6.2 Indium Phosphide. 164. 6.2.1 Damage Study in Indium Phosphide After ..... Firstly, 3x3 mm pieces are cut from a semiconductor wafer. Then two pieces are ...
The University of New South Wales Faculty of Science and technology School of Materials Science and Engineering
The Study of the Focused Ion Beam Induced Damage in Semiconductor Materials
A Thesis By Sergey Rubanov Submitted in Partial Fulfilment of the Requirements For the Degree of Doctor of Philosophy in Materials Science
2002
ii
ACKNOWLEDGMENTS
I would like firstly to express my warmest thanks to Dr Paul Munroe, my supervisor, for his close and sympathetic guidance during this research work. I am very grateful for his advice, encouragement and arrangement of financial assistance. I am grateful to Dr Anton Gutakovsky from Institute of Semiconductor Physics, Novosibirsk, Russia, who first introduced me to the amazing world of Electron Microscopy. I also wish to acknowledge him for providing a range of A3B5 specimens. The author is very grateful to Dr Yongbai Yin from Quality Semiconductor Australia and Dr Jodie Bradby from the Australian National University for providing silicon and germanium samples for this study. I would like to thank the technical staff of the Electron Microscope Unit, especially Viera Piegrova for technical assistance, and Nicholas Elea for help in solving numerous computer problems. I wish to thank my friend Bob Colvin for assistance with preparation of this manuscript. Finally, I would like to express my deepest gratitude for the endless support of my wife Galina and my son George.
iii
ABSTRACT
The focused ion beam technique (FIB) has been successfully used for TEM sample preparation for some years, especially for silicon, because of this material’s dominant role in the semiconductor industry. The FIB miller allows preparation of site-specific TEM specimens in a wide range of materials in both cross-sectional and plan-view configurations. However, the FIB milling process is accompanied by formation of damage layers in TEM samples which can limit TEM observation. No systematic study of the FIB created damage has been performed before. In the present study the structure, nature and thickness of the damage layers formed during Ga FIB milling were examined using TEM and energy dispersive x-ray spectroscopy (EDS) in a number of elemental (Si, Ge) and compound (InP, InAs and GaAs) semiconductors. It was found that the origin of the damage in the FIB prepared TEM sample is direct radiation damage rather than redepositon of atoms from the specimen. The FIB related damage appeared in two forms – the side-wall (which is typical in cross-sectional TEM specimens) and bottom-wall (which is formed in plan-view specimens). It was found that FIB milling resulted in formation of damage layers with amorphous structure for all possible beam currents for all the studied materials. Partial recrystallisation, in form of microcrystals, was also observed in the bottom-wall damage layers in all materials except GaAs. This partial recrystallisation of the amorphous damage layers was associated with dynamic annealing in these regions assisted by a high concentration of the implanted Ga atoms (above 3 at.%). The bottom-wall damage layers in Ge contained a large number of voids. The formation of these voids was associated with the supersaturation of vacancies in the near surface region and their precipitation in the form of voids due to the local heating of the sample by the ion beam. The thickness of the damage layers depended on the beam energy and was independent of ion beam current. It was also found that implantation of recoil target atoms plays a significant role in the process of the damage formation. Lighter recoil atoms penetrate iv
deeper
than heavier atoms. The thickness of the damage layers in materials with light
atoms (Si, InP) showed much thicker damage layers than with heavy atoms (Ge, GaAs, GaP). The FIB sputter rate for a number of studied materials was found to be much higher than for Si and this is inversely related to surface binding energy. The importance of Au coating to protect the specimen surface from FIB damaging was established. Cross-sectional high-resolution lattice TEM images were recorded from all that studied materials that confirm acceptable quality of FIB prepared specimens.
v
TABLE OF CONTENTS
CERTIFICATE OF ORIGINALITY
ii
ACKNOWLEDGEMENT
iii
ABSTRACT
iv
TABLE OF CONTENTS
vi
LIST OF TABLES
xi
LIST OF APPENDICES (PUBLICATIONS)
xii
PART I INTRODUCTION CHAPTER 1
INTRODUCTION
1
PART II LITERATURE SURVEY CHAPTER 2 LITERATURE SURVEY
4
2.1 Introduction
4
2.2 Transmission Electron Microscopy
5
2.3 Section of Semiconductors
7
2.4 Conventional Ion Thinning
10
2.5 Focused Ion Beam Thinning
12
2.5.1 FIB- an Overview
12
2.5.2 FIB Layout
13
2.5.3 FIB Accessories
17 vi
2.5.4 Focused Ion Beam – Sample Interaction
21
2.5.5 FIB Imaging
22
2.5.6 FIB Detectors
23
2.5.7 Conventional FIB Operation
24
2.6 FIB TEM Specimen Preparation
29
2.6.1 Introduction
29
2.6.2 Cross-Section Sample Preparation
31
2.6.3 Lift-out Technique
34
2.6.4 Plan-view Sample Preparation
36
2.6.5 Damage in FIB Specimens
37
2.7 FIB Related Damage
40
2.7.1 Introduction
40
2.7.2 Features of Ion Implantation
41
2.7.3 Ion Channelling
50
2.7.4 Ion-induced Damage
53
2.7.4.1 Introduction
53
2.7.4.2 Ion Implantation in Silicon
54
2.7.4.3 Annealing Effect
59
2.7.4.4 Ion Implantation in Germanium
62
2.7.4.5 Ion Implantation in III-V Semiconductors
64
vii
2.7.5 Waterfalling Effect
66
2.7.6 Redeposition of Targeting Material
69
2.7.7 Methods of Damage Reduction
71
2.8 Summary
73
PART III EXPERIMENTAL PROCEDURE CHAPTER 3 EXPERIMENTAL PROCEDURE
74
3.1 Introduction
74
3.2 Materials
75
3.3 TEM Investigation
77
3.4 FIB Observation and Milling Parameters
78
3.5 Primary TEM Specimen Preparation
81
3.6 Gold Sputter Coating
83
3.7 FIB Specimen Preparation Procedure
84
3.8 Sample Preparation Procedure for Damage Study
90
PART IV RESULTS AND DISCUSSION CHAPTER 4 SILICON
92
4.1 Introduction
92
4.2 Damage Study in Silicon After a 30 keV FIB Milling
93
4.3 Damage Study in Silicon After a 10 keV FIB Milling
104
4.4 Elemental Analysis of the Damage Layer in Silicon
105
viii
4.5 Study of Redeposition Effect
109
4.6 Slope Angle Measurements
115
4.7 Discussion
117
CHAPTER 5 GERMANIUM
130
5.1 Introduction
130
5.2 Study of Germanium Surface of Bulk Specimens
131
5.3 Damage Study in Germanium After a 30 keV FIB Milling
140
5.4 Damage Study in Germanium After a 10 keV FIB Milling
153
5.5 Determination of the Sputter Rate
155
5.6 Discussion
156
CHAPTER 6 III-V SEMICONDUCTORS
163
6.1 Introduction
163
6.2 Indium Phosphide
164
6.2.1 Damage Study in Indium Phosphide After a 30 keV FIB Milling
164
6.2.2 Damage Study in Indium Phosphide After a 10 keV FIB Milling 6.3 Indium Arsenide
174 177
6.3.1 Damage Study in Indium Arsenide After a 30 keV FIB Milling ix
177
6.3.2 Damage Study in Indium Arsenide After a 10 keV FIB Milling 6.3.3 Protection Properties of Au Sputter Coating Films 6.4 Gallium Arsenide
184 188 191
6.4.1 Damage Study in Gallium Arsenide After a 30 keV FIB Milling
191
6.4.2 Damage Study in Gallium Arsenide After a 10 keV FIB Milling
197
6.5 Sputter Rate Measurements
199
6.6 Discussion
201
PART V SUMMARY AND CONCLUSIONS CHAPTER 7 GENERAL DISCUSSION
211
CHAPTER 8 CONCLUSIONS
215
REFERENCES
218
APPENDICES
226
x
LIST OF TABLES
Table 3.1 Basic semiconductor materials and their fundamental properties.
76
Table 3.2 Beam current, milling spotsize and calculated Ga ion dose for the FEI xP200 workstation.
79
Table 4.1. Slope angle for different ion beam currents at 30 keV beam energy.
116
Table 4.2. Calculated data for the thickness of the damage layer for Ga+ implantation into silicon and silicon self-implantation. In the brackets is the data obtained experimentally.
128
Table 5.1. Calculated data for the thickness of the damage layer for Ga+ implantation into germanium and self-implantation of germanium recoil atoms. In the brackets is the data obtained experimentally.
161
Table 6.1. The experimentally obtained sputter rates for a 30 keV Ga FIB milling of Si, Ge and III-V semiconductors. Table 6.2. Measured thickness of the damage layer in III-V semiconductors.
200 204
Table 6.3. Calculated data for the thickness of the damage layer for Ga+ implantation into III-V semiconductors and implantation of the recoil atoms. In the brackets the data obtained experimentally is shown.
xi
209
LIST OF APPENDICES (PUBLICATIONS)
A. S. Rubanov and P.R. Mumroe, “Investigation of the structure of damage layers in TEM samples prepared using a focused ion beam”, J.Mat.Sci.Let., 20, 1181-1183 (2001). B. S Rubanov and P.R. Munroe, “Redeposition effects during the FIB milling of silicon”, Microsc.Microanal. 7 (Suppl. 2), 956-957 (2001). C. S Rubanov and P.R. Munroe, “The effect of implanted gallium on the recrystallisation of amorphous layers formed during FIB milling of silicon”, Microsc.Microanal. 7 (Suppl. 2), 958-960 (2001). D. P.R. Munroe and S. Rubanov, “The structure of damage layers in transmission electron microscope specimens in elemental and compound semiconductors after FIB fabrication”, Inst. Phys. Conf. Ser. No. 169, 515-518 (2001). E. S. Rubanov and P.R. Munroe, “The Effect of the Gold Sputter Coated Films in Minimising Damage in FIB-produced TEM Specimens”, Materials Letters, 57, 2238-2241 (2003). F. S. Rubanov and P.R. Munroe, “Ion Damage in Germanium Formed Using Gallium Ions in the FIB”, Proceedings of the 15th International Congress on EM, Durban, South Africa, 263-264 (2002).
xii
PART I INTRODUCTION CHAPTER 1 INTRODUCTION The demand for increasingly higher performance semiconductor products has stimulated the semiconductor industry to respond by producing devices with increasingly complex circuitry, more transistors in less space, large number of layers of metal, dielectric and interconnects, more interfaces, and a manufacturing process with nearly a thousand steps. Semiconductor manufacturing has rapidly shifted into a subhalf micron regime. 0.35 μm devices are in production and 0.25 μm technology is ready to be launched [1]. Although there has been no major change in the basic materials used to build these devices, many new process methods must be implemented correctly and process conditions must be controlled tightly in order to distribute the materials to the ideal positions in the right amounts. For future 0.18 mm and 0.13 mm technologies, a transition from aluminium to copper interconnections is expected, which requires implementation of many new processes, such as the use of a tantalum (Ta) barrier and electroplating. Semiconductor manufacturing has never relied so much on precise characterisation of each process step and device performance has never been so sensitive to film quality, interface structure, impurities and contamination. As all device features are being reduced in the quest for higher performance, the role of transmission electron microscopy (TEM) as a characterisation tool takes on a continually increasing importance over lower-resolution characterisation tools, such as scanning electron microscopy (SEM). TEM combined with an energy dispersive x-ray spectrometer (EDX) has great advantages in spatial resolution over the scanning electron microscope with EDX or Auger electron spectroscopy, both of which are widely used as techniques for elemental composition analysis in the field of failure analysis of integrated circuits. The Angstrom scale imaging resolution, nanometer scale chemical analysis and diffraction resolution provided by modern TEMs are particularly well suited for solving materials problems encountered during research, development, production engineering, reliability testing, and failure analysis. A critical enabling technology for the application of TEM to semiconductor-based products as the feature size shrinks below a quarter micron is the advance in specimen preparation. The
1
traditional ~100 nm thick specimen will be unsatisfactory in a growing number of applications. Using a simple geometrical model, it can be shown that the thickness of TEM specimens must shrink as the square root of the feature size reduction [2]. To meet these challenges, control of the specimen preparation process will require a new generation of polishing and ion milling tools that make use of high resolution imaging to control the ion milling process. Argon (Ar) ion milling has been the most popular method for TEM sample during last two or three decades. The focused ion beam (FIB) miller has been applied recently to prepare specimens for TEM. The FIB technique for TEM specimen preparation was first reported about ten years ago [3] and has since then become a reasonably well established method, which is extensively used for microelectronics development and failure analysis. The FIB miller allows preparing site-specific TEM specimens in a wide range of materials in both cross-sectional and planar configurations. There are several advantages in using TEM specimens fabricated by FIB: (1) a specific region with a precision of 0.1 mm range can be characterised with nanometer resolution, (2) multilayered structures which have different sputtering rates are uniformly thinned, (3) the time required for the preparation of a specimen is shorter than conventional argon ion milling, (4) a wide area ~10x10 mm can be observed. However, gallium (Ga)-based FIB systems can produce layers of damage of the order of tens of nanometers in thickness. Damage is manifested in many ways, for example ion beam radiation damage, Ga implantation, and local chemical redeposition. Ultimately this damage can consume an entire thin specimen. The thickness of damaged layers, which is induced on the surface of silicon (Si) by 30 kV Ga FIB milling has been reported to be 20-60 nm [4, 5]. However, this damage does not completely prevent TEM observation and the structure of the undamaged intermediate layer can usually be observed through the damage layers on either side of it. But this damaged layer clearly poses a serious problem for high-resolution electron microscopy (HREM). In the case of compound semiconductors, such as gallium arsenide (GaAs) or indium phosphide (InP), redeposition effects may affect the concentration of elements in the damage layer such that off-stoichiometric layers may exist. Clearly, this would affect any chemical analysis performed on these specimens. The presence of such artificial damage layers around FIB-prepared TEM samples can completely inhibit the acquisition of information about the true crystalline structure of studied materials. 2
This limitation to FIB milling requires a detailed study of the origin, structure and thickness of the damage layers in TEM samples. This study should be undertaken not only for silicon samples, but also for other semiconductor materials. For damage reduction in TEM samples, the parameters and procedures of the FIB milling process should be optimised. A technique should be established which leaves no/or minimum artefacts in the remaining material. The second part of this thesis contains a review of the present state of sectioning of semiconductors, the FIB technique and its application for TEM specimen preparation. The process of defect formation during FIB fabrication from the point of view of the physics of implantation is described. Prior studies of damage in FIB specimens are also described. The experimental part of the thesis contains the results of a detailed study of the origin and structure of damage in silicon samples prepared using FIB. Damage layers were also studied in other semiconductor materials such as Ge, GaAs, InP, and InAs. Results obtained were compared with known data and data obtained from theoretical calculations using the Monte Carlo method. The conclusions summarise the origin and structure of the damage in FIB prepared samples for different semiconductor materials and ways to minimise this damage.
3
PART II LITERAURE SURVEY CHAPTER 2 LITERATURE SURVEY 2.1 Introduction Because of its high special resolution TEM today is very widely used for semiconductor device characterisation. In the section 2.2-2.4 some features of TEM-based analysis and traditional TEM sample preparation techniques will be described. Sections 2.5-2.6 contain descriptions of the focused ion beam technique. This includes detailed description of a standard FIB system, beam-sample interactions and a review of FIB sample preparation procedures. Section 2.7 describes FIB related damage in prepared TEM specimens. The introduction of the damage during FIB specimen preparation is manifested in two ways. The first is related to the implantation of Ga ions into the specimen. The processes of formation of ion-induced damage have been described in sections 2.7.2-2.7.4. The second way of damage formation in FIB prepared samples is redeposition of sputtered material. This process has been described in section 2.7.6. The review of methods for damage minimization in FIB TEM specimens is given in section 2.7.7.
4
2.2 Transmission Electron Microscopy Transmission electron microscopy (TEM) provides a means for the characterisation of materials at high spatial resolution and high contrast through the use of diffraction contrast, micro-chemical analysis and other techniques. These capabilities have led to an increasingly essential role for TEM-based analysis in process development, defect identification, yield improvement and root-cause analysis within the electronics industry. With continuing reductions in semiconductor device dimensions, the high spatial resolution of TEM-based techniques will be required to an even greater extent. In terms of the purpose of TEM analysis, Zhang [1] classified the usage of TEM for semiconductor devices as process evaluation and failure analysis. For process evaluation, a large number of features have to be surveyed to reach a statistical conclusion about current processes, while for failure analysis, precision TEM technique has been applied to arrive at a specific area, such as a single poly gate, contact or via, to find out the root cause of the failure. The resolution of the electron microscope is limited mostly by the coefficient of spherical aberration of the objective lens. Today, the theoretical limit of 0.1 nm for modern microscopes has been achieved. This allows investigation of the structure of materials at an atomic level. Detailed image simulations for Ge [6] and Si [7] confirmed that the “atom-pair” images should be visible over a range of imaging conditions although it was found, for both materials, that these characteristic images could equally well occur near the sites of tunnels in the structure as near the atomic sites. Later results for the imaging of Si at 400 kV indicate [8] that, with the improved transfer characteristics of the newest generation of HREMs, it should be possible to obtain the correct spot separation, but only over a very restricted defocus and specimen thickness range. The high-resolution mode of imaging is useful also for the analysis of amorphous materials even though there is no long range order. An example is a gate dielectric between the single crystal silicon substrate and the polycrystalline gate electrode [1]. The amorphous gate dielectric shows a mottled contrast characteristic of amorphous materials. An important aspect of any TEM investigation is specimen preparation. To achieve optimum results during electron microscopy of semiconductor materials, it is necessary 5
to have clean, well-prepared specimens without significantly altering their microstructure or composition. For direct interpretation of high-resolution images, the specific thickness of the sample must be determined. Although various techniques and devices have been developed to improve the quality and efficiency of TEM sample preparation, it is often still the most difficult part of TEM analysis.
6
2.3 Section of Semiconductors TEM techniques are now commonly employed for investigations of semiconductor materials and devices in a large number of laboratories worldwide. There are many traditional TEM specimen preparation techniques used for semiconductor materials, including: a) Mechanical thinning to electron transparency [9], which is almost entirely dependent upon the abilities of a highly skilled technician. However, this approach may run into difficulty with mechanical support and curling of the foil around the areas of interest when it is subsequently atom-milled to produce high-resolution samples. b) A small-angle cleavage [10] or cleavage along (110) planes [11, 12], which produces samples free of radiation damage, but with a very narrow electrontransparent region. c) Ultramicrotomy [13], which can produce cross-sectional samples of ductile materials, but introduces artifacts such as high dislocation densities, knife marks, bends and tears. This technique is commonly used for biological material and polymers, but not for high modulus metals and semiconductor materials. d) Chemical thinning [14], which is useful for plan-view samples of homogeneous materials, but rarely useful for cross-sectional or heterostructures because of preferential etching. e) Chemo-mechanical polishing [15], which can produce perfect quality artifact-free samples, but is useful only for a limited number of materials, such as mercury telluride (HgTe) and cadmium telluride (CdTe). f) Lithographic and reactive ion etching [16], which can produce samples of selected regions, but requires expensive equipment and processing facilities. It is appropriate only for very difficult TEM samples such as individual devices.
7
g) Argon ion milling [17-19], which works well with many semiconductor samples. This method can be applied successfully to any material. However, preparation before ion milling is a time-consuming process because of the many steps involved. Further, argon milling is slow, it may take many hours to make a specimen. This method will be discussed in detail in section 2.4. In addition to the conventional thinning of plan-view or planar specimens, of particular significance for the study of semiconductors is the cross-sectional method of specimen preparation [19]. This is often necessary because of the complex structure of modern semiconductor materials and devices. Many of them have multilayer structures. Moreover, these layers often have different thicknesses (from delta-layers to microns), chemical compositions, crystallographical structures, resistances and other properties. A plan-view study of such semiconductors does not give exhaustive information about structures of single layers, location of defects, or interface structures. This information can be obtained only by investigation of cross-sectional semiconductor samples. Crosssectional TEM also allows the study of different physical processes to take place during semiconductor growth (e.g. morphological transformation of thin films, relaxation of elastic strain, etc.) [20, 21]. In general, cross-sectional TEM of semiconductors has the following advantages: precise measurement of layer thickness on an atomic level; determination of type and position of structural defects; and ability to investigate interface structure at atomic resolution. The typical process using argon ion milling for preparation of cross-section semiconductors involves many steps and is shown on Figure 2.1.
8
Figure 2.1. A schematic diagram of cross-section sample preparation from a semiconductor device structure [19]. Firstly, 3x3 mm pieces are cut from a semiconductor wafer. Then two pieces are glued together with epoxy resin in such a way that device structures will be inside this “sandwich”. This protects the area of interest from being destroyed during the following steps. Next thin, 100 mm, strips are cut and thinned mechanically from both sides. When the final thickness reaches 20-50 mm these slices are glued onto a 3 mm copper slot grid. Final thinning is carried out by an Ar+ ion milling until a small hole with an electron transparent wedge appears in the sample. Details of the argon ion milling are described in section 2.4.
9
2.4 Conventional Ion Thinning Ion thinning for TEM sample preparation was introduced in the early sixties [22]. Ion thinning today is the most widespread method in the final stage of specimen preparation from many semiconductor materials. Milling is based on the process of ion bombardment of the surface of the sample. The bombarding ions, usually ~ 2-10 kV Ar+, erode the target by ejecting atoms at or just below the surface of the specimen. Standard equipment for ion thinning usually consists of a vacuum chamber, a pair of adjustable ion sources (glow cathode type [22] or saddle-field type [23]), and a sample holder, which is able to rotate (Figure 2.2). The angle of ion bombardment can be changed during the working process by between 2 and 20 degrees. The beam diameter on the surface of the specimen is in the range 1-2 mm. That means that almost the entire surface of the specimen is milled at the same time. If the ion sources are adjusted correctly the highest speed of sputtering should be around the axis of rotation of the specimen. Ideally, the hole with the transparent wedge appears somewhere close to the centre of rotation.
Figure 2.2. Diagram of conventional Ar milling process [24].
10
The ion sputtering is a complex process causing various effects, both direct (e.g., implantation) and indirect (e.g., re-deposit of atoms). Ion beam sputtering can produce thin amorphous surface layers on the surface of the specimen [17]. These layers contribute, during subsequent TEM imaging, to diffuse scattering and the absorption of electrons. The effects of surface topography and damage layers can be minimised by polishing specimens down to a thickness of 20-30 mm prior to ion milling, by the use of shallow ion beam incidence angles (~4°) and by using lower ion energy (< 2.5 kV) in the final stage of milling [18, 19]. Under these conditions the common thickness of damage layers can be typically reduced below 5 nm [17], and its effect on TEM observation can mostly be neglected. Features of modern ion mills are the possibility of near-grazing angles of incidence, increased ion flux at the specimen location (desirable to compensate for low sputtering yield at low angles of incidence), programmable specimen motions, and sensorcontrolled automatic termination. Cooling during milling is not used routinely, but specimen temperatures of the order of 150°C are easily reached [25]. Diffusion, loss of volatiles and phase transformations can be suppressed by using a liquid nitrogen-cooled specimen holder and/or low energy inputs [17, 18].
11
2.5 Focused Ion Beam Thinning 2.5.1 FIB – an Overview In the case of focused ion beams, the beam of high energy ions is focused down to ~10 nm. Usually the beam is perpendicular to the specimen surface during the milling process and is applied locally around the particular area of interest. Even in the case of complex structures with different sputtering yields, the focused ion beam sputters through different phases with different speeds, but this usually does not effect the final membrane (which is parallel to the ion beam propagation) and it is uniform in thickness. The first documentation of ion emission from a liquid metal ion source (LMIS) came from the work of Krohn [26], who was attempting to develop thrusters employing charged metal droplets for use in space flights. From this observation the LMIS was developed and has gone on to find widespread application in the semiconductor and materials analysis field. FIB millers have been produced commercially for approximately ten years [27]. The overwhelming usage of these instruments has been in the semiconductor industry. In this role the FIB has developed as an increasingly important tool in defect analysis, circuit modification, mask repair [28] and for transmission electron microscope sample preparation [3, 29]. FIB provides a means of sectioning and in-situ imaging, which allows images of the substructure of the specimen to be obtained without the need to observe FIB cross sections subsequently through the scanning electron microscope. Furthermore, the use of gases in conjunction with the ion beam allows for preferential etching of materials [30] and also deposition of both metals and insulators for circuit repair [31, 32]. However, despite their attraction, FIB units are mostly found in the semiconductor industry and are relatively rare at universities especially outside of the USA. In this chapter the principles of FIB and its applications for TEM sample preparation will be described.
12
2.5.2 FIB Layout In many ways FIB can be compared to a scanning electron microscope (SEM) [33]. Both have a source that emits charged particles in a high vacuum environment. These particles are then focused onto the sample surface by lenses. This beam may be rastered across the surface of the sample for imaging (like a SEM) or held stationary for milling or analytical purposes. However, because of the nature of the ion beam, specimen-beam interactions are manifested in many different ways to an electron beam. In comparison to the SEM there are a number of important differences between the two instrument types, for example electromagnetic lenses are not of sufficient strength to focus the heavy ion beam and so electrostatic lenses are used in their place. As stated earlier, the configuration of an FIB is similar to that of an SEM. However, there are some major differences that are evident in the description below. A generalised cross- section of an FIB column can be seen in Figure 2.3.
Figure 2.3. Schematic diagram of FIB column [27]. 13
Ions are generated by a liquid metal ion source, which consists of a tungsten needle that is mounted below a liquid gallium reservoir (Figure 2.4). The LMIS provides a beam of very high brightness. A high voltage is applied to the source while the sample is at ground. Hence, ions may be accelerated down the column. The higher the acceleration voltage the faster the ions travel down the column and the greater the kinetic energy of the beam. Acceleration voltages in the current generation of FIBs may be as high as 50 kV.
Figure 2.4. Schematic diagram of LMIS [27]. The tip of the needle, which is coated in a thin film of gallium (Figure 2.4), is situated just above the extractor (Figure 2.3). The extractor is held at a higher voltage (typically ~6 kV) relative to the source, and this produces an intense electric field at the source tip. The field ionises the gallium and draws the liquid metal into a fine tip, called a Taylor cone. A tiny cusp with an end radius of about 2 nm protrudes from the end of the cone. Ion emission due to field evaporation occurs at this incipient jet and the gallium ions are emitted from the tip and accelerated down the column. The current emitted from this tip can be measured using a Faraday cup and is known as the extraction current. The suppressor uses an applied electric field (in the range of 0 to 2000 V) to work in conjunction with the extractor to maintain a constant beam current, which is important during controlled milling. Increasing or decreasing the suppressor decreases or increases 14
the extraction current; hence the extraction current may be adjusted without changing the extractor voltage [27]. Gallium is a large atom, and its wavelength when accelerated is very long. Because the force exerted by a magnetic field on a moving particle is proportional to the mass of that particle (m) divided by its charge (q), a column using magnetic lenses would have to be inordinately long to attain the focus conditions necessary for efficient FIB operation. Electrostatic lenses exert fields proportional only to the particle’s charge (q), and these field strengths can be realised with relative ease. Focusing is accomplished due to the fact that when a charged particle passes through an electrostatic field, it is deflected. The deflection is a function of the particle charge and the field potential in that area of the lens through which the particle is travelling. Consequently, ions may be brought to focus using one or more of these lenses. This is analogous to the deflection of electrons by electromagnetic lenses in the SEM or TEM. Lenses may be run in collimated or crossover mode. While it is common to run SEMs in crossover mode, FIBs are more often operated in collimated mode, where ion trajectories are parallel between lenses. A collimated mode is less subject to spherical and chromatic aberrations, but beam currents and ultimate resolution are determined by the placement of a beam limiting aperture in the beam path (usually between lens 1 and lens 2). On the other hand, operating in a collimated mode means that the images often require stigmation correction. In addition to electrostatic lenses, octopole deflectors allow for stigmation correction and beam deflection for rastering and beam blanking. Flexibility in rastering, e.g., changes in dwell time, pixel spacing, and refresh time, is very important in FIB technology as different milling operations need different beam steering conditions. In practical terms, the most important part of the column is the beam-limiting aperture. This aperture is used to control the beam diameter and the beam current incident on the sample and thus the milling rate and/or image quality. In practice, an FIB contains a wide range of apertures on a horizontal sliding bar, and the chosen aperture is brought into position by mechanical drives selected from the operating software. Apertures range in diameter from a few microns to several hundred microns. The associated beam current varies from a few pA to several nA. Small apertures minimise the beam diameter and control the beam current to a few pA. This usually means that the image quality is high (although the signal-to-noise ratio is often so low that slow scan speeds 15
are required to produce high quality images) but sputtering rates are very low. Conversely, the bigger apertures permit beam currents of several nA to interact with the specimen. Using these apertures, milling is fast - often fast enough to mill a feature before it is adequately focused. However, the beam is broad and image quality is poor [27]. Beam blanking is also extremely useful as it is not always desirable to have the ion beam impinging on the sample, due to the fact that this beam constantly removes material when interacting with the sample surface. It is also desirable to be able to have a method of monitoring beam current, and the beam may be deflected to a beam blanking aperture to accomplish both of these purposes [27].
16
2.5.3 FIB Accessories The FIB column contains a number of non-essential attachments which increase the flexibility of the instrument. These include metal deposition, gas-assisted etching and charge neutralisation. Ion-induced deposition is a process whereby gas molecules absorbed onto a sample surface may be dissociated in the presence of the ion beam to form films of material that will deposit and adhere to the substrate. Deposition may be metallic (conductive) or dielectric (insulating). Metallic films may be gold (Au), tungsten (W), platinum (Pt) or copper (Cu), whereas insulating films are generally silicon oxide (SiO 2 ) or carbon (C) based [27]. Most ion-induced metal deposition processes use an organometallic gas that requires low activation energy to cause decomposition. As the ion beam not only dissociates the gas and causes deposition, but also has the ability to etch the substrate, the operating parameters of gas flow, beam diameter and other factors such as dwell time and pixel spacing are most important in this process. The organometallic gas used to deposit W, for example, is tungsten carbonyl-W(CO) 6 . This gas is delivered through a hollow needle situated approximately 100 µm above the surface of the sample. Bulk resistivity of this deposited material (with an atomic composition of approximately 70% W and a 30% mixture of Ga, C, and oxygen (O)) is, on average, two orders of magnitude greater than pure W, but this is normally sufficient for most applications. These include repairs of metal lines, modifications using holes or straps, and reconnections of previously severed lines. These metal straps are often used in TEM preparation [29, 34, 35], which will be in described in detail in section 2.6.2. Figure 2.5 shows some platinum strips deposited on an aluminium substrate.
