapter is shown in Figure 8.6a. This
specimen is comprised of nine closely spaced transistors with the source and drain regions

Figure 8.6. Bright field TEM micrographs of the finished specimen. Diffraction contrast
due to the strain beneath the W plugs can be seen in (a). Note the continuous dark
contrast at the entrance of specimen in (b). This layer is a combination of FIB induced
damage layer and redeposited material. The doped source/drain regions are about 500nm
away from the edge of the specimen.

located about 500nm away from the vacuum region. The BF images taken from the first 6
transistors at slightly higher magnification are stitched together in Figure 8.6b. The
continuous dark layer seen in the entrance of the specimen from the vacuum, on top of the
228
oxide layer and W plugs is likely a combination of FIB damaged layer and redeposited
materials. Thickness of this layer in this sample is measured to be ~30nm. Note that this layer
is absent at the right edge of the specimen, since the specimen edge is being polished with
low current and low voltage beam at the final steps of preparation.

229
8.4. Diffraction contrast

Depositing different materials, such as silicide and tungsten, above the source/drain and
channel of the transistor (see Figure 8.1) induces stress in the underlying silicon. Locally
strained regions are formed as a result of this stress. Changes in strain, from the electron
beam point of view, mean variation in the lattice constant or, in other words, changes in the
diffraction conditions, which leads to the appearance of diffraction contrast in TEM images.

In the sample dealt with here, the W plugs induce a strong stress in the source and drain
regions, particularly where the contacts have been made. It can be seen in Figure 8.6a that by
moving towards the part of the specimen containing thicker W plugs (T6, T7, T8 and T9) the
diffraction contrast arising from the strain in the source and drain increases. The secondary
electron image in Figure 8.7 shows the contrasts arising from the strain beneath the W plugs.
This image is taken from the specimen in an intermediate stage of FIB preparation. The
specimen is milled from the backside, so the contrast due to curtaining does not interfere with
the strain contrast. The dark contrast beneath the W plugs is not the dopant contrast since it
only appears under the W plugs and when milling away the plugs, the dark contrast also
disappears.

Figure 8.7. SEM micrograph showing the contrast arising from the strain beneath the
W plugs. This contrast can be avoided by tilting the specimen.

As discussed extensively in chapter 6, electron holography is sensitive to diffraction contrast.

Minimising the diffraction contrast is vital in reducing the risk of misinterpretation of the
holographic phase images, particularly when extracting quantitative information is the aim.
Figure 8.8 shows the phase and amplitude images of T1, T4 and T7 taken exactly at the same
orientation (to bring the transistors in the field of view, the specimen was only moved in the
X and Y direction). For the transistor T4, the diffraction contrast can be clearly seen in the

230
source and drain regions in both the amplitude and phase images, while for T7 it is not easy to
see the presence of diffraction contrast by examining only the phase image. Also note that in
the same orientation no strong diffraction contrast can be seen in T1. This orientation can be
assumed to be a weakly diffracting condition for transistor T1, but not for T4 and T7. As the
strain orientation and strength varied from transistor to transistor, we had to tilt the sample for
each transistor to find the minimum diffracting conditions.

Figure 8.8. Phase and amplitude

images reconstructed from the
holograms taken from transistors
T1, T4 and T7 under the same
specimen orientation. The white
triangles in the images point to the
strong diffraction contrasts. For
transistor T4, the diffraction
contrast can be seen both in the
phase and amplitude images ((c)
and (d) respectively), while for T7
the diffraction contrast is more
distinctive in the amplitude image
(f). Note that in the same
orientation no strong diffraction
contrast can be seen in T1’s
amplitude image (b).

In some cases a very small change in specimen tilt made a significant difference in diffraction
contrast. For instance, the phase and amplitude images in Figure 8.9 are reconstructed from
the holograms taken from the transistor T3 with a tilt difference of less than 0.1o. The rather
strong diffraction contrast in Figure 8.9b has completely disappeared in Figure 8.9d by tilting
the sample less than 0.1o (measured from the diffraction pattern of the same area on the
substrate).
231
 Figure 8.9. Phase and
amplitude images from the
same transistor with less than
0.1o tilt difference. Note the
strong diffraction contrast in
the amplitude image (b) which
is disappeared by a very small
tilting of the sample.

From the above observations a few points can be highlighted:

1) Analysing the phase images obtained from two different devices with the aim of
comparing the electrostatic potential distribution within them can lead to misinterpretation, if
the corresponding amplitude images are not considered in this comparison. Before extracting
any information from the phase image, it is necessary to inspect the amplitude image to
ensure that the diffraction contrast has not perturbed the phase information.

2) The strain within the region of interest can be different from device to device, and
consequently the diffraction conditions. Thereby, suitable tilt conditions, which suppress the
diffraction contrast without significant smearing of interfaces, are different for different
samples. These conditions need to be found for each specimen individually.

3) It was found experimentally that starting from the zone axis and moving with very fine
steps along one of the systematic rows while monitoring the bright-field image taken under
the elliptical beam with the aim of minimising the diffraction contrast is a very useful method
for finding the suitable tilt conditions. Sometimes it is also helpful to slightly tilt the sample
perpendicular to the Kikuchi band; this tilt can be as small as 0.1o.

232
8.5. Charging in a MOS specimen

Reconstructed phase and amplitude images from transistor T5 are shown in Figure 8.10. The
absence of strong contrast in the amplitude image in the doped regions confirms that the
contribution of diffraction contrast to the phase image, if present, is not significant. From the
phase image, the darker p-type source and drain regions can be easily identified. Moreover,
by taking the first derivative of the phase image, which is proportional to the electric field,
the position of the junction can be delineated as illustrated in Figure 8.10c.

A phase step of ~1.7rad was measured across the junctions from the phase profile in Figure
8.10d, taken from the region marked by the dashed line in Figure 8.10b. Taking into account
the entire 400±10nm crystalline thickness of the sample, measured using CBED, this phase
step is equivalent to Vbi 0.6 (all the holograms presented in section . have been taken
with a beam energy of 300keV). Recalculation of the built-in potential by taking into account
the fraction of the potential lost due to the FIB preparation, from the curve in Figure 7.42
(chapter 7), gives Vbi 0. 7 , while the theoretical prediction value for this device is ~1 .

Figure 8.10. a) Amplitude and b) phase images of T5. c) First derivative of

the phase image (b), only in the substrate region, delineating the position of
the junction. d) Phase profile the dashed line in (b).

233
Among the consequences of electron irradiation on the measurement of built-in potential,
discussed in chapter 7, specimen charging in this sample is most problematic. The oxide
layers between the tungsten plugs are charged up under the electron illumination. The
difference between the measured potential 0.87V and its calculated value 1V is likely to be
mainly due to this charging.

8.5.1. Holographic observation of specimen charging

Figure 8.11a shows the montage of amplitude images from the first six transistors in this
specimen. The corresponding phase images are shown in the form of four-times amplified
phase contours in Figure8.11b. The fringing field due to the specimen charging can be seen in
the vacuum region. Moreover, the elliptical phase contours seen in the Si oxide regions
between the tungsten plugs are also associated with the charging of the oxide layer under the
electron beam illumination. The elliptical phase contours, first reported by Borkowski et al.
(93) as a charging effect, were also reproduced by simulation of the electrostatic potential
inside and outside a charged oxide layer. In this simulation Ubaldi et al. assumed a uniform
positive volume charge density within the oxide region (152). They showed that the position
of the ellipses is sensitive to the thickness variation of the specimen. They were also
concerned about the perturbation of the reference wave by the external fringing field (58).

In the specimen examined here, the leakage of the electric field to the vacuum region, where
the reference wave passes, is not as strong as the fringing field observed by Borkowski et al.
In preparing their specimen they thinned the sample from the oxide side; therefore, the edge
of the specimen was presumably free from any redeposited material. In our case, the
specimen is FIB milled from the backside and, as seen in Figure 8.6, a layer of redeposited
materials is present at the edge of the sample. The electric field lines are partially shielded by
this layer at the specimen edge. This is the reason why the fringing field is rather weak in the
reference wave path.

In the oxide regions in Figure 8.11b, note the elongation of the fringes towards the gate above
transistor T1 in comparison to other transistors. This elongation is associated with the absence
of W plugs on top of this transistor. The contours are elliptical above the spacer regions,
while they are dumbbell shaped above the transistors. The asymmetry of the contours above
the transistors is related to the position of the gates with respect to the W plugs.

234
 235
Figure 8.11. a) Montage of reconstructed amplitude images from the first six transistors in Figure 8.6. b) Shows four-times amplified phase contours
calculated from the phase images corresponding to the amplitude images in (a). The gates of the transistors are marked by red dots. Note the fringing
field in the vacuum region, elliptical fringes in the oxide layers between the W plugs and the aerofoil shape contours in the source/drain regions.
Another remarkable feature in Figure 8.11 is the aerofoil shape of the doped regions. Instead
of following the shape of the doped regions, shown in Figure 8.1, the contours become
narrower close to the oxide spacers. This aerofoil shape is also associated with the charging
of the oxide layers.

8.5.2. Observation of specimen charging in SEM

The features in the oxide layers in Figure 8.11b are also visible in the form of voltage contrast
in secondary electron (SE) images taken using SEM. Figure 8.12a shows an SE image taken
from the specimen in an intermediate stage of preparation. This image in the form of pseudo-
contour is shown in Figure 8.12b, where similar fringes to the phase image can clearly be
seen in the oxide regions. Moreover, the SE image in Figure 8.13 taken from the cross-
section of this TEM specimen before being thinned shows more symmetric fringes within the
oxide regions. In this view the oxide regions are more symmetric since there is neither a
transistor gate nor a spacer.

Figure 8.12. a) Secondary electron image and b) its corresponding

pseudo-contour showing the charging of the oxide layers between the
W plugs. Note that the elliptical contours in (b) are comparable with the
phase contours in Figure 8.11.

236
 Figure 8.13. As for Figure 8.12 but from the cross-section of the
specimen before being thinned. The contours within the oxide layers are
more symmetric than those in Figure 8.12. Arrows in (b) point to the
fringing fields leaked to the vacuum and the triangles in (a) show the
position of the W plugs.