17
Figure 2.5. Platinum straps deposited on an aluminium substrate [27]. In the FIB, the cumulative effects of both the positive ions entering the specimen surface, together with the negative electrons being ejected, promote charging effects. That is, there will be a large build up of positive charge in the specimen surface, and so charging effects will occur. Ultimately, this may lead to electrostatic damage in semiconductors or deflection of the ion beam. This deflection often means that milling occurs away from regions selected for milling. This can be especially problematic for TEM sample preparation. Such charging effects can be overcome by a process called 'charge neutralisation' (see Figure 2.6). A tungsten hairpin gun (often called the flood gun) is placed close to the specimen surface. The flood gun is heated, but no accelerating voltage is applied, and this emits low energy electrons, which are directed by electrostatic lenses onto the specimen surface. This supply of low energy electrons interacts with the positive charge built up in the surface of the specimen providing an antidote to charging effects. Charge neutralisation therefore allows uncoated insulators to be imaged without the charging effects normally associated with these samples. Furthermore, charge neutralisation may increase the accuracy of any milling which takes place, since the removal of any positive charge in the specimen surface limits unwanted ion beam deflection or drift (especially important in TEM preparation).
18
Figure 2.6. Schematic Diagram of Flood Gun [27]. Gas assisted etching (GAE) (also called Enhanced Etching (EE)) is widely used during circuit modification and is finding increasing usage in both TEM preparation and materials science applications [27, 30]. If a reactive gas, such as chlorine or iodine, is released near the sample surface, a monolayer of this gas will then be adsorbed onto this surface. If this surface is subsequently scanned with the ion beam, the energy of the beam breaks the molecular bond of, for example, the chlorine molecule. The chlorine ions that are formed are powerful etching agents for a number of materials. Another benefit of this is that the sputtered particles tend to be volatilised and pumped out of the system, hence lessening the likelihood of material being redeposited, especially after the drilling of deep holes. The use of the correct reactive gas can increase the rate of removal of materials up to 25 times that of etching alone. However, the most important feature of GAE is its selectivity. Consequently, if the highly selective gas etches one layer 10 to 20 times faster than another material it is much easier to remove the latter without doing significant damage to the former. A combination of GAE and selective milling using beam blanking is very effective in the exposure of deeply buried, multilayered circuitry, where it is necessary that the top layers of metal remain intact [27]. Some FIBs are dual beam instruments [36-39]; they contain a conventional electron gun, in addition to the ion source, and can allow the FIB to be used like a normal SEM. The attraction of this is that the interaction of electrons with the specimen yields 19
characteristic X-rays and this allows energy dispersive x-ray (EDS) analysis to be performed on a freshly milled surface – or without removing the sample, taking it to another instrument, finding the trench again and then imaging and analysing. Another attraction of the dual beam instrument is that the electron beam does not mill or damage the sample during location of the area of interest or mill set up [37]. High vacuum is essential for successful operation of FIB machines and the column vacuum is in the order of 1 x 10-7torr. Although column pressure may lower during deposition or gas assisted etching, it is important that an adequate base level pressure be maintained otherwise the source will become unstable and cease to emit. Larger and larger specimen chambers are increasing the demand on pumps, which are needed to maintain chamber vacuum. Pumping systems are generally, but not necessarily, dry (e.g. ion and turbo pumps) due to the need for contamination-free environments in the semiconductor industry.
20
2.5.4 Focused Ion Beam - Sample Interactions When a gallium ion is accelerated by an energy of between 5 and 50 kV, the energy is dissipated and the incident ion may implant into the specimen surface. This in turn causes the ejection of secondary electrons, secondary ions, and atoms. Atoms are displaced within the material causing lattice damage, and phonons are dissipated as heat. This process thus causes removal of material from the sample surface in the form of atoms (or neutrals), positive ions and negative ions. This process is generally known as sputtering [18]. The yield of secondary electrons to secondary ions is far greater due to the large difference in mass between these species. Neutrals are the least likely species to be generated and are not detected or used in imaging. As stated earlier, FIB systems are capable of high quality imaging, with a resolution in the region of 5 to 10 nm. This is equivalent to a reasonable tungsten or LaB 6 filament in the SEM. Image contrast mechanisms show similarities to those in the SEM, but there are fundamental differences in mechanisms between these instruments [27].
21
2.5.5 FIB Imaging Ion induced secondary electron emission results from the transfer of ion kinetic energy to electrons inside the target, either by direct inelastic collision or by recoil atoms generated in a collision cascade, and depends on many parameters. In the case of a probe with fixed energy and species, the pertinent parameters are the primary beam dose and incidence angle, and specimen mass, atomic number, composition, crystallographic structure and orientation and electrical properties. The emitted electron spectra show a characteristic intensity maximum between 1-3 eV with a low intensity tail that may extend to energies of about 50 eV. The position of the maximum for any given material is relatively insensitive to incident-ion energy. Secondary electron emission is sensitive to the presence or absence of surface oxidation states. Clean metals have the lowest electron yields, oxidised metals have intermediate yields, and insulators have the highest yields. Electron yields are highly dependent on crystal orientation and consideration of primary ion path length compared to electron escape depth is necessary. Secondary ion emission depends upon two phenomena, sputtering and ionization. Sputtering is caused by energy transfer from the incident ion to the target sample by an elastic collision cascade, resulting in atom emission from the top one or two monolayers of the material. Ionisation of sputtered atoms is complex and very sensitive to conditions at the target surface, and is thought to occur less than 1 nm from the target surface after atom emission. The characteristics of secondary ion emission are very similar to those of secondary electron imaging, but there are some notable differences. Secondary ion spectra have a maximum appearance between 5-10 eV with a tail extending out as far as 500-1000 eV. The sputtering yield for channelled ions is decreased similarly to that for electron yields, where the secondary ion emission differs greatly from the secondary electron emission in the range of yield values. Electron yields may vary slightly more than an order of magnitude, whereas ion yields from elemental targets undergoing bombardment by the same ion beam can vary by five or more orders of magnitude.
22
2.5.6 FIB Detectors A variety of detectors may be used in the detection of secondary ion and secondary electron signals. However, since the normally detected signals (ions or electrons) have different charges, it is necessary to use detectors that can be biased differently to accept these different signals. FIBs use either Microchannel plates (MCPs) or Continuous Dynode Electron Multiplier (CDM) detectors. These offer significant advantages for FIB imaging in that both positive and negative biases may be applied to the detector, and hence one unit may be used for detecting both electrons and ions.
23
2.5.7 Conventional FIB Operation The FIB is operated in much the same way that a conventional SEM is used. The sample is placed in the specimen chamber and “normal” SEM-type stubs can be used. A beam is focused onto the sample and rastered across the surface. An electron or ion signal is detected and an image is produced. Common controls are brightness, contrast, focus, magnification, stigmators and X-Y shifts. Conventional imaging is normally performed using a smaller aperture, where the beam current is small and minimal milling occurs. However, even during imaging with the smaller apertures, some milling/damage formation will still occur [27]. When an area for milling is located, a larger aperture is inserted (where the beam current is increased by 2-3 orders of magnitude). A “mill box” (usually rectangular) is then drawn onto the area of interest for milling (this is commonly done with a mouse). The beam then scans over this selected area and milling is achieved (Figure 2.7). In most FIB systems the ion beam is scanned digitally from left-to-right line-by-line from top-to-bottom of the mill box. The setting parameters of the mill are: high voltage, ion beam current (which defines a beam diameter), beam spot overlap with the neighbouring spot, dwell time of each point, size of the scanned area and required depth (Figure 2.7). The rate of milling (the sputter rate) is, of course, dependent upon the material itself. Strongly bonded materials may mill more slowly than weakly bonded ones. However, work performed on composite materials suggests that these differences are not large [27, 36, 40, 41]. The milling parameters for some common materials are stored in a so-called ‘material’ files in the FIB workstation computer software. The system uses these parameters to calculate a mill time when the pattern is defined.
24
Figure 2.7. Schematic diagram of FIB milling process. There is a range of different milling options available. These options include lines, rectangular boxes, polygonal boxes and circles (see Figure 2.8). In these modes the beam rasters rapidly over the surface of the sample, and, for example, parallel-sided mill boxes are produced.
Figure 2.8. Range of possible milling types [27].
25
Alternatively, a range of more complex mill options is available. These produce stepped trenches, which allow a single clean face to be achieved (Figure 2.9). These include options for finer, subtler mills, which allow these faces to be cleaned using lower beam currents to remove some milling artefacts. These modes are commonly used when FIB is used to reveal substructure from bulk samples. In this instance the mill is performed, then usually subjected to a clean-up mill, and the sample is then tilted about 45° to reveal the cross-section structure on the face. Cross sectioning gives the ability to proceed to a specific site on a sample, cut through a feature at that site and view the subsequent side wall to observe the buried structure. This technique is used to analyse defects in circuits or study different process steps. For study of specimen structure with higher magnification/resolution cross-sections should be prepared using TEM sample configurations. The process of TEM sample preparation will be described in detail in section 2.6.
Figure 2.9. Cross sections achieved through stepped mills [27]. Ideally, the sidewalls of an FIB milled hole should be vertical; however, it has been generally observed that the trenches tend to exhibit a gradual sloping [42, 43]. This is the result of the current-density profile of the focused ion beam. This is generally acknowledged to consist of a central, approximately Gaussian-shaped peak,
26
accompanied by broader “tails”[44, 45]. The sidewall profile inherits the shape of beam, as shown in Figure 2.10.
Figure 2.10. (a) SEM images of the end-edge view of a single pixel width line FIB mill; (b) Fitting the trenches’ cross-section profile with the Gaussian function [45]. Since the introduction of FIB specimen preparation it has been realised that the thinned slice has a slope profile, which is inherent in the beam profile [3, 46]. Rather large tilt angles for both sides have been reported, e.g,. 4-5o for GaAs by Yamaguchi et al [47] and for Si: 2o [48], 2.5o [42], 3o [49], 0.5-2o [50]. Based on modelling, an optimal tilt correction of 6o was found for W [51]. On the other hand it has been shown
27
experimentally that in open structures a slope of 1o can be obtained with a low beam current of 6 pA and 4o for a high beam current of 7000 pA [52]. The slope profile of the final transparent membrane is common for most other sample preparation techniques, which were described in section 2.3. The slope of such small range of angles (2-6o) does not significantly reduce the quality of TEM samples. But in the case of FIB sample preparation it can be completely eliminated. The specimen tilting during the final preparation was introduced to compensate this slope [47]. This method will be described in section 2.6.2.
28
2.6 TEM Specimen Preparation 2.6.1 Introduction TEM samples can be prepared readily, precisely and rapidly using the FIB [27]. Further, this mode of preparation is ideally suited to difficult materials, for example heterogenous materials [53], layered structures [36], surface-treated materials [54] and powders [55]. This is because the ion beam is able to cut through heterogenous materials relatively uniformly and it can be directed with extreme precision on to the specific area of interest. Overall, this leads to orders of magnitude increases in both sample preparation time and positional accuracy compared with conventional argon ion milling. Using the FIB, TEM specimens can be prepared in both cross-section and plan-view configuration. In the first case the final thin electron transparent membrane is normal to the specimen surface (Figure 2.11a). Cross-sectional configuration allows studying the structure of a specimen from its outer surface inwards: the structures of single layer, the location of defects and interface structures may be observed. In plan-view sample configuration, the final membrane is parallel to the surface of the specimen (Figure 2.11b). Information can be obtained about the structure of the specimen in the plane of the specimen surface.
Figure 2.11. Schematic diagram of FIB TEM sample preparation process in crosssection (a) and plan-view (b) configuration.
29
The following sections will describe both cross-section and plan-view sample preparation techniques. A relatively new method – the so-called “Lift-out technique” will be also discussed. Damage in FIB-prepared TEM samples will be described and discussed.
30
2.6.2 Cross-section TEM Sample Preparation The standard milling procedure as initially outlined by Kirk et al. [3] and subsequently further discussed by e.g. Park [46], Young et al [29] and Basile et al [56] is still the most used FIB preparation method. A detailed procedure was later discussed by Su et al [57]. Clearly, precise preparation and milling conditions vary from sample to sample. Described below is the most common methodology used to prepare a specimen from, for example, a silicon sample [5, 58]. A thin strip of silicon is cut from a substrate directly by sawing to a thickness of ~< 50 mm, by thicker sawing combined with shallow sawing along the region of interest by using a diamond dicing saw [40, 57] (Figure 2.12), or by a combination of sawing and grinding or tripod polishing [5]. In the case of using a dicing saw the first two shallow cuts (10 mm) were made from either side of the area of interest. The last cuts were then made through the entire thickness of the sample [40]. This method allows eliminating mechanical grinding of the sample.
31
Figure 2.12. Schematic diagram of mechanical sample preparation using the dicing saw [40]. Subsequently, the strip is mounted using strong glue on a half of standard TEM slot Cu grid (Figure 2.11a). During the following FIB procedures the Ga+ ion beam is focused orthogonally on the top surface of the specimen. Before milling, the surface of the region of interest is normally protected by an in-situ deposited metal layer (Pt or W). The strip of Pt, typically 20µm x 2µm x 1µm, provides increased stability for the thin section to be milled, and prevents the top of the sample being milled away. Then, a set of rectangular trenches is milled step-wise from both sides of the Pt strip with decreasing beam currents so that a thin membrane of 50-200 nm in thickness remains (Figure 2.11a). The total transparent area is generally of the order of 10x5 to 20x10 mm2. To speed up the milling of large trenches and to reduce possible redeposition, some workers [59, 60] applied gas-assisted FIB etching with a halogen gas (I 2 or Cl 2 ). 32
In order to reduce the slope of the thinned membrane, Yamaguchi et al and Tsujimoto et al [47, 53] tilted the sample 4-5o off from the beam direction.
Figure 2.13. FIB fabrication procedure. (a) Ion beam set vertical. (b) Ion beam set 5o off from vertical. (c) Ion beam set –4ooff from vertical and at the same time, FIB scanning direction tilted 3o [59]. Initially FIB milling was executed until the thickness of the slice reached less than 100 nm (Figure 2.13a). Next, the specimen was tilted mechanically 5o off from the vertical to the specimen surface. Using a beam a with 100 pA current, a sloped area was removed from one side of membrane (Figure 2.13b). In the final step (Figure 2.13c), the incident angle of the ion beam was set to –4o off the specimen normal, and simultaneously the ion beam scanning direction was tilted 3o in order to create a wedgeshaped slice. Finally, a very uniform thin membrane was obtained [59].
33
2.6.3 Lift-out Technique The lift-out technique for FIB TEM sample preparation was developed by Overwijk et al [61] and has become popular in the last few years [37, 62-68]. It has the major advantage that no sawing and grinding pre-thinning steps are involved so that preparation can be done directly on device wafers or on packaged devices. The preparation involves: the milling of two stair-step craters near the region of interest (Figure 2.14a); tilting the sample once the thinned slice is 0.5-1 mm thick and then cutting the bottom and one side fully and the other side partially (Figure 2.14b); further thinning the slice from the top from both sides of a membrane to the required thickness (Figure 2.14c); and finally cutting the partially cut side from the top (Figure 2.14d).
Figure 2.14. A schematic diagram of the FIB lift-out technique [34]. The free slice will then fall down in the milled crater and must be picked up with a fine needle (Figure 2.15) and placed on an amorphous C film, which is deposited on a TEM-
34
grid. For the pick-up, metal (W) [61, 64] or glass rods [62, 63, 65] are used.
Figure 2.15. A schematic diagram illustrating lift-out of the thin electron transparent membrane from the bulk sample [55].
Figure 2.16. The final sample configuration for lift-out technique. The sample is placed on a TEM mesh grid. Whereas high humidity is favourable for the metal needles, dry conditions are required for the glass rods as adhesion in their case is based on electrostatic attraction. Manipulation with such small objects requires very skilled operators. As well, some probability exists that a mesh of the TEM-grid will hide the transparent part of the specimen (Figure 2.16). These circumstances reduce the attraction of the lift-out technique. 35
2.6.4 Plan-view TEM Sample Preparation The majority of the TEM investigations of device structures have involved crosssectional preparation. With the increase in the number of metallisation layers, there is also a growing demand for the study of the device structures at a specific level in planview [5]. Young et al [29] first reported the use of FIB milling for plan-view preparation of GaAs. They used Cl 2 gas-assisted milling from the rear side through a 150 mm thick substrate. Later Klepeis et al [69] applied the usual FIB milling procedures from the rear side of silicon device structures for plan-view specimen preparation. They marked the area of interest with deep cuts, ground specimens from the rear side down to 3-4 mm and milled by rastering a 20-50 mm square region, centred on the region of interest delineated by the FIB marks cut from the top side. The specimen became electron transparent when there was an abrupt change in contrast within the square raster and/or the top side features became visible. Leslie et al. [48], Anderson and Klepeis [70] and Stevie et al. [63] discussed another plan-view preparation method, which involved polishing from the backside, and FIB milling from the sidewall (Figure 2.11b). In this way the thinning can be done at any required depth in the structure. It can be combined with the pick-up method [63, 71] or with conventional milling [48].
36
2.6.5 Damage in FIB Specimens Damage introduced by the Ga+ beam in the thinned membrane (Figure 2.17) has always been a major concern. Although beam-induced lattice defects have not been reported, the amorphous layer formed on the sidewalls reduces the quality of high-resolution imaging and limits the minimal useful specimen thickness [5, 72]. A schematic diagram of the damage formed in TEM sample during FIB fabrication is shown on Figure 2.17.
Figure 2.17. Diagram of FIB fabricated specimen (a) whole view, (b) magnified image with indication of wall-side damage layers and the undamaged area in the middle [60]. It is clear that the final membrane consists of an undamaged area, which represents the initial structure of the sample, and two damage layers on its both sides. During the TEM observation an electron beam, which passes through such a sandwich (that is the undamaged layer + two damage layers on both sides of membrane), contains information about the entire thickness of the membrane. It can be noted from Figure 2.17 that in case of a very thin membrane the damage can occupy the total thickness of the membrane. Thus the TEM study of the initial structure of the specimen is not possible. If damage layers are amorphous and not occupy the total thickness of the specimen (the undamaged layer exists), the TEM imaging through these damage layers will reveal the initial structure of the specimen. But the quality of TEM images will strongly depend on the thickness of the damage layer. The thicker the amorphous damage layer the more electrons will be absorbed and scattered in arbitrary directions. These will lead to a rise in intensity of background noise on TEM images and 37
significant contrast reduction [17]. If the damage is not amorphous, but contains some other FIB-produced artefacts, it can also result in additional difficulties in interpretation of TEM images. Such artefacts can be confused with the initial structure of the specimen [47]. A wide range of thicknesses, as determined by various experimental techniques, has been reported over the years for the damage layer induced by the 30 kV Ga+ ion beam. For example for Si: 20 nm [60], 25 nm [49], 28 nm [2, 73]. Kato et al. [60] have found using TEM investigation that the FIB-induced side-wall damage in silicon was about 20 nm thick irrespective of the beam current used in the final step of 30 keV FIB milling. Pantel et al. [49] have found also using TEM a 25 nm thick amorphous damage layer after 30 keV Ga+ FIB fabrication. Mardinly [2] et al. and Venables et al. [73] have reported the thickness of the damage amorphous layer to be about 28 nm. Walker and Broom [74] have studied a damage layer sandwiched between a platinum protective stripe and a silicon substrate. Using TEM cross-sectional investigation they found the thickness of amorphous damage layer 28 nm for 30 keV gallium ions and 6 nm for 10 keV gallium ions. Bender and Roussel [35] observed a damage layer in the silicon substrate, amorphised to a depth of approximately 60 nm with a dark sub-layer occurring in the top of this amorphus layer. This damage amorphous layer was situated between the undamaged silicon substrate and the platinum protective layer. The authors assumed that this dark sub-layer was associated with knocked-in Pt atoms during deposition of a protection Pt strip. Also, the damage depths calculated by Monte Carlo simulation for 30 keV gallium ions in silicon samples (angle of incidence 85o) vary widely: 10 nm [42], 15 nm [75], and 30 nm [2]. It is difficult to understand the difference in these calculated data because the authors in each case do not provide details of the calculations. For III-V materials few experimental data are reported for the thickness of the damage layers as well: 24 nm (30 keV FIB), 4 nm (10 keV) [74] for gallium arsenide and 31 nm (30 keV FIB) [59]; 40 nm (30 keV), 15 nm (10 keV) [74] for indium phosphide.
38
Such a wide range of experimental and calculated data for FIB induced damage layer thickness requires more careful investigation. It is necessary to find out the origin of damage formation (radiation damage, redeposition, or combination of both effects?) The lack of data for other semiconductor materials (InAs, Ge, etc.) should stimulate further experiments with these materials. The next section of this thesis will contain more detailed description of the processes of damage formation during FIB fabrication.
39
2.7 FIB Related Damage 2.7.1 Introduction During FIB specimen fabrication highly energetic Ga+ ions interact with the surface of the target material. An interaction of these ions with the atoms of the target causes the displacement or complete removal of atoms from the target. The process is known as sputtering or milling. The speed of milling depends on the type of ions and targeting atoms, the energy of the ions and the beam current density. The milling process is accompanied by the formation of damage layer near the milling surface and by surface relief modification caused by non-uniformed sputtering (known as the “waterfalling” effect). An investigation of damage formation and its minimisation during FIB preparation of TEM samples from semiconductor materials is the goal of this thesis. The following sections contain a review of physical processes during an ion-target interaction. This includes the physics of ion implantation, ion channelling, damage formation and redeposition of targeting material. The surface topographic artefacts such as the waterfalling effect will also be described. The final section of literature survey contains a review of methods of damage reduction during FIB TEM specimen fabrication from semiconductor materials.
40
2.7.2 Features of Ion Implantation In passing through a solid, ions interact with electrons in the solid, thereby losing energy. They can also collide with the nuclei of the atoms in the solid. Eventually, they come to rest within the solid after some distance R (termed the range). Because of collisions which occur the ion paths are not straight and thus R can have different values (Figure 2.18). The projection of R in the implantation direction is the more meaningful parameter, because it determines the implant depth and is referred to as the projected range, R p . Some ions will collide less often than average, coming to rest further than R p , whereas some will collide more and stop short of R p . The statistical fluctuations in the ion concentration along the projected range are referred to as the projected straggle (∆R p ). Ions are also scattered perpendicular to the incident direction. The resulting fluctuations in the ion concentrations in the transverse direction are called the projected transverse or lateral straggle (∆R⊥). The distribution of ions about depth R p can be approximated as Gaussian with standard deviations ∆R p and ∆R⊥ [76].
41
Figure 2.18. Schematic views of the ion range (a). The total path length R is longer than the projected range R p . (b) The stopped atom distribution is two-dimensional Gaussian. [76]. Figure 2.19 shows the projected ranges for boron (B), phosphorus (P), and arsenic (As) ions in amorphous silicon and silicon oxide. For a given energy the lighter ion has a longer range than the heavier ion. The calculated values of ∆R p and ∆R⊥ for B, P, and As ions in Si are shown in Figure 2.20. It is evident that these values follow the same dependence on the ion mass as depicted in Figure 2.19.
42
Figure 2.19. Projected range for B, P and As in Si and SiO 2 at various energies. The results pertain to amorphous silicon targets and silicon oxide [77].
43
Figure 2.20. Calculated ion projected straggle ∆R P and ion lateral struggle ∆R⊥ for B, P and As ions in Si [77]. Incident ions are slowed by nuclear collisions and coulombic interactions with the electrons. Assuming that the energy losses due to the two mechanisms are independent of each other and are additive, Lindhard, Scharft, and Schiott (LSS) have developed a theory for determining the range of an ion [78].
The energy loss per unit distance due
to nuclear and electronic collisions is given by
dE dE dE = + dx tot dx nucl dx el
(1)
where the nuclear and electronic losses depend on the ion energy. The range of the ions R, is given by
44
E
R( E ) =
E
1 dE dE ∫0 dE = N ∫0 S ( E ) dx tot
(2)
where E is energy of the incident ion, N is the number density of target atoms, and S(E) is the stopping power of the solid. If S n (E) and S e (E) are, respectively, energy losses per unit length due to nuclear collisions and coulombic interactions, then S(E)= S n (E)+S e (E).
Figure 2.21. Schematic view of ion scattering, showing the relation between impact parameter and scattering cross section [79]. A description of the scattering process by a nucleus can be provided by referring to Figure 2.21. Assume that an incoming ion, having energy E 1 and mass M 1, collides with target atoms whose mass is M 2 . The probability of having an impact parameter (p) between p and p+dp is 2ppdp, also known as the differential scattering cross section dσ. As a result of the collision, the incoming ion is scattered through an angle ϑ and the target atom is displaced from its equilibrium position as shown. The energy transferred (T) to the target atom is given by
45
T=
4M 1 M 2 θ E1 sin 2 2 (M 1 + M 2 ) 2
(3)
The energy transferred is maximum for a head-on collision (ϑ=180°). The scattering angle (ϑ) can be obtained by integrating the equation of motion for the scattering trajectory. The nuclear energy loss is given by
dE = N ∫ Tdσ =NS n ( E ) dx nucl 0 E
(4)
where dσ is the differential cross section. LSS have introduced a number of simplifications to make the above integration tractable. Using their approach, the values of (dE/dx) nucl =S n (E) for the implantation of As, P, and B into Si have been computed by Smith [77], and they are shown in Figure 2.22. In addition, the electronic losses are proportional to the velocity of the ion, i.e. √E,
dE = ke E dx el
(5)
The proportionality coefficient (k e ) is a relatively weak function of M 1 and M 2 and the atomic numbers of the incident ion and the stopping atom. The computed values of (dE/dx) el =S e (E) for the implantation of As, P, and B into Si are also shown in Figure 2.22. Note that in Figure 2.22, S n (E) increases with the mass of the implanted ion. Thus, heavy ions will transfer much more of their energy through nuclear collisions than light ions. Also, for B, S e (E) is the dominant loss mechanism over the whole energy range, while for P and As ions, S n (E) dominates for energies up to 130 and 700 keV, respectively.
46
Figure 2.22. Calculated values of dE/dx for As, P and B at various energies. The nuclear (N) and electronic components are shown [77]. The concentration profile of the implanted ions in an amorphous solid is given by [80]
N ( x) =
1 x − R 2 D0 p exp − 2 ∆R p 2p ∆R
(6)
where N(x) is the impurity concentration, D 0 is the dose (ions/cm2), x is the distance from the surface in centimetres, R p is the projected range in centimetres, and ∆R p is straggle in centimetres. The above equations ignore the effects of the transverse straggle (∆R ⊥ ) . This omission may introduce some error in determining the concentration of ions near the edges of a mask. The implantation dose D 0 is given by
D0 =
1 Idt QA ∫
(7)
47
where the integral is over time t, Q is the charge of ions, I is beam current and A the area of implantation. The profile of the ion concentration, as given by Eq. 6, is schematically shown in Figure 2.18b. The peak concentration is at R p and falls off symmetrically on either side of R p . With some dopants, considerable deviations from the Gaussian profile have been observed. These observations cannot be explained using Eq. 6, because it is based on the simple range theory. Biersack and Haggmark [81] and later Ziegler et al [82] developed a group of programs, which calculated the stopping range of ions (SRIM #) into matter using a quantum mechanical treatment of ion-atom collisions. This calculation is made very efficient by the use of statistical algorithms (Monte Carlo method), which allow the ion to make jumps between calculated collisions and then average the collision results over the intervening gap. Figure 2.23 shows results of the Monte Carlo calculation for 30 keV Ga ions trajectories in a silicon target.
Figure 2.23. Trajectory stimulations of 30 kev Ga ions in a silicon target using Monte Carlo methods [40]. #
see: www.research.ibm.com/ionbeams 48
During the collisions, the ion and atom have a screened Coulomb collision, including exchange and correlation interactions between the overlapping electron shells. The ion has long range interactions creating electron excitations and plasmons within the target. A second program, TRIM (the Transport of Ions in Matter), will accept complex targets made of compound materials with up to eight layers, each of different materials. It calculates both the final 3D distribution of the ions and also all kinetic phenomena associated with the energy loss from the ion: target damage, sputtering, ionisation, and phonon production. All target atom cascades in the target are followed in detail. However, in both programs, the target is considered amorphous with atoms at random locations, and thus the directional properties of the crystal lattice are ignored [82]. So, SRIM and TRIM are limited to “amorphous” target materials.
49
2.7.3 Ion Channelling Atoms of amorphous solids do not exhibit long range positional correlations. However, some short range correlations may exist. When the ions are incident on such a solid, the probability of the ions encountering atoms within the solid is extremely high. However, this is not generally the case with the crystalline materials. This is because the presence of three-dimensional atomic arrangements within the crystal creates open channels along certain crystallographic directions. This effect is illustrated in Figure 2.24a, which shows the perspectives of ball models of the diamond cubic lattice when viewed along the and directions. The openness that is observed along a specific direction is referred to as a “channel”. The channel width along a direction is greater than that along direction. If ions are incident along a channelling direction, some may go down the channel and hence experience fewer nuclear collisions. They are primarily slowed down by the coulombic losses, as schematically illustrated in Fig. 2.22b. As the result, they can penetrate much deeper into a crystalline solid than into an amorphous material. This effect is called ion channelling.