The boundary condition of the oxide seems to have a direct effect on the shape of the fringes.
The leakage of the electric field to the vacuum due to the charging of the oxide can also be
seen in this SE image. On the right and left side of the image the fringes leaked into vacuum
are visible (marked by arrows). They are consistent with the position of the oxide layers and
similar to the holographic observations. The oxide layers are charged up positively in the
holographic observations, while the voltage contrast in the SE image shows that these layers
have been charged negatively at the accelerating voltage of 2keV.

The observation of elliptical fringes in SEM persuades us that the similar fringes visible in
holographic phase images could not be due to the perturbation of the reference wave. Also

237
this observation confirms that the features seen in the holography of oxide layers are certainly
due to charging.

The charging behaviour in the oxide layers is not very time-dependent. By changing the scan
rate, the features in Figure 8.12 and Figure 8.13 did not change. SE images represented here
have been taken at the long frame time of 4.7s, in order to increase the SNR. Integrating the
images with a shorter frame time also results in a similar image. Therefore, the time constant
with which the accumulated charges are discharged is not short. This is consistent with the
holographic results. The fact that the conditions are stable enough that we can take holograms
with the long acquisition time means that the charging behaviour of the oxide layers is time-
independent within the time-scale of our observation.

From the SEM observation, it can also be deduced that the assumption of a uniform volume
charge throughout the entire thickness of the specimen should not be necessary for
reproducing the elliptical fringes when simulating the electric field. The elliptical features
even can be seen in the SE images taken with the low beam energy of 2keV. At this energy,
the electron beam completely stops within less than 100nm below the surface of the oxide
(based on Monte Carlo electron trajectory simulation using CASINO). Therefore, charge
should be distributed within less than 100nm of the specimen surface.

Also, although the imaging conditions used in viewing the transistor sample were the same as
the conditions for the pn junction, with almost the same dopant concentration, the source and
drain junctions were not observed in SEM. It is speculated that the charging of the oxide
layers is responsible for the absence of dopant contrast in this sample.

8.5.3. Carbon coating

a) Thin layer of carbon (~20nm)

The phase images from the oxide region above the spacer between transistors T 2 and T3
before and after coating the specimen with a thin layer of carbon (approximately 20nm) are
compared in Figure 8.14. The four-times amplified phase contours before and after carbon
coating, shown respectively in Figure 8.14c and d, demonstrate the absence of fringing field
in the vacuum. However, still the elliptical fringes can be seen in the oxide layers.

238
 Figure 8.14. Phase images before and after coating the specimen with
~20nm layer of carbon. Note that the fringing field in the vacuum seen
in (c) has disappeared in (d) after carbon coating, but the elliptical
contours in the oxide layers are still present. e) Shows the line profiles
from the dashed line marked in a) and b).

The phase profiles taken across the oxide region between the two W plugs are shown in
Figure 8.14e. The thicker W plug is brighter in the phase image, which results in a larger
phase shift. Although the absolute values of the phase profiles in these images are not reliable

239
due to the phase wrapping problem at the entrance of the specimen, the phase difference
between the two W plugs is correct. This difference remains almost unchanged as the
specimen was coated with carbon. Assuming a mean inner potential of 23.4V for tungsten
(189) and 10V for silicon oxide (190), the difference between the thicknesses of two plugs is
found to be approximately 45nm.

The maximum of phase difference in the oxide region before and after carbon coating is
approximately 1.2rad. For the 400nm thick specimen, 1.2rad decrease in the phase of the
oxide after carbon coating means 0.46V drop in the potential of the oxide (with the mean
inner potential of 10V).

Another interesting region in this specimen is the oxide layer above transistor T1. Figure 8.15
shows the phase line-scans taken from path A and B, marked in the insert figure, before and
after carbon coating. Path A is in the oxide region above the gate in T1 and path B is from the
source to the drain. The V-shape phase profile in path A, which minimises exactly over the
gate of the transistor suggests that the electric field lines in this region are terminating at the
transistor gate which has a high conductivity. The W plug of T2 is to the left of T1, while the
right side of T1 is terminating at the edge of the specimen. This can be why the slope of the
phase profile is larger on the right side than on the left. To the left of the transistor there are
the W plug and the transistor gate, where the electric field lines can mainly terminate, while
at the right side most of the lines end at the gate. Assuming that the charge is distributed
throughout the thickness of the specimen, 7. rad.μm-1 slope on the right side of the transistor
corresponds to an electric field of approximately 3×106Vm-1, for a 400nm thick specimen.
After carbon coating the electric field decreased to ~5×105Vm-1. If based on the SEM
observation we assume that the charge is confined to ~100nm of the specimen, then the
electric fields would be four times larger (~1.2×107Vm-1 before and ~2×106Vm-1 after carbon
coating). The similar behaviour also can be seen on the source and drain of the transistor.
Reducing the electric field resulted in augmentation of phase step by about 0.25rad, as seen in
the phase profile of path B. This increment in the phase step is approximately 0.1V.

240
 Figure 8.15. Phase profiles from the paths A and B, in the insert image, before and after
carbon coating. Path A is above the transistor in the oxide region and path B is from the
source to the drain of the transistor.

b) Thick layer of carbon (~60nm)

After removing the previous carbon film using plasma cleaning, we coated the specimen with
about three times thicker (~60nm) carbon layer. The phase images after carbon coating
illustrated in Figure 8.16 show no elliptical phase contours in the oxide region. The contours
are now running parallel to the edge of the specimen and no fringing fields can be seen in the
vacuum. The reason why coating the specimen with a thin layer of carbon the first time could

241
not remove the elliptical fringes can be associated with the conductivity of the carbon layer,
since this conductivity increases with increasing the thickness of carbon.

With the disappearance of the charging manifestations in the oxide regions, the phase
contours in the doped source/drain regions follow the expected dopant distribution. The
aerofoil shape observed before carbon coating in Figure 8.11 is now absent in Figure 8.16.
However, defects and non-uniformity in the carbon film, as marked in Figure 8.16c, obscure
the phase information in many areas, particularly in the doped regions. These defects in the
carbon film can be seen more easily in an out of focus image.

Figure 8.16. Phase images after coating ~60nm layer of carbon.

Neither fringing fields in the vacuum nor the elliptical contours
in the oxide layers can be seen. The triangles in (c) point to the
defects in the carbon film. The arrow in (d) points to the slight
difference in the fringes due to curtaining. Note that the aerofoil
shape contours in the source/drain regions in Figure 8.11 have
vanished here after thick carbon film deposition.

In this specimen, the curtaining effect shows itself in the oxide region as the specimen was
milled from the substrate side. The slight difference in the phase contours above the spacer,
as pointed by the arrow in Figure 8.16d is due to the curtaining effect. The thickness of the

242
oxide where protected from the Ga ions by the spacer region (left of the arrow) is slightly
thicker than other regions. The maximum of thickness variation due to the curtaining effect in
the oxide was measured from the phase variation to be less than 3nm, so its effect in our
phase profiles in Figure 8.15 and 8.14 should be negligible. The phase profiles in Figure
8.14e are reproduced in Figure 8.17 for comparing them with the phase profile taken from the
same region after coating the sample with a thicker carbon film. It can be seen that the hill
shape between the two plugs has become flat after coating the specimen with ~60nm thick
carbon film. The apex of the profile between the two plugs is decreased approximately by
2.8rad. This phase difference is equivalent to approximately 1V in the oxide if the entire
400nm thickness of the specimen is considered in the calculation. The phase gradient of
1 rad.μm-1 close to the W plugs before carbon coating, which almost disappeared after
carbon coating, suggests the electric field of ~6×106Vm-1. Both calculated 1V potential drop
in the oxide and ~6×106Vm-1 electric field increase if, based on the SEM observations, the
charge distribution is only considered in a thin layer of the specimen.

Figure 8.17. Phase profiles between the W plug of T2 and T3 before any carbon coating
(red curve), after ~20nm carbon (blue curve) and after ~60nm carbon coating (black
curve). Note the steep increase of the phase profile close to the W plugs before carbon
coating which is flattened after thick layer of carbon.

243
Although after coating a thick layer of carbon the fringes in the vacuum and elliptical
contours in the oxide layers disappeared, the parallel fringes close to the edge of the specimen
in the oxide are still inconsistent with the specimen geometry. It is tempting to associate these
fringes with thickness variation. The phase starts to decrease close to the edge of the
specimen. Assuming that this phase reduction is due to thickness variation means that the
specimen is thinner close to the edge. For the specimens prepared using conventional
methods such as mechanical polishing followed by Ar ion milling and FIB milling from the
front side (oxide side) this assumption is reasonable. However, in this specimen, which has
been thinned using back-side FIB milling, if the specimen has a wedge shape due to the shape
of the ion beam, the specimen is expected to be thicker closer to the edge.

Figure 8.18. a) Phase image and b) thickness map calculated from the
amplitude image. c) Thickness profiles of the region marked by white dashed
box in (a) from the phase (blue curve) and from the amplitude (red curve).

244
Moreover, the thickness map calculated from the amplitude image does not show such a
thickness variation. The phase image of the oxide layer between T2 and T3, and the thickness
map calculated from the corresponding amplitude image are shown in Figures 8.18a and b
respectively. The thickness profile from the dashed box in Figure 8.18b and calculated
thickness from the phase profile of the same area are shown in Figure 8.18c. No thickness
reduction can be seen close to the specimen edge in the thickness profile extracted from the
amplitude image, while converting the phase variation to thickness shows about 60nm
decrease of thickness within less than 300nm from the specimen edge. Therefore, the phase
reduction close to the edge of the specimen cannot be due to the thickness variation.

Figure 8.19. Phase image and its corresponding four-time amplified contour
map before and after carbon coating. Arrows are pointing to the transistors.
Dark source/drain regions are distinctive even in this low magnification.

After coating the specimen with such a thick layer of carbon, which completely removed the
elliptical phase contours, a low magnification hologram was also taken to examine the
flatness of the vacuum region. Surprisingly, even after coating with a thick carbon film
fringing field was still present in the vacuum region. Figure 8.19 shows low magnification
phase images taken from the specimen before any carbon coating and after coating with

245
approximately 60nm carbon film. The phase image before carbon coating in Figure 8.19a
clearly shows that the specimen has charged up, particularly at its corner where there is no W
plug (see also Figure 8.6) to stop the leakage of the electric field to the vacuum. This
charging behaviour has been diminished after carbon coating as can be seen in Figure 8.19c.
However, fringing field due to specimen charging can still be seen in the vacuum. The phase
variation close to the specimen edge after carbon coating in Figure 8.16 and Figure 8.18 is
speculated to be associated with this residual charging.

246
8.6. A semi-biased NMOS transistor

The sample examined for the electrical biasing experiments in this section are NMOSFETs
with the same structure as that shown in Figure 8.1 for the PMOSFETs except that in the
implantation of the source and drain regions a lightly doped drain (LDD) structure (see
Chapter 2) has been used. This difference is shown in Figure 8.20, where the phase,
amplitude and the four times amplified phase contours are shown both for the PMOSFET and

Figure 8.20. The phase image, amplitude image and four-times amplified phase contours of PMOS
and NMOS transistors used in this study showing the difference between these two transistors in the
source/drain regions. The presence of LDD structure in the NMOS transistor is clear in its phase
image (a) and its amplified phase contour image (c).

NMOSFET. The amplitude images show that in both cases the specimens were tilted
appropriately away from the zone axis in order to suppress the diffraction contrast. The n-
type source and drain regions in Figure 8.20a are extended under the gate by the LDD
implantations. This can be seen more clearly in the amplified phase contour image in Figure
8.20c. Also, the absence of such a structure in the PMOSETs is clear in Figure 8.20d and
Figure 8.20f.

247
The electrical connections made to the NMOS specimen for allowing electrical biasing of the
transistors are shown in Figure 8.21a. The external voltages were applied between the n-type
drain of an NMOS transistor and its p-type substrate. A tungsten line was deposited, as
explained in chapters 4 and 5, for making an electrical contact to the p-type substrate and a
gold tip, mounted on the third movable electrode of the biasing holder (see chapter 4), was
used for making contact to the tungsten plugs of the transistors. Each time during the
holography measurements that the Au tip was re-connected to the tungsten plugs, the
reliability of the contact was tested by the IV curve measurement. Such an IV curve is
represented in Figure 8.21b.

Figure 8.21. a) Schematic diagram showing how an electrical bias was applied to the T1
transistor. b) A typical IV curve measured from the biasing configuration in (a).

Similar to the specimen shown in Figure 8.6, this sample also comprises several transistors
and the gold tip could be brought in contact with any of the tungsten plugs for electrical
measurements. However, the electron holography measurement under the electrical biasing
conditions could be performed only on the closest transistor to the specimen edge (T1). In the
absence of the Au tip, any of the transistors can be captured in a hologram by positioning the
biprism in parallel with the specimen length. The vacuum area above the specimen is used as
the reference wave path in this biprism orientation. But the presence of Au tip for the
electrical biasing purpose blocks this path. Therefore, only the very end transistor, T1 in
Figure 8.21a, can be captured in a hologram. The biprism needs to be positioned in parallel
with the specimen width and the field of view has to be maximised to capture as large an area
of the transistor as possible. Figure 8.22 shows the orientation of the biprism under electrical
biasing conditions in comparison to an unbiased situation.

248
 Figure 8.22. The biprism orientation which can be used for
capturing a) any of the transistors in the specimen in the
absence of Au tip and b) the electrically biased transistor at
the specimen corner.

In the preparation of this specimen for the electrical biasing experiment, extra care was taken
to situate transistor T1 at the very edge of the sample without causing damage to it during the
preparation. It should be recalled that the specimen has been FIB milled from the backside to
avoid curtaining problem. The electrical biasing holder brought another limitation, as it is a
single tilt holder. In mounting the specimen special care was required to minimise any
misorientation. The phase, amplitude and amplified phase contour images of transistor T1,
shown in Figure 8.23, verify that all these precautions have been taken successfully during

Figure 8.23. a) Phase image, b) amplitude image and c) four-times amplified phase contours of
transistor T1 situated at the corner of the specimen under unbiased conditions.

the preparation of this sample. The n-type source and drain regions and the LDD structure
can be seen clearly in the phase image. The amplitude image shows that the diffraction
contrast is also suppressed effectively.

249
Holographic phase images obtained from transistor T1 under biasing conditions are shown in
Figure 8.24. The contrast in all of these images is scaled between 0rad (the darkest) and 30rad
(the brightest), as shown in the figure. The pn junction between the n-type drain under the last
tungsten plug (W1) and the p-type substrate was electrically biased, both forward and reverse.
It is clear in this figure that the transistor has responded to the external biasing voltages. By
increasing the reverse voltage from 0V to 4V the contrast between the n-type region and the
p-type substrate increases, which means larger phase step and therefore larger potential
barrier across the junction. Also the contrast in Figure 8.24a decreases in response to 1V
forward bias, as it lowers the potential barrier.

Figure 8.24. Phase images acquired from the transistor T1 under different biasing conditions. The pn
junction between the n+ drain region, under the W plug at the corner of the specimen, and the p-type
substrate was electrically biased, from 1V forward bias in (a) to 4V reverse bias in (f).

The two other terminals of the T1 transistor, the polysilicon gate (G) and the other tungsten
plug (W2), are electrically floated in this biasing configuration and there is a chance that they
gain a voltage during the experiment. The phase profiles taken from the oxide layer above the
gate along the path A, marked by an arrow in Figure 8.24f, are shown in Figure 8.25a as a
function of the biasing voltage. The phase profile slope, plotted in Figure 8.25b, increases
almost linearly with the biasing voltage. This means that the induced voltage on the
electrically floated tungsten plug (W2) has not been significant. Even after ion beam damage,
during the specimen preparation, the oxide layer isolates the two tungsten plugs appropriately

250
under biasing conditions. The effects of ion beam damage and material redeposition on the
oxide layer has not been destructive enough to electrically shortcut the two tungsten plugs in
this sample. This is another strong piece of evidence that FIB prepared surfaces are not
necessarily equipotential (see section 8.5 and chapter 7). If the unbiased plug (W2) is assumed
to be at zero volts when 4V is applied to the other plug (W1), the electric field between these
two tungsten plugs, inside the oxide layer, would be approximately 1.14×105 Vcm-1 since the
distance between them is approximately 350nm. This electric field can be measured from the
phase profile in Figure 8.25. The crystalline thickness of the specimen was measured using
CBED to be 470±10nm and holograms were taken at 120keV (CE = 0.00856 radV-1nm-1).
Based on that, the phase difference of about 17rad between the 0V and 4V phase profiles in
Figure 8.25a would be equivalent to approximately 1.2×105 Vcm-1. Although these two
electric field values are very close, the agreement should be regarded with caution. The
crystalline thickness was used for this calculation, while the oxide thickness might be slightly
thicker. Also, the contribution of external fringing fields to the measured phase was not
considered here.

Figure 8.25. a) Shows the phase profiles taken from the oxide layer (from path A in Figure 8.24f) as a
function of biasing voltages. b) Demonstrates that the slope of the phase profiles in the oxide layer
increases almost linearly with the biasing voltages.

In contrast to the potential induced on W2, which is assumed to be close to zero at different
external voltages, the potential at the polysilicon gate seems to be non-zero and is probably
changed by the applied biasing voltage. Variations in the phase images due to the electrical
bias can be seen more clearly in the two-times amplified phase contours in Figure 8.26. The
number of phase contours in the oxide layer increases with voltage, which demonstrates the
augmentation of the electric field in the oxide layer. If a zero volt bias was applied to the

251
polysilicon gate, the electric field in the oxide between the biased plug (W1) and the gate (G)
would have been larger than that between the gate (G) and the unbiased plug (W2). The
electric field lines would have been terminated on the gate if it were at zero volts. But, the
number of phase contours in the oxide layer between W1 and G is almost the same as that
between W2 and G (see for example Figure 8.26f). Also in the phase profiles in Figure 8.25,
no clear changes in the phase slope at the gate position can be seen. Therefore, it can be
concluded that the gate voltage is approximately biased at voltages close to half of the
external biasing voltages. For example when 4V is applied to W 1, the gate voltage is
approximately 2V and W2 at ~0V.

Figure 8.26. Showing two-times amplified phase contours obtained from the phase images in Figure
8.24.
Another notable observation in Figure 8.26 is the extension of the drain depletion region into
the source at reverse voltages of larger than 2V. In Figure 8.26d it can be seen that under the
gate in the channel region, the phase contours relating to the drain depletion region touches
those relating to the source depletion region. The phase profiles in Figure 8.27 extracted from
path B (marked in Figure 8.24f) show this phenomenon more clearly. It can be seen that at
3V and 4V reverse biases the extension of the drain depletion region strongly affects the
source depletion region. Although it is very tempting to associate this observation solely with

252
the punchthrough phenomenon (see chapter 2), it is also very likely that the external fringing
fields are being partially responsible for that. The external fringing fields and their increment
with biasing voltages can be seen in the phase contour images, in the vacuum region in
between the Au tip and the specimen.

18 Drain
Source Gate
-4V
16

14
-3 V
12
Phase profile (rad)

-2 V

LDD
10
LDD

6 -1 V
4

2
0V
0 +1V

0 50 100 150 200 250 300 350

Position (nm)
Figure 8.27. Phase profiles extracted from the path B, marked in
Figure 8.24f, at different biasing voltages.

253
8.7. Summary and discussion

Three main issues hampering quantitative electron holography of semiconductor devices have
been addressed by investigating FIB prepared MOSFET devices. Curtaining effect due to the
presence of different materials above the devices, diffraction contrast arising from the strain
induced by contacts and charging of insulator oxide layers under electron illumination are in
addition to the problems discussed in previous chapters regarding quantitative holography of
simple pn junctions. These three problems are directly associated with the complex structure
of semiconductor devices.

The curtaining effect was avoided by milling the specimen from the substrate side. Also by
depositing additional protective Pt layer on the back-side of the specimen, uniform thickness
was achieved in the direction of the Ga ion beam. Therefore, the ambiguity brought about by
the thickness variation is not an issue in the specimens prepared using the double lift-out
method.