Figure 2.24. (a) Ball model showing relative degree of “openness” of the diamond (Si) lattice when traversing in and directions. (b) Schematic representation of ion trajectories in an axial channel for various entrance angles [83]. During FIB sample preparation, the specimen usually has a random orientation (for example 100 off (110) direction in Figure 2.25). That means that ions do not initially channel. 50
Figure 2.25. Ball model showing “closing” of channels when the sample with diamond structure turned 10o of (110) direction. However, some of the ions may subsequently be scattered into a channelling direction. As a result, they may penetrate deeper than the range calculated from Eq. 2. This effect produces tails on the concentration versus depth profiles of the implanted ions (Figure 2.26).
51
Figure 2.26. Illustrations of extracted parameters which characterise the distribution of an ion implanted impurity profile in crystalline silicon [76].
52
2.7.4 Ion-Induced Damage 2.7.4.1 Introduction The physical processes of the interaction of ions with targeting material and the formation of ion-induced damage is well described in works devoted to ion implantation [77, 79, 84-89]. This process is mostly studied for silicon and less for other semiconductor materials. Although the basic picture of interaction between ions and targeting materials appears similar for most materials, but the mechanism of formation of ion-induced damage in different materials may be slightly different. Section 2.7.4.2 contains a review of the physical processes of the ion implantation and defect formation in the silicon target. In the following sections, results for ion-induced damage in other materials (Ge, III-V compounds) are presented.
53
2.7.4.2 Ion Implantation in Silicon When energetic ions enter a solid they are slowed by both electronic (inelastic) and nuclear (elastic) interactions. In semiconductors, only nuclear interactions usually generate lattice damage, whereas in insulators both processes can induce atomic displacements. If the energy transferred by the incoming ion to the host atom exceeds a limiting value (usually a few tens of volts and called the displacement energy (E d )) the atom will be dislodged from its site. Initially, point defects such as single vacancies and single interstitials are generated in crystalline silicon (c-Si) due to ion bombardment. These point defects can then agglomerate, which results in the formation of small clusters (di-vacancy, di-interstitial, tri-vacancy, etc.) as well as larger clusters (dislocations, stacking faults, voids) [90]. The displaced atom can acquire enough energy, as a result of a nuclear collision, to, in turn, displace other atoms from their lattice sites. Likewise, the incident ions and second projectile atoms will continue to cause displacement damage until their energy falls to a level where the transferred energy during nuclear collisions is less than E d . As a result of these multiple collisions, the displacement damage in ion-treated solids can be quite extensive. The extent of the damage depends on the incident ion energy (Figure 2.27) [74], ion dose, dose rate, ion mass, and temperature of implantation [83].
Figure 2.27. Calculated number of vacancies produced in silicon per incident gallium ion for different ion energies [74]. 54
The distribution of the damage produced by an incident ion will depend on whether the ion is lighter or heavier than the host atoms. Since the energy transferred in a collision is directly proportional to the mass of the ion (see Eq. 3), a light ion will transfer less energy during each collision with the lattice atom than a heavier ion. As a result, the incident ions will be scattered through large angles. The displaced lattice atoms will possess a small amount of energy and may not be able to produce additional displacement. Furthermore, the majority of the energy of the incident ion is lost by electronic collisions, so that there is relatively little crystal damage. The ion range is comparatively large, and the damage will be spread out over a large volume of the target. The damage produced by a single, light ion may thus take the form that is schematically illustrated in Figure 2.28a.
Figure 2.28. Damage due to (a) light ions and (b) heavy ions [83]. The situation with heavy ions is quite different. For them, the energy transferred to the host atoms by nuclear collisions is quite substantial, implying that the displacement atoms in turn can produce displacement damage. Also, the ions are scattered through smaller angles and the ion range is small. These factors localize the damage within a 55
smaller volume, as shown in Figure 2.28b. The stopping power for heavy ions is smaller when they are moving rapidly (Figure 2.22) so there is often a region at the surface that is relatively free of damage. The disordered regions produced by each incident ion have central cores called displacement cascades, in which the defect concentration is extremely high (Figure 2.28). The collisional or direct displacement processes which occur during the collision cascade take place on a time scale 10-11 s. Small amorphous (a) zones are eventually formed either directly near the end of a heavy ion track or by nucleation within heavily damaged regions [89]. The boundaries between the a-zones and surrounding crystal have been found to be sharp within two to three atomic distances [91]. The position of the boundary of the a-zone therefore probably corresponds to the surface inside which the critical concentration of point defects necessary for damaged crystal to transform into an amorphous state has been exceeded. If point defects are mobile, a small a- zone surrounded by crystal is inherently unstable. When the dose is increased further, more disordered regions form and eventually begin to overlap, resulting in crystalline regions surrounded by amorphous material. With a further increase in the dose, the implanted volume may become amorphous, as was shown by Narayan and Holland for the case of self-ion damage in Si with a dose 2x1014 cm-2 [92, 93]. It is also apparent from Figure 2.28 that the crystalline-to-amorphous transformation, induced by ion implantation, will occur at lower doses for heavy ions. Morehead and Crowder [89] proposed a comprehensive model for the formation of amorphous Si by ion bombardment. The sequence of events occurring as a result of a crystalline solid can be considered as follows. In the 10-13 s during which the ion comes to rest after penetrating the solids, much of the energy loss goes into atomic displacements produced by both primary and secondary impacts. A thermal spike surrounding the ion track dissipates in perhaps 10-12 s, leaving a highly disordered region of many broken bonds and displaced atoms. These displaced atoms re-form bonds and change their positions to form a relatively stable phase in a time τ of the order of ≥ 10-9 s. During this time τ, vacancies escape via thermal diffusion from the disordered core surrounding the ion track. In Figure 2.29, this core is arbitrarily represented as a cylinder (the inner one), from which insufficient vacancies have 56
escaped so that the density would have corresponded to that of crystalline Si; the stable phase formed in τ is amorphous. The outer sheath, whose width is designated δR, becomes crystalline. The ultimate stable radius of the amorphous region, R 0 -δR, and hence the number required for them to overlap and form a continuous amorphous layer, (R 0 -δR)-2, depends upon both the original average size of the damage, R 0 , and the temperature-dependent out-diffusion of vacancies which reduce the size by an amount
δR.
Figure 2.29. The damaged region surrounding the path of a high energy ion in a crystalline solid idealised as a cylinder (dotted lines). Vacancies escape from the outer sheath and only the inner core (solid lines) becomes amorphous [89]. The formation of the compact amorphous regions is local to each ion track; the temperature-dependent change in size occurs in so short a time τ that the probability of two or more ions damaging the same volume (radius ~ 1-5 nm) within time τ is essentially zero. The important variables for a given target material are the nuclear stopping power of the ion, which determines R 0 , and the temperature of the target, which determines δR. The stopping power is not strongly dependent on ion energy over a relatively wide range for most ions [89], so that ion energy is of secondary importance and primarily determines the depth of the amorphous layer and not the dose required to 57
produce it (Figure 2.30) [74]. Dose rate is important only when it is high enough to yield a steady state concentration of vacancies in the crystalline material surrounding the disordered region, which is sufficient to retard the out-diffusion of vacancies. Dose rate will be most important at temperatures such that δR~R 0 .
Figure 2.30. Silicon vacancy density (Å-1ion-1) as a function of depth at 10, 30 and 50 kV Ga ion energy [74].
58
2.7.4.3 Annealing Effect Each ion during implantation may be expected to create a damage cascade containing interstitials and vacancies along its track in the crystalline (c) matrix. If point defects are mobile, a small a-zone surrounded by crystal is inherently unstable. At liquid nitrogen (LN) temperature or below, where the mobility of the point defects created by energetic ions is limited, a buried a layer will be created when the dose is such that the a zones overlap either due to continuous formation of new a-zones or due to their growth as perhaps the point defect concentrations in remaining crystalline volumes exceed the critical level of 10-15 percent necessary for a c→ a transformation [94]. Sadana et al. [91] studied the effect of implantation temperature on the formation of amorphous layers in (111) Si. Phosphorus ions of 120 keV energy were implanted at LN, room temperature (RT) and 100o C, respectively, to a dose of 3x1014 cm-2 in each case. Using TEM they observed dark bands that represented a or heavily damaged regions appearing in LN and RT implantation cases, while no such band was present for the 100o C implant. A substrate temperature of only 100o C was enough for in situ annealing out of any small a-zones that were created during the implantation. The temperature dependence of the critical amorphisation dose is shown in Figure 3.31. Heavier ions displace a greater volume of target atoms per unit, so a higher temperature is necessary to allow complete recovery and so prevent amorphisation.
59
Figure 3.31. A plot of the critical dose necessary to make a continuous amorphous layer in silicon as a function of temperature [89]. If defects such as vacancies and interstitials (Frenkel pairs) are mobile at the implantation temperature, then significant dynamic annealing and damage annihilation can occur during the implantation process. This dynamic annealing process is much more efficient at normal implantation temperatures in some semiconductors (e.g. gallium nitride) than others such as silicon [95]. Indeed, although discrete point defects in silicon are mobile at room temperatures (RT) [96], if the density of defects in the cascade is high then discrete defects appear to be efficiently trapped within such damage, and implantation damage at room temperature is mostly stable. If no annealing of damage occurs during implantation, then amorphisation, or complete disruption of lattice order, should occur at a dose at which each atom has been displaced approximately once. At such a dose, ~ 1015 cm-2 for 100 keV Si implanted into Si at 60
liquid nitrogen temperature (LN,) the concentration of the implanted ions in the matrix of ~ 1 at.% was reported [85]. If dynamic annealing is dominant, the nature of the resulting damage will depend strongly on the implantation temperature (which controls the annealing rates) and dose rate (which controls the rate at which damage is introduced). Williams et al. [84] studied recrystallisation processes during thermal annealing for a number of group III, IV and V impurities implanted in amorphous Si. At high impurity concentrations, an amorphous to polycrystalline transformation was observed at temperatures as low as 350o C. That means that the structure of damage may strongly dependent on the impurity concentration and local specimen heating during ion implantation, which can itself be caused by the high ion dose.
61
2.7.4.4 Ion Implantation in Germanium Wang et al. [86] reported an anomalous ion induced morphological instability of a Ge surface during 1.5 MeV Kr+ ion irradiation at RT. Amorphisation in Ge samples was detectable after a Kr+ dose of 8.5x1012 ions/cm2, and was complete after 1.2x1014 ions/cm2. Continuous irradiation of the amorphised Ge resulted in a high density of small cavities. These cavities, first observed after 7x1014 ions/cm2 with an average diameter of ~3 nm, grew into large (~50 nm) irregular-shaped holes, transforming the irradiated Ge into sponge-like material after 8.5x1015 ions/cm2 (Figure 4.32).
Figure 3.32. TEM micrographs of Ge irradiated with 1.5 MeV Kr+ to 1.0x1016 ions/cm2 at room temperature; (a) before annealing; (b) after annealing at 6000 C for 10 minutes [86]. The sponge-like structure was retained after crystallisation at ~6000 C. However, the mechanism which leads to the cavity formation in amorphous Ge is still not clear. Holland et al. [97] and Appleton et al. [98] have observed a similar surface damage structure, which formed within the amorphous phase of Ge during Bi (90 keV, 1016 cm-2 and 280 keV, 4x1015 cm-2) ion implantation at room temperature. They have observed a honeycombed structure extending 260 nm from the surface and containing very large voided regions, many of which intersected the surface. The authors have proposed a possible mechanism for this unique morphological instability during ion irradiation of germanium. The possible mechanism considered was differential sputtering due to the 62
defect structures which intersected the surface. As sputtering proceeded, it will begin to uncover the regions rich in vacancy clusters. The sputtering rates between the vacancyrich regions and those that were denuded could easily have led to changes in the surface topography such as cratering. Loss of dopant atoms during formation of the damage structure and its absence in light ion implantations (small sputter rates) were consistent with this mechanism. But the exact mechanism is still uncertain. Wilson [99] used high-resolution scanning electron microscopy to observe the effects of self-ion implantation on the topography of germanium surfaces. Holes appeared in the surface at doses above 2x1015 ions/cm2 (ion energy 50 keV). These enlarged with increasing ion dose and developed into a complex cell structure. A section of the bombardment surface shows a porous layer of thickness 2.5 times the projected range (R p ) which was observed at doses just below those where changes in surface topography were first observed. It was concluded that this structure was formed when the surface, retreating as a result of sputter etching, intersected voids formed as a result of radiation damage in the surface layer.
63
2.7.4.5 Ion Implantation in III-V Semiconductors The processes of damage formation in III-V semiconductors have been less studied than for Si, but the basic principles appear similar. However in III-V semiconductors the lattice elements are distinguishable and because they recoil differently due to their different masses, local perturbations in stoichiometry are created [100]. The lighter element recoils further, leading to an excess of the heavier element near the surface (shallower than R p ) and an excess of the lighter element at greater depths (between R p and ∆R p ). Repair of the lattice during subsequent annealing requires displaced atoms to diffuse back to appropriate sites, and in III-V semiconductors the diffusion length is not long enough to accomplish complete regrowth. The displaced lattice elements are unable to move quickly enough to keep up with the growth front, leading to highly twinned material and eventually to a stop of the regrowth if the initial amorphous layer is thicker than ~ 200 nm. An evaporation of the group V element from the material upon high temperature annealing is also characteristic of compound semiconductors. Bench et al [101] studied by in-situ TEM experiment damage produced by 50 keV Ar+ and Kr+ ion implantation in GaAs at temperatures ranging from 30 to 300 K. Ion doses ranged from 2x1011 to 5x1013 ions/cm2. It was found that at low temperatures and ion doses below 1012 ions/cm2, the degree of spatial overlap of the cascades was negligible and the amorphous zones were formed from an isolated individual displacement cascade. They were stable to temperatures above 250 K. Annealing of the materials from 30 to 300 K caused the individual zones to crystallize. Pearton et al [100] found that in GaAs, amorphous layers recrystallise epitaxially during annealing at 150-200o C, but the recrystallised layer was invariably highly defective, consisting of twins, stacking faults and other defects. These defects annealed out to leave only a high density of dislocation loops in the range 400-500oC. Williams et al [95] found that during irradiation of GaAs with 100 keV Si ions to a dose of 1015 ions/cm2 significant dynamic annealing was presented. This led to recrystallisation of damaged amorphous layers in GaAs at room temperature.
64
Xiong et al [102] studied amorphisation and recrystallisation in 4 meV N+ ion implanted InP crystals by a combination of channelling Rutherford backscattering spectrometry (CRBS) and TEM. It was found that at doses below 5x1014 ions/cm2 a high density of radiation damage (amorphisation) was induced only in the region near the end of ion range, while on the samples implanted with doses over 1x1015 ions/cm2 the amorphous layers formed with their inner interface at a depth of about 4 mm, where the concentration of implanted N was at half its maximum. It was also shown that the width of the buried layer extended towards the surface with increasing implantation dose, while the inner interface depth was almost fixed, independent of the dose. Yamaguchi and Nishikawa [59] studied damage in InP after conventional and gasassisted FIB milling. In the case of conventional FIB a 31-nm-thick damage layer was observed. An electron diffraction pattern from this layer indicated that the layer was amorphous with microcrystals. Such microcrystals were found in cross-sectional TEM images of the damage layer. The diameter of these microcrystals was estimated to be 10 nm.
65
2.7.5 Waterfalling Effect Munroe [27] has described another surface artifact during FIB fabrication, the so-called “waterfalling” effect. It has been observed in a titanium nitrided steel (Figure 2.33) and a copper Raney catalyst (Figure 2.34), for example. Its nature was related to localised regions of ions becoming channelled down particular parts of the specimen. The milled surface has "rivulets" running along its face. This artefact can be easily recognised and can usually be overcome quite easily. The proposed solution is to reduce the aperture size and perform a final mill at a lower beam current. The predisposition of the sample to exhibit "waterfalling" is reduced at these lower milling rates.
Figure 3.33. Milled cross-section on titanium nitrided steel. While the coating layers can be clearly seen, artefacts (i.e. “waterfalling”) can be seen on the milled surface [27].
66
Figure 2.34. Waterfalling effects on a Cu Raney catalyst [27] Figure 2.34 shows a Cu Raney catalyst which has been given a clean-up mill. Again, a waterfalling type structure can be observed. In this instance, the highly reactive nature of the copper atoms meant that they were able to diffuse readily under the beam and local re-arrangement occurred even at very low beam currents [27]. But it is not clear why such uneven channelling could result in a differential sputtering of the specimens and produce the waterfalling type structure. The sputtering rate of the crystal during FIB milling depends on several factors (species of target atom and bombarding ions, angle of incident, and crystallographic orientation). Surface relief may be developed when an initially flat and static polycrystalline target is milled [18]. The surface retains a “memory” of the local grain structure as grains are
67
removed by sputtering. A fine-grained single-phase material undergoing sputtering will therefore tend to roughen, although it remains essentially flat. For a multiphase target, the different sputtering rates of the components favour the creation of strong surface relief.
68
2.7.6 Redeposition of Targeting Material According to the theory of ion implantation described in sections 2.7.2-2.74, energetic Ga+ ions should form damage (amorphous) layers in FIB-prepared TEM samples. The structure and thickness of such layers should depend on the acceleration voltage, dose, incident angle and target material. As well as this obvious cause of damage layers in FIB-prepared samples, some researchers [47, 103-105] have assumed that damage layers can also be formed by redeposition of the target material. However, only few of them have experimentally observed such redeposition layers. Yamaguchi et al. [47] observed both spots and rings in the electron diffraction patterns that indicated the presence of a thin polycrystalline GaAs film, which covered single-crystal GaAs specimen. They assumed that this film was possibly formed by redeposition of sputtered GaAs and/or by ion-beam-induced amorphisation and recrystallisation to form a polycrystalline film. A schematic diagram of the process of redeposition of milled material around a trench is shown in Figure 2.35. Energetic Ga+ ions knock out target atoms from near the surface of specimen. Some of them may escape the trench through the open area, while the rest reach the walls of the trench. Some may establish bonds with surface atoms and stay there. Some of Ga+ ions can be scattered back and also adsorbed on the surface of the walls. So a redeposition layer consisting of mixture of target atoms and Ga atoms may be formed around the milled trench. The process of formation of such redeposited layer is very similar to the process of sputter coating. But in the FIB this process is dynamic and some of the redeposited atoms are removed by the Ga ion beam during subsequent beam rastering. The thickness of such redeposited layers strongly depends on particular milling conditions (dimension of the trench, milling rate etc.).
69
Figure 2.35. Schematic illustration of FIB milling process with (a) low and (b) high aspect ratios. Arrows indicate the possible trajectories of removed atoms. The probability for redeposition on the walls of the trench in (b) is much higher. Cairney et al. [105], observed a redeposited layer in a FeAl-WC composite. They made a V-shape cut in the sample and using Al, Fe and W elemental maps found both tungsten and iron present at the edge of the specimen. This was thought to be a redeposited layer [105]. From this study it is not clear how such a cut was produced and what its dimensions were. The result obtained was probably connected with this artificial V-shape cut when sputtered atoms could not escape from the bottom during milling and were redeposited on the walls. Moreover the thin cleaning final cut (for what is common for all FIB sample preparation techniques, see Section 2.4.6) was not made. Prenitzer et al. [43] studied narrow and deep trenches FIB milled in Si samples and found that the sidewalls of the trenches are clearly sloped and the trench bottom littered with redeposited material. Probably the high aspects ratio limited the removal of sputtering material from the bottom of the trench. Clear evidence of the formation of damage layer by redeposition during FIB specimen preparation was not found in the literature. Present speculation about a possibility of redeposition of milled material during FIB TEM sample preparation requires additional study.
70
2.7.7 Methods of Damage Reduction Methods of damage reduction may be divided into two categories. The first is the optimisation of FIB process parameters. This includes variation of milling parameters – high voltage, beam current, angle of incident and mill configuration to achieve minimum thickness of damage in the specimen. The second is the removal of damage after FIB specimen fabrication by means of plasma cleaning or conventional ion milling, for example. Decreasing the energy of the Ga beam [106] in the final stage of sample preparation can significantly reduce the thickness of the damage layer (see Eq. 2). Thus, Jamison et al. [107] reduced the thickness of the damage amorphous layer in Si from 25 nm to 11 nm by using a 10 kV final cleaning mill. Walker and Broom [74] also used a 10 kV final mill for damage reduction during TEM specimen preparation. They reported reduction of damage layer fabricated during 30 keV FIB milling from 28 nm to 6 nm for silicon, from 24 nm to 4 nm for gallium arsenide and from 40 nm to 15 nm for indium phosphide. Langford and Petford-Long [4] described the reduction of the thickness of the damage layer by using low energy FIB cuts in the final stage of silicon sample preparation. After the first steps of sample preparation the thickness of the amorphous layer on the surface of specimen was measured as 20 nm. The final cleaning mill was produced with a 5 keV Ga+ energy and 100 pA beam current. The specimen was tilted to 10o to the incident ion beam direction. The milling was produced both with and without iodine gas assistance. In the second case the milling rate of silicon was decreased by a factor of 10. In both cases they obtained a 10 nm thick amorphous damage layer on the surface of specimen. Some workers [34, 103] believe that damage can be reduced by using a final mill at very low beam currents (e.g.< 100 pA). This approach is very doubtful because lowbeam currents can only be achieved by inserting smaller apertures. In this way the current decreases, but the central, densest, part of the beam remains and beam current density does not change. The implantation dose remains the same or may even be higher because of pixel overlapping. 71
Isbell and Fischione [108] used low energy plasma to remove carbonaceous contamination formed on a specimen under the electron beam. They found that plasma cleaning with a process gas mixture of 25 % oxygen in argon prevents such contamination from occurring and eliminates any prior contamination from the specimen. But their report did not mention any plasma cleaning parameters such as an applied voltage and plasma current. Brown and Humphreys [109] refer to the application of plasma cleaning for the final stage of FIB TEM sample preparation. A combination of oxygen and argon allowed the hydrocarbon deposit to be eliminated. Unfortunately, these authors do not show any successful result for removing FIB associated damage. Cairney and Munroe [105] also applied plasma cleaning for damage removal in FIBprepared FeAl-WC composite samples. Plasma cleaning was concluded to have little or no effect on redeposition in these composites. More successful was the application by Kato et al. [60] of conventional Ar ion thinning. An FIB-fabricated specimen was milled by a broad argon ion beam by using a Gatan JIT-100 unit with an ion beam current of 70 mA at an angle of 15 degrees to the specimen. This method was found to reduce the depth of the damage layer by about half, it was reduced to 12 nm. As reported, this damage could be further reduced by adjusting the tilt angle and beam current. This is doubtful because the depth of final damage is mostly dependent upon the energy of the argon ion beam. The authors did not mention the ion energy used for damage reduction. It is also not clear whether this 12 nm damage layer represents the prior remains of FIB induced damage or new damage produced by conventional argon milling.
72
2.8 Summary The FIB technique is very useful for TEM sample preparation in cross-section and planview configuration. The specimens can be prepared with positional placement accuracy of the order of a few tens of nanometers, which is very important for failure analysis in the semiconductor industry. But the FIB specimen milling process is accompanied by direct (radiation defects, differential sputtering) and indirect (redeposition of sputtering material) damage formation. The dynamic annealing during FIB fabrication may result in some recrystallisation of the damage layer with new artifacts like grain boundaries, stacking faults etc. This FIB related damage (amorphous layers, redeposition, topographic artefacts, etc) might dramatically limit the application of FIB prepared specimens, and samples can completely inhibit the acquisition of information about the true crystalline structure of milled materials. The effect of damage formation during FIB TEM sample fabrication has not been previously studied systematically. The FIB-related damage in different semiconductor materials has been studied here and will be described in subsequent chapters of this thesis. The experimental part of the thesis contains the results of a detailed study of the origin and structure of damage in silicon samples prepared using FIB. Damage layers were also studied in other semiconductor materials such as germanium and indium arsenide. Results obtained are compared in the discussion section of this thesis with known data and data obtained from theoretical calculation using the Monte Carlo method. The conclusions summarise the origin and structure of the damage in FIB prepared samples for different semiconductor materials and ways to minimise this damage.
73
PART III EXPERIMENTAL PROCEDURE CHAPTER 3 EXPERIMENTAL PROCEDURE 3.1 Introduction The initial goal of this study was the development of a FIB-based experimental procedure, to study FIB-related damage in a range of semiconductor materials. For this reason, in some cases the details of the experimental procedure overlaps with the results and discussion. It was shown in the literature review (Chapter 2) that interaction between the ion beam and the different target materials manifested in variations of the degree of damage. A general outline of the experimental procedure for the study of damage is provided in following sections of this chapter. But, in some cases, more detailed descriptions of the experiments performed will be given in the results and discussion section (Chapter 4) of this thesis.
74
3.2 Materials Despite the introduction of new promising materials for microelectronics such as GaAs, GaN, InP and others, silicon remains the major material used today and probably into the near future. Silicon was used as the basic material in this study. Silicon substrates Si(100) with a multi-layer structure on the top consisting of SiO 2 (15 nm), TiN (200 nm) and six layers of Si0 2 (with total thickness 200 nm), respectively were produced by Peregrine Semiconductor Australia and provided by Dr Yongbai Yin. In addition to silicon, other semiconductor materials were studied. Germanium Ge(100) and Indium Phosphide InP(100) substrates were supplied by Ms Jodie Bradby of the Electronic Materials Engineering Department of the Australian National University, Canberra. An InAs(100) substrate with a multi-layer structure deposited on the surface which consists of five alternative layers AlAs (with each thickness of 2 monolayers) and InAs (with individual thickness of first four ~ 10 nm and of fifth ~ 2 nm) respectively and GaAs(100) substrate were supplied by Dr Anton Gutakovsky of the Institute of Semiconductor Physics, Novosibirsk, Russia. The basic properties of the studied materials, which are important for understanding the damage formation during FIB fabrication, are given in Table 3.1.
75
Table 3.1 Basic semiconductor materials and their fundamental properties [110]. Element Atomic Atomic number mass (amu)
Lattice Density Displacement Surface constant (g/cm3) Energy (eV) binding (nm) energy (eV)
Melting point ,oC
Si
14
28
0.543
2.33
15
4.7
1417
Ge
32
72.6
0.566
5.33
15
3.9
937
InP
49/15
115/31
0.587
4.79
25
2.49/3.27
1062
GaAs
31/33
70/75
0.565
5.32
25
2.82/1.26
1238
InAs
49/33
115/75
0.606
5.67
25
2.49/1.26
943
76
3.3 TEM Investigation All the TEM work was carried out on a Philips CM 200 (Philips Corp., Netherlands) field emission gun (FEG) transmission electron microscope operating at 200 kV. Analytical equipment on this instrument included an EDAX mDX EDS spectrometer and a SIS Biocam CCD camera able to provide on-line imaging and analysis. The microscope has two side-entry sample holders, one-axis tilt and double-axis tilt. The specified resolution of CM 200 with Super TWIN pole pieces at 200 kV was 0.24 nm point and 0.10 nm line, respectively. The accuracy of readouts for magnification and camera lengths was ± 5%. Conventional (bright-field and dark-field), scanning transmission (STEM) and High Resolution (HREM) electron microscopy techniques were used for sample investigation. In the case of HREM 7 and 15 beams, images were taken in the vicinity to the Scherzer focus. The images were recorded using both a photo camera and a CCD camera. An EDS spectrometer was used for point elemental analysis and for elemental mapping. In this case tilting the specimen towards the EDS detector controlled the count rate. Usually the tilt was about 10o.
77
3.4 FIB Observation and Milling Parameters TEM sample preparation was carried out in a FEI xP200 Focused Ion Beam Workstation, (FEI Company, Hillsboro, OR, USA). The sample was placed inside the column in a special holder with the milling surface normal to the incident Ga ion beam. The procedure for FIB sample preparation is described in next two sections. The quality of the secondary electron images and the milling speed in FIB machines strongly depends on the Ga+ beam current density, which is manipulated by a set of beam apertures. The aperture intersects part of the beam and delivers to the specimen only its central and most intensive part. The beam current in pA gives an indication of the size of the aperture in the FIB machine. The FEI xP200 FIB user’s manual gives a set of standard milling parameters for each particular aperture (Table 3.2, columns 14). Using this data and equation 7 (Section 2.7.2), the ion doses delivered to the sample during one dwell time were calculated. Results obtained are shown in column 6 in Table 3.2. The actual ion dose may be insignificantly lower than calculated data because of the Gaussian shape of the beam profile was not taken into account in these calculations. However, the calculated data indicates that the ion dose which is delivered during one dwell time exceeds 1013 ions/cm2 for all working apertures at a 30 keV beam energy.
78
Table 3.2 Beam current, milling spotsize [111] and calculated Ga ion dose delivered to the sample during one dwell time for the FEI xP200 workstation. Beam Beam Dwell Current (pA) Voltage (kV) Time (msec)
Milling Spotsize (mm)
Milling Area Dose [Ga+ions/cm2] (mm2)
1
30
1
0.008
5x10-5
1.3x1013
4
30
1
0.012
1.1x10-4
2.3x1013
11
30
1
0.015
1.8x10-4
3.8x1013
70
30
1
0.025
4.9x10-4
8.9x1013
150
30
1
0.035
9.6x10-4
9.8x1013
350
30
1
0.055
2.4x10-3
9.1x1013
1000
30
1
0.08
5x10-3
1.25x1014
2700
30
1
0.12
1.1x10-2
1.5x1014
6600
30
1
0.27
5.7x10-2
0.72x1014
11500
30
1
0.5
2x10-1
0.36x1014
250
10
1
0.3
7x10-2
0.5x1012
For sample preparation, larger apertures (beam currents 11500 pA, 6600 pA) were used during the initial stage. Smaller apertures (beam currents 1000, 350 or 150 pA) were used for final, more precise milling or for specimen cleaning. Before each mill a milling pattern was drawn on the desired place of the specimen on the FIB computer screen. The FIB software material file for Si was used for milling in this study. The materials files contain beam parameters and sputter (or deposition) rate parameters that the system uses to calculate a mill time when a pattern is defined. For the Si material file these parameters are: 50 % beam diameter overlap, 1 msec dwell time and 0.15 mm3/nC sputter rate. The volume (and thus the depth of mill) of removed material during FIB mill is proportional to the sputter rate for a particular milling material. Because the Si 79
material file was used for all other materials the real mill depth was different from material to material. The measurements of real depth of the milled trenches and calculation the sputter rates for different materials are described in the results and discussion section. For FIB imaging, secondary electrons were detected and used in this study. All FIB images were recorded at a 30 keV ion energy and either 11 or 70 pA ion beam currents. For the study of the modification of germanium surfaces by FIB, the germanium specimen was exposed at different Ga ion beam/voltage parameters and FIB images were taken through different time intervals.