Diffraction contrast arising from the strain beneath the contacts, in the same TEM specimen
and at the same crystalline orientation, varied from one contact to another. It is very likely
that an orientation which can be considered as a weakly diffracting condition for one device
results in a strong diffraction contrast in the adjacent transistor. Therefore, it is necessary to
find the minimum diffracting condition for each device individually. Although it might be
time consuming, it is normally possible to find an orientation which suppresses the diffraction
contrast to a reasonable extent without significant smearing of interfaces.

Charging of the oxide layers located between the metal contacts was found to have a
profound influence on the measurement of dopant potentials. Four symptoms were identified
for the charged oxide layers in FIB prepared specimens: fringing field in the vacuum,
elliptical phase contours in the oxide, aero-foil shaped phase contours in the source/drain
regions and phase ramp towards the edge of the specimen.

The fringing field in the vacuum is significant if a bare oxide layer faces the vacuum region.
In our back-side milled specimen, the edge of the oxide layer was shielded by redeposited
materials, so the fringing field strength was not significant in the vacuum. However, absence
or weakness of fringing field in the small vacuum area captured in rather high magnification
holograms taken from devices does not necessarily mean that the oxide layer is not charged.

254
The appearance of elliptical phase contours in the oxide layers is another symptom of
charging. Observation of similar elliptical shapes in the SE images ruled out the effect of
perturbed reference wave in the formation of these fringes in electron holography. Also, this
observation suggests that charges in the oxide might be confined only in a thin thickness of
the oxide layer. In that case the electric field within the oxide would be comparable with the
maximum of electric field, which can be sustained by the oxide layer before breaking down.

The phase contours in the doped source/drain regions were found to be aero-foil shaped,
which become narrow close to the oxide spacers. Since phase contours in these regions
followed the expected shape of the doped regions with the elliptical fringes in the oxide
disappearing after coating the specimen with a thick layer of carbon, it can be deduced that
appearance of aero-foil shape contours in the doped regions is another symptom of charging.

The last but not least symptom of charging in the oxide, which was revealed after carbon
coating, is the phase ramp close to the edge of the specimen. It is easy to misinterpret this
ramp with the thickness variation of the specimen. It seems that it is impractical to avoid the
charging of the oxide by coating the specimen with carbon.

In the last part of this chapter, the first results of an electrically biased NMOS transistor were
presented. Forward and reverse biases were applied successfully to the junction between the
drain and the substrate of a real NMOS transistor.

255
Chapter 9

9. Conclusions and future work

9.1. Conclusions

The aim of this dissertation was to explore the application of off-axis electron holography to
the measurement of electrostatic potentials in transistors under working conditions, and to
address the associated practical challenges.

The motivation for this study was described in chapter 2. It was shown that the trend of
semiconductor device development has been towards fabricating smaller devices in size since
the invention of the transistor. The basics of MOSFET operation explained in this chapter
made it clear why miniaturisation causes problems in the transistor operation, and how
mapping electrostatic potential distributions within transistors can help overcome these
problems in the deep sub-micron regime. Existing characterisation techniques for mapping
built-in potentials were described, including both advantages and disadvantages of each
technique. Among these techniques, off-axis electron holography appears very promising.
Off-axis electron holography, described in depth in chapter 3, can provide quantitative
electrostatic potential maps of semiconductor devices under working conditions, with high
spatial and potential resolution.

The dual-beam FIB/SEM workstation was used as the primary tool to prepare the TEM
specimens in this work. Currently, the high site specificity required to prepare TEM
specimens and to make electrical contacts to the nano-scale transistor terminals can only be
achieved by employing the milling, imaging and depositing capability of dual-beam
workstations. The off-axis electron holography technique, however, is very sensitive to
artefacts and damage caused by the FIB. This matter was discussed in chapter 4.

256
The double lift-out procedure, developed during this work to prepare TEM specimens, gives
the focused ion beam access to the backside of the specimen and makes it possible to avoid
curtaining effects. The absence of curtaining effects in the TEM specimens prepared using
the double lift-out procedure was demonstrated in chapter 8 where real MOSFETs were
examined. This approach also allows electrical contacts to be made to the electron-
transparent specimen regions. Based on this ability, a novel specimen-geometry was
developed for the application of electrical biases to semiconductor devices (see chapter 4).

While developing this geometry, more reliable electrical contacts were made to silicon using
FIB-deposited W than FIB-deposited Pt (see chapter 5). This is presumably because the
percentage of desired metal in FIB-deposited W is higher than that in the FIB-deposited Pt,
resulting in a lower contact resistance between the FIB-deposited W and Si than between the
FIB-deposited Pt and Si. When electrical biasing a pn junction using this geometry, the
electrical contacts made using the FIB-deposited Pt were not good enough to transfer the
voltages to the pn junction. In such cases, when the voltage was applied, together with the
specimen drift the phase step across the junction increased irreversibly. Joule heating is likely
responsible for the specimen drift and for the increase in the phase step. This finding
confirms that low temperature annealing can be used for repairing FIB-induced damage.
However, when a similar specimen without electrical connections was annealed in the
heating holder, the increased phase step across the junction was not as symmetric as the phase
step under electrical biasing. Presumably the electrical contacts, duration of annealing, and
localization of the heat also play important roles. Further research is required to find optimum
conditions for annealing FIB-prepared TEM specimens to remove the damage induced by
FIB.

Off-axis electron holograms were acquired from a symmetric abrupt silicon pn junction under
a wide range of electrical biases. The electrostatic potential, electric field, and charge
distributions extracted from these holographic images were analysed and compared with
simulation in chapter 5. Consistent with theory, the phase step, the maximum of electric field,
and the depletion width associated with the junction increased with reverse-bias voltages.
However, the built-in potential and the specimen thickness measured from the phase-step
versus applied-voltage plot were both smaller than expected values. Also, the measured
depletion widths were much larger than those obtained from simulation. Moreover, the
measured electric field profiles were asymmetric and the measured charge density increased
with reverse-bias voltages and was larger on the n-side than on the p-side.

257
It was demonstrated that FIB-prepared specimen surfaces are not necessarily equipotential.
The electric field associated with the pn junction is found to leak to vacuum. A higher reverse
voltage yields an increased electric field across the junction and, consequently, more electric
field leakage into vacuum. Electron holographic phase images represent the projected
distribution of the electrostatic potential along the electron beam direction; therefore, the
electric fields in vacuum regions above and below the specimen also contribute to the phase
images. This explains why charge densities that are extracted from holographic phase images
increase with reverse-bias. In the vacuum region close to the specimen edge, it was observed
that the maximum of electric field in the vacuum region is slightly shifted towards the p-side.
Assuming a similar electric field distribution in the vacuum regions above and below the
specimen would explain why the measured charge density from holographic phase images is
larger on the n-side than on the p-side.

Secondary electron dopant contrast in the scanning electron microscope was used as a
complementary technique in chapter 7. Accepting, from the SEM observations, that more
secondary electrons are emitted from the p-type region than the n-type region, it is reasonable
to assume that the n- type and p- type regions charge differently under electron irradiation.
The difference in the amount of charge between the two sides of the junction could explain
the shift of the electric field in the vacuum. After sputter-coating the surface of the specimen
with Au/Pd in-situ in the dual-beam workstation (see chapter 4 for the description of the
method), the symmetry of the phase profiles across the pn junction improved significantly.
This is presumably because the deposited metallic layer on the specimen surfaces prevents
the specimen from charging, or at least prevents the accumulation of different amount of
charge on the n-side and the p-side. Also, the slope difference in the phase profiles between
the p-side and n-side disappeared completely after Au/Pd coating and phase profiles became
flat on both sides. The electric field lines terminate at the metal coated surfaces and do not
leak to vacuum. It is postulated that leaked electric fields above and below the specimen are
at least partially responsible for the slopes appearing in the phase profiles. It was shown in
chapter 6 that the slope difference between the p-side and n-side increases linearly with
increased reverse bias voltage.

The results obtained from the electrically biased specimens and the specimen coated in-situ
with metal showed that the phase step across the pn junction decreases as approaching the
specimen edge. Also it was shown in chapter 7 that the contribution of the p-type and n-type
regions to the phase step reduction is not the same. The decrease in the phase step close to the

258
specimen edge was mainly related to the phase variation on the n-type region. It is reasonable
to assume that the specimen experiences the same FIB-damage and surface termination
conditions on its top and bottom as it does at its edge. The phase step should exhibit the same
behaviour close to the top and bottom surfaces of the specimen as it does close to the
specimen edge. In other words, instead of assuming an electrically “dead layer”, a layer with
a uniform thickness and without any electrostatic potential difference between the n-type and
the p-type regions, it is postulated that the built-in potential decreases and the depletion width
increases gradually as approaching the specimen surfaces with different contributions from
the p-type and the n-type regions. This explains why the measured depletion widths are wider
and the measured built-in potentials are lower than expected values.

It was shown in chapter 7 that combining EBIC measurements with off-axis electron
holography provides a better understanding of the consequences of electron beam irradiation
on the measurement of built-in potential. In measuring the built-in potential associated with a
pn junction using off-axis electron holography, a lower potential was measured when the
electron beam current density was increased. For a constant illuminated area, a larger electron
beam current density results in a larger EBIC. In FIB-prepared specimens, re-deposited
materials and FIB-induced damage layers are likely acting like a resistor connected in parallel
with the pn junction. If the EBIC goes through this resistor, the induced voltage across the
resistor will forward bias the pn junction. It was demonstrated how the combination of
holographic and EBIC measurements can be used to estimate the magnitude of this resistor.

Using the EBIC method, the diffusion lengths of minority carriers were measured for a FIB-
prepared specimen, and found to be significantly shorter than those reported for bulk silicon.
This highlighted the effects of surface termination and FIB-induced damage. EBIC
measurements also were used to assess the role of the objective aperture in electron
holographic studies.

The experimental results presented in chapter 6 demonstrated that the measured phase step
across a silicon pn junction changes with specimen tilt angle. Although diffraction contrast
was shown to be partially responsible for the lack of consistency in the measurement of built-
in potential across the silicon homo-junction at different orientations, it was also
demonstrated that dynamical diffraction cannot be the only reason. After removing the effect
of diffraction contrast by applying an electrical bias, the measured phase step associated with
the built-in potential was still sensitive to the specimen orientation. The experimental

259
observations presented in this chapter revealed that the difference in the slope of the phase
profile between the two sides of the pn junction is also responsible for the scatter in the
measured built-in potential data, particularly when diffraction contrast is not visible in the
amplitude image.

The practical challenges encountered in examining MOSFET transistors were discussed in

chapter 8. To minimise the diffraction contrast arising from the strain beneath the transistors’
plugs, no single ideally favourable orientation which can be applied to every device exists.
For each transistor, even in the same TEM specimen, the minimum diffracting condition
should be found individually. A procedure was suggested to find an orientation where
diffraction contrast is suppressed to a reasonable extent without significant smearing of
interfaces.

The oxide layers used in the transistor structure for isolating the metal contacts charged
significantly under the electron beam. This charging has a profound influence on the
measurement of built-in potentials. In chapter 8, fringing fields in vacuum, elliptical phase
contours in the oxide, aero-foil shaped phase contours in the source/drain regions and a phase
ramp towards the specimen edge were discussed as the symptoms of oxide-layer charging.
The first three symptoms disappeared after coating the specimen with a thick layer of carbon
(~60nm) and the phase step across the pn junctions increased, but the appearance of defects
in the carbon film disturbed the phase image such that the phase steps could not be measured
accurately.

In chapter 8, electrical contacts were made successfully to the source/drain and the substrate
of a real NMOSFET, and the device was examined under electrical biasing conditions using
off-axis electron holography. Applying reverse voltages between the drain and the substrate
resulted in the extension of the depletion region around the drain into the source depletion
region. The presence of fringing fields in the phase images precluded attributing this
extension to the punchthrough phenomenon.

260
9.2. Future work

Currently, three main factors are impeding off-axis electron holography from being used
routinely in the determination of electrostatic potentials in semiconductor devices under
working conditions: surface termination and preparation damage, electron illumination, and
difficulties in making electrical contacts.

9.2.1. Surface termination and preparation damage

Bulk semiconductor devices experience surface termination and preparation damage when
they are made into TEM specimens, affecting the distribution of electrostatic potentials
within the devices. The theoretical understanding of surface effects should be furthered,
specifically for measuring built-in potentials using off-axis electron holography. In addition
to the surface termination, damage induced during specimen preparation, particularly when
using the FIB, may significantly affect the built-in potentials. It is likely that FIB irradiation
influences the distribution of electrostatic potentials differently for different regions of a
semiconductor device, depending on the material, dopant type and dopant concentration.
Understanding the consequences of surface damage on the electrostatic potentials within FIB-
prepared devices is a prerequisite for quantitative electron holography of semiconductor
devices. Elemental mapping of FIB-damaged layers, using techniques such as atom probe
tomography (APT), and using the resulting map as an input to modelling methods such as
density functional theory (DFT) should shed further light into the properties of the damaged
surface layers.

9.2.2. Electron illumination

The secondary electron emission and the electron-hole pair generation, both resulting from
exposing the specimen to the electron beam, affect the distribution of electrostatic potentials
within the device being examined. The specimen charges under electron illumination and
quantification of this beam-induced charging is part of the challenge of future studies. The
semiconductor and insulator regions of the specimen charge differently. Also, as discussed in
this thesis, the charging behaviour of the p-type and the n-type regions are different and
probably depend on the dopant concentration as well. It is also possible that the specimen
charges differently in its thicker and thinner regions, because the number of transmitted
electrons is different for these regions. Therefore, for a complicated semiconductor device
specimen, it would be naive to assume that the specimen charges uniformly under the
electron illumination. Furthermore, the possibility that the specimen charges differently at

261
different crystallographic orientations needs to be investigated. This might explain why when
diffraction contrast effects were compensated for, in the experiment presented in chapter 6,
the measured phase steps across a pn junction still changed with specimen tilt angle.

Further theoretical studies on the empirical methods suggested to mitigate the effects of
specimen charging are necessary both for understanding the electron beam induced charging
phenomena and for developing a more efficient technique for mitigating the charging effects.
Also both the in-situ metal deposition method proposed in this thesis and carbon coating
should be investigated systematically in order to determine the optimum coating conditions.
Characterising a specimen which has been examined using electron holography with non-
electron-beam-based techniques, such as scanning spreading resistance microscopy (SSPR),
may help to further the understanding of the influences of electron irradiation on the
measurement of electrostatic potentials. Also, further theoretical studies are necessary to
understand the effect of electron-hole pair generation on the electrostatic potential. It is
crucial that, in these theoretical studies, experimental conditions such as the presence of re-
deposited materials on the specimen surfaces and FIB-damaged layers are considered.

9.2.3. Making electrical contacts

Making electrical contacts to all four small terminals of a MOSFET (source, drain, gate and
substrate) when the MOSFET is made into a thin TEM specimen is another practical
challenge for examining working transistors using off-axis electron holography. The
specimen geometry developed in this thesis made it possible to make very site-specific
electrical contacts to the regions of interest. This geometry should be applicable to MOSFETs
to make all four required electrical contacts. Also, electrical biasing TEM holders with more
than four electrical connections are now commercially available. To electrically isolate FIB-
deposited W lines from each other on the specimen in this geometry, insulator materials can
be deposited using the FIB-induced deposition technique. The in-situ sputtering method in
dual-beam SEM/FIB, used in this work to deposit a layer of Au/Pd on the surface of the pn
junction specimen, can also be used for achieving higher quality metal lines and insulator
layers required for electrical wiring. A higher percentage of desired material is expected in
the deposited layers when the in-situ sputtering method is used rather than the FIB-induced
deposition method.

262
References

1. Flatow, Ira. Transistorized. http://www.pbs.org/transistor/album1/. [Online] 1999.

2. Orton, John. The story of semiconductors. Oxford, UK : Oxford University Press, 2006.

3. Ellinger, Frank. Radio frequency integrated circuits and technologies. Springer, 2008.

4. Sze, S.M. Semiconductor devices: physics and technology. John Willey & Sons, 2002.

5. The Intel 4004. http://www.intel.com/about/companyinfo/museum/exhibits/4004/facts.htm. [Online]

6. NAND Flash Memory.

http://www.toshiba.co.jp/about/press/2010_08/pr3101.htm?from=RSS_PRESS&uid=20100831-
1112e. [Online] 31 August 2010.

7. http://newsroom.intel.com/community/intel_newsroom/blog/2011/02/18/intel-to-invest-more-than-
5-billion-to-build-new-factory-in-arizona. [Online] Intel, Feb 2011.

8. Sze, S.M. Physics of semiconductor devices. John Wiley & Sons, 1981.

9. Wolf, Stanley and Tauber, Richard N. Silicon processing for the VLSI era. Lattice Press, 2000.

10. Cramming more components onto integrated circuits. Moore, G. Electronics, 1965, Vol. 38, p.
114.

11. Razavi, Behzad. Design of Analog CMOS Integrated Circuits. Thomas Casson, 2001.

12. Suppression of hot-carrier effects in submicrometer CMOS technology. Min-Liang, Chen, et al.
IEEE Transactions on Electron Devices, 1988, Vol. 35.

13. Quarter-micrometer SPI (Self-aligned Pocket Implantation) MOSFET's and its application for
low supply voltage operation. Hori, Atsushi, et al. 1, IEEE Transactions on Electron Devices, 1995,
Vol. 42, p. 78.

14. Colinge, Jean-Pierre. FinFETs and Other Multi-Gate Transistors. Springer, 2008.

15. International Technology Roadmap for Semiconductors. http://www.itrs.net. [Online] 2011.

16. Streetman, Ben G. and Banerjee, Sanjay Kumar. Solid state electronic devices. Pearson
Prentice Hall, 2006.

17. Semiconductor Measurement Technology: The Relationship Between Resistivity and Dopant
Density for Phosphorus-and Boron-Doped Silicon. Thurber, W. R., et al. National Bureau of
Standards Special Publication 400-64, 1981.

18. Kittel, Charles. Introduction to dolid state physics. John Wiley & Sons, 2005.

19. Silicon. http://www.ioffe.ru/SVA/NSM/Semicond/Si/. [Online]

20. Somodi, PK. Analysis of electron holograms of semiconductors and phase retrieval in the TEM.
PhD thesis, University of Cambridge, 2005.

21. Assessing the performance of two-dimensional dopant profiling techniques. Duhayon, N., et al. J.
Vac. Sci. Technol. B, 2004, Vol. 22, p. 385.

263
22. Status and review of two-dimensional carrier and dopant profiling using scanning probe
microscopy. Wolf, P. De, et al. J. Vac. Sci. Technol. B , 2000, Vol. 18, p. 361.

23. Depth and lateral resolution of laser-assisted atom probe microscopy of silicon revealed by
isotopic heterostructures. Shimizu, Y., et al. J. Appl. Phys., 2011, Vol. 109, p. 036102 .

24. Dopant measurements in semiconductors with atom probe tomography. Ronsheim, P. A.,
Hatzistergos, M. and Jin, S. J. Vac. Sci. Technol. B, 2010, Vol. 28, p. C1E1.

25. A review of advanced scanning probe microscope analysis of functional films and semiconductor
devices. Benstetter, Günther, Biberger, Roland and Dongping, Liu. 17, Thin Solid Films, 2009,
Vol. 517, p. 5100.

26. Two-dimensional carrier profiling on operating Si metal-oxide semiconductor field-effect

transistor by scanning capacitance microscopy. Kimura, Kenjiro, et al. 3, J. Vac. Sci. Technol. B ,
2006, Vol. 24, p. 1371.

27. Intermittent-contact scanning capacitance microscopy versus contact mode SCM applied to 2D
dopant profiling. Biberger, Roland, et al. 8-9, Microelectronics Reliability, 2008, Vol. 48, p. 1339.