80
3.5 Primary TEM Specimen Preparation First the slices 3x0.5 mm and 0.5 mm thick were cut from original substrates using a ISOMETTM Low Speed Diamond Saw (Buehler Ltd, Illinois, USA). They were mounted on a Pyrex holder with mounting wax (MWH 135-T) (Figure 3.1) by melting wax using a hotplate at a temperature around 120o C. These slices were further mechanically ground by a Model 590 Tripod Polisher (South Bay Technology, Inc., USA). The grinding process was carried out by moving the Tripod Polisher along the polishing surface with sandpaper (Wet & Dry Abrasive Paper, P1200; Carborundum Abrasives, Australia). Water was supplied on the polishing surface to remove the ground material.
Figure 3.1. Schematic Diagram of theTripod Polisher. Slices were polished from each side until a final thickness around 30-60 mm was achieved. This tool allowed either uniform or sloped samples with a thickness down to a few of microns to be obtained. However, in the case of fragile semiconductors like Ge, InAs, GaAs and GaP, mechanical grinding to a thickness of less than 20-30 mm can produce some undesirable results such as breaking the edges of the sample or even its complete destruction. A thickness of 50 mm is appropriate for FIB milling. The
81
mounting wax was then softened by heating the holder on a hot plate and samples were removed from the holder. The remains of the wax were removed from the samples by ultrasonic cleaning (Sanophon Ultrasonic Cleaner, Ultrasonic Industries, Sydney, Australia) in acetone. The final slice was glued on half of a standard 3 mm TEM copper slot grid with surface facing away from the slot (see Figure 2.11a in Section 2.6.1) for subsequent FIB milling. Very strong M-Bond 610 epoxy glue (M-line Accessories, Measurements Group, Inc., Releigh, NC, US) was used. The epoxy was cured for one hour at a temperature of 65o C. The mechanical preparation of germanium samples was found to be more complex than for the silicon samples. First, the pieces 3 mm long and 0.5 mm thick were cut from a Ge wafer using a low speed diamond saw. When these pieces were thinned by mechanical grinding using a Tripod polisher and sandpaper it was found that Ge was much softer then Si, thus the rate of grinding was higher. It was therefore not easy to control accurately the thickness of the specimen and most of it was rapidly destroyed. To overcoming this problem the pieces of TEM copper grid slot with standard thickness 50 mm were glued on the Pyrex mount around the Ge strips. Copper has a much slower thinning speed then Ge, and this prevented the extensive pressure and destruction of the Ge specimens. It was also useful for precise specimen thickness control. The process of polishing was terminated when the copper pieces started to be affected by polishing. In this case, the thickness of Ge samples was just below the thickness of the copper pieces (50 mm). Next the samples were prepared for FIB fabrication using the same procedures described previously for Si.
82
3.6 Gold Sputter Coating In some experiments pre-FIB fabrication samples or post-milling samples were sputter coated with a layer of gold. This was done using a Polaron Sputter Coater (Model E5000). The gold sputter coating was carried out at 1.4 kV voltage and 40 mA plasma current in an argon gas atmosphere with the pressure ~ 0.07 Torr. Under these conditions a uniform gold film was formed. The total thickness of the Au protective layer was typically in the range 50-150 nm. Sputter coating at higher argon pressures resulted in deposition of gold films with uneven thickness. Gold layers were used to protect the specimen surface during FIB manipulation or to protect induced damage on the specimen during subsequent FIB fabrication steps. Detailed description of the effect of the gold coating is given in the results section.
83
3.7 FIB Specimen Preparation Procedure For TEM sample preparation, methods similar to the basic procedures for FIB sample preparation described in Section 2.4 were used. In this section the technique of FIB sample preparation, which was used in this study for different materials, is described using the example of a silicon sample. Figure 3.2 illustrates the FIB image of the Si sample before FIB manipulation. The measured initial thickness of the sample was 55 mm. Scratches and cracks (areas with bright contrast, marked by arrows on Figure 3.2b) were clearly visible on both edges of the specimen surface as result of mechanical grinding. Mechanically damaged surface areas, may represent up to one third of the specimen thickness. However, there still remain sufficient undamaged regions suitable for further thinning.
Figure 3.2. FIB images of (a) the mechanically prepared Si sample in the FIB stage before FIB milling, (b) the surface of the sample after mechanical polishing (arrows indicate scratches and other damaged regions). Initially, a platinum strip was deposited on an undamaged area. The size of the strip was approximately 30x2 mm and 1 mm thick. It was deposited under a 350 pA beam current. The Pt strip is shown in Figure 3.3a. After deposition of the platinum protective strip two trenches from either side of the strip were milled using a high beam Ga ion beam current with a 6600 pA aperture. The size of these trenches were ~ 30x30 mm and 3 mm deep (Figure 3.3b).
84
Figure 3.3. FIB images of (a) Pt strip deposited on the specimen surface, (b) two milled trenches. Next, a slower cut (using a 1000 pA beam current) was made from one side to approximately half of the width of the Pt strip (Figure 3.4a). The size of this cut was 15x3 mm and depth – 3 mm. A final 1000 pA beam current cut (with dimensions 15x2 mm, depth 3 mm) was made from the other side of membrane. For the final cut the milling box was positioned very carefully to prevent the area of interest being completely milled away, but to ensure that the final membrane was thin enough to be transparent to an electron beam. After the final cut, a ~ 0.2 mm membrane was obtained (Figure 3.4b). Clear remains of the Pt protective layer on both the left and the right sides of the membrane and on the top of membrane are visible. This layer ensured that the structure of the specimen was not modified during the FIB procedures.
Figure 3.4. FIB images of the specimen after (a) the first low beam current cut, and after (b) the second final low beam cut. 85
From Figures 3.4-3.5 it can be seen that the wall and the bottom of the milled trenches in the silicon appear very smooth and very uniform.
Figure 3.5. FIB image of the final membrane. The FIB stage was tilted to 40o for better membrane observation. No visible amount of redeposited material was directly observed in or around the milled trenches when imaged in the FIB. Figure 3.6 shows a low magnification TEM image of the electron transparent area of the prepared silicon sample. The transparent (lighter) area was estimated to be 80 mm2. Two dark layers are clearly visible in the transparent membrane. EDS point analysis showed that the upper layer was the remains of the Pt protection strip and the second layer is the TiN layer showing the initial structure of the studied sample.
86
Figure 3.6. Low magnification TEM image of the final membrane of the silicon sample. The final membrane obtained clearly had a thickness adequate for conventional TEM investigation. In the higher magnification image on Figure 3.7, the details of the internal structure of the studied sample are clearly visible. The crystalline silicon substrate has characteristic diffraction contrast. The TiN layer (thickness 200 nm) has a polycrystalline structure with columnar crystals extended in the direction of growth. There were two oxide layers visible below and above the TiN layer. The absence of diffraction contrast in these layers and the corresponding diffraction pattern indicates they were amorphous in structure. The thickness of the layers measured from these TEM images were exactly in accord with the manufacturer’s data. Thus the thickness of the first oxide layer was 15 nm. The top oxide layer consists of six individual layers with a total thickness 200 nm.
87
Figure 3.7. (a) Low and (b) high magnification image of TiN-oxide-silicon structure shown in Figure 3.6. Attempts were made to obtain high-resolution images from different regions of this specimen (Figure 3.8). Only one set of lattice planes was visible in some areas of the membrane, but the contrast was weak. This result indicates that the membrane was too thick for lattice imaging and/or the specimen had significantly thick damage (amorphous) layers on both sides of the membrane.
Figure 3.8. (a) High-resolution image of silicon sample, (b) magnified image of area marked X shown in Figure 3.7.
88
In spite of the lack of high-resolution lattice images the quality of the sample obtained was reasonably good for most forms of TEM analysis. The overall structure was easily identified and characterised. The parameters of the layers measured coincided with the manufacturer’s data. The thin structures, like the 15 nm oxide layer were clearly visible.
89
3.8 Sample Preparation Procedure For Damage Study The overall process of FIB specimen preparation for TEM consists of a sequence of FIB milling steps and is described in Section 2.6. On each of these steps a trench is milled in the specimen with a different milling parameter, for example, beam current/ion energy. For investigation of the damage introduced during FIB fabrication, the experimental procedure described here was established as part of this study. Regions of the specimen affected by the FIB must first be covered with some form of protective layer which will prevent the introduction of any other damage in these regions during further specimen preparation steps. Using normal FIB thinning procedures, the prepared cross-section will contain the cross section of the damaged material. It can therefore be analysed by TEM and other associated analytical equipment to determine its thickness, structure and chemical composition. Using a FEI xP200 FIB system, with either 10 and 30 keV Ga+ ion beams, a row of trenches on the sample was milled under different beam currents ranging from 150 to 11500 pA (Figure 3. 9a).
Figure 3.9. Schematic diagram showing the TEM specimen fabrication procedure for investigation of the damage layer after FIB milling. (a) FIB milling of the trench, (b) coating with a Au film, (c) deposition of a Pt strip, (d) final thinning of the specimen. 90
The trenches were approximately 15x10 mm wide and 1-4 mm deep. The milled specimens were then removed from the FIB and sputter coated with a ~50-100 nm thick Au film to preserve the trench surfaces being affected by the FIB from further damage during the subsequent TEM specimen preparation steps (Figure 3.9b). Next the specimen was placed back inside the FIB system and the central part of the trenches were covered with 1 mm thick Pt strips using the metal deposition facility of the FIB (Figure 3.9c). The presence of these two protection layers (Au and Pt) ensured that the final TEM specimen has the unmodified original damage layer resulting from the initial milling. Cross-sectional TEM specimens of the trench walls (Figure 3.9d) were then prepared using normal FIB procedure described in section 3.6 and were studied using Philips CM 200 transmission microscope.
91
PART IV RESULTS AND DISCUSSION CHAPTER 4 SILICON 4.1 Introduction FIB systems have been successfully used for TEM sample preparation during the last decade. There are many studies devoted to the development of the FIB technique for TEM specimen preparation, especially for silicon. However, there has been no systematic study of the damage introduced during FIB specimen fabrication. This chapter of this thesis contains the results of studying the damage introduced during FIB specimen fabrication in silicon (Sections 4.2, 4.3). Section 4.4 contains elemental analysis of the damage layer in silicon. Section 4.5 describes the study of redeposition effects. The measurement of the slope angle for a range of ion beam currents is described in Section 4.6. Section 4.7 contains a discussion including comparison of the results obtained with known data.
92
4.2 Damage Study in Silicon After a 30 keV FIB Milling Using the FEI xP200 FIB system with a 30 keV Ga+ ion beam, a row of rectangular trenches in a Si sample was milled under different beam currents ranging from 150 to 11500 pA (Figure 4.1). Most of the milled trenches were approximately 15x10 mm wide and 1-3 mm deep. As will be shown, damage was formed on side-walls and the bottomwalls of these trenches during FIB milling. Following these mills, TEM cross-sections containing the wall damage were prepared using the experimental procedure which was described in Section 3.7.
Figure 4.1. FIB image of silicon with trenches milled using a 30 keV beam energy and different beam current, in the range 150 to 11500 pA. TEM was then used to study the presence of damage layers on the prepared samples between the undamaged crystalline silicon substrate and the Au protective film. A low
93
magnification cross-sectional TEM image of the walls of the trench prepared at a beam current of 2700 pA and beam energy of 30 keV is shown in Figure 4.2. In Figure 4.2 there are several layers clearly visible, including the Si-substrate, SiO 2 layers, a TiN layer and both the Au and Pt protection layers. The identification of these layers was carried out using point EDS analysis. In addition, a damaged Si layer (indicated by white arrows) may be observed between the Si-substrate and Au layers. Diffraction contours (alternation of dark and bright stripes) are visible everywhere in the crystalline silicon substrate, but they are not visible in the damage layer between the substrate and the Au protective film. The absence of diffraction contrast in the damage layer on TEM images shows that it has amorphous structure. This was confirmed by diffraction studies of the damaged layer.
Figure 4.2. Bright field TEM image of the edge of the trench in silicon sample (milled using a 2700 beam current and a 30 keV beam energy) affected by a beam “tail”. It was also found for all milled trenches that some milling takes place outside the defined FIB rastering area. Accordingly, the edges of the trenches were not sharp. A transition area (with a curved profile) from the side-wall of the trench to the unmilled 94
specimen surface is indicated by a black arrow in Figure 4.2 (the white line indicates the expected continuation of the side-wall of the trench). Sputtering from an ion beam tail presumably formed this transition area. This effect is also visible on FIB images of the milled trenches (Figure 4.1). Such transition areas were found around all milled trenches irrespective of beam current. The width of the beam tail transition region was measured as 0.65 mm from the TEM image of the trench milled using a 2700 beam current in Figure 4.2. This effect was less visible for smaller beam current apertures, e.g., narrower ion beam diameters but was nevertheless present around all milled trenches. For the 150 pA trench the width of the tail sputtering region was found to be ~ 250 nm. In Figure 4.2, the side-wall of the trench can be seen to have some slope away from the direction of Ga ion beam propagation. In this specimen (milled using a 2700 beam current and a 30 keV beam energy) the slope angle was found to be 6.5o. The slope measurements for different ion beam currents will be described in Section 4.2.5. TEM observation of the damage layer around the milled trenches showed that it had different structure and thickness in the side-wall and bottom-wall areas. A cross-section of one half of the trench milled under a 1000 pA beam current is shown in Figure 4.3. The thickness of the side-wall damaged area (marked 1 in Figure 4.3) was very uniform in thickness and structure.
95
Figure 4.3. A low magnification cross-sectional bright-field TEM image of the wall of the trench milled using a 1000 pA beam current and a 30 keV beam energy. A crosssection of one half of the milled trench is shown, containing two different damage areas: (1) side-wall area and (2) bottom-wall area. It is clearly visible that the damage in the bottom-wall area (marked 2 on Figure 4.3) is thicker than the damage in the side-wall area (marked 1). Figure 4.4a shows a higher magnification TEM image of this specimen in the region marked “1” in Figure 4.3, which is the side-wall of the trench. It may also be noted that this amorphous region is uniform in thickness along the side-wall with an average thickness of ~ 30 nm.
96
Figure 4.4. Magnified bright-field TEM image of (a) side-wall area and (b) bottomwall area of the trench shown on Figure 4.3. a-Si indicates amorphous silicon, Poly-Si indicates polycrystalline silicon region. Figure 4.4b shows a higher magnification TEM image of the bottom-wall of the initial trench (the region marked 2 in Figure 4.3). In contrast to the side-wall, the damage layer in this region was much thicker, around 56 nm, but it was also less uniform in structure. At this magnification, it may be seen to consist of two parts: an amorphous layer (~28 nm thick) on top of the silicon substrate, and a polycrystalline layer (~28 nm thick) on top of the amorphous layer (Figure 4.4b). A selected area diffraction pattern from the bottom-wall damage of the trench milled using a 1000 pA beam current and a 30 keV beam energy is shown in Figure 4.5. In addition to the diffraction spots from the crystalline silicon substrate there are clearly visible amorphous rings (indicated by black arrows) and additional spots (some are indicated by white arrows), which may be associated with diffraction from the microcrystals.
97
Figure 4.5. Selected area diffraction from the bottom-wall region of the trench milled in silicon substrate using a 1000 pA beam current and a 30 keV beam energy. Black arrows indicate broad amorphous rings. White arrows indicate some additional diffraction spots caused by micro-crystals. Figure 4.6 shows this bottom-wall at a still higher magnification. Three regions are clearly visible: the crystalline Si substrate (marked A), the amorphous region (marked B) and then a series of micro-crystallites in an amorphous matrix about 10 nm in diameter (marked C). Planes of atoms, indicative of a crystalline structure, are clearly visible in this region. In some areas these crystalline regions are larger and planes of atoms traversing this crystalline region are clearly visible (Figure 4.6).
98
Figure 4.6. High-resolution image of the bottom of a trench, milled using a 1000 pA beam current and a 30 keV beam energy, showing the silicon substrate (A), amorphous silicon (B) and silicon crystallites (C). Examination of the TEM specimens prepared from around the trenches using different beam currents (150, 350, 2700, 6600, 11500 pA for a 30 keV beam energy) indicated a similar structure, that is, uniform amorphous layers on the side-walls and both amorphous and crystalline regions on the bottom-walls of the trenches. Examples of the side-wall and the bottom-wall damage for trenches milled using the 11500 pA and 150 pA beam currents are shown in Figures 4.7 and 4.8. 99
Figure 4.7. TEM bright-field images of (a) side-wall and (b) bottom-wall of the trench milled using a 11500 pA beam current and a 30 keV beam energy.
Figure 4.8. TEM bright-field images of (a) side-wall and (b) bottom-wall of the trench milled using a 150 pA beam current and a 30 keV beam energy. The thicknesses of the amorphous layers resulting from all beam currents used for the 30 keV ion beam energy were the same as those resulting from the 1000 pA mill. That is, the thickness of the side-wall damage layer was measured to be 28±3 nm, and the thickness of bottom-wall damage area was 56±6 nm. However, the thickness of the polycrystalline region in the bottom-wall varied slightly from sample to sample and was measured in a range 22-30 nm. The degree of crystallisation in these areas was also different. In some samples, only discrete Si 100
crystallites surrounded by amorphous Si were observed (Figures 4.6-4.8). But in some samples almost perfect crystalline layers inside the bottom damage areas were found (Figure 4.9). A correlation between the ion beam current and the degree of recrystallisation in this region was not found. These results are probably connected with slightly different experimental conditions (for example, the samples could be heated differently because of small variations in ion beam currents and different heat outflow). Nevertheless recrystallised regions in bottom-wall damage layers were found on every sample.
Figure 4.9. Bright field TEM image of the bottom-wall damage layer in silicon (milled using a 6600 pA ion beam current and a 30 keV beam energy) with crystalline sub-layer (C). A indicates silicon substrate; B the amorphous part of damage layer. There is strong diffraction contrast visible in areas marked A and C in Figure 4.9, which indicates a high level of crystallinity in this layer. Some separate micro-crystals are visible as well (Figure 4.9). A magnified image of such micro-crystal (marked x in Figure 4.9) is shown in Figure 4.10.
101
Figure 4.10. High-resolution TEM image of the bottom-wall damage with a microcrystal in the trench milled using a 6600 pA beam current and a 30 keV beam energy. There is a visible atomic structure inside both the micro-crystal and the silicon substrate. Additional lines (indicated by white arrows) are visible inside the micro-crystal, which show some crystallographic defects. The interface between the crystallite and the amorphous silicon is sharp. However, the interface between the substrate and the damage layer is less sharp.
102
Figure 4.11. High-resolution TEM image of the area of transition from crystalline to amorphous structure in the bottom-wall of the trench milled using a 1000 pA beam current and a 30 keV beam energy. The crystalline to amorphous transition area is visible as a band of brighter spots on the high-resolution image of the bottom-wall damage area of the silicon matrix (Figure 4.11). It can be measured as 5-8 atomic planes in width.
103
4.3 Damage Study in Silicon After a 10 keV FIB Milling Trenches milled with a 10 kV Ga ion beam have a similar structure for both the sidewall and bottom-wall damage areas seen at 30 keV (Figure 4.12). The damage layers (indicated by arrows) are clearly visible between the Au protective film and the crystalline silicon substrate. The absence of diffraction contours in layers between the Au protective layer and the silicon substrate, as well as the absence of discrete reflections in diffraction patterns, indicates the non-crystalline structure of the damage layers.
Figure 4.12. The structure of (a) the side-wall and (b) bottom-wall of the trench milled using a 10 keV beam energy and a 250 pA Ga+ ion beam current. The side-wall damage (indicated by arrows in Figure 4.12a) is amorphous in structure with an average thickness of 13±2 nm. It is also clearly visible that the bottom-side damage again consists of two parts, that is, an amorphous layer on the top of undamaged crystalline silicon substrate. Between this layer and the Au protective layer is a visible layer with a polycrystalline structure (marked by x in Figure 4.12b). The full thickness of the bottom-wall damage area was measured as 23±2 nm. It is clear that both the side-wall and bottom-wall damage layers are about half the thickness of those observed after 30 keV milling.
104
4.4 Elemental Analysis of the Damage Layer in Silicon Using the EDS facilities of the Philips CM 200 electron microscope, elemental analysis of the damage layers was performed. Figure 4.13 shows a bright field STEM image of the bottom-wall of the trench milled using a 6600 pA Ga ion beam current and a 30 keV beam energy and the corresponding SiKα, AuMα and GaKα elemental maps. The damage layer consisting of two parts (bright amorphous and dark polycrystalline) is clearly visible in the STEM image (Figure 4.13a). Si and Au elemental maps (Figure 4.13b, c) show that interface between silicon substrate and Au film is rather sharp. The weak background noise in all areas of the specimen (Figure 4.13d) on the Ga elemental map indicates the presence of Ga atoms on both sides of the cross-sectional membrane. This Ga background noise is a result of Ga implantation during FIB preparation of the final cross-sectional TEM membrane. The thin band of high intensity indicates a high concentration of Ga atoms just under the gold protective film (Figure 4.13d).
Figure 4.13. (a) STEM image and corresponding EDS SiKα(b), AuMα (c) and GaKα(d) elemental maps of the bottom of the trench, which was milled using a 6600 pA beam current and a 30 keV beam energy. 105
It is clear from comparing the images in Figure 4.13 that the region of highest gallium intensity on the Ga elemental map corresponds to the polycrystalline layer on the STEM image (Figure 4.13a). Point EDS analysis was made from different areas of the specimen. EDS spectra from the top half and the lower half of the bottom-wall damage area and the silicon substrate are shown in Figure 4.14.
Figure 4.14. EDS spectra from the upper (a) and the lower part (b) of the bottom-wall damage of the trench milled using a 6600 pA beam current and 30 keV beam energy.
106
EDS spectra from the side-wall damage of the trench and from the undamaged Si substrate are shown in Figure 4.15.
Figure 4.15. EDS spectra from the side-wall damage (a) and from the silicon substrate (b) in the trench milled using a 6600 pA beam current and a 30 keV beam energy. As well as strong SiKα peaks on the EDS spectra in Figures 4.14-4.15 there are clearly visible GaKα and GaLα picks. The Ga concentration in the substrate was measured to be ~ 1 at. %. As stated before, this is due to damage from final preparation of the TEM 107
membrane. This data were used to correct the Ga concentration measured in the damage layers. The bottom-wall layer with the micro-crystals contained ~7 at. % Ga, the amorphous layer underneath contained ~2.8at.% Ga and the amorphous side-wall area contained ~2.5 at.% Ga.
108
4.5 Study of Redeposition Effects For a more detailed study of the origin of the damage in these specimens, the procedure for the preparation of the trenches was slightly modified. Rectangular trenches were first milled in a silicon substrate using a 2700 pA beam current and a 30 keV beam energy and then sputter coated with a Au protective film as described earlier (Figure 3.9a,b). The average size of the first trench was 10x10 mm2 and ~ 1 mm deep. The specimen was then placed back in the FIB system and a second set of trenches 5x8 mm2 wide and 0.6 mm deep was milled in the middle of the bottom-wall of the first trench using a 2700 pA beam current and a 30 keV beam energy. The specimen was sputter coated with Au protective film again and was placed back in the FIB system and the trenches were then covered with 1 mm thick and 2 mm wide Pt strips using the metal deposition facility of the FIB. The presence of these protective layers (Au and Pt) again ensured that the final TEM specimen had unmodified original damage layers resulting from the initial milling steps. Redeposited material during the milling of the second trench should therefore be visible on the TEM cross-sections between the two Au layers. Cross-sectional TEM specimens of the trench walls were then prepared using normal FIB procedures described in Section 3.8 (Fig. 3.9c,d). A schematic diagram of such a sample is shown in Figure 4.16. X-ray mapping was also used to study the distribution of elements across the different regions in this sample.
Figure 4.16. Schematic diagram of final sample with two trenches and two Au layers. 109
The damage layers are clearly visible around the first trench between the crystalline silicon substrate and the first gold protective layer. In Figure 4.17 the side-wall is marked by “1” and the bottom-wall is marked by “2”. Redeposited material was found between two Au layers only on the side-wall of first trench (Figure 4.17). A “pocket” of redeposited material between the two Au layers is clearly visible on the side-wall area in this specimen.
Figure 4.17. Cross-section TEM image of the region around the wall and bottom of the first trench milled using a 2700 pA beam current and a 30 keV beam energy. The redeposition region was found to be non-uniform in shape with a thickness up to 170 nm, which was much thicker than the typical side-wall damage layer (~30 nm) observed in silicon TEM specimens in which only a single trench was milled. The absence of diffraction contrast in this layer, as well as the absence of discrete reflections in diffraction pattern from this region (Figure 4.18), indicates its amorphous structure.
110
Figure 4.18. Selected area diffraction from the redeposited region found in the side-wall region of the first trench shown in Figure 4.17. On the bottom wall of the first trench no redeposited material was observed between the two Au layers. In this area the Au layers clearly merged (area marked “3” in Figure 4.17). The schematic diagram of the formation of redeposited regions on the walls of the trench is shown in Figure 4.19. Knocked-out silicon atoms and recoil Ga atoms escape from the trench and describe a straight path until they reach either the side-walls of the first trench or the walls of the FIB vacuum chamber. Most of these atoms establish bonds with the surfaces they impinge on. If these redeposited atoms were to remove themselves from these surfaces, they would need to overcome the energy barrier, which is known as surface binding energy.
111
Figure 4.19. Schematic diagram of the redeposition formation shown in Figure 4.17. It can be observed from the geometry of the milled tranches that knocked-out silicon atoms may be redeposited in locations which are on a “line of sight” from their original position on the milled surface of the second trench. That will be on the side-walls of first trench, but not the bottom-wall. The absence of any significant amount of redeposited materials between the two Au protective films in the bottom-wall region of the first trench (area marked “3” in Figure 4.17) confirms that no back-flow of milled atoms to this part of the specimen occurred. Figure 4.20 shows a STEM image and X-ray elemental maps from the corresponding region for SiKα, GaKα and PtLα. The redeposition “pocket” is clearly visible between two gold protective films in the side-wall of the trench on the STEM image in Figure 4.20a as well as on the Si and Ga elemental maps. It can be seen that the silicon “pocket” on the side-wall of the first trench also contains a high concentration of Ga. This suggests that the FIB milling process generates not only a backflow of Si atoms, but also backscattered Ga atoms that redeposit some distance from the region being milled (specifically the second trench).
112
Figure 4.20. (a) STEM image, (b) SiKα, (c) GaKα and PtMα elemental maps of the specimen shown in Figure 4.17. The Si map also shows the lack of any redeposited material between the two Au layers on the bottom-wall of the first trench (Figure 4.20b). It can be noted from the Ga elemental map in Figure 4.20c that there is a high concentration of implanted Ga in both the redeposition layer and in the bottom-wall of the damage layer (just underneath the Au film), which correlates with earlier observations during the damage study in silicon (Section 4.2). Point EDS analysis was done from this side-wall redeposition “pocket”. The EDS spectrum is shown in Figure 4.21.
113
Figure 4.21. EDS spectrum from redeposited region shown in Figure 4.17. As well as the SiKα peak, there are also clearly visible strong GaLα, GaKα and GaKβ peaks. Taking into account the ~ 1 at.% offset of the Ga concentration (which as described before is an integral part of the FIB preparation of a final membrane), the Ga concentration in the redeposition region was measured to be ~ 12 at.%.
114
4.6 Slope Angle Measurements In Figure 4.2 it was noted that the side-wall of milled trench sloped considerably. This was attributed to beam tail at the edge of the Gaussian distribution of the beam. The slope of the side-wall of FIB-prepared TEM specimen results in a “V” shaped crosssection of the final membrane. The thickness of the membrane will be non-uniform. The knowledge of the value of the slope angle may be important for some TEM imaging (for example for HREM). To determine the slope in the FIB-prepared TEM membrane for different ion currents (apertures) the following experiment was conducted. The already prepared TEM membrane (shown in Figure 3.5) was placed back in the FIB system. Next, a row of rectangular cuts was made normal to the plane of the membrane using different ion beam currents at 30 keV (in the current range 11 – 6500 pA). The width of these cuts was around 0.5-1 mm and they had a depth of ~ 4 mm. This sample was then studied using TEM. Results are shown in Figure 4.22.
Figure 4.22. TEM image of thinned membrane with additional FIB cuts made using different ion beam currents at 30 keV beam energy. The ion beam currents used are shown above the cuts.
115
The profiles of the cuts are clearly visible on this image. The slope angles (θ) were measured as angles between the direction of ion beam impact and the direction of straight section of side-wall of a particular cut in Figure 4.22. The measured slope angles are shown in Table 4.1. Table 4.1. Slope angle for different ion beam currents at 30 keV beam energy. Ion beam currents (pA)
Slope angle
6500
1000
350
150
70
11
70
4.50
40
30
20
1.50
The slope angles decrease with decreasing ion beam currents. The use of final mills with smallest beam current (70 or 11 pA) may result in a TEM membrane with the most uniform thickness.
116
4.7 Discussion The focused ion beam technique is a very useful tool for electron microscope sample preparation from either silicon or any other semiconductor device material. FIB allows TEM samples from silicon to be prepared with great accuracy and in a very short time (2-3 hrs). However, damage introduced during FIB fabrication reduces the quality of the sample and can pose serious problems during TEM observation and interpretation of images, especially in HREM. It was found that the FIB milling process resulted in damage on both sides of the specimen membrane. During FIB milling the ion beam is rastered digitally over the milling area. The ion dose delivered to the specimen through duration of one dwell time is shown in Table 3.2. The volume of material, V, removed by charge is defined by equation: V = S × H = Y ×Q = Y × I ×t
(8)
where S is the milling area, H is depth of milling, Y is the sputter rate, Q is the total ion charge delivered, I is the ion beam current and t is time interval. Using the sputter rate (0.15 mm3/nanocoulomb) from the “silicon material” file, the duration of one dwell time (1 msec) and the beam spot size (radius is 0.04 mm) (Table 3.2), a depth of milling for one dwell time duration (H d ) was calculated for a 30 keV beam energy and a 1000 pA beam current,
Hd =
V Y × I × t (0.15mm3 × 109 c −1 ) × (1000 × 10−12 c / sec) × (10−6 sec) = = ≈ 0.03nm (9) π × (0.04mm) 2 S S
The mill depth was calculated to be ~ 0.03 nm or about one tenth of an atomic monolayer. On the one hand that means, that when several monolayers of the specimen are milled, the specimen has been exposed to the ion beam many times and the actual ion dose which is delivered to the specimen is much higher, nominally by an order of magnitude, than that reported in Table 3.2. On the other hand, it is clear that the high radiation dose delivered through one or several interactions with the ion beam is enough to amorphise the specimen, at least to the depth R p . In other words, amorphous region will be created before sputtering occurs. To estimate the actual ion dose delivered to the 117
sample during FIB milling, a correction “exposure” coefficient (k) to the doses calculated from Table 3.2 should be introduced. The approximate value of this coefficient is calculated as the number of FIB single dwell periods which have to be delivered to the specimen for milling to a depth of one half of R p : 1 Rp k= 2 Hd
(10)
For silicon milling using a 1000 pA beam current and a 30 keV beam energy this coefficient was calculated to be ~ 430#. This gives the new values for the actual FIB dose delivered to the silicon sample. To obtain these values the ion dose for particular beam current taken from the Table 3.2 should be multiplied by coefficient k. Thus, the lowest dose was calculated to be 1.3x 1013x k =6x1015 ions/cm2 for the FIB milling using a 10 keV beam energy and a 250 pA beam current. The highest dose was calculated to be ~ 1.5x1014x k = 6x1016 ions/cm2 for FIB milling using a 30 keV beam energy and a 2700 pA beam current. The introduction of the exposure coefficient was reported for the first time in this study. The calculated actual ion dose delivered to the specimen is very important for comparison the FIB created damage with results reported in the literature for ion implantation experiments performed using conventional ion implantation methods. The interaction between the ion beam and the Si target led to the formation of a damage (amorphous) layer for all ion doses (beam currents) available in the FIB system. This means that the working FIB currents used in this study provided ion doses (the range used was 6x1015-6x1016 ions/cm2) exceeding the critical dose, which is required to turn the material into an amorphous state. Such a critical dose has been reported for 100 keV Si self-implantation to be 2x1014 ions/cm2 [93]. However, neither the critical dose for 10 keV nor for 30 keV Ga ion implantation in silicon has not been reported in the literature. It is clear from this study that this critical dose for silicon is below 6x1015 ions/cm2.
#
The depth of milling is proportional to the sputter rate. Thus, the value of the exposure coefficient for particular studied material is inversely proportional to sputter rate. 118
It was experimentally found that the thickness of the damage depends on the ion beam energy (that is, the thickness of the side-wall damage was 28 nm for the 30 keV ion beam energy and 13 nm for the 10 keV beam energy). However, it was found that the thickness of the damage layer does not depend on the ion beam current (at least in the range used in this study). Each working aperture provided gallium ion beams with sufficient dose to create the damage layer with an equilibrium thickness (H e ), whereby as much damage is removed by FIB sputtering as is created. This concept of dynamic damaging is described later in this section. The structure and thickness of the damage in FIB prepared TEM samples were found to be in good correlation with data in the literature. The authors [2, 32, 40, 44, 49, 56, 59, 60, 72, 74]] who have studied FIB-related damage in silicon TEM samples have found that the damage was amorphous in structure. The experimentally determined thickness of the side-wall damage in TEM samples after 30 keV FIB fabrication (~28 nm) coincides well with the data reported by Mardingly and Susnitzky (~ 28 nm) [2], and Venables et al. (~28 nm)[73]. Similarly, for 10 keV beam energy the obtained thickness (~ 13 nm) of the side-wall damage correlates well with data reported by Jamison et al. (~ 11 nm) [107], but is two times larger than reported by Walker and Broom (6 nm) [74]. It should be noted that Walker and Broom [74] did not use an Au sputter coating film to protect the damage layer from any other influence during final TEM specimen preparation. It is possible that the absence of this layer is the reason for the significant difference for the obtained thickness of the damage layer for a 10 keV milling. The thickness of the bottom-wall damage layer in Si after 30 keV FIB fabrication (~56 nm) is in good agreement with the only reported data obtained by Bender and Roussel (60 nm) [35]. These authors observed a damage layer in the silicon substrate, amorphised to a depth of approximately 60 nm with a dark sub-layer occurring in the top of this amorphous layer which they associated with implanted Pt atoms during FIB assisted deposition of a Pt strip. However, they did not do HREM imaging to determine if any crystallites were present.
But again no protective film (like Au sputter coating
protective film used in this study) was used to protect the damage layer. This amorphous damage layer was situated between the undamaged silicon substrate and the platinum protective layer deposited using the FIB. The authors assumed that this dark sub-layer was associated with Pt atoms implanted during deposition of the protective Pt strip, but did not perform EDS to confirm this suggestion. No other studies which have 119
investigated the bottom-wall damage layer after FIB milling were found in the literature. In this study, EDS, combined with the TEM, was used for the first time to characterise the chemistry of the FIB-produced damage layers in silicon and other semiconductors. This technique allowed measurement of the concentration of implanted Ga atoms in different areas of the damage layer. Furthermore, using the mapping facilities of the EDS system, the connection between the structure of the damage layer and the distribution of Ga implanted atoms was established. Regions with high concentrations of implanted Ga atoms were observed near the specimen surface for the first time in this study. The dependence of the thickness of the damage layer on ion beam energy correlates well with the classical model of ion implantation, which is described in Section 2.7.4 [77, 79, 84, 88, 89]. The extent of the damage depends on the incident ion energy, ion mass and the temperature of implantation. Although some researchers [47, 103, 105] believe redeposition is a possible cause of damage layers in FIB-prepared TEM samples, no experimental data, which directly confirm this suggestion, were found in the literature. Yamaguchi et al. [47] reported observation of both spots and rings in the electron diffraction patterns that indicated the presence of a thin polycrystalline GaAs film which covered single-crystalline GaAs specimen after FIB fabrication. However, it was not made clear how this polycrystalline film was formed (that is, by redeposition of sputtering
material
or/and
by
ion-induced
amorphisation
and
subsequent
recrystallisation). Other authors [105] observed the thin layer of redeposition on a wall of a V-shape FIB cut. They have shown that use of such V-shape cuts instead of large rectangular cuts may result in redeposited layers in the FIB prepared TEM sample. In this study it was for the first time shown experimentally by milling a second trench inside an original trench that redeposition of sputtering materials took place in many areas around the FIB milling area (including, for example, on the wall of the first trench (see Figures 4.17, 4.19)). However, this redeposition “pocket” area was not uniform in thickness and was on average much thicker than the normal damage observed in the silicon samples after milling of a single trench. It had an amorphous structure and, as well as silicon, contained a large amount (~12 %) of backscattered Ga atoms. In contrast, the side-wall damage in Si following the milling of a single trench was found 120
to be very uniform in thickness (~28 nm) with a very low concentration of implanted Ga atoms ~2.5 at.%. These two facts clearly indicate that the origin of the damage in TEM samples prepared using standard FIB procedures (described in Sections 2.6.2 - 2.6.4) does not relate to the redeposition of the sputtering material, but rather to direct amorphisation following interaction of the surface with the energetic gallium beam. In contrast, no visible redeposited material was found between the two Au films in the bottom wall of the first trench when a second trench was milled inside it (Figures 4.17, 4.20). This fact indicates that no significant backflow of milled atoms occurred in the vicinity of the specimen inside the FIB chamber during the specimen milling. This result has not been observed before and is very important for an explanation of the nature of the damage in FIB-prepared samples. In Section 4.2 the effect of beam “tail” on the profile of the FIB milled trench was considered. Because of the beam “tail”, the sample was milled outside the original milling area by a distance up to 0.65 mm in length. This length is more than three times the typical final thickness of a silicon membrane required for TEM examination (100-200 nm). This meant that beam rastering during the final membrane cut would also mill both sides of the membrane. Some of sputtered atoms during the final FIB cut may be redeposited on the side-wall of the membrane (parallel to the rastering area), but they may be milled away by beam “tails” during beam rastering. In this case, the final thinned membrane contains no significant amount of redeposited material because there was no significant back-flow of sputtered atoms occured. This result, which has not been discussed before, is very important for an understanding of the nature of damage in FIB-prepared silicon samples. A model of “dynamic damaging” during FIB milling is proposed here to better understand the process of damage formation. A schematic diagram of the dynamic FIB milling process and the process of damage formation is shown in Figure 4.23a.
121
Figure 4.23. Diagram of the dynamic damaging during FIB milling (a) and the related concentration profile of the implanted Ga atoms against depth (x). At a moment of time, t, the existing amorphous damage layer has a thickness H e and lies between the surface marked “1” and the undamaged substrate with the interface marked “3” (that is an overall thickness of damage1+damage2). During the time interval ∆t high-energy Ga+ ions collide with the atoms of the damage layer and remove (sputter) its top part to a depth ∆x (marked as “damage 1” in Figure 4.23.a). At the same time, some of the Ga+ ions and high-energy displaced atoms (often termed “recoil” atoms) penetrate the lower part of the damage (marked as “damage 2”) and the previously undamaged substrate (marked as “damage 3” in Figure 4.23.a) creating further damage in the former and initiating the creation of new amorphous regions in the latter. The amorphous-crystalline interface therefore moves inside the crystal to the same depth ∆x. That is, the overall thickness of the damage layer remains the same. A new position of the interface between the damaged (amorphous) layer and the undamaged crystalline substrate is marked as “4” in Figure 4.23.a. This process may be termed “dynamic damaging”. It can be noted from the diagram of the dynamic damaging process that both incident Ga+ ions and the recoil silicon atoms penetrate the undamaged crystalline substrate through the already damaged amorphous layer (damage layer 2 in Figure 4.23.a). That means that no channelling is likely to occur during the FIB milling process because no crystalline channel exists in this amorphous layer. If there is no channelling, Ga atoms scatter chaotically and the profile of gallium implanted during FIB milling will have no channelling “tail”. This result does not correlate with the suggestion that channelling 122
can play a significant role in damage formation in silicon, which has been speculated by some authors [74, 76]. Another feature of the Ga implantation also follows from consideration of the “dynamic damaging” process. According to the classical model of the ion implantation described in Section 2.7.2, the stopped atom distribution (and distribution of the vacancies produced) has a Gaussian shape with a maximum concentration of implanted atoms N(x) on the length R p [76]. Apparently, this approach is true for low implantation doses or during the initial stage of FIB fabrication. However, it is clear from consideration of the dynamic damaging model described here that during the FIB milling process such a maximum in the distribution of Ga implanted atoms moves dynamically deeper into the substrate. At the same time the upper part of the damage layer is sputtered away (damage layer 1 in Figure 4.23.a). That means that the highest Ga concentration is in the region between the surface and R p , with the maximum in the vicinity of the surface. Such a profile of Ga concentration is shown in Figure 4.23.b. This result correlates well with the experimentally obtained distribution of implanted Ga atoms in the bottom-wall damage layer described in Section 4.4. It was shown in Section 4.2 that the bottom-wall and side-wall damage in silicon after FIB milling had a different structure. The presence of a polycrystalline layer in the upper part of the bottom-wall was presumably a result of dynamic heating of the milling surface by the high-energy ion beam. This extensive heating took place only on the bottom-wall of the trench where the highest beam current density is achieved. But, the recrystallisation was observed only in the top half of the bottom-wall damage layer in silicon. The recrystallisation of amorphous silicon therefore appears to depend on the heat generated and distribution of implanted Ga atoms. Recrystallisation took place only in the upper part of the bottom-wall area with a Ga concentration of ~7 at%. In the lower part of the bottom-wall and the side-wall damage areas, the concentration was below 2.8 at% and no recrystallisation was found. These results are in good correlation with data reported by Williams et al [84]. For implanted Ga, the amorphous to polycrystalline transformation details were reported at a minimum concentration >2.5 at% and a minimum temperature >3500C. Apparently high concentrations of ionimplanted impurities have a strong influence on the crystallisation processes of
123
amorphous Si during thermal annealing. This is true generally for recrystallisation processes. The profile of the damage obtained can be explained from the general implantation model [79, 82]. Using a Monte Carlo calculation the projective range, straggle, impurity and vacancy distribution were calculated for 30 keV Ga+ implantation in a Si target. Results obtained for ions with an angle of incidence 90o are shown in Figure 4.24.
124
Figure 4.24. Monte Carlo calculation for 30 keV Ga+ implantation into the silicon target: (a) ion trajectories; (b) ion ranges; (c) vacancy distribution. The calculation was done for 120 ions implanted normal to the surface.
125
The calculated values for the ion projective range and straggle were 26.6 nm and 9.5 nm respectively. This gave a total depth of Ga penetration of 36 nm, which correlates reasonably well with experimentally obtained data for the thickness of the recrystallised layer in the bottom-wall damage layer. The high dose of Ga ions (around 1014 cm-2) during FIB specimen preparation resulted in the immediate formation of a damage layer. It should be noted that this layer was moving during the dynamic milling process together with the specimen surface. As a result the Ga ions would have interacted mostly with the already damaged (amorphised) near surface region of the target. The presence of this amorphous damage layer during the milling process hindered ion channelling. However, the experimentally obtained thickness of the damage layer in the silicon substrate was nearly twice the calculated data for projected range and vacancy distribution length. The Monte Carlo simulation method consists of following a large number of individual ion or particle “histories” in a target (Figure 4.24a) [82]. Each history begins with a given energy, position and direction. The particle is assumed to change direction and lose part of its energy as a result of binary nuclear collisions. A history is terminated when the energy drops below a pre-specified value. The final ion distribution profile is determined by assuming the distribution to be Gaussian with the mean value R p and standard deviation ∆R p also known as straggle. (Figure 2.18, 4.24b). However, it can be noted from the Figure 4.24a that in case of Monte Carlo calculation even for 120 ions, some of particles (in this case about 10 units) penetrated further than expected (projected range+straggle). This effect produces tails on the concentration versus depth profiles of the implanted ions (Figure4.24b) and on the concentration versus depth profiles of the vacancies produced. Because the straggle is a characteristic of Gaussian part of ion distribution its value does not depend on this tail part of ion distribution. The increase in number of ions in Monte Carlo calculations will also have little effect on the values of projected range and straggle. It was shown earlier in this section that during FIB milling the dose delivered to the sample is almost 3 orders of magnitude greater than the critical dose required for silicon amorphisation. So, the total number of ions which penetrate deeper than straggle may also exceed the critical dose and cause a thicker damage layer than expected. On the other hand such a high implantation dose (6x1016 ions/cm-2) may result in a large number of high-energy secondary silicon ions which may also exceed the critical dose. When the primary Ga 126
ions collided with silicon atoms at the target they transferred part of the energy according to Eq. 3 (Section 2.7.2). In the case of face-to-face collision, almost all the energy of the incident ion (up to 26 keV) can be transferred to the target atom. These secondary recoil atoms will continue to cause displacement damage until their energy falls to a level where the transferred energy during nuclear collisions is less than displacement energy E d . The Monte Carlo calculation was done for implantation of this second category of ions with an energy of 26 keV. Results obtained are shown in Figure 4.25.
127
Figure 4.25. Monte Carlo calculation for 26 keV Si+ implantation (angle of incidence 90o) into silicon target; (a) ion trajectories, (b) ion ranges, (c) vacancy distribution. In this case the calculated value for the range was 40 nm and for the straggle, 18 nm. The total thickness of the damage layer was about 58 nm. The experimental results obtained correlate very well with this calculated data. Using computer Monte Carlo simulation, the thickness of the damage layer for Ga+ implantation into silicon and silicon self-implantation was calculated also for the sidewall damage (an angle 4.50 was chosen for calculations, this angle corresponds the slope angle at 1000 pA beam current and 30 keV beam energy milling) as well as for 10 keV ion beam energy and both incidents angles. Results are shown in Table 4.2. Table 4.2. Calculated data for the thickness of the damage layer for Ga+ implantation into silicon and silicon self-implantation. In the brackets is the data obtained experimentally.
Ion
Calculated thickness (range + straggle) of damage layer in silicon (nm) 30 keV
10 keV
Bottom-wall (90o)
Side-wall (4.5o)
Bottom-wall (90o)
Side-wall (4.5o)
Ga+
36 (56)
13 (28)
16 (23)
7 (13)
Si+
58
26
23
12
It is clearly visible from Table 4.2 that the calculated thickness of the damage layer in case of silicon self-implantation is almost the same as the experimentally obtained data. That means that in silicon the secondary knocked-on atoms may penetrate much deeper and may create more damage then the original heavy Ga+ ions. The range of the measured slope angle (Section 4.6) is very similar to the data reported by other authors [42, 48, 49, 52]. It is clear from comparison of the slope angles for 128
different beam currents that employment of the lowest possible ion beam current for the final FIB mill of TEM membrane may result in the lowest slope. However, the final mill with a 1000 pA or a 350 pA ion beam current, as is usually applied, may result in a membrane slope of 4-4.50. Such a slope is smaller than the slope in the TEM samples for most other sample preparation techniques, which were described in Section 2.3. Further reduction of the ion beam current for final FIB milling leads, on one hand, to insignificant slope reduction, and, on the other hand, to a significant increase in milling time. The use of the 350 or 1000 pA beam current for the final stage of the FIB sample preparation is probably optimal. The damaged layers created during FIB preparation TEM samples may affect subsequent transmission electron microscope analysis. For example, the presence of a ~ 30 nm amorphous damage films on both sides of the FIB fabricated cross-sectional TEM specimen will limit atomic resolution, especially during HREM when a thickness of 50-100 nm is required. However, the additional polycrystalline layer observed along the trench bottom in the case of plan-view specimens may completely inhibit the acquisition of information about the true crystalline structure of such materials in this orientation. The employment of 10 keV FIB final mill may significantly reduce the thickness of the damage layer and significantly improve the quality of the TEM membrane. It is therefore suggested that when the specimen integrity is of paramount importance for HREM imaging, final 10 keV mills are used. It is also recommended that FIB preparation not be used for plan-view specimens, when investigations of specimen crystal structure are to be performed.
129
CHAPTER 5 DAMAGE STUDY IN GERMANIUM 5.1 Introduction Germanium, like silicon and gallium arsenide, is one of most important material in the semiconductor industry. As described in the literature review, the interaction of highenergy ion beams with a germanium substrate may result in the formation of high surface relief. However, no known work describing the use of the FIB for TEM sample fabrication from germanium was found in the literature. In the following sections the study of modification of a germanium surface during FIB milling and the structure of the damage layer is described. In Section 5.2 the study of germanium surfaces of bulk specimens and their modification during FIB fabrication is described. Section 5.3 contains the study of the damaged layers after FIB fabrication. Section 5.4 contains discussion focusing on comparison of the results obtained with known data.
130
5.2 Study of Germanium Surface of Bulk Specimens A row of trenches for different ion beam currents (150 – 6600 pA, at 30 keV) was milled in a germanium sample in the same manner as described for the silicon substrate in the previous chapter. However, it was found during subsequent observation that milling produced trenches with very uneven side-walls and bottom-walls (Figure 5.1). For example, the surface of the bottom-walls of these trenches exhibits a strong cellular relief.
Figure 5.1. FIB image of trenches in a germanium sample milled at 30 keV beam energy and 2700, 1000, 350 pA beam currents (from left to right). For better observation of the surfaces of the walls, the sample stage was tilted by about 450. The walls of the trenches milled using a beam current of 2700 pA are shown in Figure 5.2. Tilting of the specimen showed that the side-walls of the trenches were also very rough, and exhibited a “waterfalling” type of relief.
131
Figure 5.2. FIB image of wall surface of the trench (2700 pA beam current, 30 keV beam energy), stage tilt 450. The side-wall surface of the trench (Figure 5.2) has cellular relief and looks like a cascade of very sharp hills containing some relatively flat areas around 0.5 mm2 in size (for example marked x in Figure 5.2). Such relief was common for all used ion beam currents (150 – 6600 pA, at 30 keV). It was noted that the specimen surface around the trenches was also significantly modified during the processes associated with initial image observation and mill setup. This effect was especially visible during conventional imaging at relatively highmagnification (x25000), even under relatively small ion beam currents (70, 150 and 350 pA). In this mode of operation the ion dose delivered to the sample is comparable to the dose delivered during the milling mode (~ 1014 cm-2). To study the effect of ion beam modification of the Ge surface during FIB continuous observation, a number of FIB images were taken at intervals of five seconds. These images are shown in Figure 5.3 (the beam was held constant over this area, but images were recorded at five-second intervals). This was done at 30 keV with a beam current of ~ 350 pA.
132
Figure 5.3. Transformation of specimen and trench surfaces during 30 keV, 350 pA beam current observations. FIB images (a)-(d) were taken at five-second intervals. It is clear that modification of the sample surface around the trench occurred during the recording of these images (Figures 5.3(a)-(d)). The structure of the cells on the bottomwalls of the trenches also changed as the observation time increased. If the structure of cells mostly inherits the structure of the previous image (at five-second intervals), the structures of the bottom of the trench in the first image (a) and the last (d) are different. Magnified images of the right lower corner of these trenches are shown in Figure 5.4.
133
Figure 5.4. Magnified images of trench corner from Figure 5.3. The development of the cellular structure in Figures 5.4 (a) to 5.4 (d) is clearly visible for a particular cell marked by arrows. The shape and the size of this cell changed through images (a) to (d) in this figure. The dimensions of this cell grow from image (a) to image (c) and then diminish in image (d). Similarly, the FIB images, which were taken at a sequence of 30 sec intervals during milling with a 30 keV energy and ~ 350 pA beam current observation, (Figure 5.5a, b) show modification of the cellular structure of the bottom-wall of the trench. It can be also noted from Figure 5.5a,b that the surface of the germanium sample around the trench finally becomes similar in structure to the bottom of the trench.
134
Figure 5.5. Further development of specimen and trench surfaces during FIB observations. FIB images (a) and (b) were taken at 30-second intervals after the image in Figure 5.3d was taken; image (c) was taken after 6 min observation under 10 keV and 250 pA beam current; image (d) was taken after 2 min observation under 30 keV and 70 pA beam current. To investigate this phenomenon in further detail the effect of a reduction in both the incident ion energy and ion dose delivered to the germanium specimen was studied. For this purpose a 6-minute FIB observation under a 10 keV beam energy and a 250 pA ion beam current (ion dose 0.5x1012 cm-2) was set up with the same magnification around the same trench (as shown in figure 5.5b). Results after a six-minute observation are shown in Figure 5.5c. The cellular structure of the bottom-wall of the trench has almost disappeared. It is clear from this image that either low ion energy and/or ion dose (0.5x1012 ions/cm2) markedly improves the smoothness of the specimen surface around and inside the trench.
135
However, following a further two-minute observation under a higher ion dose (30 keV and 70 pA beam current) the surface smoothness began to deteriorate (Figure 5.5d) and the specimen surface returned to conditions similar to those shown in Figures 5.5a,b. An additional trench was milled using a beam with a 10 keV energy and 250 pA Ga+ ion beam current (ion dose 0.5x1012 ions/cm2). The size of the trench was 6.7x4.5x0.2 mm. The side-walls and bottom of this trench were noted to be relatively smooth. The FIB image of this trench is shown in Figure 5.6.
Figure 5.6. Direct (a) and tilted by 45o (b) FIB images of 0.2 mm depth trench milled at 10 keV beam energy and 250 pA beam current. To determine which of the two parameters, the ion beam energy or the ion beam current density, is more important in affecting the surface quality of Ge, a further experiment was carried out in which a sequence of mills was produced on the right half of the trench shown in Figure 5.6. All mills were 0.05 mm in depth according to the FIB software for silicon #. The first mill was carried out using a 30 keV beam energy and a 4 pA beam current (dose 2.3x1013 ions/cm2). Results are shown in Figure 5.7a,b.
#
The FIB workstation software uses a ‘silicon’ materials file (with the sputtering yield for Si) in the calculation of the milling parameters. The sputter rate for 30 keV Ga FIB milling was found to be four times greater. So the real depth of the mills was four times larger. 136
Figure 5.7. Direct (a) and tilted (b) FIB images of the trench from Figure 5 with a new 0.05 mm depth trench milled at 30 keV beam energy and 4 pA beam current. It is clear from the images in Figure 5.7 that the roughness of the new mill surface increased slightly. A further mill was fabricated using a 30 keV Ga+ beam energy and an 11 pA beam current (dose 3.8x1013 ions/cm2). The results are shown in Figure 5.8a. Again, the roughness of the right part of the trench has obviously increased.
Figure 5.8. FIB images of trenches with 11 pA (a) and 70 pA (b) mills produced in the right half of trench. A further mill was carried out using a 30 keV beam energy and a 70 pA ion beam current (ion dose 8.9x1013 ions/cm2). Here, the roughness of the specimen surface increased significantly (Figure 5.8b). The surface of the 70 pA mill was similar to those observed earlier (Figure 5.1), which were prepared using 350, 1000 and 2700 beam currents. This experiment has shown that the main factor determining germanium 137
surface roughness during FIB fabrication is the Ga ion beam current density (implantation dose). The mills made with low beam current density led to better quality surfaces of the germanium substrate. Therefore, relatively smoother milling surfaces, which are desirable for good quality TEM samples, can be obtained by using lower beam current density mills (see Table 3.2). The FIB workstation allows the lowest Ga ion dose at 10 keV beam energy and 250 pA beam current (0.5x1012 ions/cm2). But it can be noted from Figure 5.6 that even at such a low dose the surface of the side-wall in germanium is not as smooth as observed for the silicon samples (Figures 3.6, 3.7). Next, the ability to improve the smoothness of the side-walls of the trench using low ion beam current density was studied. For this purpose a much deeper (4 mm, according to FIB software) trench was milled using a 30 keV energy and a 2700 pA beam current. As expected, the side-wall surface of the trench was very rough (Figures 5.9a,b). Next, a second mill was carried out around one half of this trench. The second mill was performed using a 10 keV ion energy, a 250 pA ion beam current and to a depth of 0.3 mm depth (according to FIB software). Results are shown in Figures 5.9c,d. The effect of this mill is clearly visible in Figures 5.9c,d. Comparing images in Figures 5.9b,d it can be noticed that there is a relative improvement in the quality of the side-wall surface. Some sharp features, which appear in Figure 5.9b, have disappeared. This surface looks much smoother than before the mill. However, the wall surface is not flat; moreover, there is visible terrace-like structure on the surface.
138
Figure 5.9. FIB images of a deep trench milled at 2700 pA ion beam current (a, b) and subsequent 10 keV, 250 pA mill (c, d). The tilt of the specimen was 450 (b,d). It was also found from Figure 5.9d that the trench milled with a 10 keV beam energy was much deeper than anticipated during the mill setup. This shows that sputtering yield for Ga ions in germanium is much higher than for silicon. The sputtering yield for Ga ions in germanium can be calculated by measuring the real depth of milled trenches and is described in the next section.
139
5.3 Damage Study in Germanium After FIB Fabrication For study of the structure of the damage in germanium after FIB fabrication a set of trenches was milled on the surface of a germanium sample at 10 keV (250 pA) and 30 keV (11500, 6600, 2700, 1000, 350 and 150 pA beam currents, correspondingly). The dimensions of the trenches were around 15x10 mm2 with depth 0.2-0.7 mm #. The crosssectional TEM specimens were prepared from the walls of the trenches as described in Section 3.7. The TEM study shows a similar structure of the damage layer for all beam currents (and 30 keV Ga ions energy) used for trench milling. A TEM image of the cross-section of the 1000 pA trench walls is shown in Figure 5.10. There are clearly visible remains of both the Pt and Au protective films, the germanium substrate and a complex damage layer in between. Unlike the smooth, uniform TEM transparent membranes observed in silicon (Figures 3.6, 3.7, 4.2) there is clearly visible “waterfalling” relief of the thinned membrane. The damage layer varies greatly in structure and thickness across the thinned section. Its top part represents the structure (topology) of the surface of the milled trench and corresponds with previous FIB observations (Figures 5.1-5.3). The sputter coated Au film inherits the shape of the trench surface. Four areas with different structures of the damage layer (marked A, B, C, D) around the trench can be observed. The bottom-wall of the trench (marked D) consists of many sharp hillocks (cells) with some voids (visible as white spots in the image) with irregular shapes inside. The sidewall damage of the trench (marked C) is relatively uniform but has a few small voids marked by arrows. The top of the damage layer on the specimen surface around the edge of the trench (marked B) also has a cellular structure with many voids, but the voids appear smaller than in the region marked D. The voids almost disappear in the region marked A.
#
The FIB workstation software uses a ‘silicon’ materials file (with the sputtering yield for Si) in the calculation of the milling parameters. The sputter rate for 30 keV Ga FIB milling was found to be four times greater. So the real depth of the mills was four times larger.
140
Figure 5.10. Cross-section TEM image of the walls of the trench milled using a 1000 pA beam current and a 30 keV beam energy. Arrows indicate voids in side-wall. A higher magnification image from the bottom-wall region (marked D in Figure 5.10) is shown in Figure 5.11. It can be seen that the lower part of the damage layer (marked x) is relatively uniform in thickness. The typical crystalline diffraction contrast clearly visible in the germanium substrate in Figures 5.10, 5.11 is not visible in the lower part of the damage layer, suggesting that it is amorphous in structure. The voids were found only in the top part of the bottom-wall of the trench (marked z). The diffraction contours (bend contours and thickness fringes) were not observed in the lower part of the damage layer during the TEM study. This fact indicates a non-crystalline structure of the lower part of the bottom-wall damage.