28. Scanning spreading resistance microscopy (SSRM) 2d carrier profiling for ultra-shallow junction
characterization in deep-submicron technologies. Eyben, P., Janssens, T. and Vandervorst, W.
Materials Science and Engineering: B, 2005, Vols. 124-125, p. 45.

29. On the spatial resolution of two dimensional doping profiles as measured using secondary ion
mass spectrometry tomography. Liu, X., et al. 1, J. Vac. Sci. Technol. B, 1994, Vol. 12, p. 116.

30. Miller, M. K. Atom probe tomography: analysis at the atomic level. Kluwer Academic/Plenum
Publishers, 2000.

31. Atom Probe Tomography of Electronic Materials. Kelly, Thomas F., et al. Annual Review of
Materials Research, 2007, Vol. 37, pp. 681-727.

32. Atom probe tomography today. Cerezo, Alfred, et al. 12, Materials Today, 2007, Vol. 10, pp. 36-
42.

33. 2-D dopant profiling in VLSI devices using dopant-selective etching: an atomic force microscopy
study. Barrett, M., et al. 3, IEEE Electron Device Letters, 1995, Vol. 16.

34. Two-dimensional dopant profiling in semiconductor devices by electron holography and chemical
etching delineation techniques with the same specimen. Shaislamov, Ulugbek, et al. 10,
Microelectronics Reliability, 2008, Vol. 48.

35. Two-dimensional dopant concentration profiles from ultrashallow junction metal-oxide-

semiconductor field-effect transistors using the etch/transmission electron microscopy method. Yoo,
K. D., Marsh, C. D. and Booker, G. R. 10, Appl. Phys. Lett., 2002, Vol. 80.

36. Qua a ve ra ma g of I A /I P qua um do w 1 m a a re o u o u g dark

field electron holography. Cooper, David, et al. Appl. Phys. Lett., 2011, Vol. 99, p. 261911.

37. Medium resolution off-axis electron holography with millivolt sensitivity. Cooper, David, et al.
143501, Appl. Phys. Lett., 2007, Vol. 91. doi:10.1063/1.2794006.

38. Mapping of electrostatic potential in deep submicron CMOS devices by electron holography.
Gribelyuk, M. A., et al. Phys. Rev. Lett., 2002, Vol. 89, p. 89.

39. Electron holography on silicon microstructures and its comparison to other microscopic
techniques. Formánek, Petr and Kittler, Martin. J. Phys.: Condens. Matter, 2004, Vol. 16, p. S193.

264
40. Illumination effects in holographic imaging of the electrostatic potential of defects and pn
junctions in transmission electron microscopy. Houben, L., Luysberg, M. and Brammer, T. Phys.
Rev. B, Oct 2004, Vol. 70.

41. Myung-Geun, Han. Characterization of electrostatic potential of semiconductor devices using

off-axis electron holography. PhD thesis; Arizona State University, 2007.

42. Quantitative electron holographic tomography for the 3D characterisation of semiconductor

device structures. Twitchett-Harrison, A. C., et al. Ultramicroscopy, 2008, Vol. 108, pp. 1401-
1407.

43. Electron holography—basics and applications. Lichte, Hannes and Lehmann, Michael. Rep.
Prog. Phys., 2008, Vol. 71, p. 016102 (46pp).

44. Electron holography: Applications to materials questions. Lichte, Hannes, et al. 1, Annual
review of materials research, 2007, Annual Review of Materials Research, Vol. 37, p. 539.

45. Dunin-Borkowski, Rafal E., McCartney, Martha R. and Smith, David J. Electron Holography
of Nanostructured Materials. [book auth.] H. S. Nalwa. Encyclopedia of Nanoscience and
Nanotechnology. 2004, Vol. 3, pp. 41-99.

46. Volkl, Edgar, Allard, Lawrence F. and Joy, C. David. Introduction to electron holography.
Kluwer academic/Plenum, 1998.

47. Williams, David B. and Carter, C. Barry. Transmission Electron Microscopy: A Text for
Materials Science. 1996.

48. Twenty forms of electron holography. Cowley, J.M. 4, Ultramicroscopy, June 1992, Vol. 41, pp.
335-348.

49. Strain metrology of devices by dark-field electron holography: A new technique for mapping 2D
strain distributions. Hÿtch, Martin, et al. Electron Devices Meeting (IEDM), IEEE International,
2009. doi: 10.1109/IEDM.2009.5424424.

50. A New Microscopic Principle. Gabor, Denis. Nature , 1948, Vol. 161, pp. 777-778.
doi:10.1038/161777a0.

51. Microscopy by Reconstructed Wave-Fronts. Gabor, Denis. Proceedings of the Royal Society of
London. Series A, Mathematical and Physical Sciences, 1949, Vol. 197, pp. 454-487.

52. Tonomura, Akira. Electron Holography. Springer, 1999.

53. Reconstructed Wavefronts and Communication Theory. Leith, Emmett N. and Upatnieks, Juris.
10, J. Opt. Soc. Am., 1962, Vol. 52, pp. 1123-1128.

54. Reimer, Ludwig and Kohl, Helmut. Transmission Electron Microscopy. Springer, 2008.

55. Tutorial on off-axis electron holography. Lehmann, Michael and Lichte, Hannes. Microscopy
and microanalysis, 2002, Vol. 8.

56. Practical aspects of electron holography. Joy, David C., et al. Ultramicroscopy, 1993, Vol. 51.
ISSN 0304-3991.

57. Electron holography: optimum position of the biprism in the electron microscope. Lichte,
Hannes. Ultramicroscopy, 1996, Vol. 64.

265
58. Electron holography of long-range electrostatic fields. Matteucci, G., Missiroli, G. F. and Pozzi,
G. Advances in Imaging and Electron Physics, 2002, Vol. 122, pp. 173-249. doi:10.1016/S1076-
5670(02)80053-2.

59. Electron Holography: Phase Imaging with Nanometer Resolution. McCartney, Martha R. and
Smith, David J. Annual Review of Materials Research, 2007, Vol. 37, pp. 729-767 .

60. Off-axis electron holography in an aberration-corrected transmission electron microscope.

Lichte, Hannes, Geiger, Dorin and Linck, Martin. Philos Transact A Math Phys Eng Sci., 2009,
Vol. 28, pp. 3773-93.

61. Electron holography with a Cs-corrected transmission electron microscope. Geiger, Dorin, et al.
Microscopy and Microanalysis, 2008, Vol. 14, pp. 68-81.

62. Holographic Setup for 2D-Dopant Profiling using the Lorentz-lens. Lehmann, Michael, Brand,
Karin and Lichte, Hannes. Microscopy and Microanalysis, 2002. Vol. 8, p. Poster.

63. Electron interferometry and interference electron microscopy. Missiroli, G F, Pozzi, G. and
Valdre, U. J. Phys. E: Sci. Instrum., 1981, Vol. 14. doi:10.1088/0022-3735/14/6/001.

64. Detection limits in quantitative off-axis electron holography. Ruijter, W.J. De and Weissa, J.K.
Ultramicroscopy, 1993, Vol. 50, pp. 269-283.

65. Imaging properties and applications of slow-scan charge-coupled device cameras suitable for
electron microscopy. Ruijter, W.J. De. 3, Micron, 1995, Vol. 26, pp. 247-275.

66. Parameters for high-resolution electron holography. Lichte, Hannes. 1-4, Ultramicroscopy,
1993, Vol. 51, pp. 15-20.

67. http://www.gatan.com. [Online]

68. Voelk, Edgar. HoloWorks. 2002.

69. Holography. Holography group, University of Dresden, Triebenberg Lab, 2009.

70. Bonevich, John. HolograFREE. 2006.

71. Perkes, Paul R. and McCartney, Martha R. ASUHolography. 2002.

72. Oppenheim, Alan V. and Schafer, Ronald W. Discrete-time signal processing. 2010. ISBN
0131988425.

73. An introduction to off-axis electron holography. Midgley, P.A. Micron, 2001, Vol. 32, pp. 167-
184.

74. Volkl, E. and Lehmann, M. The reconstruction of off-axis electron holograms. [book auth.]
Edgar Volkl, Lawrence F. Allard and David C. Joy. Introduction to electron holography. 1998, p.
127.

75. Ghiglia, D. C. and Pritt, M. D. Two-dimensional Phase Unwrapping: Theory, Algorithms, and
Softwar. John Wiley and Sons, 1998. ISBN 0-471-24935-1.

76. Smith, D.J., et al. Quantitative electron holography. [book auth.] Edgar Volkl, Lawrence F.
Allard and David C. Joy. Introduction to electron holography. 1998, p. 123.

77. Egerton, R. F. Electron Energy Loss Spectruscopy in the Electron Microscope. Plenum, 1996.

266
78. Ab o u e mea ureme of orma zed ck e , /λ , from off-axis electron holography.
McCartney, M.R. and Gajdardziska-Josifovska, M. 3, Ultramicroscopy, 1994, Vol. 53, pp. 283-
289.

79. Razavy, Mohsen. Quantum theory of tunneling. Word Scientific Publication, 2003.

80. Dirac-Fock Calculations of X-ray Scattering Factors and Contributions to the Mean Inner
Potential for Electron Scattering. Rez, David, Rez, Peter and Grant, Ian. Acta Cryst., 1994, Vol.
AS0, p. 481.

81. Complex Lattice Potentials in Electron Diffraction Calculated for a Number of Crystals. Radi, G.
Acta Cryst., 1970, Vol. A26, p. 41.

82. Sample preparation for precise and quantitative electron holographic analysis of semiconductor
devices. Han, M.G., et al. Microscopy and Microanalysis, 2006, Vol. 12, pp. 295-301.

83. Electron holography study for two-dimensional dopant profile measurement with specimens
prepared by backside ion milling. Yoo, J. H., et al. Journal of Electron Microscopy, 2008, Vol. 57,
pp. 13-18.

84. Electron holography analysis of a shallow junction for planar-bulk metal-oxide-semiconductor

field-effect transistors approaching the scaling limit. Ikarashi, N., et al. J. Appl. Phys., 2008, Vol.
103.