141
Figure 5.11. Magnified TEM image of area marked D from Figure 5.10. Another example of the bottom-wall of the trench milled using a 2700 pA beam current and a 30 keV beam energy is shown in Figure 5.12. It can be noted from Figures 5.115.12 that the total thickness of the bottom-wall damage in the germanium (amorphous + porous cells) exceeds 250 nm and is independent of beam current. The size of the voids was found to be greatest on the bottom-wall of the trench. Voids were often elongated in shape. Void size was up to 150 nm in height and up to 60 nm in diameter (Figure 5.12). It was found that they were surrounded by some small size particles. These particles had darker appearance than the surrounding amorphous germanium. The typical size of the smallest particles was around 3-5 nm. The size of the largest particles exceeded ~ 30 nm.
142
Figure 5.12. TEM image of voids and surrounding particles in the bottom-wall of the trench milled using a 2700 pA beam current and a 30 keV beam energy. The magnified images of particles on the bottom wall of the trench are shown in Figure 5.13a. Letter A indicates the crystalline undamaged germanium substrate, B is the lower part of the damage layer, which is amorphous in structure, and C is the area which contains particles. A high-resolution image of one of the particles is shown in Figure 5.13b. Two sets of atomic planes are visible. It is clear that the particles have a crystalline structure. The size of this crystallite is estimated to be 10 nm. It is also clear that the matrix around the crystallites is amorphous in nature.
143
Figure 5.13. High-resolution images of the bottom-wall damage with crystallites (a), and separate crystallite (b). A selected area diffraction pattern from the bottom-wall damage area is shown in Figure 5.14. There are clearly visible rings which contain many separate diffraction spots. This type of diffraction pattern also indicates the presence of small crystallites with a random crystallographic orientation.
Figure 5.14. Selected area diffraction from the bottom-wall of the Ge damage layer shown in Figure 5.12.
144
EDS spectra were taken from the top (Figure 5.15) and the lower (Figure 5.16) parts of the bottom-wall damage area of the trench. As well as strong Ge peaks a Ga peak, is clearly visible in Figure 5.15 but is of negligible intensity in Figure 5.16. The measured concentration of Ga was ~ 6 at. % in the top part of the damage layer and ~ 2.7 at. % in the lower part.
Figure 5.15. EDS spectra from the upper part of the bottom-wall damage of the trench milled using a 1000 pA beam current and a 30 keV beam energy.
145
Figure 5.16. EDS spectra from the lower part of bottom-wall damage of the trench milled using a 1000 pA beam current and a 30 keV beam energy. The side-wall area of the trench (marked C on Figure 5.10) is more uniform in structure than the bottom-wall. A magnified image of this region is shown in Figure 5.17. Only a few elongated voids are visible (indicated by arrows). The width of these voids is much smaller than in area D and is ~ 30 nm; the height is ~100 nm.
146
Figure 5.17. Cross-section TEM image of the side-wall of the trench milled using a 1000 pA beam current and a 30 keV beam energy. The study of the side-wall damage of the trench has shown that as well as the presence of small holes (arrowed in Figure 5.10), some relatively uniform amorphous damage areas exist. The length of these uniform areas exceeds 0.5 mm, which is in good correlation with the FIB observation of bulk specimens (flat regions in the side-wall in Figures 5.2, 5.9). The magnified image of such an area is shown in Figure 5.18. There are no crystallites visible inside this damage layer. Diffraction analysis indicates its amorphous structure. The thickness of the damage layer in the side-wall region for a 30 keV milling was found to be 29 ± 3nm.
147
Figure 5.18. Magnified image of the side-wall damage of the trench milled using a 2700 pA beam current and a 30 keV beam aperture.
Figure 5.19. EDS spectra from the side-wall damage of the trench milled using a 2700 pA beam current and a 30 keV beam aperture. EDS spectra from the side-wall damage of the trench is shown in Figure 5.19. There are visible strong Ge peaks and Au peaks from the protective film. The Ga concentration in this area was found to be ~ 3 at. %. 148
From Figure 5.10 it is also clear that the FIB related damage extends far from the edge of the trench (Figures 5.10, 5.17). The surface of the germanium sample around the milled trench is also heavily damaged (Figure 5.10, areas A, B). This damage reflects the profile (“tail”) of the ion beam during the trench fabrication. The cross-sectional magnified image of region B (Figure 5.10) is shown in Figure 5.20. The damage layer also consists of two parts: lower amorphous (marked x), and upper porous (marked y). The amorphous part is relatively uniform in thickness. The average thickness of this amorphous part was around 25 nm. The porous structure of the damage layer is clearly visible on the surface of the specimen near the edge of the trench (area marked B on Figures 5.10, 5.17 and 5.20). The size of voids increases from ~ 3 nm to 30-50 nm near the edge of the trench. The voids appear to be irregular in shape.
Figure 5.20. The cross-section TEM image of the specimen surface near the edge of the trench milled using a 1000 pA beam current and a 30 keV beam energy. The crystallites are also clearly visible around the voids in the upper part of the damage layer in this area (marked y in Figures 5.20, 5.21).
The thickness of the
porous/crystallite part of the damage layer is greatest at the edge of the trench and
149
gradually (Figures 5.20, 5.21) decreases to zero at a distance around 3 mm from the edge of the trench (Figure 5.10, area marked “z” in Figure 5.21).
Figure 5.21. Magnified TEM image from the area marked “A” in Figure 5.10. The voids in this region are around 3-30 nm in diameter. The voids and crystallites were not observed in the damage layer on the surface of the germanium specimen far from the edge of the trench (area marked “z” in Figure 5.21). This damage (amorphous) layer is shown in Figure 5.22. No crystallites are visible. The damage layer is relatively uniform. The interface between the Au protective layer and the amorphous damage layer is rather sharp. The thickness of the amorphous layer in germanium was found to be 38±4 nm for 30 keV Ga+ ion beams and independent as a function of a beam current.
150
Figure 5.22. The cross-section TEM image of damage in the germanium substrate in the area around 3 mm from the edge of the trench milled using a 1000 pA beam current and a 30 keV beam energy. The interface between the amorphous damage and the germanium substrate is not so sharp. The magnified high-resolution image of this interface is shown in Figure 5.23. Two sets of atomic planes are clearly visible. The transition area from crystalline to amorphous is in the range of 6-8 atomic planes.
151
Figure 5.23. High-resolution image of the crystalline to amorphous transition area in the bottom-wall damage of the trench milled in the germanium substrate using a 1000 pA beam current and a 30 keV beam energy. It can be noted from the Figure 5.23 that the presence of the damage layers (with a total thickness ~ 58 nm) on both sides of the germanium membrane still allow a good quality high-resolution image to be obtained. The contrast is high, and the columns of germanium atoms are clearly visible around the substrate. This image corresponds to the structure of the germanium lattice in the [110] direction.
152
5.4 Damage Study in Germanium After a 10 keV FIB Milling The cross-sectional structure of the damage in 10 keV-milled trenches was found to be significantly different to those milled at 30 keV (Figure 5.26). The cross-sectional TEM image of the bottom-wall of the trench exhibited a markedly wavy shape (marked A on Figure 5.26) with a period of about 0.3 mm. The results obtained correlated well with the earlier FIB observation of the bulk specimen (Figure 5.9) and ion beam trace (a spot size to be 0.3 mm for a 250 pA beam current and a 10 keV beam energy (Table 2.2)). During milling in this particular case the beam movement (rastering) was from top to bottom and from left to right. This presumably resulted in the bottom-wall of this trench taking the shape as it did.
Figure 5.26. The cross-section TEM image of the 10 keV trench in the germanium substrate. The magnified images of the bottom-wall area marked A and side-wall area marked B are shown in Figure 5.27.The top part of the damage layer on the bottom-wall of the trench also consisted of small holes and fine crystallite particles (Figure 5.27a). The total thickness (amorphous part + holes/crystallite part) of the damage layer of the
153
bottom-wall of the trench was found to be 30±3 nm. The side-wall damage was found to be amorphous in structure with a thickness of 12±2 nm (Figure 5.27b).
Figure 5.27. Magnified cross-sectional images of (a) the bottom-wall and (b) the sidewall of the trench milled using the 250 pA beam current and the 10 keV beam energy. The damage in the germanium sample away from the milled trench was found to be uniform in thickness (area marked B on Figure 5.26) and amorphous. Only a few holes and crystallites were found near the edge of the trench. The thickness of the amorphous damage area was found to be 23±3 nm.
154
5.5 Determination of the Sputter Rate The sputter rate for the FIB milling of germanium samples using a gallium beam is not known. During the milling of the trenches described in the previous sections, the FIB silicon material file (with a sputter rate 0.15 mm3/nanocoloumb for silicon, 30 keV beam energy) was used for the trench depth calculation. In these experiments the depth of the trench shown in Figure 5.10 was set up during FIB milling to be 0.6 mm according to the FIB software for the silicon material file. Using the cross-sectional TEM image (Figure 5.10) the actual depth of the trench was measured as ~ 2.4 mm. That means that the sputter rate for the 30 keV Ga FIB milling of germanium is ~ 4 times higher than that of silicon. Its value was calculated to be ~ 0.6 microns3/nanocoulomb. During the FIB milling of the 10 keV trench (Figure 5.26), the chosen depth was 0.2 mm. The actual depth of the 10 keV trench was measured in cross-sectional images as ~ 0.77 mm. This, again, means that the sputtering yield for gallium ion beams in germanium was ~ 4 times greater than for silicon.
155
5.6 Discussion From the point of view of FIB TEM specimen fabrication, germanium has received almost no attention. However, the investigation of the FIB milling processes in germanium in this study led to the observation of several new features, which were not observed in other similar materials (in silicon, for example). The first feature of note is the high sputtering yield of the germanium samples. The sputter rate for germanium has not been reported before. The sputtering yield for germanium was found to be ~ 4 times greater than for silicon. That is, the sputter rate for FIB milling of germanium using a 30 keV Ga ion beam is ~ 0.6 microns3/nanocoulomb, which is the volume of material removed per unit of charge. The Ge atoms displaced by the energetic Ga ions must overcome some energy barrier to leave the surface of the target. This barrier is known as the surface binding energy. It can be noted from Table 2.1 that the surface binding energy is highest in silicon (4.7 eV), but less in germanium (3.9 eV) and lower again in III-V semiconductors. The surface binding energy is the key parameter for the sputter rate in the particular material. The lower surface binding energy increases the number of atoms leaving the target and results in a higher sputter rate. Ultimately, the obtained sputter rate may be used for the creation of a germanium materials file for use in the FIB system. The second notable feature was the formation of very rough surfaces with a cellular structure during FIB milling of bulk samples with an ion dose greater than 3.4x1015 ions/cm2#. The surface of the milled trenches had a sponge-like (cellular) appearance. These results are similar to those obtained by several authors who studied ion implantation in germanium. Wilson [99], using the SEM, observed the formation of a complex cell structure in self-implanted germanium at a threshold ion dose above ~ 2x1015 ions/cm-2 for a 50 keV ion energy. Similarly, Holland et al. [97] observed cellular relief of the damage layer in germanium after a dose of 5x1015 ions/cm-2 for 120 keV In ions. Appleton et al. [98], using cross-sectional TEM, found that columnar voids erupted through the surface of Bi ion-implanted amorphous Ge (dose 4x1015 ions/cm2, 280 keV ion energy). Despite the difference between the ion energies and ion species #
The exposed ion dose was calculated using exposure coefficient for Ge. The sputter rate in Ge is four times greater than in Si. Thus, the exposure coefficient is four times lower. It was calculated to be ~ 83. 156
used in the reported experiments and experiments performed in this study, the value of the critical ion dose for the void formation correlates very well. It was found from TEM studies that the bottom-wall of the trenches exhibited cellular relief that consisted of many voids with an average diameter of ~ 60 nm. TEM showed that the intersection of some of these voids with the surface of the specimen produced the cellular (or sponge-like) structures on the specimen surface. Similar relief on germanium surfaces after high dose ion implantation has been observed before [97-99], but the origin of such relief was not understood. The cross-section TEM investigations performed in this study showed that this sponge-like structure was present only in the top part of the bottom-wall damage layer induced during the FIB milling process. In contrast, the lower part of the damage layer was completely amorphous with a relatively sharp interface between it and the crystalline substrate (with a transition region of 6-8 atomic planes). Such voids were not found in damage layers in other semiconductors (for example in silicon). According to Monte Carlo calculations each 30 keV Ga ion produces 820 vacancies in the germanium sample (in comparison 520 vacancies/ion for silicon). The calculated total volume of vacancies produced per Ga ion in germanium can be estimated to be ~ 140 nm3. This higher density of vacancies presumably contributed to the nucleation of voids in the amorphised Ge during Ga irradiation. Nevertheless, the exact mechanism responsible for void formation is still not clear. It may be possible that the supersaturation of vacancies in the near surface region leads to precipitation of these voids. Moreover, the local heating of the specimen during FIB milling increases mobility of the vacancies. It was also found during this study that the voids in the damage layer of these trenches were surrounded by small crystallites. Although the voids inside the amorphous damage layer have been observed by some authors [86, 97-99], the presence of crystalline particles in these areas in germanium was not previously reported. In these studies the damage layer was studied after ion implantation of Ge, In, or Bi ions. But it is not clear if the microcrystals were not found there because these workers did not perform HREM studies or because the formation of microcrystals in damaged layers depends on the type of implanted atoms or the implantation conditions used. As described earlier, such crystallites, as a result of the recrystallisation process, were also found in top part of the bottom-wall damage layer in silicon after Ga FIB milling (Section 4.2.1). The micro157
crystals in germanium were also found only in the upper part of the bottom-wall damage where the concentration of implanted Ga atoms was found to be ~ 6 at.%, significantly higher than in the lower part of the layer. Again, it would seem that the presence of a large amount of Ga-implanted atoms, combined with local heating, assisted the recrystallisation process. This partial recrystallisation again occurred only in the upper part of the damage layer and mostly in the bottom of the trench where the highest ion beam density generated extensive heating of the sample. During trench milling the focused ion beam is rastered around the bottom of the trench and the side-walls of the trench are mostly affected by the beam “tail” with a much lower beam current density. So, extensive heating only occurs in the bottom of the trench. The crystallites were found inside the amorphous germanium matrix around the voids. This means that some recrystallisation of the originally amorphous damage area occurred during or immediately after FIB milling. Apparently, the process of recrystallisation was accompanied by a morphological transformation of the specimen surface. Local heating of the sample generated the formation of germanium crystallites, which in turn generated diffusion of displaced germanium atoms to the centres of crystallization. This, in turn, generated both the diffusion of vacancies away from the crystallites and the further growth of the voids. The voids in the vicinity of the specimen surface are an effective sink for the diffusion of these vacancies. This growth caused the transformation of the germanium surface. A study of the side-wall structure in the bulk germanium samples has not been reported before. The side-walls of the milled trenches in the bulk specimen also exhibited cellular relief containing some relatively flat areas around 0.3 mm2 in size (Figure 5.2). These flat areas corresponded to uniform amorphous damage areas, which were observed in the cross-sectional TEM images (Figure 5.10) of the side-wall of the trench (length up to 0.5 mm, thickness ~ 29 nm, for 30 keV beam energy). The small size of these flat areas in FIB-prepared thin membrane limits the total area available for TEM imaging. The amorphous side-wall damage with the thickness around 29 nm on either side of the TEM membrane partially affected analysis. A uniform amorphous damage layer was also found on the surface of the germanium sample, several microns from the trench edge. Apparently, this damage layer was created by beam “tails”. This effect of the beam “tails” has not been reported before and 158
is apparently connected not only with the beam distribution, but also with the high sputter rate for germanium samples. Although the crystalline germanium became amorphous in this region, no voids were found. That means that the ion dose was below the critical dose which results in void formation. The thickness of this amorphous layer on the germanium surface was measured to be 38±4 nm for a 30 keV Ga+ ion beam. The thickness of the damage layers observed by TEM was compared with the calculated value according to the classical model of ion implantation (range + straggle). Calculations were made using the TRIM program for Ga ion traces, range, straggle and vacancy distribution. The results are shown in Figure 5.28 and summarised in Table 5.1. The thickness of the side-wall damage layers was calculated to be 13 nm (30 keV beam energy) and 7 nm (10 keV beam energy). For these calculations the angle of incidence was taken to be 4.5o, which corresponds to the average slope angle in FIB-prepared TEM samples. For 10 keV the thickness of the bottom-wall damage layer was calculated to be 11 nm. Calculated data for total penetration depth (range+straggle) for 30 keV Ga ions in Ge was found to be ~ 26 nm, which was about 25 % less than the experimental data.
159
Figure 5.28. Monte Carlo calculations of (a) ion trajectories, (b) range and (c) vacancy formation for 30 keV 120 Ga ions on the germanium target (angle of incidence 90o).
160
Table 5.1. Calculated data for the thickness of the damage layer for Ga+ implantation into germanium and self-implantation of germanium recoil atoms. In the brackets is the data obtained experimentally.
Ion
Calculated thickness (range + straggle) of damage layer in germanium (nm) 30 keV
10 keV
Bottom-wall (90o)
Side-wall (4.5o)
Bottom-wall (90o)
Side-wall (4.5o)
Ga+
26 (38)
13 (29)
11 (23)
7 (12)
Ge+
25
14
11
6.5
This difference cannot be explained using a model of self-implantation of secondary knocked-out (recoil) atoms. As shown earlier, this model was successfully applied to describe the thickness of the amorphous layers for silicon specimens. However, because Ga and Ge are neighbours in the Periodic Table, they have similar mass and size. According to Eq.3 in Section 2.7.2, for a head-on collision almost all the initial energy may be transferred from the 30 keV Ga ion to a target Ge atom. Monte Carlo calculations give data for ion range, straggle and the density of vacancies for 30 keV Ge during self-implantation similar to those calculated for Ga (Table 5.1). However, the difference between the experimental and theoretical data can be explained by using the following dynamic approach. The Monte Carlo model describes accurately the events in the ion collision cascade for target material with a particular density (in the case of Ge 5.35 g/cm3, Table 3.1) [82]. However, during the FIB milling process the initially formed damage layer is not static and moves dynamically inside the target (the proposed model of the “dynamic damaging” was described in detail in the silicon section). Some of energetic ions collide with the atoms in the near-surface region of the amorphous damage layer and knock them out. However, other ions penetrate through this amorphous layer and cause further amorphisation of the crystalline germanium substrate. It can be noted from the data obtained by Monte Carlo calculations that germanium has a very large number of vacancies per incident Ga ion. That means that
161
the density of the amorphous Ge formed by Ga irradiation is much less than for the original Ge target. Obviously, Ga ions lose less energy when they penetrate such a heavily damaged, vacancy-rich area, and this can produce displacement events at a greater distance than in a fully crystalline Ge matrix. The observation of the considerable thickness of the damage layer correlated well with the reduction of density of the near surface region during ion implantation. This has not been reported in the literature before. It has also been shown that the reduction in ion dose dramatically reduces the roughness of the damage on the surface of germanium samples (the size and the height of the cells). Using a cleaning mill with a 4 pA beam (exposure dose 1.9x 1015 ions/cm2) aperture produced a relatively smooth surface of germanium sample. However, the thickness of the damage layer remained the same. Using the final mills of TEM membrane with a 10 keV energy and a 250 pA beam current further decreased the cellular structure of the surface and decreased the total thickness of the damage layer in germanium by a factor of two. This results, on the one hand, in an increase in the size of the area suitable for TEM observation and, on the other hand, in an increase in the quality of the TEM samples. However, the use of a 10 keV energy FIB milling may result in an increase in milling time.
162
CHAPTER 6 III-V SEMICONDUCTORS 6.1 Introduction Unlike Si, the FIB fabrication of TEM specimens from III-V semiconductors has not been extensively studied. Only a few studies were found in the literature relating to the FIB fabrication of InP and GaAs (Section 2). Further, no known work employing the FIB technique for TEM sample preparation from other prospective III-V semiconductors such as InAs, GaN or InSb was found in the literature. It was shown in previous sections that damage layers formed during FIB milling may have different structure and thickness from material to material, i.e., between Si and Ge. Moreover in the case of FIB milling of germanium the bottom-wall damage had a complex structure with an amorphous matrix, micro-crystals and voids. Such damage layers in TEM membranes may cause some confusion during TEM investigations. In the following sections the employment of the experimental procedure, which was used to study the damage in silicon and germanium, is described for a number of III-V semiconductors. Section 6.2 describes the study of the FIB related damage in InP, Section 6.3 in InAs and Section 6.4 in GaAs. Section 6.5 describes the determination of sputter rates for a number of III-V semiconductors. Obtained results are discussed and compared with known data in Section 6.6.
163
6.2 Indium Phosphide 6.2.1 Damage Study in Indium Phosphide After a 30 keV FIB Milling An InP specimen was mechanically ground to a thickness of ~ 25 mm. Then a row of trenches was milled in this specimen using a range of beam currents (from 150 pA to 6600 pA with a beam energy of 30 keV). The size of the trenches was around 20x10 mm2 with a depth 0.2-0.6 mm #. Examples of these FIB milled trenches are shown in Figure 6.1. FIB observation of the milled trenches in InP showed that the bottom-walls and side-walls of the trenches were very smooth. No significant milling relief was found (Figures 6.1- 6.2). The smoothness of the walls of the trench was very similar to the walls observed in silicon (Section 4.2).
Figure 6.1. FIB image of milled trenches in InP sample milled using a 2700 pA and a 1000 pA beam currents at 30 keV. #
The FIB workstation software uses a ‘silicon material’ file (with the sputtering yield for Si) in the milling parameters calculation. The sputter rate for 30 keV Ga FIB milling of a InP sample was found to be nine times greater. So the real depth of the mills was nine times larger.
164
The effect of beam “tails”, described in Section 4.2, is clearly visible on the tilted (angle 45o) FIB image of the trench milled with a 2700 pA beam current (Figure 6.2). Clearly, the left and the right edges of trench (marked with letters A and B) were milled away by beam “tails”. The profile of the left and right side-walls of the trench exhibits a Gaussian shape.
Figure 6.2. Tilted (450) image of a 2700 pA beam current milled trench, a 30 keV beam energy. Arrows indicate a sinusoidal relief. On the upper part of the side-wall of the trench a small amount of sinusoidal relief (marked by arrows) can be observed. This is connected with the nature of the beam rastering during the milling process. Such relief was not found in the lower part of the side-wall or the bottom-wall of the milled trenches. To protect the superlattice on the surface of these specimens they were initially sputter coated with thick (around 200 nm) Au films.
Using the experimental procedure
described in detail in Section 3.3, cross-sectional TEM samples containing walls of the 165
trenches were prepared. A low magnification cross-sectional image of the side-wall and bottom-wall of the trench milled using a 1000 pA beam current and a 30 keV beam energy is shown in Figure 6.3.
Figure 6.3. A low magnification cross-sectional TEM image of the trench milled in InP sample using a 1000 pA beam current and a 30 keV beam energy. Similar to the previously studied Si and Ge samples, the InP substrate, Pt and Au protective films are readily identified. The relief of the side-wall surface is very flat. Unlike Si, the relief of the surface of the bottom of the trench is not so flat on a macro level. The reason for this is not clear. The TEM contrast around the InP substrate is very similar to that observed in the Si substrate (Figure 4.3). The “waterfalling” relief of the thinned membrane, which was visible in all the Ge specimens, was not found in the InP TEM specimens. The effect of the beam “tail” which was clearly visible in Figure 6.3 or in the Si or Ge samples is not visible in this image because, in this particular case, the sample was initially sputter coated with a Au protective film. The damaged layers (marked by arrows in Figure 6.3) are also clearly visible.
166
Figure 6.4 shows a TEM cross-section of the side-wall of the trench milled using a 1000 pA ion beam current. The protective Pt, Au films and the InP substrate are indicated on this image.
Figure 6.4. TEM cross-section image of the side-wall of the trench prepared using a 1000 pA beam current and a 30 keV ion energy. The side-wall damage layer, which is clearly visible between the InP substrate and Au protective film, is uniform in structure and thickness. A high-resolution TEM image of this damage layer is shown in Figure 6.5. The interface between the Au protective film and the damage layer is sharp. Atomic planes are clearly visible in the InP substrate. The side-wall damage layer has an amorphous structure similar to that observed in Si and Ge. Again no crystalline clusters are visible in the side-wall damage layer.
167
Figure 6.5. HREM image of the damage layer in the side-wall of trench milled with a 1000 pA beam current and a 30 keV beam energy. The interface between the damage layer and the crystalline substrate is less sharp. The transition between the substrate and the amorphous damage layer is visible as a dark band (arrowed). The diffraction contrast from this area indicates a large number of crystal defects, although this area is not completely amorphous. Some atomic planes are visible in this region. The thickness of this transition zone is typically 5-9 atomic planes. The thickness of the amorphous side-wall for 30 keV trenches was measured to be 33±3 nm for all beam currents used (the thickness of the side-wall damage at 30 keV was found early to be 28 nm for Si and 29 nm for Ge). Examples of the bottom-wall damaged area of the trench milled using the 1000 and the 150 pA beam currents are shown in Figure 6.6a,b.
168
Figure 6.6. TEM cross-section images of the bottom-wall damage of the trenches milled with a 30 keV ion energy, beam current 1000 pA (a) and 150 pA (b). The structures in these images are easily identified. The Pt and Au protective films are indicated. The interface between the Au film and the damage layer is again sharp. However, there are also visible dark clusters (arrowed) in the upper part of the damage near by the interface with the Au film. This structure of the bottom-side damage is typical for trenches milled in InP samples using a 30 keV energy and a range of ion beam currents. The thickness of the bottom-wall damage was measured as 62±6 nm, and was independent of beam current. The structure of one of the clusters visible in Figure 6.6a is shown in the highresolution image in Figure 6.7. Two sets of atomic planes are visible inside the cluster which indicates its crystalline structure. The arrows on Figure 6.7 indicate the bottom and top boundaries of this micro-crystal.
169
Figure 6.7. The magnified HREM image of the cluster in the bottom-wall damage of the trench milled using a 1000 pA beam current and a 30 keV beam energy. The size of the micro-crystal is estimated at ~ 25 nm in width and ~15 nm in height. The micro-crystal is surrounded by amorphous matrix. The presence of micro-crystals in the bottom-wall damage was also confirmed by selected-area diffraction patterns, which were taken from this area (Figure 6.8). As well as continuous rings, there are also clearly visible discrete spots inside the rings which are associated with diffraction of the micro-crystals.
170
Figure 6.8. A selected area diffraction pattern from the bottom-wall damage in the InP sample milled using a 1000 pA beam current and a 30 keV beam energy. Using the EDS facilities interfaced to the Philips CM200 electron microscope, elemental maps around the damage area were recorded and the concentration of Ga atoms in different parts of the damage layer was measured. Figure 6.9 shows elemental maps recorded around the bottom-wall area of the trench milled using a 1000 pA beam current.
171
Figure 6.9. STEM image (a) of the bottom-wall of the trench milled using a 1000 pA beam current and a 30 keV beam energy and corresponding EDS elemental maps: Au L (b), Ga L α (c), Pt L α (d), P K α (e), In L α (f). Comparing the elemental maps in Figure 6.9, a high concentration of Ga atoms is noted in the Pt film and in the InP substrate directly under the Au protective film. The latter is indicated by arrows on the Ga map (Figure 6.9c). This region corresponded to the upper crystalline part of the bottom-wall damage.
172
Figure 6.10. EDS spectrum taken from the upper part of the bottom-wall damage of the trench milled using a 1000 pA beam current and a 30 keV beam energy in the InP sample. The concentration of Ga atoms in this region was determined by EDS analysis to be ~ 6.5 at.% in the upper part (Figure 6.10), ~2.5 at.% in the lower part of the bottom-wall of the trench, and ~ 2.8 at.% in the side-wall damage of the trench. As before, corrections were made to allow for Ga implantation as a result of the preparation of the cross-sectional TEM specimen from the milled trench.
173
6.2.2 Damage Study in Indium Phosphide After a 10 keV FIB Milling A similar structure of damage was found in a trench milled using a 10 keV ion beam energy and a 250 pA beam current. The cross-sectional TEM image of the side-wall of this trench is shown in Figure 6.11a.
Figure 6.11. (a) The side-wall damage area of the trench milled using 10 keV FIB; (b) high-resolution image of this damage area. The arrows indicate the damage layer. The thickness was measured as 15±2 nm. The high-resolution images of the side-wall showed a structure which was typical for amorphous films. This indicates an amorphous structure of the damage in the side-wall of the trench milled using FIB at 10 keV beam energy and 250 pA beam current. A cross-sectional TEM image of the bottom-wall of this trench is shown in Figure 6.12a. The arrows in this image indicate the damage layer. The total thickness of the bottom-wall damage formed using a 10 keV ion beam energy was measured as 32±3 nm. The thickness is notably larger than those observed earlier in silicon (~ 26 nm) and germanium (~ 23 nm) for 10 keV milling.
174
Figure 6.12. (a) The TEM image of the bottom-wall damage of the trench milled using a 10 keV beam energy and a 250 pA beam current; (b) magnified image of its upper part. The upper part of the bottom-wall damage layer also consisted of dark clusters. An example of such clusters near the area marked X is shown in the magnified image in Figure 6.12b. A high-resolution image of such a cluster is shown in Figure 6.13.
Figure 6.13. Magnified high-resolution image of a crystallite cluster in the bottom-wall damage area shown in Figure 6.12a. 175
One set of atomic planes is visible inside the cluster in this high-resolution image. This indicates that the clusters have crystalline structure. The diameter of these nano-crystals was ~ 12 nm. They were surrounded by an amorphous matrix. Such nano-crystals were found in the upper part of the bottom-wall damage of the trench after 10 keV FIB milling. The concentration of Ga atoms was obtained using EDS spectra, which were recorded from different parts of the damage layer. The concentration of Ga atoms was measured at around 5 at% in the upper part, 2 at% in the lower part of the bottom-side damage and 2 at% in the side-wall of the trench. Again, corrections were made to allow for Ga implantation as a result of the FIB preparation of the TEM specimen.
176
6.3 Indium Arsenide 6.3.1 Damage Study in Indium Arsenide After a 30 keV FIB Milling The FIB related damage in InAs was studied in the same manner as for InP. A row of trenches was milled in this specimen using a range of beam currents (from 150 pA to 6600 pA) for 30 keV and a 250 pA beam current for a 10 keV FIB operating voltage. The size of the trenches was around 20x10 mm2 with a nominal depth of 0.2-0.6 mm (according to the FIB software) #. A cross-sectional TEM image of the trench milled using a 150 pA beam current and a 30 keV beam energy is shown in Figure 6.14.
Figure 6.14. A cross-sectional TEM image of the trench milled in In As sample using a 150 pA beam current and a 30 keV beam energy. #
The FIB workstation software uses a ‘silicon material’ file (with the sputtering yield for Si) in the milling parameters calculation. The sputter rate for 30 keV Ga FIB milling of a InAs sample was found to be 13 times greater. So the real depth of the mills was thirteen times larger. 177
The protective Au and Pt films as well as the InAs substrate are marked in this figure. The TEM contrast around the InAs substrate is very similar to that observed in Si (Figure 4.3) and the InP substrate (Figure 6.3). The “waterfalling” relief of the thinned membrane, which was visible in all Ge specimens, was not found in the InAs TEM specimens. The study of the side-wall damage of the trenches milled using 30 keV ion beam and different beam currents in InAs sample displayed the same thickness for every ion beam current. The example of the side-wall damage of the trench milled using a 150 pA ion beam current is shown in Figure 6.15. The Au and Pt protective films may be readily identified. The damage layer (indicated by arrows) is clearly visible between the crystalline InAs substrate and Au-protective film. The thickness of the side-wall damage for each trench milled using a 30 keV ion beam (a beam current range 150 – 6600 pA) was found to be 20 ±2 nm.
Figure 6.15. TEM image of side-wall damage of the trench milled in InAs sample using a 30 keV operating voltage and a 150 pA beam current. 178
No diffraction contours were found in the damage during the tilting of the specimen. The absence of the diffraction contours as well as the absence of discrete spots in the ring diffraction patterns indicated the amorphous structure of the side-wall damage of the trench milled in InAs sample. No crystallites were found in this damage layer. An example of the bottom-wall damage of the trench milled using a 1000 pA beam current is shown in Figure 6.16a.
Figure 6.16. A cross-sectional TEM image (a) and HREM image (b) of the bottom-wall damage of the trench in a InAs sample milled using a 1000 pA ion beam current and a 30 keV beam energy. The TEM study identified the Au protective film and a crystalline InAs substrate and a damage layer below the Au film (Figure 6.16a). The thickness of the bottom-wall damage for all 30 keV milled trenches was measured to be 42±4 nm. There are clearly visible crystallites inside the damage layer in the TEM image and in the magnified image of the upper part of the damage layer in Figure 6.16b. Some of these crystallites are marked in these images by arrows. The selected area diffraction that was taken from a specimen area containing crystalline InAs substrate and the bottom-wall damage also confirmed the presence of crystallites in the damage layer (Figure 6.17). As well as diffraction spots from the crystalline substrate there are clearly visible continuous rings (marked by arrows) with many individual spots. The structure
179
of the diffraction rings confirmed the presence of both amorphous and polycrystalline regions inside the damage layer.
Figure 6.17. The selected area diffraction from the specimen region containing crystalline InAs substrate and the bottom-wall damage layer shown in Figure 6.16b. Such rings were not found on selected area diffraction images taken from undamaged crystalline InAs substrate. The presence of these rings in Figure 6.17 indicated the polycrystalline structure of the bottom-side damage of the milled trenches. Also clearly visible are marked rings that intersect the diffraction spots of the crystalline InAs matrix. The first three rings were identified as (111), (200) and (022). The distribution of the intensity in the rings was found to be uniform. This fact indicates the absence of any preferential crystallite orientation.
180
The crystallites were also found in Figure 6.16a in some areas of the lower part of the bottom-wall damage near the crystalline substrate (for example marked by arrows). The magnified high-resolution image of such an area of the bottom-wall damage is shown on Figure 6.17. It is distinctly visible that the upper part of the damage layer (marked A) is totally polycrystalline. The lower part (marked B) contains many discrete crystallites (marked by arrows) surrounded by an amorphous matrix.
Figure 6.17. HREM image of the bottom-wall damage of the trench milled using a 1000 pA beam current and a 30 keV beam energy. Arrows indicate separate crystallites in the lower part of the damage. A magnified high-resolution image of the one of these crystallites is shown on Figure 6.18. Its width and height were measured to be ~ 20 and 10 nm respectively. However, the interface between this crystallite and amorphous matrix was not sharp. The 181
thickness of the transition area from crystalline to amorphous structure was found to be between 3 and 5 monolayers.
Figure 6.18. HREM image of a discrete crystallite in the lower part of the bottom-wall damage of the trench milled using a 1000 pA ion beam current and a 30 keV beam energy in the InAs sample. Two sets of {111} atomic planes and separate spots, which represent the projection of atomic columns in the [022] direction in InAs, are clearly visible in this figure. This fact indicates the good quality of this FIB-prepared specimen in spite of presence of the damage layers (with thickness ~20 nm each) on both sides of the TEM membrane. EDS spectra were recorded from different parts of the damage of the milled trenches. The EDS spectrum taken from the upper part of the bottom-wall damage of the trench is shown in Figure 6.19. As well as As and In peaks, a Ga Kα peak is also clearly visible.
182
Figure 6.19. EDS spectra taken from the upper part of the bottom-side damage of the trench milled using a 1000 pA ion beam current and a 30 keV beam energy in InAs sample. The concentration of Ga atoms in the crystalline InAs substrate was found to be 1.4 at.%. This concentration relates to the Ga atoms which were implanted in the TEM transparent membrane during specimen preparation of cross-section. To obtain a correct concentration of Ga atoms, this data was substracted from experimentally measured Ga atom concentration taken from different parts of the damage. The concentration of Ga atoms in the upper part of the bottom-wall damage layer was found to be ~ 5.5 at.% and in the lower part to be ~ 2.2 at.%. The concentration of Ga atoms in the side-wall damage of the trench was found to be ~ 2.4 at.%.
183
6.3.2 Damage Study in Indium Arsenide After a 10 keV FIB Milling The structure of the damage layer around the trench milled using 10 keV voltage and 250 pA beam current is shown in Figure 6.20. The Pt and Au protective films are shown as well as the InAs substrate. The damage layers are indicated from both sides by arrows.
Figure 6.20. TEM images of side-wall (a) and bottom-wall (b) of the damage of the trench milled using a 10 keV energy and a 250 pA beam current. No crystallites, diffractions contours, or discrete spots on selected area diffraction patterns were found in the side-wall damage (Figure 6.20a). It is amorphous in structure. The high-resolution image of side-wall damage of the FIB milled trench in InAs substrate is shown in Figure 6.21. Two sets of atomic plains are clearly visible in the InAs substrate. The damage layer, which is indicated by arrows, has an appearance which is typical for the amorphous structure. The transition area from crystalline InAs substrate to amorphous damage layer was found to be 1-2 nm. The thickness of the sidewall damage for 10 keV FIB milling was found to be ~ 11 nm.
184
Figure 6.21. HREM image of the side-wall damage of the trench milled using a 10 keV energy and a 250 beam current. The bottom-wall damage layer of the trench was much thicker than the side-wall. Its thickness was found to be 20 ± 2 nm. Crystallites were again found in the bottom-wall of the trench milled using a 10 keV energy and 250 pA ion beam current (Figure 6.20b). These crystallites (marked by arrows) are clearly visible in the magnified highresolution image shown in Figure 6.22. The crystallites were up to 10 nm in diameter. Some crystallites were observed to have almost coalesced, but some were separated by an amorphous matrix. There are different sets of atomic planes visible in the crystallites in this high-resolution image, which indicates the random crystallographic orientation of these crystallites.
185
Figure 6.22. HREM image of the bottom-wall damage fabricated using a 10 keV ion beam and a 250 pA beam current. Arrows indicate discrete crystallites. A selected area diffraction was taken from an area containing both the bottom-wall damage and the crystalline InAs substrate (Figure 6.23). The substrate orientation was found from the arrangement of diffraction spots to be close to the [011] zone axis. As well as diffraction spots relating to the crystalline InAs substrate, there are clearly visible rings which are typical for a fine polycrystalline structure. These rings consisted of many discrete spots. Each of this spots was formed as results of Bragg diffraction of the electron beam inside a crystallite.
186
Figure 6.23. Selected area diffraction from the region of the trench (containing bottomwall damage and crystalline substrate) milled in InAs using a 250 pA beam current and a 10 keV beam energy. The intensity of the rings was found to be uniform, which confirmed random orientation of the crystallites. The rings intersected with the diffraction spots from the crystalline InAs matrix and were identified as (111), (200), (022) and (311). The intersection of these rings and the spots from the InAs substrate indicates that the lattice parameter of crystallites is very similar to the InAs substrate.
187
6.3.3 Protection Properties of Au Sputter Coating Films Au sputter-coated films have been used throughout this study to protect the FIB created damage layers from any alteration during the subsequent FIB preparation of the TEM specimens. During standard FIB TEM specimen fabrication steps, which are described in Sections 2 and 3, a 1 mm thick Pt strip is usually deposited on the area of particular interest using the deposition facilities of the FIB system. If the Au protective film is thin or non-existent then high-energy Ga ions and recoil Pt atoms may penetrate through this film into the specimen and cause damage in the top surface of the substrate. To study this possible effect an InAs specimen, with an AlAs-InAs superlattice grown on the substrate, was sputter coated with an Au film. However, in this experiment, the sputter coating was produced with a vacuum pressure higher than usual (~ 0.05 Torr). It was found that in this case the Au film was not uniform in structure and thickness. A cross-sectional TEM image of the Au sputter-coated film on the InAs specimen is shown in Figure 6.24.
Figure 6.24. A cross-sectional TEM image of the Au sputter coated protective film on an InAs substrate. Numbers 1, 2 indicate the voids in Au layer. The Pt, Au films and the InAs substrate are indicated in Figure 6.24. The thickness of the Au protective film is not uniform, with a maximum around 200 nm and a minimum (areas marked “A” and “B”) around 15 nm. There are also visible a number of voids (for example, marked “1” and “2”) in the film. The AlAs-InAs superlattice appears as 188
the alternation of individual AlAs and InAs layers. The AlAs layers appear as white lines. The strong contrast arises because the difference in structure factors of the individual AlAs and InAs layers. However, it was found that contrast on individual layers of the superlattice was not evident in the regions (marked “A” and “B” in Figure 6.24) where the thickness of Au film was minimal. The magnified image of the area marked “A” in Figure 6.24 is shown in Figure 6.25.
Figure 6.25. The magnified TEM image of the area marked “A” in Figure 6.24. It can be seen that the Au film in the middle of the Figure 6.25 has a gap with length around 20 nm. The gap is filled by Pt. The superlattice was intermixed not only under the gap but also under the thin Au film. The thickness of the Au film in the area marked “x” was measured to be ~15 nm and the dimension of the superlattice affected by Ga ions penetrating through the Au protective film was found to be around 25 nm. It can be 189
noted that the superlattice under the Au film with voids (regions marked “1” and “2” in Figure 6.24) was not be altered. The total thickness of the Au film above and below this void was found to be around 60 nm. It can be concluded that if the thickness of the Au sputter coated film is larger than ~ 60 nm, the original structure of the specimens will be protected from any alteration during FIB TEM preparation.
190
6.4. Gallium Arsenide 6.4.1 Damage Study in Gallium Arsenide After a 30 keV FIB Milling The FIB related damage in the GaAs was studied in the same manner as for InAs. Every trench milled in GaAs sample using a 30 keV ion energy and a particular beam current (the range of beam currents was from 150 pA to 6600 pA) had a similar thickness and structure of damage. The size of the trenches was around 20x10 mm2, with a depth of 0.2-0.6 mm (according to FIB software) #. A cross-sectional low-magnification TEM image of the trench milled using a 350 pA beam current and a 30 keV beam energy is shown in Figure 6.26.
Figure 6.26. A cross-sectional TEM image of the trench milled in GaAs substrate using a 350 pA beam current and a 30 keV beam energy. #
The FIB workstation software uses a ‘silicon material’ file (with the sputtering yield for Si) in the milling parameters calculation. The sputter rate for 30 keV Ga FIB milling of a GaAs sample was found to be six times greater. So the real depth of the mills was six times larger.
191
The TEM contrast around the GaAs substrate is very similar to that observed earlier in other III-V semiconductors (Figures 6.3, 6,14) and the Si substrate (Figure 4.3). A “waterfalling” relief of the thinned membrane, which was found in Ge specimens (Figure 5.10), was not visible in the GaAs TEM specimens. The magnified images of the side-wall and the bottom-wall damage of the trench milled using a 350 pA beam current are shown in Figure 6.27.
Figure 6.27. TEM images of side-wall (a) and bottom-wall (b) damage of the trench milled in GaAs sample using a 30 keV energy and a 350 pA beam current. The Pt and Au protective films and crystalline GaAs substrate are easy recognised and are marked in these images. The damage layer between the crystalline substrate and the Au protective film is also clearly visible. TEM study of the area around the damage in GaAs shows the absence of diffraction contours inside the damage layer. This, together with the absence of discrete reflections in selected area diffraction patterns, indicates its non-crystalline structure. The side-wall (Figure 6.27a) and the bottom-wall damage (Figure 6.27b) had a very uniform structure and thickness. However, in contrast to the other materials studied in this thesis, no crystallites were found in the bottom-wall damage. An example of the side-wall and the bottom-wall damage in the FIB milled trenches in GaAs sample prepared using a 30 keV ion energy and a 1000 pA beam currents is shown in Figures 6.28. 192
Figure 6.28. TEM images of side-wall (a) and bottom-wall (b) of the damage of the trench milled in GaAs sample using a 1000 pA beam current and a 30 keV beam energy. Arrows indicate the damage layer in this figure. Again crystallographic diffraction contrast was not visible in the damage layer in either of these images. A selected area diffraction pattern from a region containing the bottom-wall damage layer and GaAs substrate of the trench milled using a 1000 pA beam current is shown in Figure 6.28.
Figure 6.28. A selected area diffraction pattern from the bottom-wall of the trench milled using a 1000 pA beam current and a 30 keV beam energy.
193
Apart from individual spots from Bragg diffraction from the crystalline GaAs substrate, there are clearly visible continuous amorphous diffraction rings from the bottom-wall damage layer. Discrete spots from microcrystals, similar to those found in InAs and InP, were not found. The magnified image of the bottom-wall damage layer (from Figure 6.27b) is shown in Figure 6.29. The damage layer with sharp interfaces is distinctly visible in this image. No crystallites were found there. Letters A and B in Figure 6.29 indicate areas in the upper and the lower parts of the damage, respectively.
Figure 6.29. Magnified HREM image of the bottom-side damage of the trench milled in GaAs using a 1000 pA beam current and a30 keV beam energy. Label A indicates the upper part of the damage layer. Label B indicates the lower part of the damage layer.
194
The magnified HREM images of the areas marked A and B are shown in Figure 6.30. Atomic planes are not visible in the upper part of the damage layer (Figure 6.30a) and are only visible in the GaAs substrate near region B in the lower part of the damage layer (Figure 6.30b).
Figure 6.30. Magnified HREM images of the upper (a) and lower part of the bottomwall damage in GaAs shown on Figure 6.29. Another example of the damage around the trench milled in GaAs using 30 keV FIB is shown in Figure 6.31. This trench was milled using a 2700 pA ion beam current. The side-wall damage layer (indicated by arrows in Figure 6.31a) is uniform in thickness and exhibits typical amorphous contrast. The HREM image of the bottom-wall damage (Figure 6.31b) also shows its amorphous structure. The magnified HREM images of the upper part and the lower part of the bottom-wall damage are shown in Figure 6.32a,b. Again, no crystallites were found in the bottom-wall damage layer.
195
Figure 6.31. TEM images of (a) side-wall and (b) bottom-wall damage of the trench milled in GaAs sample using a 2700 pA beam current and a 30 keV beam energy. Label A indicates the upper part of the damage layer. Label B indicates the lower part of the damage layer.
Figure 6.32. Magnified HREM images of the (a) upper and (b) lower part of the bottom-wall damage in GaAs shown in Figure 6.31. Every trench milled in GaAs using a 30 keV FIB energy and over the range of beam currents used showed a similar thickness and structure of the damage layer. The thickness of the side-wall damage in GaAs was measured to be 22±2 nm and the thickness of the bottom-wall to be 42±4 nm.
196
6.4.2 Damage Study in Gallium Arsenide After a 10 keV FIB Milling The study of the damage layer around the trench in GaAs milled using a 10 keV ion energy and a 250 pA beam current also showed that all the damage layers were amorphous in structure. The TEM cross-section image of the side-wall damage is shown on Figure 6.33. No diffraction contours were observed inside this damage layer during the TEM study. The side-wall damage was found to be very uniform in structure (completely amorphous) and thickness. The thickness was measured to be ~ 11 nm. The thickness of the amorphous-to-crystalline transition area was found to be ~ 2 nm.
Figure 6.33. The TEM image of the side-wall of the trench milled in GaAs using a 10 keV ion energy and a 250 pA ion beam current.
197
A high-resolution TEM image of the bottom-wall damage of the trench in GaAs milled using 10 keV energy and 250 pA beam current is shown in Figure 6.34. The atomic planes were only observed in the crystalline GaAs substrate.
Figure 6.34. The HREM image of the bottom-wall of the trench milled in GaAs using 10 keV ion energy and a 250 pA ion beam current. No crystallites were found inside the damage layer. The damage layer had a typically amorphous appearance. The interface between the damage layer and the GaAs substrate was found to be quite sharp. The thickness of the amorphous to crystalline transition zone was measured to be ~ 2 nm. EDS analysis was not useful on these GaAs specimens. Because the Ga atoms are native in the GaAs lattice, the concentration of the implanted Ga atoms cannot be accurately determined.
198
6.5 Determination of the Sputter Rate During milling of the trenches in III-V specimens described in the previous sections the FIB silicon material file (with a sputter rate 0.15 mm3/nanocoloumb for silicon, 30 keV beam energy) was used for the trench depth calculation. The sputter rate for 30 keV Ga FIB milling of InP, InAs and GaAs samples was not known. Real depth of the trenches was measured using preparation of section of the trenches and TEM imaging. For example, for InP the depth of the trench shown in Figure 6.3 was set up during FIB milling to be 0.2 mm according to the FIB software using the “silicon material” file. Using the cross-sectional TEM image (Figure 6.3) the real depth of the trench was measured to be ~ 1.8 mm. That means that the sputter rate for the 30 keV Ga FIB milling of InP is ~ 9 times higher than that of silicon. On this basis its value was calculated to be ~ 1.35 microns3/nanocoulomb. Again, in the case of InAs the trench shown in Figure 6.14, the depth of the trench was set up during FIB milling to be 0.5 mm according to the FIB software using the “silicon material” file. From this cross-sectional TEM image (Figure 6.14), the real depth of the trench was measured to be ~ 6.5 mm. That means that sputter rate for the 30 keV Ga FIB milling of InAs is ~ 13 times higher than that of silicon. Its value was calculated to be ~ 1.95 microns3/nanocoulomb. For the trench shown in Figure 6.24 the depth of the trench in the GaAs substrate was set up during FIB milling to be 0.4 mm, according to the FIB software for “silicon material” file. Using the cross-sectional TEM image (Figure 6.26), the real depth of the trench was measured as ~ 2.5 mm. That means that sputter rate for the 30 keV Ga FIB milling of gallium arsenide is ~ 6 times higher than that of silicon. Its value was calculated to be ~ 0.9 microns3/nanocoulomb. The sputter rates for materials studied in this thesis are summarised in Table 6.1
199
Table 6.1. The experimentally obtained sputter rates for a 30 keV Ga FIB milling of Si, Ge and III-V semiconductors. Substrate materials
Sputter rate
Si
Ge
InP
InAs
GaAs
0.15
0.6
1.35
1.95
0.9
(mm3/nanocoulomb) The value for sputter rate can be used for the creation of FIB “material files” for materials studied in this thesis.
200
6.6 Discussion Unlike silicon, the FIB fabrication of TEM specimens from III-V semiconductors has not been extensively studied. Only few data was found in literature relating to the FIB milling of InP and GaAs. Walker and Broom [74] reported measurements of the thickness of the FIB-fabricated damage layer in GaAs and InP. Yamaguchi and Nishikawa also studied FIB-related damage layers in GaAs [47] and InP [59]. No reports of any study of the FIB milling process and the FIB related damage in InAs were found in the literature. The study of the FIB milling process showed that the sputter rate for the III-V compound was much higher than for silicon and germanium. Thus, the TEM specimens from III-V materials may be prepared very rapidly using the FIB system. The sputter rate compared with silicon, was 6 times greater in GaAs, 9 times greater in InP and 13 times in InAs. The sputter rate for these materials was not found either in the literature or in the FIB workstation manual. The data obtained is new and may be used for creation of “material files” for these materials. During the FIB milling high-energy Ga ions collide with the target atoms and displace them. The displaced target atoms may leave the sample if their kinetic energy is higher than the surface binding energy. If the surface binding energy is high then less atoms can leave the target and the sputter rate is lower. Unlike elemental semiconductors (Ge, Si), the surface binding energy for III-V semiconductors is slightly different for each species in the compound (Table 3.1). So one of the atomic species in III-V semiconductors can escape more easily than the other. For example, in GaAs the displaced As atom (the surface binding energy to be 1.2 eV) can easily escape, compared with the displaced Ga atoms (the surface binding energy to be 2.8 eV). This may result in different to the undamaged material stoichiometry of the damage layer. The measured sputter rate as a function of the surface binding energy is shown in Figure 6.35.
201
Sputter Rate (a.u.)
14 12 10 8 6 4 2 0 4.7
3.9
2.8/1.2
2.5/3.2
2.5/1.3
Si
Ge
GaAs
InP
InAs
Surface Binding Energy (eV)
Figure 6.35. The sputter rate for a 30 keV Ga FIB milling for studied semiconductors as a function of surface binding. The sputter rate for Si was taken as “1”. For III-V semiconductors the surface binding energy is shown for both component atoms. It is clear from Figure 6.35 that the value of the sputter rate is inversely related to the surface binding energy. The sputter rate is lowest in silicon and is highest in indium arsenide. In general, the practical employment of FIB for TEM specimen preparation from InP, InAs and GaAs showed quite good results. The quality of the TEM specimens was found to be very similar to the Si specimens. High-resolution lattice images were obtained from almost all specimens prepared using a 30 keV beam energy. This fact indicates that on the one hand this technique allows specimens with a thickness thin enough for HREM imaging and, on the other hand, that the thickness of damage on both sides of the membrane is less than a critical thickness to completely prevent HREM observations. The structure of the damage layers in InP and InAs was found to be similar to Si. The side-wall damage in these materials had an amorphous structure. The structure of bottom-wall damage had a more complex structure. In addition to an amorphous bottom layer, its upper part contains crystallites. The presence of such micro-crystals in bottomwall damage layers in InP were also reported by Yamaguchi and Nishikawa [59]. Such microcrystals was also found in this study in the bottom-wall damage layers in both silicon (Section 4) and germanium (Section 5).
202
The process of “dynamic damaging” was described in detail in the silicon section. On the one hand the FIB milling resulted in the formation of amorphous damage layers in every material studied for all ion beam currents used in this study. That means that the ion dose exceeded the critical dose to turn the material into the amorphous state. On the other hand, the high-energy Ga FIB resulted in local heating of the milled area of the sample, which stimulated thermal diffusion of the vacancies and the interstitial atoms. The recrystallisation of the amorphous damage occurred because of local dynamic annealing and damage annihilation. For both compounds a high concentration of Ga atoms was found which presumably acted as an agent for recrystallisation. As noted earlier, the recrystallisation of the amorphous damage layer was observed before in Si doped with Ga atoms by Williams et al [84] at temperature as low as 350o C. In contrast, microcrystals were not found in the bottom-wall damage in GaAs. That means that no recrystallisation occurred during FIB fabrication in this material. These results for GaAs are not consistent with results reported by Pearton et al. [100]. These workers found that, for GaAs, amorphous layers recrystallise epitaxially during static annealing at temperatures as low as 150-200o C, although the time of this heating was not reported. However, during FIB milling the dynamic annealing occurred during the milling periods with very short dwell times (a single dwell time is ~ 1 msec, see Table 3.2). This suggests that there was insufficient time for recrystallisation of GaAs during FIB milling. The profile of the temperature of the surface of the samples during the FIB milling is not known. But, because of the nature of the FIB milling process (dwell rastering) the heating occurs locally and during very short period of time. It should also be noted that the process of recrystallisation may be inhibited in GaAs not only by the very short period of heating, which limits diffusion of point defects and specimen atoms, but also by a lack of the impurity atoms. The impurity Ga atoms (with a concentration > 3 at.%) apparently played the role of catalyst in the recrystallisation process in InAs and InP, but not in GaAs where they were native to the lattice. The experimentally measured thicknesses of the damage layer in the studied III-V compounds are summarised in Table 6.2.
203
Table 6.2. Measured thicknesses of the damage layer in III-V semiconductors.
Materials
Thickness of damage layer (nm) 30 keV
10 keV
Bottom-wall
Side-wall
Bottom-wall
Side-wall
InP
62
33
32
15
InAs
42
20
20
11
GaAs
42
21
20
11
The study of the FIB-related damage layers in indium arsenide has not been reported before. Therefore, the data of the thickness of the damage layers in the InAs obtained in this study are totally new. The obtained thickness of the side-wall damage in InP at 30 keV correlates well with data reported by Yamaguchi and Nishikawa (31 nm for a 30 keV beam energy) [59]. However, the thicknesses of the bottom-wall damage layers in GaAs and InP obtained in this study are not consistent with the data reported by Walker and Broom [74] for a 10 keV beam energy (4 nm for GaAs and 15 nm for InP) and for a 30 keV beam energy (24 for GaAs and 40 nm for InP). However, these authors used only Pt FIB assisted deposited films to protect the damage layer and did not use an Au protective film. It is probable that during deposition of this Pt film without Au protection, the bottom-wall surface was damaged and its thickness incorrectly determined. This may possibly have led to errors in the determination of the damage layer thickness. The importance of employment of the Au sputter coating films applied prior to FIB milling to protect the original surface structure of the specimens was shown for the first time in this study in the example of the InAs sample. It was shown that, if the thickness of the Au sputter coating film is below ~ 60 nm, high-energy Ga ions can penetrate through this film and cause damage in the top surface of the substrate. In the case of a ~15 nm thin Au film deposited on the superlattice structure, the subsequent employment of FIB for Pt deposition resulted in destruction of the periodic structure of a superlattice
204
to a depth of up to 25 nm. The application of protective Au sputter-coating films has not been reported before. It is clear from the Table 6.2 that the thickness of the damage layers in GaAs and InAs is very similar, but it is almost 50 % higher for InP. In many III-V materials the two atomic species have different mass and because of this, may they recoil unequally (Eq. 3) so that local alteration of stoichiometry is created. The lighter element recoils further, leading to an excess of the heavier element near the surface. Such difference in atomic mass of the two species is most distinct for InP, rather than for InAs and GaAs (Table 3.1). The atomic masses of the components of InP are 115 amu for In and 31 amu for P. From the point of view of the classical theory of implantation (Section 2.7.4), Ga ions are light with respect to In and are heavy with respect to P. So Ga ions will lose their energy in InP by both electronic and nuclear collisions (Eq. 1). In the case of head-to-head collision of 30 keV Ga ions and P atoms in InP, the P atoms can be knocked-out with a transferred energy of up to 26 keV (Eq. 3). In the case of head-to-head collisions with In atoms, the transferred energy may be up to 28 keV (Eq. 3). These high-energy P atoms may penetrate much deeper and cause
more damage
than the original Ga ions or the secondary In ions. The results of Monte Carlo calculations (SRIM) for implantation of 30 keV Ga ions, 26 keV P ions and 28keV In ions in an InP substrate are shown in Figure 6.36-6.38. Comparing these images it was found that recoil P atoms (Figure 6.37) penetrate much deeper creating damage than either the original Ga ions (Figure 6.36) or In recoil atoms (Figure 6.38). From these calculations the thickness of the damage layer (range+straggle) which may be produced was found for Ga ions to be 30 nm, for In ions to be 23 nm and for P ions to be 53 nm. The data for P is the closest to the experimentally obtained data (62 nm). This fact indicates that the self-implantation of secondary knocked-out ions plays a significant role in the damage formation in semiconductor materials.
205
Figure 6.36. Monte Carlo computer calculations of (a) ion trajectories, (b) range and vacancies formation for 30 keV 120 Ga+ ions implanted normal to the surface of the InP target.
206
Figure 6.37. Monte Carlo computer calculations of (a) ion trajectories, (b) range and vacancies formation for 26 keV, 120 P+ ions implanted normal to the surface in the InP target.
207
Figure 6.38. Monte Carlo computer calculations of (a) ion trajectories, (b) range and vacancies formation for 28 keV, 120 In+ ions implanted normal to the surface of the InP target.
208
For both InAs and GaAs materials, Monte Carlo calculations give data for the bottomwall damage layer thickness. For GaAs, it is 20 nm for secondary 27 keV As ions. For InAs it is 19 nm for secondary 28 keV In ions and 22 nm for 27 keV As ions. The calculated data for the thickness of the damage layers in the studied III-V semiconductors for both Ga and recoil ions are summarised in Table 6.3. Table 6.3. Calculated data for the thickness of the damage layer for Ga+ implantation into III-V semiconductors and implantation of the recoil atoms. In the brackets the data obtained experimentally are shown.