85. Shallow source/drain extensions for deep submicron MOSFETs using spin-on-dopants. Gopalan,
C., et al. IEEE Transactions on Electron Devices, 2003, Vol. 50, pp. 1277-1283.

86. Two-dimensional mapping of the electrostatic potential in transistors by electron holography.

Rau, W. D., et al. Phys. Rev. Lett., 1999, Vol. 82, pp. 2614-2617.

87. Specimen preparation for electron holography of semiconductor devices. Formanek, P. and
Bugiel, E. Ultramicroscopy, 2006, Vol. 106, pp. 365-375.

88. Giannuzzi, Lucille A. and Stevie, Fred A. Introduction to Focused Ion Beams; Instrumentation,
Theory, Techniques and Practice. Springer, 2005.

89. A review of focused ion beam applications in microsystem technology. Reyntjens, S. and Puers,
R. J. Micromech. Microeng., 2001, Vol. 11, pp. 287-300.

90. Nanometer scale patterning using focused ion beam milling. Petit, D., et al. Rev. Sci. Instrum.,
2005, Vol. 76.

91. TEM Study of Surface Damage and Profile of a FIB-Prepared Si Sample. Wang, SX. Microscopy
and microanalysis, 2004, Vol. 10, p. 1158.

92. A review of focused ion beam milling techniques for TEM specimen preparation. Giannuzzi, L.
A. and Stevie, F. A. Micron, 1999, Vol. 30, pp. 197-204.

93. Conventional and back-side focused ion beam milling for off-axis electron holography of
electrostatic potentials in transistors. Dunin-Borkowski, R. E., et al. Ultramicroscopy, 2005, Vol.
103, pp. 67-81.

94. Development of Sample Preparation Method for Observation of Dopant Profile by Electron
Holography. Sato, T, et al. S2, Microscopy and microanalysis, 2008, Vol. 14, p. 836.

95. Side-milling technique of preparing device cross-sections for electron holography based on a
focused ion beam micro-sampling system. Wang, Z., Kato, T. and Hirayama, T. Journal of Electron
Microscopy, 2004, Vol. 53, pp. 489-91.

267
96. Evaluation of top, angle, and side cleaned FIB samples for TEM analysis. Montoya, E., et al.
Microscopy Research and Technique, 2007, Vol. 70, pp. 1060-1071.

97. Issues Affecting Quantitative Evaluation of Dopant Profiles Using Electron Holography.
McCartney, M. R. S02, Microscopy and microanalysis, 2003, Vol. 9, p. 776.

98. 2D-mapping of dopant distribution in deep sub micron CMOS devices by electron holography
using adapted FIB-preparation. Lenk, A., Lichte, H. and Muehle, U. J. of Electron Microsc., 2005,
Vol. 54, pp. 351-359.

99. A critical literature review of focused electron beam induced deposition. Dorp, W. F. van and
Hagen, C. W. J. Appl. Phys., 2008, Vol. 104.

100. Surface damage induced by focused-ion-beam milling. in a Si/Si p-n junction cross-sectional
specimen. Wang, Z. G., et al. Applied Surface Science, 2005, Vol. 241, pp. 80-86.

101. Investigations on the topology of structures milled and etched by focused ion beams. Lipp, S., et
al. J. Vac. Sci. Technol. B, 1996, Vol. 14, pp. 3996-3999.

102. A plasma-polymerized protective film for transmission electron microscopy specimen

preparation by focused ion beam etching. Kato, N. I., Miura, N. and Tsutsui, N. J. Vac. Sci.
Technol. A, 1998, Vol. 16, pp. 1127-1130.

103. FIB-induced damage in silicon. Rubanov, S. Journal of microscopy, 2004, p. 213.

104. FIB Damage in Silicon: Amorphization or Redeposition? Rajsiri, S., et al. S02, Microscopy and
microanalysis, 2002, Vol. 8.

105. Transmission electron microscope specimen preparation of metal matrix composites using the
focused ion beam miller. Cairney, J. M., Smith, R. D. and Munroe, P. R. Microscopy and
Microanalysis, 2000, Vol. 6, pp. 452-462.

106. Experimental off-axis electron holography of focused ion beam-prepared Si p-n junctions with
different dopant concentrations. Cooper, David, et al. J. Appl. Phys., 2008, Vol. 104, p. 064513.

107. Reducing FIB damage using low energy ions. Giannuzzi, LA. S02, Microscopy and
microanalysis, 2006, Vol. 12, p. 1260.

108. Sample Preparation for Aberration Corrected Microscopy. Genç, A. S2, Microscopy and
microanalysis, 2008, Vol. 14, p. 998.

109. Advances in TEM Sample Preparation Using a Focused Ion Beam. Leer, B.V. S2, Microscopy
and microanalysis, 2008, Vol. 14, p. 380.

110. Reducing focused ion beam damage to transmission electron microscopy samples. Kato, N. I.
Journal of Electron Microscopy, 2004, Vol. 53, pp. 451-458.

111. Side-wall damage in a transmission electron microscopy specimen of crystalline Si prepared by

focused ion beam etching. Kato, N. I., Kohno, Y. and Saka, H. J. Vac. Sci. Technol. A, 1999, Vol.
17, pp. 1201-1204.

112. Assessment of surface damage and sidewall implantation in AlGaN-based high electron mobility
transistor devices caused during focused-ion-beam milling. Cullen, D. A. and Smith, D. J. J. Appl.
Phys, 2008, Vol. 104.

113. Reduction of electrical damage in specimens prepared using focused ion beam milling for dopant
profiling using off-axis electron holography. Cooper, D., Truche, R. and Rouviere, J. L.
Ultramicroscopy, April 2008, Vol. 108, pp. 488-493.

268
114. Improvement in electron holographic phase images of focused-ion-beam-milled GaAs and Si p-n
junctions by in situ annealing. Cooper, David, et al. Appl. Phys. Lett., 2006, Vol. 88, p. 063510.

115. Electron holographic mapping of two-dimensional doping areas in cross-sectional device

specimens preparted by the lift-out technique based on a focused ion beam. Wang, Z. G., et al.
Journal of Electron Microscopy, 2004, Vol. 53, pp. 115-119.

116. Quantitative analysis of 2-d electrostatic potential distributions in 90-nm Si pMOSFETs using
off-axis electron holography. Han, Myung-Geun, et al. 12, IEEE Transactions on Electron Devices,
2007, Vol. 54, pp. 3336-3341 .

117. Novel scheme for the preparation of transmission electron microscopy specimens with a focused
ion beam. Overwijk, M. H. F., Vandenheuvel, F. C. and Bullelieuwma, C. W. T. J. Vac. Sci.
Technol. B, 1993, Vol. 11, pp. 2021-2024.

118. Cooper, David. Off-axis electron holography of focused ion beam prepared semiconductor
devices. PhD thesis; University of Cambridge, 2006.

119. Avoiding the curtaining effect: Backside milling by FIB INLO. Schwarz, Stephen M., et al. S2,
2003, Vol. 9.

120. A New method for investigating the electric field regions of p n junctions. Titchmarsh, J. M.,
Lapworth, A. J. and Booker, G. R. Physica Status Solidi. B, 1969, Vol. 34, p. K83.

121. In situ electron holographic analysis of biased Si n(+)-p junctions. Han, M.-G., Smith, D. J.
and McCartney, M. R. Appl. Phys. Lett., 2008, Vol. 92, p. 143502.

122. Off-axis electron holography of unbiased and reverse-biased focused ion beam milled Si p-n
junctions. Twitchett, AC, et al. 1, Microsc Microanal., 2005, Vol. 11, pp. 66-78.

123. Quantitative off-axis electron holography of GaAs p-n junctions prepared by focused ion beam
milling. Cooper, D, et al. 1, J. Microsc., 2009, Vol. 233, pp. 102-113.

124. Electron Holographic Observations of the Electrostatic Field Associated with Thin Reverse-
Biased p-n Junctions. Frabboni, S., et al. Phys. Rev. Lett., 1985, Vol. 55, pp. 2196-2199.

125. Off-axis electron holography of electrostatic potentials in unbiased and reverse biased focused
ion beam milled semiconductor devices. Twitchett, A. C., et al. J. Microsc., 2004, Vol. 214, pp. 287-
296.

126. A versatile three-contact electrical biasing transmission electron microscope specimenholder for
electron holography and electron tomography of working devices. Kasama, Takeshi, et al. Mater.
Res. Soc. Symp. Proc., 2006, Vol. 907E.

127. Liu, Lewis Zhen-Yu. Advanced Transmission Electron Microscopy of GaN-based Materials
and Devices. PhD thesis, University of Cambridge, 2011.

128. Off-axis electron holography of Si semiconductors prepared using FIB. Cooper, D, et al. 16th
International Conference on Microscopy of Semiconducting Materials, 2010.

129. http://www.okmetic.com/. [Online]

130. Liu, Xian. Arsenic-doping of silicon by molecular beam epitaxy. PhD thesis; Stanford
University, 2003.

131. Vasileska, Dragica and Klimeck, Gerhard. Tutorial for PADRE Based Simulation Tools.
http://nanohub.org/resources/7223, 2009.

269
132. Vasileska, Dragica, et al. PN Junction Lab. 2011. DOI: 10254/nanohub-r229.20.

133. Structural and electrical studies of conductive nanowires prepared by focused ion beam induced
deposition. Reguer, A., et al. J. Vac. Sci. Technol. B, 2008, Vol. 26, p. 175.

134. Electric field induced motion of metallic droplets: Application to submicron contactor.
Dallaporta, H., et al. J. Vac. Sci. Technol. B, 2010, Vol. 28.

135. Predel, B. Ga-Pt (Gallium-Platinum). Madelung, O. (ed.). SpringerMaterials - The Landolt-

Börnstein Database (http://www.springermaterials.com).

136. —. Ga-W (Gallium-Tungsten). Madelung, O. (ed.). Springer Materials - The Landolt-Börnstein

Database (http://www.springermaterials.com). DOI: 10.1007/10501684_1413.

137. Contact resistance of focused ion beam deposited platinum and tungsten films to silicon. J.
DeMarco, Anthony and Melngailis, John. J. Vac. Sci. Technol. B, 2001, Vol. 19, p. 2543.