Target material
Ion
Calculated thickness (range + straggle) of damage layer in silicon (nm) 30 keV
InP
InAs
GaAs
10 keV
Bottom-wall (90o)
Side-wall (4.5o)
Bottom-wall (90o)
Side-wall (4.5o)
30 (62)
18 (33)
14 (32)
8 (15)
P
53
32
22
14
In
23
11
12
6
Ga
24(42)
13(20)
12(20)
7(11)
As
22
12
10
6
In
19
9
9
5
Ga
25(42)
12(21)
12(20)
5(11)
As
20
10
10
5
Ga
It can be concluded comparing the data in Table 6.3 that the experimentally obtained thicknesses of the damage layer are somewhat larger than the calculated data even taking into account the effect of recoil atoms. A probable reason for the disparity between the calculated data and the experimental data is that during FIB fabrication the process of ion implantation is going through an already heavily damaged layer (dynamic 209
damaging, see Section 4.2). This damage layer has an amorphous structure and the density of the material in this layer is less than the crystalline undamaged state, which is assumed in the Monte Carlo calculations. That means that the Ga and recoil ions may penetrate deeper and cause the creation of a thicker damage layer than predicted by Monte Carlo calculations.
210
PART V SUMMARY AND CONCLUSIONS CHAPTER 7 GENERAL DISCUSSION During the last decade, progress in the semiconductor industry has resulted in production of devices with increasingly complex circuitry. That is, more transistors in less space, a larger number of layers, more interfaces, and manufacturing processes with a large number of precise steps. TEM is particularly well suited for solving material problems encountered during research, development, production engineering, and failure analysis of device materials. The demand for increasingly higher performance semiconductor products has stimulated the development of the FIB system as an advanced tool for TEM sample preparation. The introduction and development of the FIB systems in the early 90’s coincided with progress in the computer industry. The development of FIB systems was mostly focused in the direction of integration with device development rather than on gaining a fuller understanding of the physics of the FIB milling process. The current generation of FIB system is a complex device, with operations which are completely controlled by a powerful computer. Users of such FIB systems today control FIB fabrication by communicating with very user-friendly graphical interface of an integrated computer. The straightforward nature of these manipulations of FIB fabrication has resulted in the illusion of simplicity of the milling process, and a lack of systematic study of the complex physical processes which occur during FIB milling. At the commencement of this project the focused ion beam technique had been successfully used for TEM sample preparation for some years, especially for silicon. However, it was known that the FIB fabrication process of TEM samples resulted in the formation of a damage layer on both sides of the specimen. However, no systematic study had been performed to determine the nature and the origins of the damage in FIB prepared TEM samples. The origins of damage in FIB-milled specimens are not fully understood. Some workers have argued that the damage is due entirely to direct radiation damage; others have assumed the damage is related to redeposition or a combination of direct radiation damage and redeposition.
211
Moreover, several authors, in a few published studies, have provided different data regarding the thickness of the damage layer in Si and GaAs [2, 4, 5, 74]. Also, few if any studies have been done for other semiconductor materials like Ge, GaAs, GaN, InAs. In this project a systematic study of the process of FIB preparation of TEM samples and the process of damage formation for Si, Ge, GaAs, InAs and InP was carried out. An experimental procedure was established to study damage layers after FIB fabrication. This involved the deposition of sputter-coated Au films to protect and localise the previously fabricated damage. It was shown for the first time in this study that very thin (around 15 nm) Au protective films do not protect the structure of specimens from Ga ion implantation. Therefore to protect the studied structure completly from any alteration during the FIB TEM preparation steps the thickness of the Au protective films should exceed ~ 60 nm. It was found during this study that the speed of FIB milling (that is, sputter rate) was different for every material examined. The sputter rate in any given material was found to be inversely related to the surface binding energy. The sputter rate was obtained for the first time for Ge, GaAs, InP and InAs. The TEM specimens from these materials may be prepared faster than from Si because of their much higher sputter rate. This data can be used for creation of FIB system “material” files for all the studied materials. The process of TEM sample preparation involves FIB milling of a set of rectangular trenches from both sides of the area of interest until a thin electron transparent membrane is achieved. It was shown that because of the presence of a beam “tail” (because the profile of the current beam density has a Gaussian shape) that milling occurred outside the defined milling area. In this way when the FIB final mill is performed on one side of a thin membrane, the other side of the membrane may also be affected by the beam “tail”. Using the experimental procedure established in this thesis the damage was studied around single trenches milled under different conditions (ion beam energy and current) in different semiconductor materials. It was found that damage (amorphous) layers existed for the range of beam currents studied. That means that the ion dose during FIB 212
fabrication exceeded the critical dose necessary to create amorphous material. In this case the thickness of the damage layers was dependent on the ion beam energy, but was independent of the ion beam current. Thus reduction of the ion beam energy from 30 to 10 keV may result in a ~ 50 % reduction of the thickness of the damage layer. Two different kinds of damage layers were observed around the milled trenches. The first kind was the side-wall damage layer which was very uniform in thickness and in structure (amorphous). These amorphous side-wall damage layers were observed on both sides of the FIB prepared cross-sectional TEM samples. The second kind was the bottom-wall damage, which was uniform in thickness (except for the germanium samples), but non-uniform in structure with microcrystals inside an amorphous matrix (except for GaAs). The structure of the bottom-wall damage layer in germanium was found to be different. The lower part of the bottom-wall damage had an amorphous structure, but the upper part of the damage contained many large voids, which were surrounded by micro-crystals. It may be possible that this was associated with the formation of a high concentration of vacancies which underwent diffusion during the local heating of the germanium surface during FIB fabrication. This bottom-wall damage layer may be present on the surface of electron transparent membranes in the case of plan-view TEM sample and will likely cause confusion during further TEM study. In general, the FIB produced damage layers were initially sputter-coated with Au protective films. These films not only protected the damage from any alteration during subsequent TEM specimen preparation, but also delineated surface layers, which made it easy to identify the damage layers, and more accurately measure their thickness. The thickness of the damage layers was measured for all studied materials. The thickness of the damage layers in Ge and InAs has not been reported before. This study has shown that the thickness of the FIB produced damage layers in different materials depends on the atomic mass of the component atoms. In both silicon and indium phosphide, which contain lighter atoms, the damage layers were found to be significantly thicker than in the other studied semiconductors. In both these materials low atom mass species were present. It was established that implantation of the light
213
recoil atoms plays a significant role in the damage formation during FIB TEM specimen fabrication. Using the experimental procedure described in Section 4.5 the region of redeposited material was localised only on the side-wall of a previously milled trench. Its structure, thickness and composition were studied using TEM and EDS and compared with sidewall damage from a single trench. It was shown that no back-flow of sputtered atoms occurs during FIB milling. Any sputtered atoms which locate to the walls of the membrane during milling are quickly sputtered again by the beam “tail” during FIB rastering. So,the nature of the damage in the FIB-prepared TEM sample is direct radiation damage rather than redeposition of atoms in the specimen. The damage formation process during the FIB milling, which was studied in this thesis, was found to be dependent on several factors: ion beam energy, ion beam current, the atomic mass of component atoms and surface binding energy. However, in general, the cross-sectional TEM specimens prepared using FIB technique from a number of the studied elemental and compound semiconductor materials were of good quality. Highresolution TEM lattice images were obtained for all materials. The FIB-produced amorphous damage layers on both sides of the electron transparent TEM membrane, but this did not prevent TEM investigation of the studied materials with the highest spatial resolution. The additional reduction of the ion beam energy from 30 keV to 10 keV in the final stages of FIB milling may result in reduction of the thickness the damage layers by a factor of two.
214
CHAPTER 8 CONCLUSIONS The FIB technique was applied to TEM sample preparation from elemental and compound semiconductors. The conclusions may be summarised as follows: 1. The FIB technique may be used to prepare TEM specimens from different semiconductor materials, especially in the cross-sectional configuration. Large (~5x20 mm2), uniformly thin electron-transparent membranes may be prepared from these materials in cross-sectional configuration using a FIB in less than about two hours. 2. The FIB sputtering rate for a number of semiconductor materials was found to be much higher than for Si. It was 4 times higher for Ge, 6 times for GaAs, 9 times for InP and 13 times for InAs. It was found that the value of the sputter rate is inversely related to the surface binding energy. These data can be used for the creation of FIB “material” files for these semiconductors. As a consequence the TEM specimens from these materials can be prepared more quickly than from Si. 3. The process of FIB sample preparation involves the milling of a set of rectangular trenches until a thin electron-transparent membrane is formed in the area of particular interest. Using especially designed experimental procedures, which involves the initial deposition of Au sputter-coated films, the damage layers were studied around the walls of the trenches prepared using different milling conditions (that is, variation of ion beam current and high voltage). In all cases the amorphous damage layer was found around the surface of the milled trenches. That means that for all working FIB-beam conditions used, the ion dose exceeded the critical dose for turning material into an amorphous state. 4.
The damage layer had a different thickness and structure in the side-wall from those in the bottom-wall of the milled trench. The side-wall damage layer for all studied materials was found to be amorphous in structure. It was found that the origin of the side-wall damage layer present in the FIB prepared TEM specimens was direct radiation damage from the Ga beam, not redeposition from the 215
specimen. Amorphous damage layers were formed on both sides of the TEM specimen. Any back-scattered atoms produced during the milling process, which were redeposited on the specimen surface during FIB milling process, were probably re-milled by the ion beam “tail” during beam rastering. 5. The thickness of the damage layer in a particular material depends on the energy of the Ga ion beam, but is independent of the ion beam current for the entire range of the FIB-system working currents used. The thickness of the damage layer may be significantly decreased by using lower ion beam energy on the final stage of the FIB sample preparation. 6. The thickness of the damage layers in different materials depends on the atomic mass of the atoms in the materials. The head-to-head collisions between the Ga ions and the target atoms create high-energy recoil atoms. Implantation of lighter recoil atoms results in much thicker damage layers in both Si and InP than in the other studied materials 7. It was found that the recrystallisation of the amorphous damage layer took place on the bottom-wall damage layer of the milled trenches in all studied materials except GaAs. The high density of ion beam current caused local heating of the sample near the milling surface and as result some recrystallisation of the damage layer took place. 8.
The presence of the impurity atoms (implanted Ga atoms) assisted in the recrystallisation process. The micro-crystallites were found only in the upper part of the bottom-wall of the damage layer where a high concentration of the implanted Ga atoms (~6-7 at.%) was measured. However, the recrystallisation was not found in the damage layers in GaAs. This was associated with the absence of impurity atoms (the implanted Ga atoms are native to the GaAs lattice). The possible recrystallisation of the damage layer must be taken into account in the case of plan-view TEM sample preparation. It was also found that the profile of concentration of implanted Ga atoms was different from the theoretically predicted Gaussian shape (which has maximum on the length of the
216
projective range). The highest concentration of implanted Ga atoms was found to be nearer the milling surface. 9. The processes of FIB sample milling and damaging occurred through a preexisting damage layer, which was formed during the initial stage of the milling process. Thus, Ga ions pass through an amorphous layer, so ion channelling does not play any significant role during FIB specimen fabrication. 10. The high-energy Ga ions may penetrate through thin Au sputter coating protective film and alter the initial structure of the specimens or studied damaged layers. To protect the original structure of the sample the thickness of the Au films should exceed about - 60 nm. 11. The surface of the Ge sample after FIB fabrication had strong cellular relief. The upper part of the damage layer in the bottom-wall consisted of many voids and micro-crystallites. These voids were probably formed through an agglomeration of vacancies. 12. The possible formation of the bottom-wall damage layer containing voids and micro-crystals during plan-view TEM sample preparation from Si, Ge, InAs and InP may cause difficulties during TEM study of these materials. 13. The use of FIB milling with a 30 keV ion beam energy allows good quality cross-sectional TEM specimens from binary and compound semiconductors to be prepared. The presence of amorphous damage layers on both sides of the specimen membrane does not significantly reduce the quality of the specimens. High-resolution lattice images were recorded on all studied materials prepared using 30 keV FIB milling. The use of low beam energy for final milling may minimize the created damage. Thus, the 10 keV cleaning mill may additionally reduce the thickness of the damage in a TEM sample by a factor of two and result in increases in the quality of the specimens.
217
REFERENCES 1.
Zhang, H., Application of TEM on sub-half micron semiconductor devices. Mat. Res. Soc. Proc., 1998. 523: p. 45--59.
2.
Mardinly, J. and D. Susnitzky, Transmission electron microscopy of semiconductor based products. Mat. Res. Soc. Symp. Proc., 1998. 523: p. 3-12.
3.
Kirk, E.C.G., D.A. Williams, and H. Ahmed, Cross-sectional transmission electron microscopy of precisely selected regions from semiconductor devices. Inst. Phys. Conf. Ser., 1989. 100: p. 501-506.
4.
Langford, R.M. and A.K. Petford-Long, Reduction of the side-wall damage and gallium consentration of focused ion beam prepared TEM cross sections. Proceedings of EUREM 12, 2000: p. 557-558.
5.
Bender, H., Focused ion beam specimen preparation for transmission electron microscopy studies of ULSI devices. Inst. Phys. Conf. Ser., 1999. 164: p. 593602.
6.
Spence, J.C.H., M.A. O'Keefe, and H. Kolar, Optic, 1977. 49: p. 309.
7.
Glaisher, R.W., A.E.C. Spargo, and D.J. Smith, A theoretical analysis of HREM imaging for [110] tetrahedral semiconductors. Ultramicroscopy, 1989. 27: p. 1934.
8.
Hutchison, J.L., T. Honda, and E.D. Boyes, JEOL News, 1986. E24: p. 9.
9.
Klepeis, S.J., J.P. Benedict, and R.M. Anderson, Mater. Res. Soc. Symp. Proc., 1988. 115: p. 81.
10.
McCaffrey, Small-angle cleavage of semiconductors for transmission electron microscopy. Ultramicroscopy, 1991. 38: p. 149-157.
11.
Hetherington, C.J.D., Mater. Res. Soc. Symp. Proc., 1988. 115: p. 143.
12.
Boone, T. and S. Nakahara, Mater. Res. Soc. Symp. Proc., 1988. 115: p. 81.
13.
Mails, T.F. and D. Steele, Mater. Res. Soc. Symp. Proc., 1988. 199: p. 3.
14.
Ellington, M.B., Mater. Res. Soc. Symp. Proc., 1988. 115: p. 265.
218
15.
Sabinina, I.V. and A.K. Gutakovsky, Preparation of TEM samples from compound semiconductors by chemomechanical polishing. Ultramicroscopy, 1992. 45: p. 411-415.
16.
Wetzel, J.T. and D.A. Danner, Mater. Res. Soc. Symp. Proc., 1988. 115: p. 253.
17.
Chew, N.G. and A.G. Cullis, The preparation of transmission electron microscope specimens from compound semiconductors by ion milling. Ultramicroscopy, 1987. 23: p. 175-198.
18.
Barber, D.J., Development of Ion-Beam Milling as a Major Tool for Electron Microscopy. Asia/Pacific Microscopy and Analysis, 1999. 5: p. 5-8.
19.
Rubanov, S.V., A.K. Gutakovsky, and S.M. Pintus, Preparation of cross-sections from hetero-epitaxial structures for transmission electron microscopy. Pribori i technika experimenta, 1989(3): p. 190-192.
20.
Karasev, V.Y., et al., HREM of epitaxial layers in the InAs/GaAs system. Ultramicroscopy, 1991. 35: p. 11-18.
21.
Lenkeit, K., et al., Study of InGaAs-GaAs strained-layer superlattices by TEM and RBS techniques. Phys. Stat. Sol. (a), 1989. 115: p. 413-425.
22.
Paulus, M. and F. Reverchon, Dispositif de bombardement ionique pour preparations micrographiques. J. Phys. Radium, 1961. 22: p. 103-107.
23.
Barna, A., Proc. EUREM 84, 1984. 1: p. 107-108.
24.
Tomas, G. and M. Goringe, Transmission electron microscopy of materials. 1979, New York: John Wiley & Sons.
25.
Kim, M.J. and R.W. Carpenter, TEM specimen heating during ion beam thinning: microstructural instability. Ultramicroscopy, 1987. 21: p. 327-334.
26.
Krohn, V.E., Progress in Astronautics and Rocketry. 1961, New York: Academic Press. 73-79.
27.
Munroe, P., Principles and aplications of focused ion beam milling. preprint, UNSW, Sydney, 1999.
28.
Gamo, K., Nanofabrication by FIB. Microelectronic Engineering, 1996. 32: p. 159-171.
29.
Young, R.J., et al., Fabrication of planar and cross-sectional TEM specimens using a focused ion beam. Mat. Res. Soc. Proc., 1990. 199: p. 205-216.
219
30.
Alani, R., R.J. Mitro, and K. Ogura, Reactiv ion beam etching (RIBE) technique and instrumentation for SEM specimen preparation of semiconductors. Proc. Microscopy and Microanalysis, 1999. 5(2): p. 912-913.
31.
Puretz, J. and L.W. Swanson, Focused ion beam deposition of Pt containing films. J. Vac. Sci. Technol. B, 1992. 10(6): p. 2695-2698.
32.
Young, R.J. and J. Puretz, Focused Ion beam insulator deposition. J. Vac. Sci. Technol. B, 1995. 13(6): p. 2576-2579.
33.
Rowlands, N. and P. Munroe, FIB for the Evaluation of Non-Semiconductor Materials. Microstructural Science, 1998. 26: p. 233-241.
34.
Giannuzzi, L.A. and F.A. Stevie, A review of focused ion beam milling techniques for TEM specimen preparation. Micron, 1999. 30: p. 197-204.
35.
Bender, H. and P. Roussel, Cross-sectional transmission electron microscopy and focused ion beam study of advanced silicon devices. Inst. Phys. Conf. Ser., 1997. 157: p. 465-468.
36.
Krueger, R., Dual-column (FIB-SEM) wafer applications. Micron, 1999. 30: p. 221-226.
37.
Xu, Y., C. Schwappach, and R. Cervantes, Precision TEM specimen preparation for integrated circuits using dual-beam FIB lift-out technique. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 516-517.
38.
Young, R.J., Atomation of focused ion beam (FIB) sample preparation. Microsc. Microanal., 2000. 6(2): p. 512-513.
39.
Ma, Z., et al., Novel preparation of precision planar TEM specimens for integrated circuits using dual-beam focused ion beam. Proc. Microscopy and Microanalysis, 1999. 5(2): p. 904-905.
40.
Ishitani, T., T. Yaguchi, and H. Koike, Focused Ion Beam system for TEM sample preparation. Hitachi Review, 1996. 45(1): p. 19-24.
41.
Pidduck, A.J., et al., Focused ion beam microscopy investigatio of InGaP/GaAs heterojunction bipolar transistors. Inst. Phys. Conf. Ser., 1999. 164: p. 585-588.
42.
Ishitani, T., et al., Transmission electron microscope sample preparation using a focused ion beam. J. Electron. Microsc., 1994. 43: p. 322-326.
43.
Prenitzer, B.I., et al., Practical Aspects of FIB milling: Understanding ion beam/material interactions. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 502-503. 220
44.
Templeton, I.M. and H.G. Champion, Focused ion-beam line profiles: A study of some factors affecting beam broadening. J. Vac. Sci. Technol. B, 1995. 13(6): p. 2603-2606.
45.
Vasile, M.J., J. Xie, and R. Nassar, Depth control of focused ion-beam milling from a numerical model of the sputter process. J. Vac. Sci. Technol. B, 1999. 17(6): p. 3085-3090.
46.
Park, K., Cross-sectional TEM specimen preparation of semiconductor devices by focused ion beam etching. Mat. Res. Soc. Proc., 1990. 199: p. 271-280.
47.
Yamaguchi, A., M. Shibata, and T. Hashinaga, Transmission electron microscopy specimen preparation technique using focused ion beam fabrication: Application to GaAs metal-semiconductor field effect transistors. J. Vac. Sci. Technol. B, 1993. 11(6): p. 2016-2020.
48.
Leslie, A.J., et al., TEM sample preparation using FIB: Practical problems and artifacts. Proc. 21nd Int. Symp. Testing and Failure Analysis, ISTFA95, 1995: p. 353-362.
49.
Pantel, R., G. Auvert, and G. Mascarin, Microelec. Engineering, 1997. 37/38: p. 49.
50.
Young, R.J., et al., Proc. 24nd Int. Symp. Testing and Failure Analysis, ISTFA98, 1998: p. 329-336.
51.
Ishitani, T., et al., J. Vac. Sci. Technol. B, 1998. 16: p. 1907.
52.
Lipp, S., et al., Investigations on the topology of structures milled and etched by focused ion beams. J. Vac. Sci. Technol. B, 1996. 14(6): p. 3966-3999.
53.
Tsujimoto, K., et al., Transmission electron microscopy of thin-film transistors on glass substrates. Thin solid Films, 1998. 319: p. 106-109.
54.
More, K.L., et al., TEM specimen preparation of oxidized Ni-base alloys using the focused ion bean (FIB) technique. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 540-541.
55.
Prenitzer, B.I., et al., Transmission Electron Microscope Specimen Preparation of Zn Powders Using the Focused Ion Beam Lift-Out Technique. Metallurgical and Materials Transactions A, 1998. 29A: p. 2399-2406.
56.
Basile, D.P., et al., FIBxTEM - focused ion beam milling for TEM sample praparation. Mat. Res. Soc. Proc., 1992. 254: p. 23-41.
221
57.
Su, D.H.I., et al., A detailed procedure for reliable preparation of TEM samples using FIB milling. Mat. Res. Soc. Symp. Proc., 1997. 480: p. 105-116.
58.
Hudson, M.V., A technique To Prepare Cross Sections Of Semiconductor Devices In Small Samples For Transmission Electron Microscopy. Microsc. Microanal., 2000. 6(Suppl 2: Proceedings): p. 498-499.
59.
Yamaguchi, A. and T. Nishikawa Low-damage specimen preparation technique for transmission electron microscopy using iodine gas-assisted focused ion beam milling. J. Vac. Sci. Technol. B, 1995. 13(3): p. 962-966.
60.
Kato, N.I., K. Tsujimoto, and N. Miura, Reduction of the damage induced in a FIB-fabricated X-TEM specimen. Mat. Res. Symp. Proc., 1998. 523: p. 39-44.
61.
Overwijk, M.H., F.C. van den Heuvel, and C.W. Bulle-Lieuwma, Novel scheme for the preparation of transmission electron microscopy specimens with a focused ion beam. J. Vac. Sci. Technol. B, 1993. 11(6): p. 2021-2024.
62.
Shaapur, F., Mat. Res. Soc. Proc., 1997. 480: p. 173.
63.
Stevie, F.A., et al., AIP Conference Proc., 1998. CP449: p. 868.
64.
Herlinger, L.R., S. Chevacharoenkul, and D.C. Ervin, TEM sample preparation using FIB and probe manipulator. Proc. 22nd Int. Symp. Testing and Failure Analysis, ISTFA96, 1996: p. 199-205.
65.
Giannuzzi, L.A., et al., Focused ion beam milling and micromanipulation lift-out for site specific cross-section TEM specimen preparation. Mat. Res. Soc. Proc., 1997. 1997(480): p. 19.
66.
Giannuzzi, L.A. and F.A. Stevie, Revisiting the FIB TEM lift-out specimen preparation trchnique. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 508509.
67.
Kamino, T., et al., Site specific TEM specimen preparation using an FIB/TEM system. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 510-511.
68.
Lomness, K.J., L.A. Giannuzzi, and M.D. Hampton, Site specific TEM analysis of micrometer-sized particles with the FIB lift-out technique. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 518-519.
69.
Klepeis, S., et al., FIB dimpling: A method for preparing plan-view TEM specimens. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 506-507.
222
70.
Anderson, R. and K.S. J, Combined tripod polishing and FIB method for preparing semiconductor plan view specimen. Mat. Res. Soc. Proc., 1997. 480: p. 187.
71.
Stevie, F.A., et al., The FIB lift-out specimen preparation technique for TEM analyses and beyond: SEM, Auger, STEM, and SIMS applications. Proc. Microscopy and Microanalysis, 1999. 5(2): p. 888-889.
72.
Chu, C.H., et al., Structural damage induced by Ga+ focused ion beam implantation in (001) Si. J. Vac. Sci. Technol. B, 1991. 9(6): p. 3451-3454.
73.
Venables, D., D.W. Susnitzky, and A.J. Mardingly, AIP Conference Proc., 1998. CP449: p. 667.
74.
Walker, J.F. and R.F. Broom, Surface damage of semiconductor TEM samples prepared by focused ion beams. Inst. Phys. Conf. Ser., 1997. 157: p. 473-478.
75.
Ishitani, T., Monte Carlo simulation of ion bombardment at low glancing angles. Jpn. J. Appl. Phys., 1995. 34: p. 3303-3306.
76.
Sze, S.M., Semiconductor Devices, Physics and Technology. 1985, New York: Wiley.
77.
Smith, B., Ion implantation range data for silicon and germanium device technologies. 1997, Oregon: Forest Grove.
78.
Lindhard, J., M. Scharff, and H. Schiott, Range Concepts and Heavy Ion Ranges. Mat.-Fys. Medd. K Dan. Vidensk. Selsk., 1963. 33(14): p. 1.
79.
Sze, S.M., VLSI technology. 1983, New York: McGraw-Hill.
80.
Ghandhi, S.K., VLSI Fabrication Principles. 1983, New York: Wiley.
81.
Biersack, J.P. and L. Haggmark, Nucl. Instr. and Meth., 1980. 174: p. 1980.
82.
Ziegler, J.F., J.P. Biersack, and U. Littmark, The stopping and range of ions in solids. 1985, New York: Pergamon Press.
83.
Jackson, K.A., ed. Processing of Semiconductors. Materials Science and Technology. Vol. 16. 1996: New York.
84.
Williams, J.S., et al., Amorphous to crystalline phase transformations in high dose ion implanted silicon. Mat. Res. Soc. Symp. Proc., 1990. 157: p. 629-634.
85.
Williams, J.S., Ion inplantation of semiconductors. Materials Science and Engineering, 1998. A253: p. 8-15. 223
86.
Wang, L.M., R.C. Birtcher, and L.E. Rehn, Effects of ion dose on morphology and crystallization of amorphous germanium. Mat. Res. Soc. Symp. Proc., 1990. 157: p. 659-664.
87.
Washburn, J., Formation and effects of secondary defects in ion implanted silicon, in Defects in Semiconductors, N.a. Tan, Editor. 1981, North-Holland, Inc.
88.
Seidel, T.E., G.A. Pasteur, and J.C.C. Tsai, Visible interference effects in silicon caused by high-current -high-dose implantation. Appl. Phys. Lett., 1976. 29(10): p. 648-651.
89.
Morehead, F.F. and B.L. Crowder, A model for the formation of amorphous Si by ion bombardment. Radiation effects, 1970. 6: p. 27-32.
90.
Motooka, T., The role of defects during amorphization and crystallization processes in ion implanted Si. Materials Science and Engineering, 1998. A253: p. 42-49.
91.
Sadana, D.K., et al., Damage and in situ annealing during ion implantation. Mat. Res. Soc. Symp. Proc., 1983. 14: p. 511-516.
92.
Narayan, J. and J. Fletcher, Radiation damage and its annealing in semiconductors, in Defects in semiconductors, J. Narayan and Tan, Editors. 1981, North-Holland, Inc. p. 191-207.
93.
Narayan, J. and O.W. Holland, J. Appl. Phys., 1984. 56: p. 2913.
94.
Christel, L.A., J.F. Gibbons, and T.W. Sigmon, J. App. Phys., 1981. 52: p. 7143.
95.
Williams, J.S., et al., Dynamic annealing and amorphous phase formation in Si, GaAs and AlGaAs under ion irradiation. Mat. Res. Soc. Symp. Proc., 1994. 316: p. 15-25.
96.
Svensson, B.G., C. Jagadish, and J.S. Williams, Phys. Rev. Lett., 1993. 71: p. 1960.
97.
Holland, O.W., B.R. Appleton, and J. Narayan, Ion implantation damage and annealing in germanium. J. Appl. Phys., 1983. 54(5): p. 2295-2301.
98.
Appleton, B.R., et al., Characterisation of damage in ion implanted Ge. Appl. Phys. Lett., 1982. 41(8): p. 711-712.
99.
Wilson, I.H., The effects of self-ion bombardment (30-500 keV) on the surface topography of single-crystal germanium. J. Appl. Phys., 1982. 53(3): p. 16981705. 224
100.
Pearton, S.J., W.S. Hobson, and C.R. Abernathy, Ion implantation processing of GaAs and related compounds. Mat. Res. Soc. Symp. Proc., 1989. 147: p. 261272.
101.
Bench, M.W., I.M. Robertson, and M.A. Kirk, In situ TEM observations of heavy ion damage in gallium arsenide. Mat. Res. Soc. Symp. Proc., 1988. 100: p. 293-298.
102.
Xiong, F., et al., Amorphization and recrystallization in MeV ion implanted InP crystals. Mat. Res. Soc. Symp. Proc., 1988. 100: p.105-111.
103.
Urbanik, C.A., et al., Characterization of FIB damage in silicon. Proc. Microscopy and Microanalysis, 1999. 5(2): p. 740-741.
104.
Cairney, J.M., R.D. Smith, and P.R. Munroe, Transmission electron microscope specimen preparation of metal matrix composites using the focused ion beam miller. Proc. Microscopy and Microanalysis, 1999. 5(2): p. 892-893.
105.
Cairney, J.M. and P.R. Munroe, Redeposition effects in TEM sample preparation of FeAl-based metal matrix composites using the focused ion beam miller. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 514-515.
106.
Ishitani, T. and T. Yaguchi, Microsc. Res. Techn., 1996. 35: p. 320.
107.
Jamison, R.B., et al., Effects of ion species and energy of the amorphization of Si during FIB TEM sample preparation as determined by computational and experimental methods. Proc. Microscopy and Microanalysis, 2000. 6(2): p. 526527.
108.
Isabell, T.C. and P.E. Fischione, Application of plasma cleaning for electron microscopy of semiconductor materials. Mat. Res. Soc. Symp. Proc., 1998. 523: p. 31-38.
109.
Brown, P.D. and C.J. Humphreys, Electron microscopy, electrical activity, artefacts and the assessment of semiconductor epitaxial growth. Mat. Res. Soc. Symp. Proc., 1998. 523: p. 207-223.
110.
Kolesnikov, V.G., ed. Electronics. 1991, Encyclopaedia: Moscow.
111.
FIB FEI xP200 Workstation User's Guide. 1996.
225