138. Gevers, R. Modern diffraction and imaging techniques in material science. North-Holland,
1970.

139. Accurate measurements of mean inner potential of crystal wedges using digital electron
holograms. Gajdardziska-Josifovska, M., et al. 3, Ultramicroscopy, 1993, Vol. 50, p. 28.

140. Determination of the mean inner potential in III–V semiconductors by electron holography.
Kruse, P., Rosenauer, A. and Gerthsen, D. 1, Ultramicroscopy, 2003, Vol. 96, pp. 11-16.

141. Transmission electron holography of silicon nanospheres with surface oxide layers. Wang, Y.
C., et al. Appl. Phys. Lett., 1997, Vol. 70, p. 1296 .

142. O e average Cou omb o e a (Φ ) a d co ra o e e ec ro de y cry a .

O'Keeffe, M. and Spence, J. C. H. A, Acta Crystallographica, 1994, Vol. 50.

143. Digital techniques in electron off-axis holography. Ade, G. Advances in Electronics and
Electron Physics, 1994, Vol. 89, pp. 1-51.

144. Ashcroft, Neil W. and Mermin, D. David. Solid state physics. 1976.

145. The effect of dynamical scattering in off-axis holographic mean inner potential and inelastic
mean free path measurements. Lubk, Axel, Wolf, Daniel and Lichte, Hannes. 5, Ultramicroscopy,
2010, Vol. 110, pp. 438-446.

146. Method of off-axis electron holography and investigations of the phase structure in crystals.
Hanszen, KJ. J. Phys. D: Appl. Phys., 1986, Vol. 19, p. 373.

147. Spence, J.C.H. and Zuo, J.M. Electron Microdiffraction. Plenum, 1992.

148. Peng, L.M., Dudarev, S.L. and Whelan, M.J. High-Energy Electron Diffraction and
Microscopy. Oxford Science, 2004.

149. On specimen tilt for electron holography of semiconductor devices. Formanek, P. and Bugiel,
E. Ultramicroscopy, 2006, Vol. 106, pp. 292-300.

150. Dopant profiling in semiconductor devices using electron holography: influence of specimen
orientation. Brand, K., et al. Durban : Proceedings of the 15th International Congress on Electron
Microscopy ICEM15, 2002. Vol. I, p. 37.

151. Mapping the electrostatic potential across AlGaN/AlN/GaN heterostructures using electron
holography. Wu, Z.H., et al. Appl. Phys. Lett., 2007, Vol. 90, p. 032101.

270
152. Interpretation of electron beam induced charging of oxide layers in a transistor studied using
electron holography. Ubaldi, F., et al. Oxford, England : 16th International Conference on
Microscopy of Semiconducting Materials, 2010.

153. Characterization of charging in semiconductor device materials by electron holography.

McCartney, M. R. Journal of Electron Microscopy, 2005, Vol. 54, pp. 239-242.

154. Quantitative analysis of one-dimensional dopant profile by electron holography. McCartney,

M. R., et al. Appl. Phys. Lett., 2002, Vol. 80, pp. 3213-3215.

155. The influence of electron irradiation on electron holography of focused ion beam milled GaAs p-
n junctions. Cooper, D., et al. J. Appl. Phys., 1 May 2007, Vol. 101.

156. Transmission electron microscopy observations of p-n junctions. Merli, P. G., Missiroli, G. F.
and Pozzi, G. Physica Status Solidi a-Applied Research, 1975, Vol. 30, pp. 699-711.

157. Influence of charged oxide layers on TEM imaging of reverse-biased p-n junctions. Beleggia,
M., et al. Phys. Rev. B, 2003, Vol. 67.

158. Electron microscopy of reverse biased p-n junctions. Beleggia, M., et al. Micron, June 2000,
Vol. 31, pp. 231-236.

159. Experimental characterization and mitigation of specimen charging on thin films with one
conducting layer. Downing, K. H., McCartney, M. R. and Glaeser, R. M. Microscopy and
Microanalysis, Dec 2004, Vol. 10, pp. 783-789.

160. The surface of evaporated carbon films is an insulating, high-bandgap material. Larson, D. M.,
Downing, K. H. and Glaeser, R. M. Journal of Structural Biology, May 2011, Vol. 174, pp. 420-
423.

161. Electron holography of thin amorphous carbon films: Measurement of the mean inner potential
and a thickness-independent phase shift. M. Wanner, D. Bach, D. Gerthsen, R. Werner, and B.
Tesche. Ultramicroscopy, March 2006, Vol. 106, pp. 341-345.

162. Effects of beam-specimen interaction on the observation of reverse-biased p-n junctions by

electron interferometry. Fazzini, P. F., et al. Phys. Rev. B, 2005, Vol. 72.

163. Electron holography of doped semiconductors: when does it work and is it quantitative? Dunin-
Borkowski, R. E., et al. Oxford, England : 14th Conference on Microscopy of Semiconducting
Materials, 2005. pp. 203-212.

164. On the reliability of quantitative phase measurements by low magnification off-axis image plane
electron holography. Frost, B.G. and Voelkl, E. Ultramicroscopy, 1998, Vol. 72, p. 101.

165. An electron holographic study of electric charging and electric charge distributions. Frost, B.G.
Ultramicroscopy, 1998, Vol. 75, pp. 105-113.

166. Tiemeijer, Peterl. Monochromator manual Titan. FEI, 2008.

167. Charge collection scanning electron microscopy. Leamy, H. J. J. Appl. Phys. , 1982, Vol. 53,
pp. R51-R80.

168. Schroder, Dieter K. Semiconductor material and device characterization. John Wiley and
Sons, 2006.

169. Direct measurement of electron beam induced currents in p-type silicon. Han, M.-G., et al.
Solid-State Electronics, Aug 2010, Vol. 54, pp. 777-780.

271
170. Hontgas, C. H. Investigation of PN junction delineation resolution using electron beam induced
current. PhD thesis, University of Central Florida, 2007.

171. Specimen charging on thin films with one conducting layer: Discussion of physical principles.
Glaeser, R. M. and Downing, K. H. Microscopy and Microanalysis, Dec 2004, Vol. 10, pp. 790-
796.

172. Mechanism for secondary electron dopant contrast in the SEM. Sealy, C. P., Castell, M. R. and
Wilshaw, P. R. Journal of Electron Microscopy, 2000, Vol. 49, pp. 311-321.

173. Energy selective scanning electron microscopy to reduce the effect of contamination layers on
scanning electron microscope dopant mapping. Rodenburg, C., et al. Ultramicroscopy, 2010, Vol.
110, pp. 1185-1191.

174. Material contrast in SEM: Fermi energy and work function effects. Cazaux, J. Ultramicroscopy,
Feb 2010, Vol. 110, pp. 242-253.

175. Static and dynamic charges: Changing perspectives and aims in electron microscopy. Howie, A.
Microscopy and Microanalysis, Dec 2004, Vol. 10, pp. 685-690.

176. Why is it possible to detect doped regions of semiconductors in low voltage SEM: a review and
update. El-Gomati, M., et al. Surf. Interface Anal., Nov 2005, Vol. 37, pp. 901-911.

177. Field-emission SEM imaging of compositional and doping layer semiconductor superlattices.
Perovic, D. D., et al. Ultramicroscopy, 1995, Vol. 58, pp. 104-113.

178. A quantitative model for doping contrast in the scanning electron microscope using calculated
potential distributions and Monte Carlo simulations. Chee, A. K. W., et al. J. Appl. Phys., 2011,
Vol. 109.

179. Effect of experimental parameters on doping contrast of Si p-n junctions in a FEG-SEM.

Kazemian, P., Rodenburg, C. and Humphreys, C. J. Microelectronic Engineering, June 2004, Vol.
73, pp. 948-953.

180. Site-selective dopant profiling of p-n junction specimens in the dual-beam FIB/SEM system.
Chee, K. W. A., et al. Oxford, England : 16th International Conference on Microscopy of
Semiconducting Materials, 2010.

181. Progress towards site-specific dopant profiling in the scanning electron microscope. M. A. E.
Jepson, B. J. Inkson, R. Beanland, A. K. W. Chee, C. J. Humphreys, and C. Rodenburg. Oxford,
ENGLAND : 16th International Conference on Microscopy of Semiconducting Materials, 2010.

182. Ultrashallow Si p+–n junction fabrication by low energy Ga+ focused ion beam implantation.
Steckl, A. J., Mogul, H. C. and Mogren, S. M. 6, J. Vac. Sci. Technol. B, 1990, Vol. 8.

183. Electron holography characterization of ultra shallow junctions in 30-nm-gate-length metal-

oxide-semiconductor field-effect transistors. Ikarashi, N., et al. Japanese Journal of Applied Physics,
2008, Vol. 47, pp. 2365-2368.

184. Electron holography for analysis of deep submicron devices: Present status and challenges.
Ikarashi, N., et al. J. Vac. Sci. Technol. B, 2010, Vol. 28, pp. C1D5-C1D10.

185. Off -Axis Electron Holography for 2D Dopant Profiling of Ultra-Shallow Junctions. Frost, B.
G., Joy, D. C. and Thesen, A. e-J. Surf. Sci. Nanotech., 2004, Vol. 2, p. 119.

186. Electron holographic characterization of electrostatic potential distributions in a transistor

sample fabricated by focused ion beam. Wang, Z. G., et al. Appl. Phys. Lett., 2002, Vol. 80, pp.
246-248.

272
187. Two-dimensional electrical characterization of ultrashallow source/drain extensions for
nanoscale MOSFETs. Singisetti, U., et al. Superlattices and Microstructures, 2003, Vol. 34, pp. 301-
310.

188. El-Kareh, Badih. Fundumentals of semiconductor processing technologies. Kluwer Academic,

1995.

189. Interpretable resolution of 0.2 nm at 100 kV using electron holography. Harscher, A., Lang, G.
and Lichte, H. Ultramicroscopy, 1995, Vol. 58, pp. 79-86.

190. Transmission electron holography of silicon nanospheres with surface oxide layers. Wang, Y.
C., et al. Appl. Phys. Lett., 1997, Vol. 70, p. 1296 .

273